North American Journal of Fisheries Management Raft and Floating

This article was downloaded by: [Department Of Fisheries] On: 20 July 2015, At: 19:14 Publisher: Taylor & Francis Informa Ltd Registered in England and Wales Registered Number: 1072954 Registered office: 5 Howick Place, London, SW1P 1WG

North American Journal of Fisheries Management Publication details, including instructions for authors and subscription information: http://www.tandfonline.com/loi/ujfm20 Raft and Floating Radio Frequency Identification (RFID) Antenna Systems for Detecting and Estimating Abundance of PIT-tagged Fish in Rivers Eric R. Fethermana, Brian W. Avilab & Dana L. Winkelmanc a Colorado Parks and Wildlife, 317 West Prospect Road, Fort Collins, Colorado 80526, USA b Colorado State University, Colorado Cooperative Fish and Wildlife Research Unit, Department of Fish, Wildlife and Conservation Biology, Colorado State University, Fort Collins, 80523, Colorado, USA c U.S. Geological Survey, Colorado Cooperative Fish and Wildlife Research Unit, Department of Fish, Wildlife and Conservation Biology, Colorado State University, Fort Collins, 80523, Click for updates Colorado, USA Published online: 10 Oct 2014.

To cite this article: Eric R. Fetherman, Brian W. Avila & Dana L. Winkelman (2014) Raft and Floating Radio Frequency Identification (RFID) Antenna Systems for Detecting and Estimating Abundance of PIT-tagged Fish in Rivers, North American Journal of Fisheries Management, 34:6, 1065-1077, DOI: 10.1080/02755947.2014.943859 To link to this article: http://dx.doi.org/10.1080/02755947.2014.943859

PLEASE SCROLL DOWN FOR ARTICLE

Taylor & Francis makes every effort to ensure the accuracy of all the information (the “Content”) contained in the publications on our platform. However, Taylor & Francis, our agents, and our licensors make no representations or warranties whatsoever as to the accuracy, completeness, or suitability for any purpose of the Content. Any opinions and views expressed in this publication are the opinions and views of the authors, and are not the views of or endorsed by Taylor & Francis. The accuracy of the Content should not be relied upon and should be independently verified with primary sources of information. Taylor and Francis shall not be liable for any losses, actions, claims, proceedings, demands, costs, expenses, damages, and other liabilities whatsoever or howsoever caused arising directly or indirectly in connection with, in relation to or arising out of the use of the Content. This article may be used for research, teaching, and private study purposes. Any substantial or systematic reproduction, redistribution, reselling, loan, sub-licensing, systematic supply, or distribution in any form to anyone is expressly forbidden. Terms & Conditions of access and use can be found at http:// www.tandfonline.com/page/terms-and-conditions North American Journal of Fisheries Management 34:1065–1077, 2014 C American Fisheries Society 2014 ISSN: 0275-5947 print / 1548-8675 online DOI: 10.1080/02755947.2014.943859

ARTICLE

Raft and Floating Radio Frequency Identiﬁcation (RFID) Antenna Systems for Detecting and Estimating Abundance of PIT-tagged Fish in Rivers

Eric R. Fetherman* Colorado Parks and Wildlife, 317 West Prospect Road, Fort Collins, Colorado 80526, USA Brian W. Avila Colorado State University, Colorado Cooperative Fish and Wildlife Research Unit, Department of Fish, Wildlife and Conservation Biology, Colorado State University, Fort Collins, Colorado 80523, USA Dana L. Winkelman U.S. Geological Survey, Colorado Cooperative Fish and Wildlife Research Unit, Department of Fish, Wildlife and Conservation Biology, Colorado State University, Fort Collins, Colorado 80523, USA

Abstract Portable radio frequency identification (RFID) PIT tag antenna systems are increasingly being used in studies examining aquatic animal movement, survival, and habitat use, and their design flexibility permits application in a wide variety of settings. We describe the construction, use, and performance of two portable floating RFID PIT tag antenna systems designed to detect fish that were unavailable for recapture using stationary antennas or electrofishing. A raft antenna system was designed to detect and locate PIT-tagged fish in relatively long (i.e., ≥10 km) river reaches, and consisted of two antennas: (1) a horizontal antenna (4 × 1.2 m) installed on the bottom of the raft and used to detect fish in shallower river reaches (<1 m), and (2) a vertical antenna (2.7 × 1.2 m) for detecting fish in deeper pools (≥1 m). Detection distances of the horizontal antenna were between 0.7 and 1.0 m, and detection probability was 0.32 ± 0.02 (mean ± SE) in a field test using rocks marked with 32-mm PIT tags. Detection probability of PIT-tagged fish in the Cache la Poudre River, Colorado, using the raft antenna system, which covered 21% of the wetted area, was 0.14 ± 0.14. A shore-deployed floating antenna (14.6 × 0.6 m), which covered 100% of the wetted area, was designed for use by two operators for detecting and locating PIT-tagged fish in shorter (i.e., <2 km) river reaches. Detection distances of the shore-deployed floating antenna were between 0.7 and 0.8 m, and detection probabilities during field deployment in the St. Vrain River exceeded 0.52. The shore-deployed floating antenna was also used to estimate abundance of PIT-tagged fish. Results suggest that the shore-deployed floating antenna could be used as an

Downloaded by [Department Of Fisheries] at 19:14 20 July 2015 alternative to estimating abundance using traditional sampling methods such as electroﬁshing.

Passive integrated transponder tag technology has many ad- behavior studies, especially those examining habitat selection vantages over traditional marking techniques. They allow indi- or movement (Nunnallee et al. 1998; Zydlewski et al. 2006; vidual identification, have an infinite life as long as the tag is not Bond et al. 2007; Compton et al. 2008; Connolly et al. 2008; damaged, are easily applied and well retained, and have minimal Aymes and Rives 2009). effects on growth and survival of fishes (Gries and Letcher 2002; Stationary antennas are typically used to detect PIT-tagged Zydlewski et al. 2006; Ficke et al. 2012). Traditionally, the utility fish, but the use of portable antennas is becoming more com- of PIT tagging has been limited to physical recapture events mon. Portable antennas have been used in studies examining using methods such as electrofishing (Zydlewski et al. 2006). aquatic animal movement, survival, and habitat use, and their Stationary antennas allow passive recaptures of PIT-tagged design flexibility permits application in a wide variety of set- fish and have recently been used to detect PIT-tagged fish in tings. Initial technological advances in portable PIT tag antenna

*Corresponding author: [email protected] Received December 19, 2013; accepted July 1, 2014 1065 1066 FETHERMAN ET AL.

systems enabled effective detection in shallow, wadable rivers reach of the Cache la Poudre River, Colorado. The array was of salmonid fishes such as Atlantic Salmon Salmo salar (Rous- constructed using a 4.9-m, self-bailing, inflatable river raft. The sel et al. 2000; Zydlewski et al. 2001), Brown Trout S. trutta first antenna was installed in the bottom of the raft (horizontal) (Cucherousset et al. 2005), and steelhead Oncorhynchus mykiss and was designed for continuous deployment to detect fish in (anadromous Rainbow Trout) (Hill et al. 2006). Additionally, shallower (<1 m) sections of the river. The second antenna was a portable antennas have been developed to detect and locate dropper antenna (vertical) designed for intermittent deployment age-0 Northern Pike Esox lucius (Cucherousset et al. 2007) to detect fish in deeper (≥1 m) pools. and various age-classes of European Eel Anguilla anguilla and Both antennas consisted of two, loosely bound, continu- Common Dace Leuciscus leuciscus (Cucherousset et al. 2010). ous loops of 12-gauge thermoplastic, high-heat-resistant, nylon- Previous designs of portable antenna systems have limited their coated (THHN) wire. Binding of the wire occurred at specific effectiveness to shallow, wadable floodplains or small streams; attachment points with the raft (horizontal) or support beams however, a boat-mounted antenna system was recently devel- used to maintain antenna shape (vertical). Between these points, oped for monitoring mussel populations in larger, nonwadable the wire formed loose gaps of varying distances. The horizon- rivers (Fischer et al. 2012). tal antenna was an elliptical antenna 4 × 1.2 m in size and Portable antenna systems are limited by factors affecting de- located in the self-bailing channel of the raft (Figure 1). An- tection efficiency including tag size, power source, tag orienta- tenna shape and loop proximity were maintained by threading tion, antenna proximity if multiple antennas are used (Zydlewski the wire through sections of flexible plastic tubing secured to et al. 2006), tag collision (Axel et al. 2005; O’Donnell et al. the self-bailing holes in the floor of the raft with soft nylon 2010), and disruption of the magnetic field by the presence of cord. The vertical antenna was a 2.7- × 1.2-m rectangle, main- metal (Greenberg and Giller 2000; Bond et al. 2007). For exam- tained by four, 19-mm PVC crossbeams secured to the antenna ple, tag orientation relative to the antenna field affects detection wire with expandable spray-foam insulation (Figure 1). Holes and is higher when the tag is oriented perpendicular rather than were drilled in each crossbeam to allow water entry and 51-mm parallel to the antenna (Nunnallee et al. 1998; Morhardt et al. PVC caps filled with cement were attached to the lowermost 2000; Zydlewski et al. 2006; Compton et al. 2008; Aymes and crossbeam for ballast. Foam pipe insulation placed on the first Rives 2009). Disruption due to antenna proximity can be re- crossbeam allowed the antenna to hang vertically in the water duced through the use of multiplexers (Aymes and Rives 2009), column while remaining at the water surface without becom- and disruption from metal can be reduced by utilizing nonin- ing completely submerged. Connectors produced for welding ductive materials such as epoxy coil encasements or nylon nuts applications were used to allow for a quick disconnection of and bolts (Fischer et al. 2012). Potential limitations should be the vertical antenna in the event that the antenna got entangled accounted for during the design process, and we elaborate on and caught on submerged rocks or vegetation while deployed. the ways in which these limitations were accounted for in our The vertical antenna design facilitated easy deployment at the portable antenna designs. head of a pool, retrieval at the tail end of the pool, and onboard We describe the design and construction of two, portable, storage in an accordion-like fashion for swift deployment on floating, radio frequency identification (RFID) PIT tag antenna subsequent occasions. systems: a raft antenna system and a shore-deployed floating The horizontal and vertical antennas were both connected antenna system. To assess the performance of each antenna sys- to an Oregon RFID half-duplex (HDX) multiplex reader (al- tem, we estimated detection distance and detection probability ternating read cycle of six times per second for each antenna), using both experimental and field data. Our research objective which helped prevent proximity detection errors (Aymes and

Downloaded by [Department Of Fisheries] at 19:14 20 July 2015 for the raft antenna system was to determine the location and Rives 2009). The HDX reader stored detections for the array fate of PIT-tagged fish in relatively long (i.e., ≥10 km) river along with date and time of detection. Two, 12-V, marine, deep- reaches. For the shore-deployed floating antenna system, our cycle batteries, connected in parallel, powered the raft antenna research objectives were twofold. First, we wanted to compare system. Batteries, tuner boxes, and the reader were placed in abundance estimates to those obtained via electrofishing to de- plastic, top-locking containers and strapped to a rigid plastic termine whether the antenna could be used as an alternative for deck located on the floor of the raft, which prevented equipment estimating abundance of PIT-tagged fish. Second, we wanted to shifts and submersion during deployment (Figure 1). determine the location and fate of fish that were not available Detection distance.—To measure the detection distance of for recapture via electrofishing due to movement from release the horizontal antenna, the raft was elevated on stands, and locations in shorter (i.e., <2 km) river reaches. detection distances were measured by running a 32-mm PIT tag in a perpendicular and parallel orientation past the antenna on ◦ METHODS horizontal, vertical, and 45 detection planes. The horizontal detection plane was defined as the plane extending 180◦ to the Raft Antenna System sides of the antenna, the same plane in which the antenna existed Design and construction.—We designed a two-antenna ar- lying flat on the bottom of the raft, and was used to simulate ray for detecting PIT-tagged fish in a relatively long (≥10 km) detection of PIT-tagged fish near the water surface or in shallow PORTABLE RFID ANTENNA SYSTEMS 1067 Downloaded by [Department Of Fisheries] at 19:14 20 July 2015

FIGURE 1. (A) Schematic representation of the raft antenna system (not to scale). Two continuous loops of 12-gauge THHN wire were used to create the horizontal (heavy dotted line) and vertical (heavy solid line) antennas. Both antennas were connected to tuning boxes (T), which were in turn connected to a multiplex reader (MR) and batteries (B) housed inside plastic, top-locking containers (light gray) and strapped to a rigid plastic deck (dark gray). (B) Diagram of the shore-deployed ﬂoating antenna system (not to scale). The antenna consisted of a single loop of 8-gauge, multistrand, speaker wire connected to a tuner box (T), which interfaced with the reader (R) and battery (B) enclosed in the sling-load backpack.

water on the edges of the river. The vertical detection plane was of the antenna detection field. When the tag was detected, a defined as the plane directly under the antenna, 90◦ to the bottom piezoelectric buzzer connected to the reader produced an audible of the raft, and was used to simulate detection when passing beep. Maximum continuous detection distance was defined as directly over a PIT-tagged fish. The 45◦ detection plane was the distance between when a beep was heard for every movement defined as the plane extending at a 45◦ angle to the bottom of the of the tag past the antenna (100% detection rate) and when a raft and was used to simulate detection of fish on the periphery lack of beep indicated that detections were being missed. The 1068 FETHERMAN ET AL.

presence of a diamond-plated, floor-covering, aluminum oar TABLE 1. Placement of rocks marked with PIT tags, with regard to transect frame was thought to affect the horizontal array as metal within number (T), depth (cm), and distance from center (DFC; cm), of the 50 rocks (two on each of 25 transects) used in the raft antenna system detection probability the detection field had been shown to disrupt both detection experiment conducted in the Parvin Lake inlet stream. occurrence and detection distance (Greenberg and Giller 2000; Bond et al. 2007). To test this, maximum detection distances T Depth DFC T Depth DFC T Depth DFC were measured with and without the frame and compared. To evaluate the effects of the oar frame, tag orientation, and 1 38.1 53.3 10 34.3 94.0 19 5.1 346.7 detection plane on maximum detection distance, we used a gen- 1 38.1 190.5 10 20.3 82.6 19 30.5 19.1 eral linear model (GLM) as implemented in SAS ProcGLM 2 64.8 22.9 11 16.5 330.2 20 25.4 171.5 (SAS Institute 2010). We considered an intercept-only model 2 25.4 257.8 11 34.3 25.4 20 22.9 224.8 as well as models that included effects of oar frame only, tag 3 53.3 20.3 12 45.7 180.3 21 25.4 36.8 orientation only, detection plane only, additive effects between 3 35.6 147.3 12 30.5 53.3 21 12.7 450.9 oar frame, orientation, and detection plane, and an interaction 4 44.5 121.9 13 33.0 104.1 22 15.2 237.5 model. Models were ranked using Akaike’s information crite- 4 48.3 138.4 13 30.5 195.6 22 53.3 616.0 rion corrected for small sample sizes (AIC ), compared using 5 52.1 85.1 14 33.0 190.5 23 38.1 2.5 c 5 48.3 82.6 14 39.4 144.8 23 27.9 271.8 AICc differences (AICc) and weights (wi), and we report parameter estimates and associated SEs from the top-supported 6 33.0 269.2 15 41.9 180.3 24 47.0 22.9 models (Burnham and Anderson 2002). 6 39.4 55.9 15 35.6 94.0 24 26.7 205.7 Detection probability.—To determine the detection proba- 7 44.5 90.2 16 22.9 342.9 25 33.0 381.0 bility of the horizontal antenna under riverine conditions, we 7 25.4 214.6 16 30.5 358.1 25 17.8 15.2 epoxied fifty 32-mm HDX PIT tags to rocks and placed them 8 17.8 299.7 17 35.6 108.0 in a 95.5-m (5.6 m average width) reach of an inlet stream lo- 8 25.4 35.6 17 30.5 257.8 cated at Parvin Lake, Red Feather Lakes, Colorado. We divided 9 17.8 194.3 18 22.9 298.5 the reach into transects (N = 25) and two rocks were placed 9 44.5 293.3 18 29.2 158.8 on each transect. The first transect was located 6.8 m downstream from the raft put-in site, which allowed the raft to be underway before detection occurred. Subsequent transects were in program MARK (White and Burnham 1999). The Huggins located at random distances from the first by using a random closed capture–recapture estimator differs from a traditional number generator. Depths and locations of the rocks were cho- closed capture–recapture estimator in that only two parameters, sen deliberately to provide a variety of distances and depths for capture or detection probability (p) and recapture probability analysis. Rocks were placed such that PIT tags were oriented (c) are included in the likelihood; N is conditioned out of the parallel to the banks, similar to the orientation of fish facing likelihood and estimated as a derived parameter using estimates into the current. Distance from the south bank and water depth of p (Huggins 1989). This quality allows individual covariates was recorded for each tag, and the metric distance from center affecting p to be included in Huggins estimator (Huggins 1991). (DFC) was calculated by dividing the transect length in half and Primary assumptions are that tags are not lost, are correctly subtracting the distance from south bank (Table 1). identified, and that the system is closed. Ten passes were conducted to estimate detection probability. Encounter histories were constructed for each rock such that A crew of three was used to operate the raft, and the raft was if it were detected on a pass, it would be given a value of 1, and

Downloaded by [Department Of Fisheries] at 19:14 20 July 2015 maneuvered down the center of the inlet stream on each pass to if it were not detected, it would be given a value of 0. Depth reduce bias because the raft operators were the same people that and DFC were included as individual covariates for each rock, placed the tagged rocks. Additionally, approximately the same and models in which p was constant, varied by depth, DFC, course was followed on all 10 passes to reduce bias. The raft was or the additive combination of the two, were included in the maneuvered at a 45◦ angle to the banks, providing a detection model set; p equaled c in all models, as c was not expected to be field roughly 4.2 m wide, including both the 1.2- and 4-m axes affected by previous detection. Models were ranked using AICc, of the antenna, and allowing raft operators to maneuver through compared using AICc and wi (Burnham and Anderson 2002), meanders and avoid obstacles. Unfortunately, the shallow nature and a model-averaged parameter estimate and unconditional SE of the inlet stream precluded the use of the vertical antenna in were reported (Anderson 2008). Information from all models the detection probability field test. Detection probability for the with wi > 0 were included in the model-averaged parameter raft antenna system, including continuous deployment of the estimate. In addition, cumulative AICc weights were used to horizontal antenna and intermittent deployment of the vertical assess the relative importance of each covariate. antenna, was later estimated following deployment and PIT- Location and fate of PIT-tagged fish.—Our objective was to tagged fish detection in the Cache la Poudre River. detect and locate PIT-tagged fish that had been marked and re- Detection probability (p) for the horizontal antenna was es- leased in the Cache la Poudre River near Glen Echo, Colorado, timated using the Huggins closed capture–recapture estimator during a previous study (Fetherman 2013). We were primarily PORTABLE RFID ANTENNA SYSTEMS 1069

FIGURE 2. Study area in the Cache la Poudre River, Colorado. The map shows the reach in which the raft antenna system was deployed to determine PIT-tagged fish location and fate, the locations where Brown Trout were and were not removed from the river as part of the study conducted by Fetherman (2013), the stationary antenna locations, and the four study segments (CPR1, CPR2, CPR3, and CPR4) in which abundance was estimated using both the shore-deployed floating antenna system and electrofishing.

interested in fish that were not available for detection by station- also documented a decrease in detection distance associated ary antennas deployed in the river because they had migrated with the metal oar frame (see below). The raft antenna system from the initial study sections, or they remained in the study was deployed in low-water conditions, which were conducive sections and could not be detected by the stationary antennas. to higher detection probabilities but made maneuvering the raft During the initial tagging event (Fetherman 2013), we PIT- difficult; however, we attempted to maneuver the raft within the tagged 5,271 Rainbow Trout and Brown Trout with 32-mm HDX river’s thalweg. The horizontal antenna was deployed contin- tags, inserted posterior to the pectoral fin through the midventral uously to detect fish in shallower reaches (<1 m), while the body wall into the peritoneal cavity using a hypodermic needle vertical antenna was intermittently deployed in pools ≥ 1m Downloaded by [Department Of Fisheries] at 19:14 20 July 2015 (Prentice et al. 1990; Acolas et al. 2007). At the time of tagging, deep. Rainbow Trout averaged 177 mm TL, ranging from 116 to Two passes were made through the 11.3-km reach on sub- 236 mm TL, and Brown Trout averaged 274 mm TL, ranging sequent days. Raft course, restricted within the deeper water from 107 to 500 mm TL. Fish were marked 1 year prior to raft of the thalweg, and vertical antenna deployment within deep antenna deployment. pools were similar among the passes. Before the start of each The raft antenna system was deployed in an 11.3-km section pass, operators’ watches were synchronized with the PIT tag of the Cache la Poudre River (19.6 m average width; Figure 2). reader clock. We recorded start and stop times, as well as times Prior to deployment, the horizontal and vertical antennas were at which recognizable landmark features adjacent to the river tuned and tested with a PIT tag to ensure proper operation. A were passed, allowing us to pair PIT tag detection times and crew of six was used to maneuver the raft: a captain, four pad- locations for analysis. dlers, and a person to operate the antenna equipment and deploy Following deployment in the Cache la Poudre River, two- the vertical antenna in pools. Four paddlers were needed because pass detection data from the horizontal and vertical antennas previous research indicated that a metal oar frame would inter- were pooled to obtain an overall estimate of p and c for the fere with the operation of the antennas (Greenberg and Giller raft antenna system using the Huggins closed capture–recapture 2000; Bond et al. 2007). In assessing our antenna systems, we estimator in program MARK. To meet the closure assumption, 1070 FETHERMAN ET AL.

passes were made on consecutive days over a relatively long maneuvered downstream, allowing the river current to carry the stretch of river, making it unlikely that fish died or moved out of antenna over the majority of obstacles, primarily large boulders. the study section between passes. Although fish were previously Detection distance.—Detection distance was tested by run- marked with PIT tags and released, the raft antenna deployment ning a 32-mm PIT tag over the antenna in the horizontal, vertical, was treated as a traditional mark–recapture study in which the and 45◦ detection planes, holding the tag at both a perpendicular first pass was the “mark” pass and the second pass was the and parallel orientation to the antenna. Both sides of the antenna “recapture” pass. Encounter histories were constructed for each were tested to determine whether there were differences in de- fish such that fish encountered only on the first pass had a history tection symmetry. When the tag was detected by the reader, a of “10,” fish encountered only on the second pass had a history piezoelectric buzzer attached to the reader produced an audible of “01,” and fish encountered on both passes had a history of beep. Maximum continuous detection distance was determined “11.” To estimate p and c, the model set included models in as the distance between when a beep was heard for every move- which p and c were constant and equal, constant but not equal, ment of the tag past the antenna (100% detection rate); when or varied by species. Models were ranked using AICc, compared an audible beep was not produced, this indicated that detections using AICc and wi, and we report model-averaged parameter where being missed. estimates and associated SEs (models with wi > 0). To evaluate antenna symmetry and influence of detection Locations of PIT-tagged fish detected by the raft antenna plane on maximum detection distance, we used a general linear system were determined by comparing the times at which the model (GLM) as implemented in SAS ProcGLM. We considered fish were detected to the times at which recognizable landmark an intercept-only model, as well as models that included effects features adjacent to the river were passed. To determine fate, of side only and detection plane only and models with additive tag numbers detected by the raft antenna system were compared and interactive effects between side and detection plane. Models with the release information associated with the initial PIT tag- were ranked using AICc and compared using AICc and wi, and ging event and the tag numbers obtained from the stationary we report parameter estimates and associated SEs from the top antennas deployed in the Cache la Poudre River. Numbers of supported models. fish detected by the raft antenna system, fish location, and fish Abundance estimation.—One of our primary objectives was fate are reported. to determine whether the shore-deployed floating antenna could be used to estimate abundance of PIT-tagged fish. To accom- Shore-deployed Floating Antenna System plish this, we compared estimates obtained from antenna de- Design and construction.—We designed a river-spanning, tection data with traditional electrofishing estimates. We de- shore-deployed, floating antenna system for detecting PIT- ployed the antenna in 10 river segments, four segments in the tagged fish in shorter (i.e., <2 km) river reaches. The antenna Cache la Poudre River (Figure 2) and six segments in the St. was designed to not only determine the location and fate of fish Vrain River, Lyons, Colorado (Figure 3). Passive integrated that were not available for detection via electrofishing, but also transponder-tagged fish within the Cache la Poudre River were to test its utility for use in place of traditional methods, such as the same as those previously described for the raft antenna sys- electrofishing, for estimating the abundance of PIT-tagged fish. tem (N = 5,271). Rainbow Trout and Brown Trout in the St. The shore-deployed floating antenna was rectangular in Vrain River were PIT-tagged and released 1 year prior to abun- shape (14.6 × 0.6 m) and consisted of a single loop of in- dance estimation (N = 1,569). At the time of tagging, Rainbow sulated, 8-gauge, multistrand, copper speaker wire. Antenna Trout averaged 192 mm TL, ranging from 113 to 272 mm TL, shape was maintained by threading the wire through foam pipe and Brown Trout averaged 190 mm TL, ranging from 120 to

Downloaded by [Department Of Fisheries] at 19:14 20 July 2015 insulation (used for flotation) and 13-mm PVC crossbeams, lo- 480 mm TL. Fish that were PIT-tagged within the St. Vrain cated every 1.8 m along the length of the antenna (Figure 1). River were being used to evaluate and compare movement rates Floating nylon rope was threaded through the upstream side of through a whitewater park reach (WWP) containing artificially the foam pipe insulation, allowing operators to maintain tension constructed flow-control structures and through a natural reach and antenna shape during deployment. An Oregon RFID HDX (natural) where no such flow-control structures existed. single reader and tuner box, located in the top compartment of a Sampling segments varied in length, width, and habitat char- plastic-framed, sling-load pack and a 12-V, marine, deep-cycle acteristics. In the Cache la Poudre River, the two most upstream battery, secured to the pack via the sling, were used to power segments (CPR1: 114 m long × 17 m wide; CPR2: 165 × the antenna (Figure 1). 12 m) were characterized by slower-velocity pool habitat with Antenna design facilitated two-person deployment, with both higher-velocity riffles on the upstream end. The two most operators walking along the banks of the river to avoid scaring downstream segments (CPR3: 91 × 21 m; CPR4: 124 × 19 m) fish out of the study segments by wading during deployment. were both characterized by moderate riffle habitat with higher- One person carried the sling-load pack and was the primary velocity riffle habitat on the upstream end (Figure 2). The Cache guide for the antenna. The second person retained tension on la Poudre River was sampled at an average discharge of 3.4 m3/s, the nylon rope in order to maintain antenna shape and guide and average segment depth was 0.6 m. Segments within the the antenna over obstacles. The fully extended antenna was St. Vrain River consisted of naturally occurring and artificially PORTABLE RFID ANTENNA SYSTEMS 1071

FIGURE 3. Study area in the St. Vrain River, Colorado. The map shows the two 0.8-km reaches in which the shore-deployed floating antenna system was deployed to determine PIT-tagged fish location and fate, and the six study segments in which abundance was estimated using both the shore-deployed floating antenna system and electrofishing.

constructed pools of varying depth. The three artificial pools Abundance estimates of PIT-tagged fish per segment were (lower, middle, and upper WWP) were 1.6, 1.4, and 2.1 m obtained using the Huggins closed capture–recapture estimator deep, respectively, and the three natural pools (lower, middle, in program MARK. Encounter histories and model set were and upper natural) were 0.5, 0.4, and 1.0 m deep, respectively constructed similar to those described for estimating detection (Figure 3). Average width of segments was 7.6 m, and the probability of the raft antenna system. Data for each segment segments were sampled at an average discharge of 0.7 m3/s. was analyzed separately (10 analyses in total). Models were Fish abundance was estimated by making two passes with ranked using AICc and compared using AICc and wi, and we the fully extended, shore-deployed, floating antenna through a report model-averaged parameter estimates and associated SEs > Downloaded by [Department Of Fisheries] at 19:14 20 July 2015 study segment; the antenna was folded up and returned to the (models with wi 0). top of the segment between passes. The first pass with the an- Estimates of PIT-tagged fish abundance obtained from the tenna was considered the “mark” pass, while the second pass shore-deployed floating antenna were compared with those ob- acted as the “recapture” pass. In the St. Vrain River, closure tained from the same study segment using electrofishing. Im- for the estimates was achieved by using block nets at both the mediately following antenna deployment, two-pass (Cache la upstream and downstream ends of the study segments. In the Poudre River) or three-pass (St. Vrain River) removal abun- Cache la Poudre River, we used stream features, such as high- dance estimates were conducted in the same segment using a velocity riffles on the upstream end of the segments, to restrict bank electrofishing unit with four electrodes. All fish captured fish movement during the estimates. Use of natural barriers is were weighed, measured, and scanned for PIT tags using a hand- the protocol currently used by Colorado Parks and Wildlife to held reader. Tag numbers were compared with those recorded conduct electrofishing population estimates, and we wanted to by the shore-deployed floating antenna to determine whether use a similar protocol when deploying the antenna. However, the same fish were detected by both gears. we realized that we were assuming closure and it was not guar- Abundance estimates were similarly obtained from the elec- anteed. As in the St. Vrain River, passes were conducted one trofishing data using the Huggins closed capture–recapture es- immediately after the other, and wading was avoided to prevent timator in program MARK, with the exception that c was fixed scaring fish from the segment. to zero in all models because fish were removed after capture 1072 FETHERMAN ET AL.

TABLE 2. Model selection results for factors inﬂuencing maximum detection distance for the horizontal antenna of the raft antenna system. The maximized log-likelihood [log(L)], the number of parameters (K) in each model, and AICc values are shown. Models are ranked by their AICc differences (i) relative to the best model in the set and Akaike weights (wi) quantify the probability that a particular model is the best model in the set given the data and the model set.

2 Model R log(L) K AICc i wi Orientation × Frame × Plane 0.90 −207.46 12 441.30 0.00 1.00 Orientation + Plane 0.87 −225.11 5 460.65 19.35 0.00 Orientation + Frame + Plane 0.87 −224.22 7 463.25 21.96 0.00 Orientation 0.84 −237.02 2 478.13 36.83 0.00 Orientation + Frame 0.84 −236.27 4 480.82 39.52 0.00 Intercept only 0.00 −370.24 1 742.52 301.22 0.00 Plane 0.02 −368.50 3 743.16 301.86 0.00 Frame 0.00 −370.13 2 744.34 303.04 0.00 Frame + Plane 0.03 −368.38 5 747.19 305.89 0.00

(Hense et al. 2010; Saunders et al. 2011). Encounter histories tagging event and tag numbers of fish captured via electrofish- were constructed for only PIT-tagged fish and were constructed ing. Numbers of fish detected by the shore-deployed floating such that fish captured on the first pass had a history of “10” and antenna system, fish location, and fish fate are reported. fish caught on the second pass had a history of “01.” Rainbow Detection probability for the shore-deployed floating antenna Trout and Brown Trout were included as groups in the same was estimated for each reach following deployment. Encounter analysis and p was modeled as constant or varied by length, histories and model sets were constructed similar to those de- species, or both. Models were ranked using AICc and compared scribed for estimating detection probability of the raft antenna using AICc and wi, and we report model-averaged parameter system. Models were ranked using AICc and compared using estimates and associated 95% CIs (models with wi > 0). Abun- AICc and wi, and we report model-averaged parameter esti- dance estimates obtained with the two gear types were compared mates and associated SEs (models with wi > 0). within each study segment, and differences between the gears were determined by a lack of overlap in 95% CIs. RESULTS Location and fate of PIT-tagged fish.—Another primary objective of the shore-deployed floating antenna was to determine Raft Antenna System the location and fate of PIT-tagged fish that were not available Detection distance.—Tag orientation, oar frame presence, for recapture via electrofishing in the St. Vrain River study seg- and detection plane had the largest influence on hand-held PIT ments. To address this objective we deployed the antenna in tag maximum detection distance of the horizontal antenna of two 0.8-km reaches of the St. Vrain River (Figure 3). In the the raft antenna system (AICc weight = 1.00; Table 2). Average natural reach, the antenna was deployed just upstream from the detection distance was greater for tags oriented perpendicular upper natural study segment to just downstream from the lower than for those oriented parallel to the antenna. The presence natural study segment. In the WWP reach, the antenna was de- of the oar frame affected detection distance along the vertical, ployed just upstream from the upper WWP study segment to ◦

Downloaded by [Department Of Fisheries] at 19:14 20 July 2015 horizontal, and 45 detection planes for tags oriented perpendic- just downstream from the lower WWP study segment. All six of ular to the antenna. Detection distance in the vertical detection the previously described study segments were contained within plane was affected by the presence of the oar frame for tags ori- the two reaches. Shallow riffle habitat constituted the majority ented parallel to the antenna, but were not affected by oar frame of the habitat between study segments in both reaches. presence in the horizontal or 45◦ detection planes (Figure 4). Two passes with the antenna were made through each reach. Detection probability.—The use of PIT-tagged rocks in the Due to the narrow width of the reaches, the array was deployed Parvin Lake inlet stream showed that both DFC and depth af- ◦ at a 45 angle to the stream banks, allowing full extension of fected p of the horizontal antenna of the raft antenna system the array for proper tuning. Before the start of each pass, op- (model = DFC + Depth; AICc weight = 0.99). A negative re- erators’ watches were synchronized with the reader clock. We lationship was observed between DFC and detection probability recorded start and stop times, as well as times at which recog- (βˆ =−0.011 ± 0.001) indicating that the farther a tag was from nizable landmark features adjacent to or within the river were the center of the inlet stream, and consequently the center of the passed, allowing us to pair PIT tag detection times and locations antenna’s detection field, the less likely it was to be detected. for analysis; average deployment time was roughly 45 min per Interestingly, there was a positive relationship between p and pass. To determine PIT-tagged fish fate, tag numbers were com- depth (βˆ = 0.02 ± 0.005). This relationship likely occurred pared with the release information associated with the initial PIT because tags located near the center of the inlet stream were PORTABLE RFID ANTENNA SYSTEMS 1073

Shore-deployed Floating Antenna System Detection distance.—Detection plane had the largest influence on hand-held PIT tag maximum detection distance for the shore-deployed floating antenna (model 1 = Plane, AICc weight = 0.48; model 2 = Side + Plane, AICc weight = 0.44; Table 3). On average, tags in the vertical detection plane were detected at a greater distance (79.9 ± 0.8 cm) than tags in the horizontal or 45◦ detection planes. Horizontal (71.6 ± 0.7 cm) and 45◦ (72.3 ± 0.8 cm) detection planes exhibited average maximum detection distances. There was some evidence that detection distances were not symmetrical about the antenna (model 2 = Side + Plane, AICc weight 0.44; Table 3). However, average maximum detection distances did not appear to differ between the right (75.3 ± 1.2 cm) and left (74.0 ± 0.9 cm) FIGURE 4. Maximum detection distances for the horizontal antenna of the sides of the antenna. raft antenna system with and without the aluminum oar frame and with tags oriented perpendicular and parallel to the antenna in the vertical (V), horizontal Abundance estimation.— Abundances of PIT-tagged fish ob- (H), and 45◦ detection planes. Error bars represent SE. tained from the shore-deployed floating antenna were similar to those obtained via electrofishing, as indicated by overlapping 95% CIs, in two of the four study segments in the Cache la deeper than those placed closer to the banks and, thus, more Poudre River (CPR2 and CPR3; Figure 5). However, it is im- likely to be detected regardless of depth. The model-averaged portant to note the 95% CIs overlapped in these two segments estimate of p for the horizontal antenna was 0.32 ± 0.02 (mean because of large 95% CIs for the electrofishing estimates (CPR2) ± SE). or the antenna estimates (CPR3), suggesting that the estimates Location and fate of PIT-tagged fish.—A total of 44 PIT- were fairly imprecise, likely due to the low numbers of PIT- tagged fish (32 Rainbow Trout and 12 Brown Trout) were de- tagged fish detected in these segments. In CPR2, 19 PIT tags tected by the 4.2-m-wide antenna detection field (21% of the were detected by the antenna, whereas eight PIT-tagged fish wetted area) over the 11.3-km reach of the Cache la Poudre were detected via electrofishing; four PIT-tagged fish were de- River. Two unique fish were detected by the intermittently detected by both gear types. In CPR3, seven PIT tags were detected ployed vertical antenna, whereas 42 unique fish were detected by the antenna, whereas five PIT-tagged fish were detected via by the continuously deployed horizontal antenna. Three fish electrofishing; two PIT-tagged fish were detected by both gear were detected on both passes, which allowed both p and c to be types. estimated for the raft antenna system. Detection probability of Abundance estimates differed in two of the four study seg- the raft antenna system, including continuous deployment of the ments in the Cache la Poudre River (CPR1 and CPR4; Figure 5). horizontal antenna and intermittent deployment of the vertical The estimate of PIT-tagged fish abundance in CPR4 was higher antenna, was 0.14 ± 0.14, and for c was 0.13 ± 0.07. with electrofishing than with the antenna, which was potentially Twenty-seven of the fish detected had never been detected a function of differences in maneuverability of the gear types by the stationary antennas after their release 1 year prior to de- around the large boulders located within this study segment. In ployment of the raft antenna system. Of these, 15 were detected CPR4, six PIT tags were detected by the antenna, whereas 11 Downloaded by [Department Of Fisheries] at 19:14 20 July 2015 in their release location and had never moved past the station- PIT-tagged fish were detected via electrofishing; two PIT-tagged ary antennas. The remaining 12 fish had moved upstream or fish were detected by both gear types. Abundance estimation downstream, but were not detected by the stationary antennas. via electrofishing was not possible in CPR1 due to depletion The complementary use of the raft antenna system enabled us failure (i.e., three PIT-tagged fish caught on both passes); how- to locate and determine the fate of those fish not detected by the ever, abundance estimates were obtainable using the antenna stationary antennas. (Figure 5). In CPR1, 11 PIT tags were detected by the antenna, The other 17 fish detected by the raft antenna system had whereas six PIT-tagged fish were detected via electrofishing; been previously detected by the stationary antennas. Detections five PIT-tagged fish were detected by both gear types. at stationary antennas occurred between 8 and 12 months prior The shore-deployed floating antenna failed to obtain compa- to detection by the raft antenna system, and use of the system rable abundance estimates to those obtained by electrofishing allowed us to confirm the new locations of these fish. Fish moved in the WWP study segments in the St. Vrain River (Figure 5), both upstream and downstream from their release locations. which was likely a function of segment depth. The shallowest On average, fish were located within 1.2 km of the locations study segment (middle WWP, 1.4 m) exceeded the maximum from which they had been released; however, several fish were read range of the array by 0.6 m. All three PIT tags detected by detected farther from their release location including one fish the antenna in the lower and middle WWP segments were not located over 4 km upstream from its release site. detected via electrofishing. 1074 FETHERMAN ET AL.

TABLE 3. Model selection results for factors inﬂuencing maximum detection distance for the shore-deployed ﬂoating antenna system. The maximized log- likelihood [log(L)], the number of parameters (K) in each model, and AICc values are shown. Models are ranked by their AICc differences (i) relative to the best model in the set and Akaike weights (wi) quantify the probability that a particular model is the best model in the set given the data and the model set.

2 Model R log(L) K AICc i wi Plane 0.64 −43.43 3 93.49 0.00 0.48 Side + Plane 0.66 −42.27 4 93.63 0.14 0.44 Side × Plane 0.68 −41.33 6 97.06 3.57 0.08 Intercept only 0.00 −64.98 1 132.05 38.56 0.00 Side 0.02 −64.57 2 133.45 39.96 0.00

Similar abundance estimates were obtained in the lower nat- of the antenna by 0.2 m, the abundance estimate obtained us- ural and middle natural study segments in the St. Vrain River ing the antenna was lower than that obtained by electrofishing (Figure 5). Maximum pool depth in these study segments did (Figure 5). Twelve PIT tags were detected by the antenna in not exceed 0.5 m. In the lower natural segment, 20 PIT tags the upper natural segment, whereas 21 PIT-tagged fish were de- were detected by the antenna, whereas 18 PIT-tagged fish were tected via electrofishing; only two PIT-tagged fish were detected detected via electrofishing; six PIT-tagged fish were detected by both gear types. by both gear types. In the middle natural segment, 11 PIT tags Location and fate of PIT-tagged fish.—Thirty-two PIT- were detected by the antenna, whereas 10 PIT-tagged fish were tagged fish—16 Rainbow Trout and 16 Brown Trout—were detected via electrofishing; four PIT-tagged fish were detected detected by the shore-deployed floating antenna, which covered by both gear types. In the upper natural segment, which had a 100% of the wetted area within the 0.8-km natural reach of maximum depth of 1 m and exceeded the maximum read range the St. Vrain River. In the 0.8-km WWP reach, 49 PIT-tagged fish were detected by the shore-deployed floating antenna: 18 Rainbow Trout and 31 Brown Trout. Estimated p for the shore- deployed floating antenna did not differ between the reaches: p = 0.52 ± 0.15 in the natural pool reach and p = 0.60 ± 0.10 in the WWP reach. Of the 32 fish detected by the shore-deployed floating antenna in the natural reach of the St. Vrain River, 12 were detected in the location to which they had been released 6 to 12 months prior to antenna deployment. Eighteen of the 32 fish that had been released within the reach exhibited upstream and downstream movements, and were detected in locations that differed from their release location. Six of these fish were located within the lower, middle, or upper natural segments and were available for detection, whereas the other 12 fish were located between study segments and were not available for detection via elec-

Downloaded by [Department Of Fisheries] at 19:14 20 July 2015 trofishing. The new locations of these fish would not have been known had they not been detected by the shore-deployed floating antenna. Finally, two of the fish that had been released in the WWP reach 12 months earlier were located between study segments and were not available for detection via electrofishing. The detection of these fish confirmed that fish were able to navigate the artificial structures of the WWP and move upstream, an observation that was only made possible because the locations of these fish were confirmed by the shore-deployed floating antenna. Of the 49 fish detected by the shore-deployed floating antenna in the WWP reach of the St. Vrain River, 31 were detected in FIGURE 5. Estimated number of PIT-tagged salmonids per section estimated the location in which they had been released. Twelve of the via electrofishing (Cache la Poudre River: two passes; St. Vrain River: three passes) and the shore-deployed floating antenna (two passes) within the study 49 fish were detected downstream from where they had been segments in (A) the Cache la Poudre River and (B) the St. Vrain River. Error released within the WWP reach. Two fish had been released bars represent 95% CI. in the natural reach upstream from the WWP reach and had PORTABLE RFID ANTENNA SYSTEMS 1075

exhibited downstream movement into the WWP reach. Four distance from the detection field of the raft, which suggests that fish were detected upstream from the location in which they had if coverage had been wider, p would have been higher. Coverage been released, again confirming that fish were able to navigate was less in the Cache la Poudre River where the raft antenna the artificial structures of the WWP. All four were located be- system covered only 21% of the wetted area. This resulted in tween the WWP study sections and as such were not available alowerp (0.14) than that of the Parvin Lake field test. Rafting for detection via electrofishing. Therefore, the shore-deployed technique, e.g., pointing the nose of the raft straight downriver floating antenna helped confirm upstream movement through versus maintaining the raft at a 45◦ angle to the banks, could the WWP structures, which would not have been confirmed also affect coverage and therefore p of the raft antenna system. using traditional sampling methods. In contrast, the shore-deployed floating antenna system, which covered 100% of the wetted area, produced higher estimates of p during deployment, and p was similar between the DISCUSSION natural and WWP reaches in the St. Vrain River. Similarities in p Both our raft and shore-deployed floating antenna systems were likely a result of where fish were detected in these reaches. allowed us to determine the location and fate of PIT-tagged fish In the natural reach, depths rarely exceeded the read range of that had not moved, moved but had not been detected by station- the antenna, which allowed nearly all fish in the reach to be ary antennas or electrofishing, or had migrated from the release available for capture on one or both passes. In the WWP reach, location entirely. Our raft antenna system was deployed over fish were most commonly detected in the shallow riffle habitat an 11.3-km section of the Cache la Poudre River, detecting fish between the deeper study segments, the depth of which did not in locations that would not have otherwise been sampled using exceed the read range of the antenna. Estimation of p requires traditional sampling methods in shorter study segments. Our that at least one fish needs to be detected on both passes (White shore-deployed floating antenna allowed us to obtain estimates et al. 1982). In this study repeat detections on subsequent passes of PIT-tagged fish abundance in shorter study segments in both occurred in the shallower sections of the WWP reach. Fish in the rivers and determine the location and fate of fish in two 0.8- deeper study segments were generally unavailable for detection km reaches of the St. Vrain River. Overall, the raft and shore- on either pass, which likely artificially inflated the estimates of deployed floating antenna systems have overcome limitations p in the WWP reach. Overall, p could be increased by increasing recognized with other portable systems, namely antenna size, the number of passes made through a study section. However, a stream distance surveyed, coverage, and detection distances. balance is needed between the number of passes made, the time Detection distances for both antenna systems were between it takes to make a pass, and the information gained by adding 0.7 and 1.0 m and can be partially attributed to using 32-mm more passes to the study. tags (Zydlewski et al. 2006). Other studies have used 12- Only two fish were detected by the vertical antenna of the or 23-mm tags, which resulted in lower detection distances raft antenna system, potentially contributing to the low p.Lin- (Roussel et al. 2000; Zydlewski et al. 2001; Cucherousset et al. nansaari and Cunjak (2007) suggested that there may be a 2005; Hill et al. 2006). However, one possible disadvantage of fright bias associated with larger submerged antennas if fish greater detection distance is an increased chance of multiple are tracked in their active state. Therefore, our vertical antenna tags being present in the detection field of the array, resulting may have caused some behavioral avoidance when deployed in in no tags being detected (tag collision; Axel et al. 2005; pools. In addition, the vertical antenna was tuned fully extended O’Donnell et al. 2010). In addition, greater detection distance prior to deployment. Antenna shape may have differed when could result in the detection of ghost tags, i.e., tags lodged in deployed in pools, or intermittent deployment and storage may

Downloaded by [Department Of Fisheries] at 19:14 20 July 2015 the substrate through a combination of tag loss, predation, and have caused the antenna to become detuned, thereby reducing natural mortality (O’Donnell et al. 2010). Finally, although detection distance of the vertical antenna upon deployment. The detection distances were greater than other portable antenna small number of fish detected by the vertical antenna suggests designs, both the raft and the shore-deployed floating antenna that antenna contributed little to the overall detection of fish systems were still limited to use in shallow (<1m)river by the raft antenna system. However, the utility of the vertical reaches. The utility of both antenna systems for use in deeper antenna could be increased by correcting issues regarding de- river reaches could be increased by manipulating voltage or tuning of the antenna by constructing a more rigid frame that adjusting wire arrangement (e.g., multiple loops or adjusting maintains antenna shape during deployment. spacing between loops) to increase their detection distance. Despite potential issues with detection probability, the ability Despite greater detection distances, detection probability for to determine the location and fate of PIT-tagged fish over short the raft antenna system was relatively low. Low p could be a (0.8 km) and long (11.3 km) distances is a major advantage of result of the coverage of the raft antenna system relative to these antenna systems relative to other portable antenna designs. the width of the river. In the Parvin Lake field test, the raft Most portable antenna designs, with the exception of the boat- antenna system covered an average of 32% of the wetted area mounted antenna for monitoring mussels (Fischer et al. 2012), of the inlet stream. Consequently, p for the horizontal antenna have been constrained to use in shallow, wadable streams; as was 0.32, and modeling indicated that p was affected by tag a result, survey length was limited by the length of river an 1076 FETHERMAN ET AL.

operator could walk. Here, we demonstrated that both antenna also illicit an avoidance response, especially when conditions systems could detect and locate fish that had migrated from are such that shadows are cast by the antenna (e.g., sunny days; smaller study segments and were no longer available for detec- Ellis et al. 2013), fish are less likely, relative to smaller designs, tion by stationary antennas or electrofishing. In several cases, to move completely out of the detection field due to antenna we were able to document upstream movement of fish through coverage (i.e., 100% of the wetted area). The effect of operator structures thought to be a barrier to movement (i.e., WWP struc- experience is also reduced due to antenna coverage as the op- tures in the St. Vrain River), and over long distances (up to 4 km erator is not required to identify specific locations or habitats in the Cache la Poudre River). However, little inference can be to sample. However, fish located directly behind large boulders made regarding fish fate from tags that were detected in the or other obstacles may not be detected by the floating antenna same location or downstream from where they had been re- when passing over these obstacles. leased. Ghost tags (O’Donnell et al. 2010) could potentially be The design flexibility of these antenna systems provides an detected by these antenna systems, leading to incorrect inter- opportunity to potentially combine designs and create a larger pretations of the data regarding fish location and fate. Ghost tag detection field for greater river coverage. For example, wing- detections cannot be removed from the data without locating like floating antennas could be combined with the raft antenna the tags using a smaller wand-type antenna or electrofishing to system to create a larger array that could cover more of the confirm whether tags were retained or lost by the fish to which wetted area over long distances. Multiplexers, or a well-designed they are associated. master–slave set-up, could be used to power the system and Portable PIT-tag antenna systems are fairly accurate in prevent proximity detection errors (Aymes and Rives 2009). estimating abundance of PIT-tagged fish in small streams However, the larger size of the system could potentially result in (O’Donnell et al. 2010; Sloat et al. 2011). Portable antennas a greater chance of entanglement with obstacles such as boulders have the advantage of allowing frequent sampling for abun- or submerged trees, and this would need to be considered during dance estimation and fish location without subjecting the fish to the design of these larger systems. excessive handling stress or mortality, and by minimizing dis- Our portable antenna systems provide noninvasive methods turbance to individuals (Sloat et al. 2011). However, this feature that minimize disturbance to individual fish for determining the also excludes the ability to examine fish for growth or physi- fate of PIT-tagged fish in both small (hundreds of meters) and ological parameters (Zydlewski et al. 2001) or to estimate the large (kilometers) river reaches. Through the use of marker tags, overall abundance (marked and unmarked) of fish within a des- accurate timing devices, and submeter GPS, the location of fish ignated area. Our results suggest that if estimates of tagged fish can be determined fairly accurately using these systems. In addi- are desired, and handling fish (beyond tagging) is not neces- tion, the shore-deployed floating antenna provides an alternative sary to collect individual information (e.g., fish size or signs of to traditional sampling methods for estimating PIT-tagged fish disease), portable antennas such as the shore-deployed floating abundance. More research is needed to examine the effects of antenna present an alternative to traditional sampling methods ghost tags on inferring the fate of fish without physical recap- such as electrofishing. However, estimates may be imprecise. In tures, to assess fish behavioral responses to antenna systems, addition, detection distance limitations must be considered, as and to determine ways to increase p and reduce variability in abundance may be greatly underestimated in deeper study seg- abundance estimates. ments, e.g., those of the WWP reach. Finally, our data regarding the small number of tags detected by both gears suggest that ghost tags (O’Donnell et al. 2010) could influence abundance ACKNOWLEDGMENTS

Downloaded by [Department Of Fisheries] at 19:14 20 July 2015 estimates obtained with the shore-deployed floating antenna This work was sponsored in part by Colorado Parks and system. Wildlife and the Colorado Cooperative Fish and Wildlife Re- The shore-deployed floating antenna system overcomes some search Unit at Colorado State University, and funding was pro- of the limitations observed with other antenna systems, such as vided in part by the Federal Aid in Sport Fish Restoration pro- antenna size, or those caused by lack of operator experience gram, project F-394R. We thank G. Schisler, K. Davies, M. and fish behavior (O’Donnell et al. 2010). Many of the previ- Kondratieff, and C. Praamsma of Colorado Parks and Wildlife, ously described portable antenna systems were small, designed and L. Bailey, G. Fraser, and N. Shannon for their help with to be operated by one person in a small stream (Roussel et al. antenna design and deployment. Any use of trade, firm, or prod- 2000; Cucherousset et al. 2005; Hill et al. 2006), and as such, uct names is for descriptive purposes only and does not imply antenna coverage was small relative to the width of the river. endorsement by the U.S. Government. This study was approved Our shore-deployed floating antenna is the largest two-person under Colorado State University Institutional Animal Care and portable antenna described to date, as previous two-person an- Use Committee 10-1956A. tennas did not exceed 5 m in length (Linnansaari and Cunjak 2007). Submersion of the antenna is not required, theoretically REFERENCES reducing the chance of a behavioral response to an antenna lo- Acolas, M. L., J. -M. Roussel, J. M. Lebel, and J. L. Bagliniere.` 2007. Laboratory cated within the water column. Although overhead stimuli may experiment on survival, growth and tag retention following PIT injection into PORTABLE RFID ANTENNA SYSTEMS 1077

the body cavity of juvenile Brown Trout (Salmo trutta). Fisheries Research Hense, Z., R. W. Martin, and J. T. Petty. 2010. Electrofishing capture efficien- 86:280–284. cies for common stream fish species to support watershed-scale studies in Anderson, D. R. 2008. Model based inference in the life sciences: a primer on the central Appalachians. North American Journal of Fisheries Management evidence. Springer, New York. 30:1041–1050. Axel, G. A., E. F. Prentice, and B. P. Sandford. 2005. PIT-tag detection system Hill, S. A., G. B. Zydlewski, J. D. Zydlewski, and J. M. Gasvoda. 2006. Develop- for larger-diameter juvenile fish bypass pipes at Columbia River basin hydro- ment and evaluation of portable PIT tag detection units: PITpacks. Fisheries electric dams. North American Journal of Fisheries Management 25:646–651. Research 77:102–109. Aymes, J. C., and J. Rives. 2009. Detection efficiency of multiplexed passive Huggins, R. M. 1989. On the statistical analysis of capture experiments. integrated transponder antennas is influenced by environmental conditions Biometrika 76:133–140. and fish swimming behavior. Ecology of Freshwater Fish 18:507–513. Huggins, R. M. 1991. Some practical aspects of a conditional likelihood ap- Bond, M. H., C. V. Hanson, R. Baertsch, S. A. Hayes, and R. B. MacFarlane. proach to capture experiments. Biometrics 47:725–732. 2007. A new low cost instream antenna system for tracking passive integrated Linnansaari, T. P., and R. A. Cunjak. 2007. The performance and efficacy of a transponder (PIT)-tagged fish in small streams. Transactions of the American two-person operated portable PIT-antenna for monitoring spatial distribution Fisheries Society 136:562–566. of stream fish populations. River Research and Applications 23:559–564. Burnham, K. P., and D. R. Anderson. 2002. Model selection and multimodel Morhardt, J. E., D. Bishir, C. I. Handlin, and S. D. Mulder. 2000. A portable inference: a practical information-theoretic approach, 2nd edition. Springer- system for reading large passive integrated transponder tags from wild trout. Verlag, New York. North American Journal of Fisheries Management 20:276–283. Compton, R. I., W. A. Hubert, F. J. Rahel, M. C. Quist, and M. R. Bower. 2008. Nunnallee, E. P., E. F. Prentice, B. F. Jonasson, and W. Patten. 1998. Evaluation Influences of fragmentation on three species of native warmwater fish in a of a flat-plate PIT tag interrogation system at Bonneville Dam. Aquaculture Colorado River basin headwater stream system, Wyoming. North American Engineering 17:261–272. Journal of Fisheries Management 28:1733–1743. O’Donnell, M. J., G. E. Horton, and B. H. Letcher. 2010. Use of portable an- Connolly, P. J., I. G. Jezorek, K. D. Martens, and E. F. Prentice. 2008. Measuring tennas to estimate abundance of PIT-tagged fish in small streams: factors the performance of two stationary interrogation systems for detecting down- affecting detection probability. North American Journal of Fisheries Man- stream and upstream movement of PIT-tagged salmonids. North American agement 30:323–336. Journal of Fisheries Management 28:402–417. Prentice, E. F., T. A. Flagg, C. S. McCutcheon, D. F. Brastow, and D. C. Cross. Cucherousset, J., J. R. Britton, W. R. C. Beaumont, M. Nyqvist, K. Sievers, 1990. Equipment, methods, and an automated data-entry station for PIT and R. E. Gozlan. 2010. Determining the effect of species, environmental tagging. Pages 335–340 in N. C. Parker, A. E. Giorgi, R. C. Heidinger, D. B. conditions and tracking method on the detection efficiency of portable PIT Jester Jr., E. D. Prince, and G. A. Winans, editors. Fish marking techniques. telemetry. Journal of Fish Biology 76:1035–1045. American Fisheries Society, Symposium 7, Bethesda, Maryland. Cucherousset, J., J.-M. Paillisson, and J.-M. Roussel. 2007. Using PIT technol- Roussel, J.-M., A. Haro, and R. A. Cunjak. 2000. Field test of a new method ogy to study the fate of hatchery-reared YOY Northern Pike released into for tracking small fishes in shallow rivers using passive integrated transpon- shallow vegetated areas. Fisheries Research 85:159–164. der (PIT) technology. Canadian Journal of Fisheries and Aquatic Sciences Cucherousset, J., J.-M. Roussel, R. Keeler, R. A. Cunjak, and R. Stump. 2005. 57:1326–1329. The use of two new portable 12-mm PIT tag detectors to track small fish SAS Institute. 2010. SAS system software, release 9.2. SAS Institute, Cary, in shallow streams. North American Journal of Fisheries Management 25: North Carolina. 270–274. Saunders, W. C., K. D. Fausch, and G. C. White. 2011. Accurate estimation of Ellis, T. R., T. Linnansaari, and R. A. Cunjak. 2013. Passive integrated transpon- salmonid abundance in small streams using nighttime removal electrofish- der (PIT) tracking versus snorkeling: quantification of fright bias and com- ing: an evaluation using marked fish. North American Journal of Fisheries parison of techniques in habitat use studies. Transactions of the American Management 31:403–415. Fisheries Society 142:660–670. Sloat, M. R., P.F. Baker, and F. K. Ligon. 2011. Estimating habitat-specific abun- Fetherman, E. R. 2013. Introduction and management of Myxobolus cerebralis- dances of PIT-tagged juvenile salmonids using mobile antennas: a compari- resistant Rainbow Trout in Colorado. Doctoral dissertation. Colorado State son with standard electrofishing techniques in small streams. North American University, Fort Collins. Journal of Fisheries Management 31:986–993. Ficke, A. D., M. C. Kondratieff, and C. A. Myrick. 2012. The effects of PIT White, G. C., D. R. Anderson, K. P. Burnham, and D. L. Otis. 1982. Capture- tagging on the swimming performance and survival of three nonsalmonid recapture and removal methods for sampling closed populations. Los Alamos freshwater fishes. Ecological Engineering 48:86–91. National Laboratory, LA-8787-NERP, Los Alamos, New Mexico. Downloaded by [Department Of Fisheries] at 19:14 20 July 2015 Fischer, J. R., T. E. Neebling, and M. C. Quist. 2012. Development and eval- White, G. C., and K. P. Burnham. 1999. Program MARK: survival esti- uation of a boat-mounted RFID antenna for monitoring freshwater mussels. mation from populations of marked animals. Bird Study 46(Supplement): Freshwater Science 31:148–153. 120–138. Greenberg, L. A., and P. S. Giller. 2000. The potential of flat-bed passive Zydlewski, G. B., A. Haro, K. G. Whalen, and S. D. McCormick. 2001. Perfor- integrated transponder antennae for studying habitat use by stream fishes. mance of stationary and portable passive transponder detection systems for Ecology of Freshwater Fish 9:74–80. monitoring of fish movements. Journal of Fish Biology 58:1471–1475. Gries, G., and B. H. Letcher. 2002. Tag retention and survival of age-0 Atlantic Zydlewski, G. B., G. Horton, T. Dubreuil, B. Letcher, S. Casey, and J. Zydlewski. Salmon following surgical implantation with passive integrated transponder 2006. Remote monitoring of fish in small streams: a unified approach using tags. North American Journal of Fisheries Management 22:219–222. PIT tags. Fisheries 31:492–502. This article was downloaded by: [Department Of Fisheries] On: 20 July 2015, At: 19:14 Publisher: Taylor & Francis Informa Ltd Registered in England and Wales Registered Number: 1072954 Registered office: 5 Howick Place, London, SW1P 1WG

North American Journal of Fisheries Management Publication details, including instructions for authors and subscription information: http://www.tandfonline.com/loi/ujfm20 A Market Model of Eastern Bering Sea Alaska Pollock: Sensitivity to Fluctuations in Catch and Some Consequences of the American Fisheries Act James W. Stronga & Keith R. Criddlea a School of Fisheries and Ocean Sciences, University of Alaska Fairbanks, 17101 Point Lena Loop Road, Juneau, Alaska 99801, USA Published online: 16 Oct 2014.

Click for updates

To cite this article: James W. Strong & Keith R. Criddle (2014) A Market Model of Eastern Bering Sea Alaska Pollock: Sensitivity to Fluctuations in Catch and Some Consequences of the American Fisheries Act, North American Journal of Fisheries Management, 34:6, 1078-1094, DOI: 10.1080/02755947.2014.944678 To link to this article: http://dx.doi.org/10.1080/02755947.2014.944678

PLEASE SCROLL DOWN FOR ARTICLE

Taylor & Francis makes every effort to ensure the accuracy of all the information (the “Content”) contained in the publications on our platform. However, Taylor & Francis, our agents, and our licensors make no representations or warranties whatsoever as to the accuracy, completeness, or suitability for any purpose of the Content. Any opinions and views expressed in this publication are the opinions and views of the authors, and are not the views of or endorsed by Taylor & Francis. The accuracy of the Content should not be relied upon and should be independently verified with primary sources of information. Taylor and Francis shall not be liable for any losses, actions, claims, proceedings, demands, costs, expenses, damages, and other liabilities whatsoever or howsoever caused arising directly or indirectly in connection with, in relation to or arising out of the use of the Content. This article may be used for research, teaching, and private study purposes. Any substantial or systematic reproduction, redistribution, reselling, loan, sub-licensing, systematic supply, or distribution in any form to anyone is expressly forbidden. Terms & Conditions of access and use can be found at http:// www.tandfonline.com/page/terms-and-conditions North American Journal of Fisheries Management 34:1078–1094, 2014 C American Fisheries Society 2014 ISSN: 0275-5947 print / 1548-8675 online DOI: 10.1080/02755947.2014.944678

ARTICLE

A Market Model of Eastern Bering Sea Alaska Pollock: Sensitivity to Fluctuations in Catch and Some Consequences of the American Fisheries Act

James W. Strong and Keith R. Criddle* School of Fisheries and Ocean Sciences, University of Alaska Fairbanks, 17101 Point Lena Loop Road, Juneau, Alaska 99801, USA

Abstract A simultaneous-equation equilibrium model of domestic and international markets for fillets, surimi, and roe of Alaska pollock (i.e., Walleye Pollock Gadus chalcogrammus) is used to examine the combined effect of variability in landings and changes in product mix on first wholesale prices and revenues. From the commencement of American- ization in the mid-1970s through the joint venture era of the 1980s and following full Americanization in early 1990s, U.S. producers of Alaska pollock focused mainly on selling surimi and roe to Japan. Passage of the American Fisheries Act of 1998 (AFA) unleashed an economic transformation that more than doubled fishery revenues without increasing exploitation rates or catches. The increased value came about because freedom from the race for fish allowed fishers and processors to increase product quality, improve product recovery rates, diversify their mix of product forms, and develop new markets in the USA and Europe. Despite the size and importance of the Alaska pollock fishery—among U.S. fisheries, it consistently ranks number one by volume and in the top five by value—little attention has been given to the development of a formal model of Alaska pollock markets or the market impacts of the AFA. The results of the model indicate that first wholesale revenues are maximized when harvests are maximized, but the value of the harvest is particularly sensitive to changes in the demand for Alaska pollock harvested in Russian waters, price differentials between fillet and surimi, and Japanese inventories of Alaska pollock products. Comparative static simulations comparing pre- and post-AFA fishery product recovery rates and product mix indicate revenues would have increased by 62% from 1998 to 2007.

The eastern Bering Sea fishery for Alaska pollock (i.e., Wall- the fishery. This led to rapid increases in the number of U.S.- Downloaded by [Department Of Fisheries] at 19:14 20 July 2015 eye Pollock Gadus chalcogrammus) yields gross ex-vessel rev- flagged harvesters and catcher–processors and the development enues of over US$300 million and a first wholesale value of over of extensive shore-based processing capacity in the 1980s and $1.2 billion. Over the last 30 years, annual total allowable catch 1990s. In the mid-1980s, the total allowance level of foreign (TAC) has ranged from 813,000 metric tons to 1,494,900 metric fishing was zeroed-out in favor of allocations to joint ventures. tons, making the Alaska pollock fishery the largest U.S. fishery By 1991, the joint-venture allocations had also been eliminated in tonnage. During the 1960s and 1970s, Alaska pollock was to make room for harvesting by U.S.-flagged catcher vessels that largely harvested by a Japanese mothership fleet that extracted delivered to U.S.-flagged motherships or processing facilities in roe and processed the meat into surimi, a fish-based protein the USA and for harvesting by U.S.-flagged catcher–processors paste. A mostly Russian mothership fleet produced a frozen and U.S.-flagged catcher vessels that delivered to them. Con- headed and gutted (H&G) product destined for consumption tinued increases in harvesting and processing capacity during in the Eastern Bloc. Passage of the Fishery Conservation and the 1990s precipitated a race for fish and engendered heated po- Management Act in 1976 stimulated an “Americanization” of litical battles over allocations between the at-sea (motherships,

*Corresponding author: [email protected] Received June 21, 2013; accepted July 1, 2014

1078 ALASKA POLLOCK MARKETS 1079

catcher–processors, and associated catcher vessels) and inshore range of catches, implying that ex-vessel revenues could be max- sectors (Strong and Criddle 2013). While a few processors in- imized by harvesting the maximum permissible level of catch. vested in fillet technology, roe and surimi remained the primary However, because the AFA led to substantial reorganization of products. During the race for fish, it was not advantageous for production, that model is unlikely to provide a good representa- processors to produce time-intensive product such as fillets. In- tion of the effect of variation in catch limits under current market stead, they focused on high-throughput production methods to conditions. Fell (2008) used cointegration methods to examine accommodate the time-compressed volumes of catch that re- the responsiveness of Alaska pollock processors to changes in sulted from the race for fish. surimi and fillet prices before and after implementation of the These conditions changed following passage of the Amer- AFA and found that price responsiveness increased in both the ican Fisheries Act (AFA) in 1998. The AFA established per- inshore and at-sea sectors after the AFA was implemented. Mor- manent sector allocation shares for each fleet sector and autho- rison Paul et al. (2009) used a revenue function model to examine rized the formation of harvesting cooperatives in which fishing which product form was chosen by catcher–processors pre- and operations privately negotiated contracts that guaranteed each post-AFA and found that the mix of regular and deep-skin fillets, vessel a portion of the TAC allocated to that sector (Criddle surimi, and roe was responsive to variations in product prices and Macinko 2000; Matulich et al. 2001; Anderson 2002). The and increased operational flexibility that followed implementa- catcher–processor sector and catcher vessels that delivered to tion of the AFA. These choices of product form, in turn, affected them formed cooperatives in 1999. Catcher vessels that deliv- operation choices related to tow duration, crewing, and vessel ered to inshore processors formed cooperatives in 2000, as did power. the catcher vessels that delivered to motherships. Of the Alaska This study uses monthly time series data from 2000 to 2008 pollock allocated to these groups, the catcher–processor sector to extend the single product–single market model of Herrmann received 40%, the inshore sector 50%, and the mothership sector et al. (1996) to a represent post-AFA demand and allocation of 10%. The AFA also expanded the Western Alaska Community fillets, roe, and surimi to markets in Japan, Europe, and the USA. Development Quota (CDQ) program from an allocation of 7.5% In contrast to the cointegration approach applied in Fell (2008) of the eastern Bering Sea pollock TAC to 10% of the TAC for to identify structural breaks before and after the implementation each eastern Bering Sea groundfish species. The six CDQ enti- of AFA, and in contrast to the revenue function approach used ties represent 65 western Alaska communities and were created by Morrison Paul et al. (2009) to explicate the joint effects of to provide opportunities for community residents to benefit from product mix and productivity that arose from implementation of the wealth being derived from exploitation of eastern Bering the AFA, this study uses a structural model of post-AFA markets Sea fishery resources (Ginter 1995). These groups have, until to account for the effects of variability in landings and changes recently, leased their quota to AFA-qualified fishing vessels op- product demand on the mix of products produced, patterns of erating in the at-sea and inshore sectors. Under the AFA Alaska trade, and first wholesale prices and revenues. Simulations with pollock operations have enjoyed increased operational stability pre- and post-AFA product mixes, recovery rates, and markets and increased revenues (Felthoven 2002; Fell 2008; Wilen and based on the estimated model are used to explore the effects Richardson 2008; Morrison Paul et al. 2009). The AFA enabled on revenue of moving from a race-for-fish fishery to operating processors to look beyond simply processing the largest number under a cooperative structure. of fish before the end of the fishing season and to instead focus on maximizing revenues through increased capital expenditures into technology, expanded global markets, and optimized prod- METHODS Downloaded by [Department Of Fisheries] at 19:14 20 July 2015 uct mix. The Alaska pollock fishery is the largest U.S. fishery in vol- Model Specification ume and among the top five in value. Despite its relative impor- The five most important markets for Alaska pollock prod- tance, markets for Alaska pollock products have received very ucts were modeled simultaneously: fillet sales to the U.S. and little attention in formal models. While Hiatt et al. (2010) pro- Europe, surimi shipped to Japan and U.S., and roe wholesaled vides an excellent descriptive overview of Alaska pollock prod- to Japan. The first four of these markets were represented by ucts and their markets, it does not model supply-and-demand inverse demand equations representing amount of the product relationships that drive observed patterns of production or trade and its respective price in that market and allocation functions flows and is thus ill suited for representing future production or that distributed quantities of product among competing markets. value. Using data from 1986 to 1993, Herrmann et al. (1996) de- The roe market was also represented by an inverse demand func- veloped a simultaneous equation model of the Japanese demand tion. Although U.S. export statistics suggest that 34% of Alaska for U.S. surimi, the U.S. supply of surimi to Japan, inventory pollock roe is exported to Korea, Japanese import statistics in- holdings of surimi in Japan and the United States, and U.S. ex- dicate that 85% of that roe is re-exported to Japan. Thus, over vessel price for Alaska pollock. They found that although surimi 95% of U.S. Alaska pollock roe production (2000–2008) was exports to Japan were unresponsive to changes in export price, ultimately sold into the Japanese market. The demand equations ex-vessel demand was inelastic over the biologically plausible were specified as inverse (price dependent) demands because, 1080 STRONG AND CRIDDLE

in a typical year, fishers harvest the entire TAC, which is ex- Market Clearing Identities ogenously determined, and global prices adjust to the quantity Market clearing identities are equations that transform input produced. Each demand equation includes per capita imports, data series into appropriate forms for inclusion in the behavioral substitute prices, and per capita income. The producer price equations. These adjustments account for changes in price due index for intermediate foods and feeds was used to normalize to inflation and fluctuating exchange rates. Quantities supplied prices to 2008 levels. The equation system was modeled using were not adjusted to account for changes in human populations monthly data for 2000–2008. due to the relatively short time span modeled. The first 12 market The supply of Alaska pollock of U.S. origin was treated as clearing identities included in the model are used to obtain time a stochastic process moderated by variations in survival and series for the endogenous variables (Table 3). The 13th identity growth of Alaska pollock, various environmental factors, and adjusts prices to 2008 price levels to eliminate the influence of changes in harvests. The upper bound on U.S. harvests (the inflation. The final six identities describe the time series used for TAC) is set by the National Marine Fisheries Service (NMFS) Alaska pollock fillets and surimi allocated to the United States. on behalf of the U.S. Secretary of Commerce and based on recommendations of the North Pacific Fishery Management Allocation Equations Council (NPFMC). The NPFMC, in turn, is advised by a Sci- Allocation of Alaska pollock fillets to the United States.—The entific and Statistical Committee invested with authority (under United States and Europe are the two largest markets for fillets the Magnuson–Stevens Fishery Conservation and Management produced from U.S. harvests of Alaska pollock. Nearly 44% of Act) to set binding upper limits for the acceptable biological all fillets produced from 2000 to 2008 were retained for sale in catch and the overfishing level, both of which serve as upper the U.S. market. Equation (1) represents the quantity of these fillet US bounds on the TAC. In practice, the TAC is often well below fillets, Qt , as a linear function of monthly dummy vari- the overfishing level because the Bering Sea and Aleutian Is- ables, mi ; the U.S. real price for single frozen Alaska pollock US land groundfish fisheries are subject to an aggregate cap of 2 fillet fillet blocks, Pt ; the real price of U.S. Alaska pollock fillet million metric tons. The Scientific and Statistical Committee US→EU fillet acts as a peer-review panel for stock assessments prepared by exports to Europe, Pt ; the quantity of Alaska pollock fillet SS NMFS staff who develop models that incorporate information fillets produced by the inshore processing sector, Qt ;the from the fishery and fishery-independent surveys to estimate quantity of Alaska pollock fillets produced by the at-sea pro- fillet AS and project recruitment. While the fishery occasionally harvests cessing sector, Qt ; the lagged real U.S. import price for IM→US less than the TAC, landings are generally within a few percent- fillet Alaska pollock fillets, Pt−1 ; and the lagged quantity of ages of the TAC; thus, the supply of Alaska pollock is not well fillet US Alaska pollock fillets allocated to the USA, Q − : represented by a traditional supply function. Instead, the dispo- t 1 sition of surimi to the USA and Japan and the disposition of 11 US fillets to the USA and Europe were represented by allocation fillet US = β + β + β fillet Qt 10 1i mi 112 Pt equations that each considered own price, prices in compet- i=1 ing markets, levels of production in the at-sea processing and US→EU + β fillet + β fillet SS shore-based processing sectors, and monthly seasonal variables. 113 Pt 114 Qt Additional variables, such as lagged quantities and competing IM→US product prices, were included in some of the allocation equa- + β fillet AS + β fillet 115 Qt 116 Pt−1 tions to account for variability due to other factors. Linear and fillet US Downloaded by [Department Of Fisheries] at 19:14 20 July 2015 + β . nonlinear model forms were explored for each of the behavioral 117 Qt−1 (1) equations. The ultimate choice of functional form was based on goodness of fit for the individual equations and on performance The coefficients βni represent unknown parameters to be esti- of the overall equation system. Data sources used in the model mated, where the subscript associates the coefficient with the are listed in Table 1. Import and export statistics for the USA ith explanatory variable in the nth equation. Because there are were obtained from the U.S. International Trade Commission, no public data on the monthly quantities of U.S.-origin Alaska which maintains records of export and import quantities and pollock fillets sold within the United States, a proxy time series prices. Japanese imports and exports of roe and surimi, as well was constructed based on publically available data and advice as Japanese consumption statistics and welfare measures, were from industry representatives. These representatives indicated obtained from Trade Statistics of Japan. European import and that most pin-bone-out and pin-bone-in fillets are sold within export statistics were obtained from the Eurostat database. Pro- 90 d and that deep-skin fillets are nearly all sold within 180 d. duction data for the at-sea and inshore sectors of the U.S. Alaska A complete description of the identity (number 19) is listed in pollock fishery were derived from the NMFS catch account- Table 3. ing system (P. Britza, NMFS/Alaska Regional Office, personal While some U.S.-origin Alaska pollock are headed and gut- communication). Variable definitions and sources are listed in ted, reprocessed in China, and then reimported into the USA, Table 2. a majority of Alaska pollock fillets imported into the USA are ALASKA POLLOCK MARKETS 1081

TABLE 1. Data sources, frequencies, and ranges for information used in the market model of eastern Bering Sea Alaska pollock ﬁsheries.

Number Type Description Source Frequency and range 1 Japanese roe consumption Total value and quantity Trade statistics of Japan Monthly, 1999–2008 purchased per household. 2 Roe auction prices Seattle roe auction prices. BANR reports Intermittent spring and fall, 1999–2009 3 Japanese surimi and roe Amount in inventory. Trade statistics of Japan Monthly, 1992–2009 cold-storage holdings 4 Roe, fillet, and surimi Quantity produced by. NOAA, personal Monthly, 1998–2008 production inshore and at-sea sector. communication 5 Japanese roe import quantity Average price and quantity of Trade statistics of Japan Monthly, 1999–2008 and price imports from Russia, USA, and Korea. 6 Japanese surimi import Average price and quantity of Trade statistics of Japan Monthly, 2000–2008 quantity and price imports from all countries. 7 U.S. surimi import and export Average price and quantity U.S. International Trade Monthly, 1992–2009 quantity and price for individual and total Commission countries. 8 U.S. fillet price 66-lb block, US$/kg Urner Barry Monthly, 1998–2008 9 U.S. fillet export and import Average price and quantity U.S. International Trade Monthly, 1992–2009 quantity and price for individual and total Commission countries. 10 European fillet and surimi Average price and quantity of Eurostat Monthly, 1999–2009 import quantity and price imports from USA, Russia, China, and Japan. 11 U.S. per capita income Real terms U.S. Bureau of Labor Monthly, 1999–2008 statistics 12 Japanese index of real Real terms Trade statistics of Japan Monthly, 1999–2008 consumption expenditures 13 German per capita Income Real terms Eurostat Monthly, 1999–2008 14 U.S. halibut fillet prices 10–20 lb, US$/kg Urner Barry Monthly, 1998–2008 15 U.S. producer price index Intermediate foods and feeds U.S. Bureau of Labor Monthly, 1998–2008 statistics 16 Exchange rates U.S. dollar, yen, and euro Board of Governors of the Monthly, 1998–2008 U.S. Federal Reserve

Downloaded by [Department Of Fisheries] at 19:14 20 July 2015 fillet US→EU twice-frozen fillets of Russian origin processed in China. There- Alaska pollock fillet exports to Europe, Qt , as a linear fore the lagged real price of Alaska pollock imports to the function of monthly dummy variables, mi ; the real price of U.S.- USA was included in equation (1) to represent the impact of US→EU origin Alaska pollock fillet exports to Europe, fillet P ;the Russian-origin Alaska pollock fillet imports on sales of U.S.- t real U.S. price (from Urner Barry reports) for single, frozen, origin Alaska pollock fillets in the United States. The lagged US fillet quantity of U.S.-origin Alaska pollock fillets allocated to the Alaska pollock fillet blocks, Pt ; the lagged quantity of fillet SS USA represents the effect that past fillet sales to the United fillets produced by the inshore sector, Qt−1; and the lagged fillet AS States have on current sales through, for example, establish- quantity of fillets produced by the at-sea sector, Qt−1: ment of long-term contracts with major purchasers. 11 Allocation of Alaska pollock fillets to Europe.—The sales US→EU fillet US→EU = β + β + β fillet volume of U.S.-origin Alaska pollock fillets exported to Europe Qt 20 2i mi 212 Pt exceeded the volume of Alaska pollock fillets retained in the i=1 United States for the first time in 2002 and has frequently done US +β fillet + β fillet SS so since 2004. Overall, from 2000 to 2008, more than 49% 213 Pt 214 Qt−1 of U.S.-origin Alaska pollock fillets were sold into European + β fillet AS . markets. Equation (2) represents the quantity of U.S.-origin 215 Qt−1 (2) 1082 STRONG AND CRIDDLE

TABLE 2. Variable deﬁnitions and sources of information used in the market model of eastern Bering Sea Alaska pollock ﬁsheries. Source numbers refer to those listed in Table 1.

Variable Deﬁnition Source

fillet US Pt U.S. Urner Barry price of single frozen Alaska pollock fillet block. 8 fillet US→EU Pt Price of U.S.-origin Alaska pollock fillet exports to Europe. 9 fillet SS Qt Alaska pollock fillets produced by the inshore sector. 4 fillet AS Qt Alaska pollock fillets produced by the at-sea sector. 4 fillet IM→US Pt U.S. import price of Alaska pollock fillets. 9 fillet US→EU Qt U.S.-origin Alaska pollock fillets exported to Europe. 9 fillet totalEx Qt Total exports of U.S.-origin Alaska pollock fillets. 7 fillet total Qt Total quantity of Alaska pollock fillets produced. 4 surimi US→JN Qt U.S.-origin Alaska pollock surimi imported to Japan. 6 surimi US→JN P(JP )t Price of U.S.-origin Alaska pollock surimi imported to Japan. 6 surimi SS Qt Alaska pollock surimi produced in the inshore sector. 4 surimi AS Qt Alaska pollock surimi produced in the at-sea sector. 4 surimi US Pt U.S. surimi import price. 7 surimi totalEx Qt Total exports of U.S.-origin Alaska pollock surimi. 7 surimi total Qt Total quantity of U.S.-origin Alaska pollock surimi. 4 I1 and I2 Indicator variables for missing observations of U.S.-origin Alaska pollock surimi prices. US Inct Real U.S. per capita disposable personal income. 11 fillet € CR→EU P(EU )t European import price of Alaska pollock fillets from China and Russia. 10 € GE Inc( EU )t Index of disposable income of Germany in euros. 13 €→$ Ext Europe–U.S. exchange rate: one euro to U.S. dollars. 16 →$ Ext Japan–U.S. exchange rate: one U.S. dollar to Japanese yen. 16 roe US→JN P( JP )t Japanese import price of U.S.-origin Alaska pollock roe. 5 roe SS Qt Alaska pollock roe produced in the inshore sector. 4 roe AS Qt Alaska pollock roe produced in the at-sea sector. 4 roe CR→JN P ( JP )t Japanese import price of Alaska pollock roe from China and Russia. 5 JN Inct Japanese index of real consumption expenditures. 12 roe JN St Japanese cold-storage holdings of Alaska pollock roe. 3 I3 Indicator variable for missing observations of Alaska pollock roe prices in Japan. surimi JN St Japanese cold-storage holdings of Alaska pollock surimi. 3 surimi other→JN P( JP )t Japanese import price of surimi other than U.S.-origin Alaska pollock surimi. 6 surimi € US→EU P ( EU )t European import price of U.S.-origin Alaska pollock surimi. 10 halibut US Pt Price of halibut in the USA. 14 I5and I6 Indicator variable for data point with no observations on U.S.-origin Alaska

Downloaded by [Department Of Fisheries] at 19:14 20 July 2015 pollock surimi prices. roe US→JN Qt Japan imports of Alaska pollock roe. 5 US = PPIt U.S. producer price index for intermediate foods and feeds (base 1982). 15

Lagged quantities of fillets produced by the inshore and at-sea surimi SS surimi produced in the inshore sector, Qt−2; the two-period sectors were used to represent the difference in time between lagged quantity of Alaska pollock surimi produced in the at-sea fillet production and their distribution to Europe. surimi AS sector, Qt−2; and the real U.S. price (from Urner Barry Allocation of Alaska pollock surimi to Japan.—The primary US fillet market for Alaska pollock surimi is Japan, which purchased reports) for single, frozen, Alaska pollock fillet blocks, Pt : over 63% of total U.S.-origin Alaska pollock surimi from 2000 11 US→JN to 2008. Equation (3) models the quantity of Alaska pollock → surimi QUS JN = β + β m + β surimi P surimi imported from the USA by Japan, surimi QUS→JN,as t 30 3i i 312 t t i=1 a linear function of monthly dummy variables, mi ; the real +β surimi Q SS + β surimi Q AS price of Alaska pollock surimi imports by Japan from the USA, 313 t−2 314 t−2 US→JN US surimi + β fillet . Pt ; the two-period lagged quantity of Alaska pollock 315 Pt (3) ALASKA POLLOCK MARKETS 1083

TABLE 3. Market-clearing identities in the market model of eastern Bering Sea Alaska pollock ﬁsheries.

Identity Equation

US fillet = fillet US ∗ 2008 US 1. Price of U.S.-origin Alaska pollock fillets in the USA. Pt Pt PPIt

US→EU fillet = fillet US→EU ∗ 2008 US 2. Price of U.S.-origin Alaska pollock fillets exported to Pt Pt PPIt Europe. IM→US fillet = fillet IM→US ∗ 2008 US 3. U.S. import price for Alaska pollock fillets. Pt Pt PPIt US→JN surimi = surimi US→JN / →$ ∗ 2008 US 4. Japanese import price for U.S.-origin Alaska pollock surimi. Pt P (JP )t Ext PPIt

US surimi = surimi US ∗ 2008 US 5. U.S. import price of surimi. Pt Pt PPIt CR→EU fillet = fillet € CR→EU ∗ €→$ ∗ 2008 US 6. European import price for Alaska pollock fillets from China Pt P (EU )t Ext PPIt and Russia. US→JN roe = roe US→JN / →$ ∗ 2008 US 7. Japanese import price for U.S.-origin Alaska pollock roe. Pt P (JP )t Ext PPIt

roe total = roe SS + roe AS 8. Total quantity of roe produced. Qt Qt Qt CR→JN roe = roe CR→JN / →$ ∗ 2008 US 9. Japanese import price for Alaska pollock roe from China Pt P (JP )t Ext PPIt and Russia. other→JN surimi = 10. Japanese import price for surimi excluding U.S.-origin Pt → → imports. surimi other JN / $ ∗ 2008 US P (JP )t Ext PPIt US halibut = halibut US ∗ 2008 US 11. U.S. halibut price. Pt Pt PPIt US→EU surimi = surimi € US→EU ∗ €→$ ∗ 2008 US 12. European import price for U.S.-origin Alaska pollock Pt P (EU )t Ext PPIt surimi. 2008 US = US/2008Mean US 13. 2008 producer price index for intermediate food and feeds. PPIt PPIt PPI

surimi totalUS = surimi total − surimi totalEx 14. Annual quantity of Alaska pollock surimi not exported in year Qt year Qt year Qt the year of month t. surimi prop = surimi totalUS/surimi total Downloaded by [Department Of Fisheries] at 19:14 20 July 2015 15. Ratio of annual Alaska pollock surimi not exported to total year Qt year Qt year Qt annual production of Alaska pollock surimi in the year of month t. surimi QUS = (0.2 ∗ surimi Qtotal ∗ surimi Q prop) + (0.4 ∗ surimi Qtotal 16. Quantity of U.S.-origin Alaska pollock surimi allocated to t t−1 year t−1 t−2 ∗surimi Q prop) + (0.4 ∗ surimi Qtotal ∗ surimi Q prop) the USA. year t−2 t−3 year t−3

fillet totalUS = fillet total − fillet totalEx 17. Quantity of U.S.-origin Alaska pollock fillets not exported. Qt Qt Qt

fillet prop = fillet totalUS/fillet total 18. Ratio of annual Alaska pollock fillet not exported to total year Qt year Qt year Qt annual fillet production in the year of month t. fillet US = . ∗ fillet total ∗ fillet prop + . ∗ fillet total Qt (0 35 Qt−1 year Qt−1 ) (0 4 Qt−2 ∗fillet Q prop) + (0.1 ∗ fillet Qtotal ∗ fillet Q prop) 19. Quantity of U.S.-origin Alaska pollock fillets allocated to year t−2 t−3 year t−3 +(0.05 ∗ fillet Qtotal ∗ fillet Q prop) + (0.05 ∗ fillet Qtotal the USA. t−4 year t−4 t−5 ∗fillet prop + . ∗ fillet total ∗ fillet prop year Qt−5 ) (0 05 Qt−6 year Qt−6 ) 1084 STRONG AND CRIDDLE

Japanese import data are used in place of U.S. export data be- linear function of monthly dummy variables, mi ; the quantity of cause the latter fails to account for the portion of Japan-bound U.S.-origin Alaska pollock fillets allocated to the U.S. market, fillet US surimi that is first exported to Korea and held in bonded ware- Qt ; the real price of Alaska pollock fillets imported into IM→US houses for reshipment to Japan. To account for the lag time fillet the USA, Pt ; the lagged, real first wholesale price of between production and arrival in Japan, the model includes a US fillet 2-month lag on surimi production. The real U.S. price (from Alaska pollock fillet blocks in the USA, Pt−1; and real U.S. US Urner Barry reports) for single, frozen, Alaska pollock fillet per capita disposable personal income, Inct : blocks was included to account for the opportunity cost of pro- 11 ducing high-grade surimi for Japan in terms of forgone produc- US fillet = β + β + β fillet US tion of fillets. Pt 50 5i mi 512 Qt Allocation of Alaska pollock surimi to the United States.— i=1 The United States was the secondmost important market for IM→US US +β fillet P + β fillet P Alaska pollock surimi over the period 2000–2008 when the 513 t 514 t−1 U.S. share was 21% of U.S.-origin Alaska pollock surimi out- US + β515 Inc . (5) put. Equation (4) expresses the quantity of U.S.-origin Alaska t surimi US pollock surimi allocated to the USA, Qt , as a linear func- Although the model was estimated with several different white- tion of monthly dummy variables, mi ; the real price of all surimi fish prices including Pacific Cod G. macrocephalus, Atlantic US surimi Cod G. morhua, and tilapia Oreochromis spp., estimation with imported into the USA, Pt ; the quantity of Alaska pollock surimi SS these parameters did not produce statistically significant or theo- surimi produced in the inshore sector, Qt ; the quantity of surimi AS retically consistent results. Instead, the model used the real price Alaska pollock surimi produced in the at-sea sector, Qt ; the lagged quantity of Alaska pollock surimi allocated to the of Alaska pollock fillets imported into the United States as a sub- surimi US stitute for U.S.-origin Alaska pollock fillets. These imports are USA, Qt−1; and two indicator variables, I1 and I2,usedto remove the influence of 2 months when there were no recorded largely comprised of twice-frozen, Russian-origin fillets pro- surimi imports to the United States: cessed in China. This is consistent with the opinion of Doug Christensen, president of the U.S. Surimi Commission, who be- 11 US lieves the decline in Russian-origin Alaska pollock harvests was surimi US = β + β + β surimi Qt 40 4i mi 412 Pt one of the primary reasons for the success of the U.S. Alaska i=1 pollock fishery following the passage of the AFA. He believes +β surimi SS + β surimi AS 413 Qt 414 Qt that future changes in Russian production will significantly influence the profitability of the U.S. fishery (D. Christensen, U.S. +β surimi QUS + β (I ) + β (I ) . (4) 415 t−1 416 1 417 2 Surimi Commission, personal communication). Inverse demand for Alaska pollock fillets in Europe.— Although there are no public data on monthly quantities of U.S.- Equation (6) characterizes the real price of U.S.-origin Alaska origin Alaska pollock surimi sold within the United States, a US→EU fillet proxy time series described in Table 3 (identity 16) was con- pollock fillet exports to Europe. Where Pt is repre- structed based on advice from industry representatives who sug- sented as a linear function of monthly dummy variables, mi ; gested that most surimi sold into the U.S. market is sold within the quantity of U.S.-origin Alaska pollock fillet exports to Eu- fillet US→EU 90 d of its production. Japanese surimi price was provision- rope, Qt ; the real price of European imports of Alaska CR→EU

Downloaded by [Department Of Fisheries] at 19:14 20 July 2015 ally included in equation (4), but was dropped because the es- pollock from China and Russia, fillet P ; an index of dis- timated coefficient was not statistically significant and its sign t posable income in Germany, IncGE; and the euro to U.S. dollar was inconsistent with accepted microeconomic theory. This out- t exchange rate, Ex→$: come is not surprising because the model does not differentiate t

between high-grade surimi produced from flesh and recovery- 11 US→EU grade surimi produced from trimmings and stickwater. Most fillet fillet US→EU P = β60 + β6i mi + β612 Q high-grade surimi of U.S. origin is sold to the Japanese market, t t i=1 leaving lower-grade surimi for sale to less-discerning markets. CR→EU +β fillet + β GE Consequently, it is not surprising that the U.S. market was not 613 Pt 614 Inct significantly impacted by changes in the Japanese import price +β →$ . of U.S.-origin Alaska pollock surimi. 615 Ext (6) Inverse Demand Equations As with equation (5), the prices of several substitute fish species Inverse demand for Alaska pollock fillets in the United were included in initial model specifications but were dropped States.—Equation (5) represents the real U.S. price (from Urner US from this equation because the estimated coefficients were fillet Barry reports) for Alaska pollock fillet blocks, Pt ,asa not statistically significant. Instead, the real price of European ALASKA POLLOCK MARKETS 1085

imports of Alaska pollock fillets from China and Russia was cho- surimi JN Japanese cold-storage holdings of surimi, St−3: sen as a substitute because Russian-origin Alaska pollock fillets provide the most direct competition with U.S.-origin Alaska pol- 11 US→JN lock fillets in Europe. Large quantities of those fillets are derived surimi = β + β + β surimi US→JN Pt 80 8i mi 812 Qt from minimally processed fish shipped to China for reprocess- i=1 ing and then shipped into the European market as twice-frozen other→JN +β surimi fillets. An index of disposable income in Germany was chosen 813 Pt because Germany was the largest European importer of U.S.- US→JN origin Alaska pollock fillets over the modeled time period. The + β surimi 814 Pt−2 euro to U.S. dollar exchange rate is included to represent the +β JN + β surimi JN . portion of variability in European prices of Alaska pollock fillets 815 Inct 816 St−3 (8) that is due to exchange rate fluctuations. Inverse demand for Alaska pollock roe in Japan.—No alloca- The two-period lagged Japanese import real price for U.S.- tion equation was modeled for the Japanese demand for Alaska origin Alaska pollock surimi is used to capture unknown market pollock roe because over 95% of all U.S.-origin Alaska pollock forces that affect the Japanese surimi price. The three-period roe is exported to Japan. Instead, equation (7) models the real Japanese cold-storage holdings of surimi is included to reflect its price of Japanese imports of U.S.-origin Alaska pollock roe, impact on price in subsequent periods because negotiations over US→JN roe the price of surimi between the U.S. Surimi Commission and Pt , as a linear function of monthly dummy variables, Japanese surimi processors often take place before the season mi ; the quantity of U.S.-origin Alaska pollock roe produced, roe total begins; therefore, current prices reflect to some degree the cold- Qt ; the real price of Japanese imports of Alaska pollock CR→JN storage holdings of prior periods. roe roe from China and Russia, Pt ; the Japanese index of Inverse demand for Alaska pollock surimi in the United JN real consumption expenditures, Inct ; the quantity of Alaska States.—Equation (9) represents the real wholesale price of roe JN pollock roe held in Japanese cold-storage facilities, St ; and US U.S.-origin Alaska pollock surimi in the USA, surimi P ,as an indicator variable, I3, used to eliminate the influence of one t missing observation: a linear function of monthly dummy variables, mi ; the quantity of U.S.-origin Alaska pollock surimi allocated to the surimi QUS Hippoglos- 11 USA, t ; the real price of Pacific Halibut US→JN halibut US roe = β + β + β roe total sus stenolepis in the USA, Pt ; the lagged real whole- Pt 70 7i mi 712 Qt i=1 sale price of U.S.-origin Alaska pollock surimi in the USA, US CR→JN surimi +β roe + β JN Pt−1; the real U.S. per capita disposable personal income, 713 Pt 714 Inct US Inct ; the real price of U.S.-origin Alaska pollock surimi ex- US→EU + β roe JN + β . surimi 715 St 716 (I3) (7) ports to Europe, Pt ; the four-period, lagged, real wholesale price of U.S.-origin Alaska pollock surimi in the US Because Alaska pollock roe is consumed almost exclusively surimi USA, Pt−4; and two indicator variables, I4 and I5: in Japan, roe prices are influenced by anticipated Russian and

U.S. production and Japanese inventories (American Seafoods 11 US surimi surimi US Downloaded by [Department Of Fisheries] at 19:14 20 July 2015 Group 2006). Therefore, the Japanese import price of Russian- P = β90 + β9i mi + β912 Q origin Alaska pollock roe and Japanese cold-storage holdings t t i=1 of Alaska pollock roe have been included in the model to repre- US US +β halibut + β surimi sent their respective impacts on Japanese import prices of U.S. 913 Pt 914 Pt−1 Alaska pollock roe. US→EU Inverse demand for Alaska pollock surimi in Japan.— +β US + β surimi 915 Inct 916 Pt Equation (8) describes the inverse demand for U.S.-origin US→JN surimi US surimi +β917 P − + β918 (I4) + β919 (I5) . (9) Alaska pollock surimi imports in Japan, Pt ,asa t 4 linear function of monthly dummy variables, mi ; the quantity of Japanese imports of Alaska pollock surimi from the USA, The allocation of Alaska pollock surimi to the United States surimi US→JN Qt ; the average real price of all other surimi im- has varied widely, ranging from 34% of production in 2002 other→JN to 6% in 2005. Because there is no public time series of U.S. ported into Japan, surimi P ; the two-period, lagged, real t surimi prices, the model assumes that the U.S. wholesale price price of Japanese imports of Alaska pollock surimi from the US→JN of Alaska pollock surimi is proportional to the average U.S. surimi USA, Pt−2 ; the Japanese index of real consumption import price of all surimi. The model uses the import price of JN expenditures, Inct ; and the three-period lagged quantity of all surimi rather than just Alaska pollock surimi, because most of 1086 STRONG AND CRIDDLE

the surimi purchased by the U.S. market is of intermediate and estimator could not be used. While two-stage least squares recovery grades, and U.S. surimi purchasers do not generally (2SLS) and three-stage least squares (3SLS) are possible op- place a premium on Alaska pollock products. The one-period tions, 2SLS assumes that there is no covariance among equa- and four-period lagged prices account for autocorrelations that tions; thus, an iterated 3SLS approach is preferred when the represent the role of long-term contracts and establishment of disturbances are contemporaneously correlated, as they were relationships with major buyers. The indicator variables are used expected to be in this case (Zellner and Theil 1962; Henningsen to account for two periods, February 2001 and May 2006, for and Hamann 2007). Coefficient estimates and corresponding which there were no observations of first wholesale prices in the asymptotic P-values are reported in Table 4. data set for the U.S. import price of all surimi. Goodness-of-fit statistics were used to evaluate the predictive Roughly 10% of Alaska pollock surimi produced in the accuracy of the system of equations as a whole because statis- United States was exported to Europe during the 2000–2008 pe- tics that characterize the performance of individual equations riod. A majority of that quantity was exported in later years when do not account for how the equations interact and how those in- U.S. exports of Alaska pollock surimi to Europe approached and, teractions affect overall system performance. Historic dynamic in 2009, exceeded exports to Japan. But because this was a re- simulations were conducted for 2000–2008. The goodness-of- cent development there were not enough observations to model fit statistics considered included the coefficient of variation, CV the emergence of this market. Instead, the model includes the (100·SD/mean); the correlation between the actual and predicted real price of European imports of U.S.-origin Alaska pollock values of the endogenous variables, r; the Theil inequality co- surimi to represent the increasing importance of the European efficient, and U1 (Theil 1966). market. Total Alaska pollock revenue.—The model was designed to Relating the U.S. TAC for Alaska Pollock to the Allocation highlight trade flows and product demands in the most important Equations markets for U.S.-origin Alaska pollock. Quantities and prices Because the TAC varies as a function of stock abundance, from the five most important markets were estimated as part of a it is necessary to relate changes in the TAC to changes in the simultaneous equation system. While not explicitly represented production of roe, fillet, and surimi, which can then be used in the estimated equation system, the first wholesale value of with the estimated coefficients to conduct comparative static Alaska pollock products can be derived from estimates of prices simulations that describe how changes in stock abundance af- and allocations from the primary markets and used to create fect markets and value. Relating surimi and fillet production estimates of first wholesale revenues. Equation (10) represents to catch requires several considerations. First, flesh recovery real first wholesale revenues for U.S.-origin Alaska pollock, for fillets and surimi is largely dependent on fish size. Age- Pollock TRt , as the sumproduct of fillet allocations and prices in 3 and age-4 Alaska pollock, which average 345 to 548 g, are the USA and Europe, surimi allocations and prices in the USA considered ideal for producing surimi. Larger fish are ideal for and Japan, and quantities and prices of Japanese roe imports: producing fillets. Catches of small fish can decrease recovery from an average of 29–31% to an average of as little as 24– US 25%. Recovery rates can be further depressed when the catch TRPollock = fillet QUS × fillet P t t t includes large numbers of very small fish (age 2), which are typ- US→EU ically ground into fish meal. Alaska pollock larger than 750 g + fillet US→EU × fillet Qt Pt tend to yield lower-grade surimi because their flesh strength declines (S. Wilt, Alyeska Seafoods, personal communication). US→JN → Downloaded by [Department Of Fisheries] at 19:14 20 July 2015 + surimi US JN × surimi Qt Pt Moreover, because the Alaska pollock fishery faces a cap on bycatch of Chinook Salmon Oncorhynchus tshawytscha and US fixed and dynamic spatial closures intended to reduce bycatch + surimi QUS × surimi P t t of other species, the fishery may be constrained from fishing US→JN in areas with preferred sizes of Alaska pollock. As a result, + roe US→JN × roe . Qt Pt (10) changes in the size distribution of the catch can significantly affect the mix of processed products independent of market considerations. Coefficient Estimates and Model Performance Another consideration is that there are limits to the flexibil- Coefficient estimates for the system of nine equations were ity that processors have in shifting between fillet and surimi obtained using the open source program Systemfit in the R pro- production. Trimmings and stickwater, along with small fish, gramming language (Henningsen and Hamann 2007). System- can be used to produce surimi but cannot be shifted into fil- fit uses an iterative three-stage least-squares (3SLS) approach lets (Park 2005; D. Abbasian, Trident Seafoods, personal com- (Zellner and Theil 1962). All nine equations in the specified sys- munication; Christensen, personal communication; J. Jacobs, tem are overidentified. Consequently an indirect least-squares American Seafoods, personal communication). Consequently, ALASKA POLLOCK MARKETS 1087

TABLE 4. Coefﬁcient estimates with corresponding P-values (in parentheses) and summary statistics from equations used in the market model of eastern Bering Sea Alaska pollock ﬁsheries. Equation numbers refer to those presented in the text.

Equation βn0 βn1 βn2 βn3 βn4 1 –3,334,730 (<0.01) –91,632 (0.79) –359,214 (0.44) 633,345 (0.11) –67,905 (0.84) 2 4,579,960 (0.09) –552,692 (0.66) 2,747,550 (0.04) –2,150,530 (0.23) –6,363,480 (<0.01) 3 18,244,300 (<0.01) 231,429 (0.90) 981,453 (0.61) 9,281,380 (<0.01) 2,529,950 (0.21) 4 –1,937,260 (<0.01) –352,360 (0.54) –481,554 (0.62) 1,914,120 (0.02) 3,420,430 (<0.01) 5 –1.285 (<0.01) –0.018 (0.78) 0.104 (0.14) 0.084 (0.28) 0.014 (0.83) 6 –3.123 (<0.01) 0.047 (0.70) 0.091 (0.52) 0.210 (0.18) –0.141 (0.26) 7 6.968 (0.78) 1.374 (0.55) 4.924 (0.31) 9.153 (0.02) 7.760 (<0.01) 8 2.437 (0.18) –0.462 (0.02) –0.378 (0.04) 0.425 (0.05) 0.362 (0.16) 9 4.960 (0.03) 0.075 (0.81) –0.0024 (0.99) 0.183 (0.58) 0.680 (0.05) 11 3.062 (<0.01) –0.394 (<0.01) –0.801 (<0.01) –0.740 (<0.01) –0.662 (<0.01) Equation βn5 βn6 βn7 βn8 βn9 1 –909,894 (<0.01) –44,909 (0.90) –57,792 (0.91) 367,504 (0.47) –259,613 (0.57) 2 –738,716 (0.55) –159,983 (0.90) 4,595,410 (<0.01) –1,482,110 (0.46) –2,630,240 (0.19) 3 –4,356,780 (0.01) 1,284,130 (0.45) 1,642,130 (0.35) 5,109,160 (<0.01) 4,865,250 (<0.01) 4 –122,711 (0.83) –715,066 (0.20) 375,974 (0.64) 1,016,020 (0.27) 1,653,210 (0.05) 5 –0.035 (0.59) –0.012 (0.85) 0.099 (0.18) 0.275 (<0.01) 0.197 (0.03) 6 –0.173 (0.16) –0.171 (0.17) 0.122 (0.43) 0.199 (0.25) 0.306 (0.07) 7 8.185 (<0.01) 8.562 (<0.01) 8.771 (<0. 01) 8.022 (<0.01) 9.112 (<0.01) 8 –0.209 (0.20) –0.229 (0.19) –0.259 (0.14) –0.032 (0.85) 0.226 (0.32) 9 –0.295 (0.36) 0.233 (0.45) 0.233 (0.45) 0.183 (0.57) 0.294 (0.41) 11 –0.681 (<0.01) –0.941 (<0.01) –1.050 (<0.01) 0.235 (<0.01) –0.460 (<0.01) Equation βn10 βn11 βn12 βn13 βn14 βn15 1 –63,100 (0.87) –323,840 (0.33) 677,624 (<0.01) –494,360 (0.01) 0.158 (<0.01) 0.233 (<0.01) 2 –3,279,040 (0.06) –2,040,040 (0.13) 2,272,130 (<0.01) –2,971,050 (<0.01) 0.372 (<0.01) 0.564 (<0.01) 3 5,397,310 (<0.01) 406,270 (0.83) 479,587 (0.71) 0.375 (<0.01) 0.319 (0.02) –4,914,180 (0.06) 4 1,359,520 (0.03) 755,711 (0.17) 87,489 (0.63 0.199 (<0.01) 0.034 (0.53) 0.783 (<0.01) 5 0.144 (0.06) 0.033 (0.62) –2.28×10−8 (<0.01) 0.203 (<0.01) 0.827 (<0.01) 4.60×10−5 (<0.01) 6 0.195 (0.17) 0.114 (0.36) –1.98×10−8 (0.08) 1.608 (<0.01) 7.68×10−5 (0.65) 1.065 (<0.01) 7 5.996 (<0.01) 5.617 (<0.01) –6.1×10−7 (0.17) 0.686 (<0.01) 0.154 (0.52) –9.1× 10−7 (<0.01) 8 0.459 (0.10) 0.172 (0.38) –4.76×10−8 (0.02) 0.663 (<0.01) 0.444 (<0.01) –0.013 (0.50) 9 –0.105 (0.77) 0.179 (0.57) –2.56×10−8 (0.52) 0.183 (0.12) 0.270 (<0.01) 0.382 (<0.01) 11 0.641 (<0.01) 2 Equation βn16 βn17 βn18 βn19 df R 1 856,770 (<0.01) 0.617 (<0.01) 90 0.96 2 92 0.81 3 92 0.82 4 –883,162 (0.46) 783,296 (0.50) 90 0.88

Downloaded by [Department Of Fisheries] at 19:14 20 July 2015 5 92 0.89 6 92 0.77 7 –12.90 (<0.01) 91 0.76 8 –2.32×10−8 (<0.01) 91 0.80 9 –1.95×10−4 (0.01) 0.330 (<0.01) 0.501 (0.44) –0.059 (0.93) 88 0.56 11 1,741 0.78

no matter what the price differential is, a portion of the total re- lines that can be switched between surimi and fillet produc- covery will invariably go into surimi. Product flexibility is also tion. Some minimum amount of Alaska pollock must be pro- limited by long-term contracts, which can constrain processors cessed through the surimi lines to avoid shutdown or reductions from responding to short-term changes in demand. Additional in surimi quality. Time spent in refrigerated fish holds aboard technical constraints arise in limits to the interchangeability of catcher vessels affects flesh recovery and product quality; thus, processing lines. Processors typically maintain dedicated pro- increased travel distances from fishing grounds, especially in cessing lines devoted to surimi or fillets, as well as processing the summer and fall (B season) negatively impact flesh recovery 1088 STRONG AND CRIDDLE

in shore-based processors and lead to an increase in surimi pro- eries production and trade data). Based on Hiatt et al. (2010), duction relative to fillet production. (Note: Beginning in 1992, surimi produced at sea was modeled with a 23% price premium. the Alaska pollock TAC was apportioned between a late winter Although the inshore sector has greater flexibility to direct fil- A season and a summer–fall B season [NMFS 1999]. Because let production into different product forms, the at-sea sector Alaska pollock spawning peaks in late winter and early spring, produces a higher proportion of high-value, deep-skin fillets. roe content is higher in the A season than it is in the B season.) Nevertheless, because the model does not differentiate between Flesh recovery rates calculated for inshore and at-sea fillet types, fillet prices were treated as equal across the sectors. (catcher–processor and mothership) sectors and by season for Data from 1,752 actual transactions (from 2000 to 2006 and 1998 were based on Hiatt and Terry (2000) and source 4 (Ta- 2008) were used to model roe price as a log-linear function of ble 2). For 2007, estimates of flesh recovery rates were based the year of the transaction for 2001–2006 and 2008, Yt , and three on catches reported in Hiatt et al. (2010) and source 4 (Table 2). binary variables representing harvest sector: St , (0 for inshore, The 2007 fishing season was chosen instead of 2008 due to and 1 for at sea), season, Bt , (0 for A season, 1 for B season,), the 2008 financial crisis, which caused an unexpected shock and grade, Gt (0 for below standard grade, 1 for standard grade): to the system. In 2007, catches by the Ocean Phoenix were not included in the at-sea harvest because the Ocean Phoenix 7 Roe BANR = β + β +β +β +β . had been converted to produce H&G instead of fillets. Inciden- loge Pt 110 11i Yt 118 St 119 Bt 1110Gt tal catches of Alaska pollock in nontarget fisheries were also i=1 omitted because they are a small component of total catch and (11) they enter the market as low-valued product forms. Based on The transactions used to estimate the parameters of equation these assumptions, the 1998 late winter (A season) flesh recov- (11) represented 45% (82,914 metric tons) of the 181,242 metric ery rate was 19.0% for the inshore sector and 14.3% for the tons of roe produced from 2000 to 2006 and in 2008. The data at-sea sector, and was 25.8% and 24.5%, respectively, in 2007. were compiled from BANR market reports over 2000–2006 and In 1998, B season recovery rates were 23.2% for the inshore from Seafood News reports in 2008, and included a variety sector and 14.98% for the at-sea sector, and 25.2% and 26.8%, of factors that contributed to the determination of the different respectively, in the 2007 fishing season. To account for the pro- values of different characteristics of the roe. The 2007 data were duction of recovery-grade surimi that cannot be alternatively not included because BANR reports were discontinued in 2006 used as fillets, we assigned a production rate of 3.5% in 1998 and because 2007 data from Seafood News reported transaction and 4.25% in 2007, equivalent to the ratio of surimi produced prices but did not report corresponding quantities. A weighted that did not go into the Japanese market and the sum of U.S. least-squares approach was used to account for differences in catches of Alaska pollock. For the 2007 simulations, based on the volumes of reported transactions. The log-linear functional the advice of industry representatives, one-third of the remain- form was used to represent percent changes in price associated ing flesh recovery was allocated to surimi and one-third was with processing roe in the at-sea and inshore sectors. allocated to fillets. For the 1998 simulations, 55% and 75% from the at-sea and inshore sectors, respectively, were allocated Comparative Static Simulation to surimi, and 10% from both sectors was allocated to fillet. Comparative static simulations were conducted to quantify The balance was allocated according to highest price—surimi the additional revenue derived from optimized product mix and exports to Japan, fillet exports to Europe, or U.S. fillet sales. increased recovery rates consequent to the transition from a Similarly, based on advice of industry representatives, roe re- race-for-fish to the AFA cooperative management regime. Sim-

Downloaded by [Department Of Fisheries] at 19:14 20 July 2015 covery rates during the A season were assumed to be 1.91% ulations were also conducted over a wide range of harvest levels for the inshore sector and 2.38% for the at-sea sector in 1998; to determine the effect of changing the TAC on product prices 2007 rates were assumed to be 4.12% and 4.93%, respectively, and revenues, as well as to examine potential TAC levels where for those sectors. In the 1998 B season, roe recovery rates were sectors may decide to harvest less than their full allocation. Con- modeled as 0.008% for the inshore sector and 0.009% for the at- ditions reflective of 2007 were used as a basis for simulations sea sector, and 0.60% and 0.38% for those sectors, respectively, and sensitivity analyses of revenue because they preceded the in 2007. 2008–2010 recession and because the B-season inshore TAC was underharvested in 2007 due to a large TAC, low surimi Price Differences between Sectors prices, high fuel prices, and an unexpected reduction in CPUE To examine sector specific revenues within the comparative near the end of the season. To explore the effects of transition to static simulations, it was necessary to differentiate product val- AFA cooperatives, simulations based on 2007 market conditions ues between the at-sea and inshore sectors. Differences between were run using the 1998 recovery rates and markets. Although surimi prices in the inshore and at-sea sectors can be substantial. some Alaska pollock roe, surimi, and fillet products are sold Bill Atkinson’s News Report (BANR) market reports suggest into markets not described within the model, the 1998 and 2007 that average prices differ by 20% to 33% in favor of at-sea simulations assumed the entire roe, fillet, and surimi production product (BANR is a commercial publication that reports fish- was allocated to the five markets described above. ALASKA POLLOCK MARKETS 1089

The simulations are structured around first wholesale rev- TABLE 5. Goodness-of-fit statistics from equations used in the market model enues, rather than ex-vessel prices, because ex-vessel prices of eastern Bering Sea Alaska pollock fisheries: CV is the coefficient of variation · are not defined for catcher–processors and because ex-vessel (100 SD/mean), r is the correlation between the estimated and predicted values of the dependent variables, and U1 is the Theil inequality coefficient. Equation transaction data on the mothership sector is confidential due to numbers refer to those presented in the text. the small number of firms. In addition, the available ex-vessel data fail to account for roe bonuses, end-of-year profit sharing Equation Variable CV rU1 bonuses, vertical integration between processors and the owned 1 U.S. fillet allocation 13.0% 0.98 0.10 or leased catcher vessels that deliver to them, and lease and 2 European fillet allocation 47.0% 0.90 0.31 profit sharing arrangements that have emerged under the AFA 3 Japanese surimi allocation 32.3% 0.90 0.24 cooperative and CDQ structures. 4 U.S. surimi allocation 34.6% 0.94 0.24 5 U.S. fillet demand 3.7% 0.94 0.03 RESULTS 6 European Union fillet demand 8.0% 0.88 0.07 7 Japanese roe demand 23.2% 0.87 0.20 Allocation Equations 8 Japanese surimi demand 11.2% 0.90 0.10 Coefficient estimates with corresponding P-values and sum- 9 U.S. surimi demand 26.2% 0.75 0.22 mary statistics for individual equations (equations 1–9) in the 10 Total revenue 14.2% 0.97 0.12 market model are reported in Table 4. The estimated allocation of U.S.-origin Alaska pollock fillets to the United States (equation 1) accounts for nearly all (R2 = 0.96) of the observed variation in the quantity of fillets allocated to the U.S. market. (R2 = 0.81) of the observed variation in exports of U.S.-origin In addition, the estimated coefficients on all price and quantity Alaska pollock fillets to Europe. The allocation of U.S.-origin variables included in equation (1) were statistically significant Alaska pollock fillets to Europe is more sensitive to variations in at a 1% level (see Tables 4, 5). The results indicate that the allo- prices in Europe and the United States than is the allocation of cation of U.S.-origin Alaska pollock fillets to the United States U.S.-origin Alaska pollock fillets to the USA (equation 1). This is inelastic to variations in prices in the USA and Europe: a 1% is likely due to higher competition from Russian-origin Alaska increase in the U.S. price of U.S.-origin Alaska pollock fillets pollock fillets in the European market. Thus, a 1% increase in leads to a 0.48% increase in the allocation of U.S.-origin Alaska the European price of U.S.-origin Alaska pollock fillets leads to pollock fillets to the USA, while a 1% increase in the European a 1.32% increase in the allocation of U.S.-origin Alaska pollock price of U.S.-origin Alaska pollock fillets leads to a 0.31% de- fillets to Europe, while a 1% increase in the U.S. price of U.S.- crease in the allocation of U.S.-origin Alaska pollock fillets to origin Alaska pollock leads to 1.94% decrease in the allocation the USA. That is, although the allocation of U.S.-origin Alaska of U.S.-origin Alaska pollock fillets to Europe. European de- pollock fillets to the United States is affected by fluctuations mand for U.S.-origin Alaska pollock fillets has increased over in U.S. and European market prices, some U.S. buyers have the time, fueled, in part, by appreciation of the euro against the long-term contracts that inure them to short-term fluctuations in U.S. dollar. In addition, because the European market for U.S.- price. In addition, in interviews U.S. processors indicated that a origin Alaska pollock fillets is more recent than the U.S. market majority of their Alaska pollock fillet output is presold and thus for U.S.-origin Alaska pollock fillets, European buyers are less unresponsive to short-term price fluctuations. Changes in the dependent on supplies. As determined in equation (1), European production of Alaska pollock fillets by the at-sea and inshore response to the production decisions of at-sea and inshore sec-

Downloaded by [Department Of Fisheries] at 19:14 20 July 2015 sectors have different effects on the allocation of U.S.-origin tors differed. A 1% increase in the production of Alaska pollock Alaska pollock fillets to the United States. A 1% increase in the fillets in the inshore sector leads to a 0.34% increase in the quan- production of Alaska pollock fillets in the at-sea sector leads to tity of U.S.-origin Alaska pollock fillets allocated to European a 0.28% increase in the quantity of U.S.-origin Alaska pollock markets, while a similar production increase in the at-sea sector fillets allocated to U.S. markets, whereas a similar increase in leads to a 0.64% increase in the quantity of U.S.-origin Alaska inshore production only leads to a 0.16% increase in the amount pollock fillets allocated to European markets. of fillets allocated to the USA, results that are likely due to The allocation of U.S.-origin Alaska pollock surimi to Japan higher fillet recovery rates in the at-sea sector that stem from (equation 3) depended on Japanese import prices for U.S.- fresher product. Moreover, the production responses for both origin Alaska pollock surimi, U.S. output of Alaska pollock sectors would be higher if lagged allocations had not also been surimi, U.S. prices of U.S.-origin Alaska pollock fillets, and included in the model. seasonal dummy variables. This equation accounted for a large The allocation of U.S.-origin Alaska pollock fillets to Europe portion (R2 = 0.82) of the observed variation in allocations (equation 2) is explained by European and U.S. prices, U.S. pro- of U.S.-origin Alaska pollock surimi to Japan. Although there duction, and seasonal dummy variables. Coefficient estimates, was a high P-value (0.70) for Japanese surimi prices, various P-values, and model performance statistics are reported in Ta- BANR reports indicate that industry members focus on U.S. ble 4 and Table 5. The included variables accounted for most prices for U.S.-origin Alaska pollock fillets and TAC when 1090 STRONG AND CRIDDLE

describing the production of higher-grade surimi for Japan. The cient for at-sea production was not statistically significant in the model estimates suggest that the amount of U.S.-origin Alaska equation, the coefficient for inshore production was statistically pollock surimi exported to Japan is responsive to changes in significant at a 1% level. The inshore sector, with most catches U.S. prices of U.S.-origin Alaska pollock fillets; a 1% increase being processed 24–72 h postharvest, produces a lower overall in the U.S. price of U.S.-origin Alaska pollock fillets led to grade of surimi and consequently supplies more surimi to the a 1.92% decrease in the allocation of U.S.-origin Alaska pol- U.S. market than does the at-sea sector. lock surimi to Japan. This is because the high-quality surimi desired in the Japanese market requires high-quality Alaska Demand Equations pollock flesh that might otherwise be used in fillet production. The U.S. and European inverse demand equations for U.S.- Changes in inshore and at-sea production of U.S.-origin Alaska origin Alaska pollock fillets (equations 5 and 6) performed well pollock surimi had similar effects on the allocation of U.S.- (R2 = 0.89 for equation 5 and R2 = 0.77 for equation 6). The origin Alaska pollock surimi to Japan. A 1% increase in pro- observed and estimated prices are depicted in Figure 1a, b. The duction by the inshore sector resulted in a 0.29% increase in the influence of twice-frozen Alaska pollock fillets, lagged prices allocation of U.S.-origin Alaska pollock surimi to Japan, while for U.S.-origin Alaska pollock fillets, U.S.-origin Alaska pol- a similar increase in production by the at-sea sector resulted in lock fillet allocations, and U.S. per capita income were all sta- a 0.26% increase in the allocation of U.S.-origin Alaska pollock tistically significant at a 1% level. Changes in lagged prices of surimi to Japan. While this may be surprising, given that most U.S.-origin Alaska pollock fillets had the greatest effect on cur- of the highest-grade surimi is produced in the at-sea sector, it rent prices of U.S.-origin Alaska pollock fillets; a 1% change in should be recognized that three of the four major shore-based the lagged price led to a 0.83% change in the current price. The processors, Alyeska, Unisea, and Westward, are subsidiaries of European allocation of U.S.-origin Alaska pollock fillets, and Japanese seafood conglomerates. Consequently the inshore sec- the euro/U.S. dollar exchange rate were statistically significant tor can be expected to be particularly influential in determining at a 1% level, the European import prices for Alaska pollock Japanese imports. The price of U.S. surimi in the United States fillets from Russia and China was statistically significant at a was omitted from this equation because the U.S. demand for 10% level. Despite its low P-value, the European import price surimi was not influential in determining the allocation of U.S.- for Alaska pollock fillets from Russia and China was retained origin Alaska pollock surimi to Japan. It is likely that this is be- in the model because it increased performance of the overall cause the U.S. market generally receives recovery-grade surimi, equation system. German real disposable income was not statis- while the Japanese market sources recovery-grade surimi from tically significant. The European import price for Alaska pollock a variety of other fisheries. Furthermore, the U.S. market ap- fillets from Russia and China was particularly influential on Eu- pears to act as a secondary market that is able to absorb surimi ropean prices for U.S.-origin Alaska pollock; a 1% change in that would otherwise depress Japanese surimi prices. Consis- the import price engendered a 1.44% change in the price of tent with this observation about the role of the U.S. market for U.S.-origin Alaska pollock fillet exports to Europe. With the surimi, tests of the model with and without the U.S. surimi European market turning to U.S.-origin Alaska pollock fillets in as an explanatory variable suggested that U.S. surimi prices the early 2000s in response to reduced Russian output, the mar- were not influential in determining the allocation of surimi to ket has become increasingly dependent on the U.S. supply. The Japan. high elasticity of U.S. prices to the price of Russian and Chinese Equation (4), the allocation of U.S.-origin Alaska pollock imports suggests, however, that a large rise in Russian catches surimi to the USA, also performed well (R2 = 0.88), particu- of Alaska pollock and the resulting price decrease would have a

Downloaded by [Department Of Fisheries] at 19:14 20 July 2015 larly considering that the data were constructed from time se- significant adverse impact on U.S. producers of Alaska pollock ries of production and exports adjusted to reflect expert opinion fillets. about temporal relationships between production and distribu- The Japanese import price of U.S.-origin Alaska pollock roe tion from cold-storage holdings. As with equation (3), the price (equation 7) also performed well (R2 = 0.76). The observed of U.S. surimi in the United States was initially included in the and estimated prices are depicted in Figure 1c. Japanese cold- model but was omitted from the final model because the esti- storage holdings and Japanese imports of Alaska pollock roe mated coefficient was not statistically significant, and tests of from other countries were both statistically significant at a 1% the model with and without this explanatory variable indicated level (P-value < 0.01). Although interviews with industry mem- that the model performed better when this variable was omitted bers discounted the credibility of Japanese cold-storage hold- from equation (4). It is likely that this is because the U.S. market ings data, the results suggest that the monthly reported values generally receives recovery-grade surimi, and similar grades of accurately reflect factors that influence purchasers’ perceptions. surimi are available from many other sources. Of highest im- The model found that a 1% change in cold-storage holdings portance was the lagged U.S. allocation of U.S.-origin Alaska leads to a 1.2% change in roe prices, which accords with indus- pollock surimi to the USA, which led to a change of 0.79% try reports (e.g., BANR) that suggest that cold-storage holdings in the U.S. allocation of U.S.-origin Alaska pollock surimi for are an important factor in the determination of roe auction prices. every 1% change in the lagged quantity. Although the coeffi- On the other hand, Japanese roe consumption statistics, which ALASKA POLLOCK MARKETS 1091 Downloaded by [Department Of Fisheries] at 19:14 20 July 2015

FIGURE 1. Performance of the five estimated inverse demand equations: (a) price of U.S.-origin Alaska pollock fillets in the USA; (b) price of U.S.-origin Alaska pollock fillets in Europe; (c) price of U.S.-origin Alaska pollock roe in Japan; (d) price of U.S.-origin Alaska pollock surimi in Japan; and (e) price of Alaska pollock surimi in the USA. Time-series observations are represented as points, estimates are represented as thick solid lines, upper and lower 95% confidence limits are represented as thin solid lines, and upper and lower 95% prediction limits are represented as dotted lines.

were considered important by industry members in determin- roe continue to be consumed, the market clearing price has ing the direction of the market, were not statistically significant declined. and did not provide theoretically consistent results. Changes in The performance of equation (8), the Japanese import price of Japanese cold-storage holdings appear to reflect other under- U.S.-origin Alaska pollock surimi, is depicted in Figure 1d. The lying market influences not specifically modeled in the equa- coefficient of determination for this equation was 0.80. Statisti- tion. For example, demand for Alaska pollock roe appears to cally significant variables included the allocation of U.S.-origin be declining and more is being sold at grocery stores than at Alaska pollock surimi to Japan, the Japanese price of surimi specialty shops. Although similar quantities of Alaska pollock from other sources, lag-three Japanese cold-storage holdings, 1092 STRONG AND CRIDDLE

and the lag-two Japanese import price of U.S.-origin Alaska price differential was applied to the roe prices used for the at-sea pollock surimi. The Japanese price for U.S.-origin Alaska sector. Model results are included in Table 4. pollock surimi involves negotiations that often extend over several months between the U.S. Surimi Commission and Japanese Goodness-of-Fit Statistics buyers. Prices are often agreed to before fishing begins, thus the The goodness-of-fit statistics (Table 5) indicate that the importance of including a 2-month lag for Japanese import price model closely fits the historic observations. The lowest cor- and a 3-month lag on Japanese cold-storage holdings. As was the relation between actual and estimated values of the dependent case for the Japanese roe demand equation, Japanese consump- variables was 0.75 (U.S. surimi prices), while six of the nine tion of surimi was dropped from this equation because it was correlation coefficients were 0.90 or greater. In addition, the cor- not statistically significant and exhibited theoretically implausi- relation between actual and estimated total revenue was 0.97. ble behavior. Because Japanese cold-storage holdings of Alaska The CV was somewhat more variable between equations, par- pollock surimi are moderately collinear with Japanese surimi ticularly in the allocation equations, varying from 3.7% in the consumption, omission of the consumption statistics could con- U.S. fillet price equation to 47.0% in the European fillet im- found estimates of the influence of cold-storage holdings but port equation. The Theil U1-statistics indicate that the model should not adversely affect estimates of Japanese demand for provides a reliable fit to the historical data, with statistics for U.S.-origin Alaska pollock surimi. An increase of 1% in the individual equations ranging from a low of 0.03 in the U.S. fil- Japanese price of non-U.S. surimi imports is estimated to stim- let price equation to a high of 0.31 in the Europe fillet import ulate a 0.69% increase in the price of Japanese imports of equation. Although the goodness-of-fit statistics show reduced U.S.-origin Alaska pollock surimi, underscoring Japanese con- reliability in equations (2), (3), and (4), the overall results were sumers’ decreased willingness to pay for high-quality and high- favorable. Overall U.S. revenue from the five product markets priced surimi. The Japanese index of real disposable income, was particularly successful, with a CV of 14.2% and Theil’s while not statistically significant (P-value = 0.50), is consistent U1 of 0.12, indicating that errors in the individual equations with the behavior of an inferior product. Surimi appears to be a did not harm the overall predictive capability of the system of substitute for higher-priced seafoods and, according to industry equations. members, surimi demand increases during weak economic periods. This was the case in the latter half of 2008 at the onset of the global market crisis, when Japanese surimi prices increased. DISCUSSION The U.S. price of U.S.-origin Alaska pollock surimi (equa- In 2007, 1.332 million metric tons of Alaska pollock was har- tion 9) did not perform as well as the other eight equations (R2 vested from the eastern Bering Sea. This catch had a first whole- = 0.56); see Figure 1e. This was likely due, in part, to the lack sale value of $1.170 billion and an average value per kilogram of direct data on domestic sales and inventories and because the of $0.91 based on the product forms, prices, and markets mod- U.S. surimi market is the least important of the five markets eled. Model-based estimates of $1.189 billion in first wholesale modeled. Nevertheless, inclusion of this equation led to signifi- value and an average value of $0.93/kg do not differ significantly cant improvements in the performance of other equations in the or appreciably from the reported values. Moreover, model esti- equation system and of the equation system as a whole. Similar mates exclude catches and production associated with the moth- to the role of income in the Japanese surimi demand equation, ership Ocean Phoenix, which produces headed-and-gutted fish there is a negative coefficient associated with U.S. per capita rather than surimi or fillets; inclusion of the Ocean Phoenix data disposable income, consistent with the expected behavior of an would have reduced model estimates of price and first whole-

Downloaded by [Department Of Fisheries] at 19:14 20 July 2015 inferior product, except that in this case, the estimated coeffi- sale value. Based on the modeled relationships, revenue from cient is statistically significant. As described in equation (4), the roe, fillets, and surimi would be maximized at a harvest level U.S. market for surimi fluctuates widely, and surimi generally of 4.987 million metric tons. The U.S Alaska pollock stock has sold as a substitute for higher-priced seafood products such as never been and seems unlikely to ever be large enough to sup- crab, lobster, and scallops. Consequently a 1% decrease in the port a sustainable harvest larger than 1.5 million metric tons; U.S. per capita disposable income is associated with a 2.49% thus, it seems likely that the fishery will operate in an inelas- increase in the price of U.S.-origin Alaska pollock surimi. The tic portion of the combined demand function for surimi, fillets, price of halibut served as a proxy for the price of other seafoods and roe. This outcome contrasts with results in Herrmann et al. that act as substitutes for Alaska pollock surimi in the U.S. mar- (1996) that reported that surimi revenue in the pre-AFA fishery ket; a 1% increase in this price led to a 0.39% increase in the was maximized at a harvest level of 1.7 million metric tons. This price of surimi, reemphasizing that U.S. consumers treat surimi difference reflects the importance of fillet and roe as components as an inferior substitute for high-value seafoods. of the demand for Alaska pollock and changes in market struc- Equation (11) performed reasonably well (R2 = 0.78). All ture that have developed in the aftermath of the AFA. Instead independent variables were statistically significant at a 1% level. of maximizing revenue by maximizing catch, industry looks to The model suggests that the expected additional value derived produce the highest return per kilogram harvested. As reported from harvesting Alaska pollock at sea is 23.5%; this estimated in Fell (2008), processors are now able to shift production into ALASKA POLLOCK MARKETS 1093

With the fishery operating in the inelastic portion of the demand curve, it makes sense to maximize catch up to the limits of sustainability. However, the inshore sector appears to be vulnerable to reductions in first wholesale prices and increases in fuel costs, especially during the B season, where under conditions such as were present in 2007, first wholesale value declines from $0.814/kg to $0.785/kg. Because revenues to U.S. fishers are responsive to changes in Russian harvests of Alaska pollock, Russia’s upward-trending catches and the recent Marine Stew- ardship Council certification of the Sea of Okhotsk fishery for Alaska pollock could depress prices of U.S.-origin Alaska pollock products. Such a drop in value could again lead the inshore sector to underharvest a portion of their B-season catch share or to incur short-term loss in order to satisfy contracted deliveries. FIGURE 2. Simulated 2007 wholesale revenue curves over a range of harvest levels for Alaska pollock in the eastern Bering Sea. The inshore sector includes The ability to switch between product forms is a result of catches that are delivered to inshore processors. The at-sea sector includes increased investment and revenues, largely attributable to the catcher–processors (vessels that both catch and process at sea) and mothership switch from the race-for-fish structure to that of AFA cooper- operations (catches that are delivered to at-sea processors). ative management. The fishery has rapidly expanded into new markets, optimized first wholesale revenues across a mix of different product forms in response to changes in market de- roe, fillets, and surimi, and maximized recovery rates among mand. This ability to diversify over products and markets helps this suite of product forms. To quantity the financial level of explain why first wholesale value has become less vulnerable the AFA’s impact, simulations were conducted comparing 1998 to year-to-year changes in the TAC—as the TAC changes, pro- and 2007 fishing seasons given a TAC of 1.394 million met- ducers respond by reallocating production to take advantage of ric tons (the 2007 TAC) and assuming that the entire TAC is differences in the elasticity of demand across product forms and harvested and sold under 2007 market conditions and denomi- markets. nated in 2007 base year value, but with 1998 (or 2007) recovery Revenue also varies widely by sector. For example, Criddle rates and production mix. The simulation results estimate that and Strong (2013) report that at a TAC of 1.394 million metric under these conditions, fishery revenues would have increased tons (the 2007 TAC), under conditions such that all the TAC is from $0.795 billion in 1998 to $1.289 billion in 2007. Of this, harvested, inshore wholesale revenue is estimated to be $551 $376 million is attributable to increased recovery rates and $63 million and the at-sea sector wholesale revenue is estimated to million is attributable to the production of more valuable prod- be $738 million (Figure 2). Under 2007 market conditions, the uct forms; an additional $55 million arises from an interaction at-sea sector held a 15% advantage in value per kilogram rel- between increased recovery rates and the production of more ative to the inshore sector; the inshore sector was estimated to valuable products. The largest source of increased value was derive $0.91/kg, while the at-sea sector was estimated to derive due to increased fillet production. The simulation suggests that, $1.04/kg (Criddle and Strong 2013). Under those conditions, under the assumed conditions, the value of fillets would increase it is estimated—based on their ownership in harvesting ves- from $229 million in 1998 to $649 million in 2007. In fact, fil- sels and royalty payments received from leases of their Alaska let production increased substantially after passage of the AFA

Downloaded by [Department Of Fisheries] at 19:14 20 July 2015 pollock quota—that CDQ entities received over $311.8 million when processors ceased to be driven to maximize throughput (over 24% of the total first wholesale revenue generated in this and could instead invest in fillet equipment to meet increased fishery). Since 2000, virtually the entire CDQ quota has been fillet demand (Felthoven 2002; Morrison Paul et al. 2009). The leased to the at-sea sector as that sector is willing to offer higher simulation also suggests that, under the assumed conditions, lease prices and other benefits to their CDQ partners because roe production would double and revenues would increase by at-sea operation offers increased value per metric ton harvested. $261 million due to improved roe recovery rates. Surprisingly, Despite a lack of direct information about the cost structure of even with the increased emphasis on fillet product, investment the inshore and at-sea sectors, Criddle and Strong (2013) sug- in technology allowed processors to produce recovery-grade gest that the 2007 season provides insights into absolute costs surimi from trimmings and waste water and so total surimi pro- in the inshore sector and relative costs in the at-sea sector. In duction only dropped by 13%. The elevated revenues and overall 2007, in face of unanticipated high fuel costs and low CPUE in stability made possible by product diversification have allowed the late B season, the inshore sector harvested only 90% of its companies to weather poor economic conditions, including the B-season quota. In that same year, the at-sea sector harvested conditions in the 2007 B season that led to a 10% underharvest its entire B-season quota as well as the entire CDQ quota. This of the inshore sector quota. is particularly notable because the at-sea sector paid an average With all the turmoil this fishery has endured over its short 35- royalty of $340/metric ton to their CDQ partners. year history, the expected conditions seen in the future appear 1094 STRONG AND CRIDDLE

relatively benign. The fishery has endured far greater challenges Criddle, K. R., and S. Macinko. 2000. A requiem for the IFQ in US fisheries? than short-term potential losses to a single sector. The current Marine Policy 24:461–469. management, governance, and market structures have played Criddle, K. R., and J. Strong. 2013. Dysfunction by design: the effects of limitations on transferability of catch shares in the Alaska Pollock fishery. vital roles in the success of the fishery. The AFA regulatory Marine Policy 40:91–99. regime has increased revenues by an estimated 62% and created Fell, H. 2008. Rights-based management and Alaska Pollock processors’ supply. long-term value to a fishery characterized in the 1990s by its American Journal of Agricultural Economics 90:579–592. rapid succession of bankruptcies and a race for fish. It remains Felthoven, R. 2002. Effects of the American Fisheries Act on capacity, utiliza- to be seen, however, with rising fuel prices and changes in the tion, and technical efficiency. Marine Resource Economics 17:181–205. Ginter, J. J. C. 1995. The Alaska community development quota fisheries man- timing and location of fishing that will result from regulatory agement program. Ocean and Coastal Management 28:147–163. requirements to reduce salmon bycatch, if the inshore sector’s Henningsen, A., and J. Hamann. 2007. Systemfit: a package for estimating B-season quota will remain profitable. The AFA, despite all its systems of simultaneous equations in R. Journal of Statistical Software 23:1– positive impacts, does not provide the flexibility to adapt to these 40. potential changes. With the possibility of increased shares of the Herrmann, M., K. R. Criddle, E. M. Feller, and J. A. Greenberg. 1996. Estimated economic impacts of potential policy changes affecting the total allowable TAC being left unharvested, amending the AFA to allow leasing catch for Walleye Pollock. North American Journal of Fisheries Management catch shares between sectors could be to the advantage of both 16:770–782. sectors and to the CDQ entities and the western Alaskans they Hiatt, T., M. Dalton, R. Felthoven, B. Fissel, B. Garber-Yonts, A. Haynie, represent. S. Kasperski, D. Lew, C. Package, J. Sepez, and C. Seung. 2010. Stock Model results will aid fishery managers in analyzing the po- assessment and fishery evaluation report for the groundfish fisheries of the Gulf of Alaska and Bering Sea/Aleutian Islands Area: economic sta- tential impact of future policies and understanding the external tus of the groundfish fisheries off Alaska, 2009. National Oceanic and economic pressures on the fishery and its various sectors. Fur- Atmospheric Administration–Fisheries, Alaska Fishery Science Center, thermore, model results can be used to assist fishery managers Seattle. in taking an ecosystem-based management approach that seeks Hiatt, T., and Terry, J. 2000. Stock assessment and fishery evaluation report for to balance conservation and socioeconomic objectives as re- the groundfish fisheries of the Gulf of Alaska and Bering Sea/Aleutian Islands Area: economic status of the groundfish fisheries off Alaska, 1999. National quired under the Magnuson–Stevens Fishery Conservation and Oceanic and Atmospheric Administration–Fisheries, Alaska Fishery Science Management Act of 2007. Center, Seattle. Matulich, S. C., M. Sever, and F. Inaba. 2001. Fisheries cooperatives as alterna- ACKNOWLEDGMENTS tives to ITQs: implications of the American Fisheries Act. Marine Resource Economics 16:1–16. This paper is the result of research supported by the National Morrison Paul, C. J., M. O. Torres, and R. G. Felthoven. 2009. Fishing revenue, Marine Fisheries Service Fisheries Development and Research productivity, and product choice in the Alaskan Pollock fishery. Environmen- Program, under award NA08NMF4270420, the Pollock Conser- tal and Resource Economics 44:457–474. vation Cooperative Research Center, the North Pacific Research NMFS (National Marine Fisheries Service). 1999. Revised final reasonable and Board (publication number 494), and the University of Alaska prudent alternatives for the Pollock fisheries in the Bering Sea and Aleutian Islands and Gulf of Alaska with supporting documentation. NMFS, Juneau, Fairbanks with funds appropriated by the state. All opinions are AK. those of the authors. Park, J. W. 2005. Surimi and surimi seafood. CRC Press, Boca Raton, Florida. Strong, J., and K. R. Criddle. 2013. Fishing for Pollock in a sea of change: a historical analysis of the Bering Sea Pollock fishery. University of Alaska REFERENCES Sea Grant Press, Fairbanks. American Seafoods Group. 2006. Form S-4 registration statement under the Theil, H. 1966. Applied economic forecasting. North-Holland Publishing, Am- Securities Act of 1933. Securities and Exchange Commission, Washington, sterdam. Downloaded by [Department Of Fisheries] at 19:14 20 July 2015 D.C. Wilen, J. E., and E. J. Richardson. 2008. Rent generation in the Alaskan Pollock Anderson, L. G. 2002. A microeconomic analysis of the formation and conservation cooperative. FAO Fisheries Technical Paper 504:361–368. potential reorganization of AFA coops. Marine Resource Economics 17: Zellner, A., and H. Theil. 1962. Three-stage least squares: simultaneous esti- 207–224. mation of simultaneous equations. Econometrica 30:54–78. This article was downloaded by: [Department Of Fisheries] On: 20 July 2015, At: 19:14 Publisher: Taylor & Francis Informa Ltd Registered in England and Wales Registered Number: 1072954 Registered office: 5 Howick Place, London, SW1P 1WG

North American Journal of Fisheries Management Publication details, including instructions for authors and subscription information: http://www.tandfonline.com/loi/ujfm20 Reducing Bias and Filling in Spatial Gaps in Fishery- Dependent Catch-per-Unit-Effort Data by Geostatistical Prediction, I. Methodology and Simulation John F. Waltera, John M. Hoenigb & Mary C. Christmanc a National Oceanic and Atmospheric Administration–Fisheries, Southeast Fisheries Science Center, 75 Virginia Beach Drive, Miami, Florida 33149, USA b Virginia Institute of Marine Science, College of William and Mary, Post Office Box 1364, Gloucester Point, Virginia 23062, USA c MCC Statistical Consulting LLC, 2219 Northwest 23rd Terrace, Gainesville, Florida 32605, USA Click for updates Published online: 31 Oct 2014.

To cite this article: John F. Walter, John M. Hoenig & Mary C. Christman (2014) Reducing Bias and Filling in Spatial Gaps in Fishery-Dependent Catch-per-Unit-Effort Data by Geostatistical Prediction, I. Methodology and Simulation, North American Journal of Fisheries Management, 34:6, 1095-1107, DOI: 10.1080/02755947.2014.932865 To link to this article: http://dx.doi.org/10.1080/02755947.2014.932865

PLEASE SCROLL DOWN FOR ARTICLE

Taylor & Francis makes every effort to ensure the accuracy of all the information (the “Content”) contained in the publications on our platform. However, Taylor & Francis, our agents, and our licensors make no representations or warranties whatsoever as to the accuracy, completeness, or suitability for any purpose of the Content. Any opinions and views expressed in this publication are the opinions and views of the authors, and are not the views of or endorsed by Taylor & Francis. The accuracy of the Content should not be relied upon and should be independently verified with primary sources of information. Taylor and Francis shall not be liable for any losses, actions, claims, proceedings, demands, costs, expenses, damages, and other liabilities whatsoever or howsoever caused arising directly or indirectly in connection with, in relation to or arising out of the use of the Content. This article may be used for research, teaching, and private study purposes. Any substantial or systematic reproduction, redistribution, reselling, loan, sub-licensing, systematic supply, or distribution in any form to anyone is expressly forbidden. Terms & Conditions of access and use can be found at http:// www.tandfonline.com/page/terms-and-conditions North American Journal of Fisheries Management 34:1095–1107, 2014 C American Fisheries Society 2014 ISSN: 0275-5947 print / 1548-8675 online DOI: 10.1080/02755947.2014.932865

ARTICLE

Reducing Bias and Filling in Spatial Gaps in Fishery-Dependent Catch-per-Unit-Effort Data by Geostatistical Prediction, I. Methodology and Simulation

John F. Walter* National Oceanic and Atmospheric Administration–Fisheries, Southeast Fisheries Science Center, 75 Virginia Beach Drive, Miami, Florida 33149, USA John M. Hoenig Virginia Institute of Marine Science, College of William and Mary, Post Ofﬁce Box 1364, Gloucester Point, Virginia 23062, USA Mary C. Christman MCC Statistical Consulting LLC, 2219 Northwest 23rd Terrace, Gainesville, Florida 32605, USA

Abstract Geostatistical prediction can address two difficult issues in interpreting fishery-dependent catch per unit effort (CPUE): the lack of a sampling design and the need to fill spatial gaps. In this paper we demonstrate the spatial weighting properties of geostatistics for treating data collected without a sampling design or with a selection bias, two basic traits of fishery-dependent data. We then examine the bias and precision of geostatistical prediction of CPUE based on fishery-dependent data through simulation. We create data sets with known variograms, sample them with a preference for sites with high abundance, and then estimate variograms and CPUE as the geostatistical mean relative abundance. The variograms obtained from the simulated fishery samples correctly estimated the range but underestimated the sill, and the geostatistical mean substantially improved the estimation of CPUE over the arithmetic mean. Though the geostatistical mean still overestimated the true value, the error was primarily due to prediction into unsampled locations, where predictions revert toward the arithmetic mean. The geostatistical variance at a point, which is a function of spatial autocorrelation and the location of adjacent samples, provides a measure of uncertainty. This variance measures the degree to which predictions are derived from nearby data versus distant observations, which translates the spatial extent of extrapolation into probabilistic terms. In conjunction with conventional standardization methods that account for factors affecting catchability, geostatistical prediction Downloaded by [Department Of Fisheries] at 19:14 20 July 2015 provides an additional tool that reduces but does not eliminate biases inherent in fishery-dependent data and supports the need to predict CPUE in unsampled areas.

Indices of catch and bycatch per unit effort (CPUE) de- However, the utility of indices derives exclusively from the rived from ﬁshery-dependent data represent essential mea- degree to which CPUE is proportional to abundance. Since sures of relative abundance for stock assessments. Fishery- ﬁshery data are collected with the goal of economic re- dependent information drives assessment results and forms turn rather than unbiasedly indexing stock status, the data the basis for stock status conclusions for highly migratory create considerable problems in interpreting the relation- species, species lacking surveys. and almost without excep- ship between CPUE and abundance (Paloheimo and Dickie tion, bycatch and discard estimates (Maunder et al. 2006). 1964).

*Corresponding author: [email protected] Received July 26, 2013; accepted June 2, 2014 1095 1096 WALTER ET AL.

Fishery data are collected with fishing gears that change over can be plotted and predictions restricted to areas of desired pre- time, by fishers who alter their behavior in response to condi- cision, thus providing a probabilistic framework for evaluating tions, and generally in a manner inconsistent with design-based the spatial extent of adequate prediction. Depending upon the random sampling. In contrast, scientific surveys generally em- spatial extent of the samples this may extend to the population ploy statistical designs that ensure unbiased estimates of mean spatial coverage, but more likely it is smaller, particularly with and variance (Thompson 2002). fishery dependent data. Third, the spatial weighting properties of To account for the absence of scientific control over how sam- the kriging predictor can reduce the bias that targeted, clustered, ples are collected in fishery data, researchers have employed and nonrandom data selection introduces. many techniques to standardize the data in the attempt to re- Geostatistical methods have found widespread use in fish- cover the true relative abundance signal in the data (Campbell eries applications, particularly as alternatives to design-based 2004). These methods usually are some form of generalized lin- estimators for survey data (Warren 1998; Rivoirard et al. 2000; ear (GLM) or mixed-effect models (GLMM) and other variants Petitgas 2001). This role is complementary to current GLM- (Maunder and Punt 2004). These models can account for differ- based standardization methods (Campbell 2004) in that factors ences in vessels, gear, season, fishing method, and other factors that affect catchability may be accounted for in the geostatistical which affect catchability and thus the interpretation of CPUE. analysis and prediction, such as when using universal kriging. However, most statistical standardization methods further Geostatistics involves a two-step process in which a model of assume independent and identically distributed observations at spatial autocorrelation (known as the variogram) is first esti- some level, though this can sometimes be relaxed using mixed- mated and then used to predict abundance over the area of effect models. Furthermore, they assume that the observed sam- interest (Matheron 1963; Cressie 1993). ples provide unbiased inference for the unobserved or unsam- In this paper we employ simulated fisheries data to demon- pled locations. These methods also provide little capacity to strate the valuable role that geostatistics can play in at least par- fill in gaps in sample observations short of assuming that the tial improvement of fishery-dependent CPUE indices. Specifi- mean of the sampled locations represents the unsampled areas cally, we consider situations in which observations are numerous (Walters 2003). and clumped (where abundance is high) and ones in which ob- The lack of a sampling design, coupled with the selection bias servations are sparse (where abundance is low). Next we test, of fishery-dependent data, motivates alternative approaches to through simulation of fisheries data, the performance of the estimation of CPUE. Our approach is based on model-based kriging estimator as a measure of single-year CPUE versus a inference methods (Sarndal¨ et al. 1992; Gregoire 1998) because standard linear model. the sample from which we desire a CPUE index is not a true probabilistic sample. Using design-based methods with nonrandom samples provides a possibly biased estimate of mean CPUE METHODS with a certainly biased estimate of its standard error, where the For model-based inference, we start with a model bias is due to improper designation of a sampling design. The of the population from which we observed a sam- { , ,..., } model-based approach may still be biased, but we believe that ple, s (x1) s (x2) s (x N ) at, say, N locations { , ,..., } the bias can be reduced under certain conditions with designa- x1 x2 x N . Using a geostatistical approach, we as- { ∈ } tion of an appropriate model and possibly additional sampling. sume that the model is S (x) ; x D , where S is a random One class of model-based inference is geostatistical predic- field or process that is stationary and isotropic and x is a tion, which exploits the property of spatial autocorrelation to location in the spatial domain of interest, D. The assumptions

Downloaded by [Department Of Fisheries] at 19:14 20 July 2015 obtain a model of the process and uses this autocorrelation for of stationarity and isotropy imply that interpolation. Geostatistical predictive methods make three par- (1) E [S (x)] = µ, that is, the mean of the random field is ticularly useful contributions to obtaining more accurate relative constant and does not depend on location; and abundance information from fishery-dependent data. First, these (2) Cov [S (x1) , S (x2)] = C( x1 − x2 ), that is, the covari- methods provide a statistically based means of interpolation with ance between values of the random field at any two locations the spatial extent of interpolation determined by the strength of in D depends only on the covariance function C(·) and the autocorrelation. This spatial gap filling is necessary when we Euclidean distance between the locations but not the direc- desire maps and local predictions of CPUE. It is also critical tion. C (0) is defined as the variance of the random field. when there are spatial changes in the fishery such that CPUE trends in the fished areas may not be representative of abundance The population mean is defined as the integral in unfished areas (Walters 2003; Carruthers et al. 2011). Second, geostatistical prediction provides a measure of prediction error, S (x) dx µ = x∈D , (1) which is critical for determining the quality of a prediction. The |D| geostatistical prediction variance is location dependent, that is, it is a function of the amount of adjacent spatial information used where |D| is the area (or volume) of the spatial region for which in a prediction (Warren 1998). Contours of the kriging variance the mean is desired. In our approach, we shall estimate the mean GEOSTATISTICAL PREDICTION I 1097

by ordinary kriging and approximate the integral using a grid of |N(h)| = the total number of sample pairs for a distance prediction locations that span the domain, i.e., we approximate interval h. µ using ThesummationisoverN(h) ≡ (i, j):dist(xi − x j ) = h , ⎛ ⎞ where dist is the Euclidean distance between the two points. N ∗ Multiple distance classes (h) are defined and the semivariance ∗ −1 ⎝ ⎠ µˆ = (N ) Sˆ (x0 j) , (2) for all pairs within each distance class is calculated to obtain an j=1 empirical variogram. This is then used to estimate the parameters of a variogram functional model. ∗ where N is the number of grid nodes for which estimates are Geostatistical prediction or kriging at a prediction point x0 ∗ desired, {x0 j, j = 1, 2,...,N } are the grid node locations at starts by solving a system of N linear equations subject to several which predictions are obtained from kriging, and Sˆ (x0 j)isthe constraints, where N is the number of observations. In the pro- predicted value for S(·) at prediction location x0 j . This estimator cess of kriging, the matrix A, defined below, is inverted and the is a refinement over using the sample mean to estimate µ because optimal weights used in the prediction at point x0 are obtained it uses the predicted values Sˆ (x0 j ), which are functions of nearby by the equation observations rather than of the global sample mean. Model-based geostatistics.—Geostatistics involves first ob- Aλ0 = b0, (5) taining a model, the variogram, of spatial autocorrelation of a process and then using this model to predict abundance and in- where terpolate to create maps (Matheron 1963; Cressie 1993; Webster ⎡ ⎤ γ , γ , ... γ , and Oliver 2001). The variogram represents how spatial auto- (x1 x1) (x1 x2) (x1 xN )1 ⎢ γ , γ , ... γ , ⎥ correlation declines with distance. The empirical variogram is ⎢ (x2 x1) (x1 x2) (x2 xN )1⎥ ⎢ ...... ⎥ estimated from the data, and then a theoretical functional form ⎢ ⎥ = ⎢ ...... ⎥ , is fit to the data. Variograms are usually characterized by three A ⎢ ⎥ ⎢ ...... ⎥ parameters: the range of autocorrelation, the sill, and the nugget ⎢ ⎥ ⎣ γ , γ , ... γ , ⎦ (Webster and Oliver 2001). The range defines the spatial dis- (xN x1) (xN x2) (xN xN )1 ... tance over which observations are correlated. The sill represents 11 10 the population variance or the variance of observations when the distance between the location of observations is greater than the ⎡ ⎤ ⎡ ⎤ λ γ , range, and the nugget represents fine-scale variability such as 1 (x1 x0) ⎢ λ ⎥ ⎢ γ , ⎥ measurement error. The theoretical variogram is the variance of ⎢ 2 ⎥ ⎢ (x2 x0) ⎥ ⎢ ⎥ ⎢ ⎥ the difference of pairs of data points as a function of distance: ⎢ . ⎥ ⎢ . ⎥ ⎢ ⎥ ⎢ ⎥ λ0 = ⎢ . ⎥ , and b0= ⎢ . ⎥ , ⎢ . ⎥ ⎢ . ⎥ var[S(x) − S(x + h)] = E{S(x) − S(x + h)}2, ⎢ ⎥ ⎢ ⎥ ⎣ λ ⎦ ⎣ γ(x , x ) ⎦ = 2γ(h)(3) n N 0 ψ(x0) 1

where where A is the symmetric N + 1 × N + 1 matrix of semivar- S(x) = the value of a random function at point x; iograms between observed points xi and xj, λ0 is an N + 1 × S(x + h) = the value of a random function at point x + Downloaded by [Department Of Fisheries] at 19:14 20 July 2015 ψ 1 vector of weights assigned to data points and (x0)istheLa- distance h; and N grange multiplier that ensures that Aλ0 = b0 = λi = 1, and γ(h) = the semivariance. i 1 b0 is the N + 1 × 1 vector of semivariogram values between all Under second-order stationarity, there is a relationship between observed points and the prediction point, x0 (Cressie 1993). The the covariance function mentioned earlier and the variogram predicted value at point x0, Sˆ (x0), is then obtained as a weighted function. The empirical variogram is the sample estimate of sum of the sample values (s={s(x1), s(x2), ..., s(xN )}: the theoretical variogram using following formula (Matheron 1963): ˆ = λ , S(x0) 0s (6) 1 2γˆ(h) = s(x ) − s(x ) 2, (4) and the kriging variance is obtained by |N(h)| i j N(h) σ2 ˆ = λ . ˆ S(x0) 0b0 (7) where 2γˆ(h) = the estimated semivariance for a distance interval h; The kriging weights (λ0) are assigned to each observation in the s(xi) = the sample value at point xi; sample to obtain a prediction at a single point or for a grid of s(xj) = the sample value at a point xj; and prediction points within the spatial domain with the average of 1098 WALTER ET AL.

the prediction points over the grid providing an estimate of the 400 and then multiplying the vector by the lower triangle of overall mean. In contrast to an arithmetic mean, which would the decomposed covariance matrix using the Cholesky decom- weight all samples equally, the kriging weights provide a spatial position method (Davis 1987). The values for the mean and weighting of samples that has several desirable properties: variance were chosen to obtain a coefficient of variation of approximately 0.67, as is commonly found in fisheries data. An (1) weights for samples closer to the prediction point are higher example of the resulting autocorrelated simulated vector that (2) observations close in space receive less weight since they are was then assigned to the 20 × 20 grid is given in Figure 1a. We spatial replicates and provide similar information; isolated intentionally did not impose any differences in catchability re- observations receive greater weight lated to differences in targeting, fishing power, or other factors, (3) in the absence of autocorrelation all predictions are the as these would complicate the analyses by requiring that these arithmetic mean differences be standardized a priori. Such situations would be (4) beyond the range of autocorrelation predictions tend toward likely to occur in practice, however. the arithmetic mean For each simulation cycle we chose an initial set of 20 sam- (5) the kriging variance (equation 4) provides a measure of ple locations (Figure 1b) with probabilities of selection based prediction uncertainty; predictions beyond the range have a upon the 1st and 3rd quartiles of the data with 1, 15, and 84% variance equal to the variogram sill. probabilities of selection for the 1st, 2nd, and 3rd quartiles, re- In this paper we use ordinary kriging to calculate the spa- spectively. These samples were chosen from a 100 × 100 grid tial average of a domain where the estimate is the average of of locations within the originally created data so that it would be the kriged predictions for a set of points within the predic- unlikely that any point would correspond exactly to the initial tion domain. As we also want to evaluate prediction accuracy, data grid (which is also the prediction grid). This allowed for we predict on a series of grid points distinct from the sample some straying from high abundance, as may be the case when points. For all analyses we use functions from the S-PLUS spa- fishers select initial fishing areas on the basis of some prior, but tial statistics package (Kaluzny et al. 1997), and simulations are imperfect, knowledge. programmed in S-PLUS 2000 (Insightful). After selection of an initial sample, subsequent locations Simulations.—To evaluate the performance of geostatistical were chosen based upon the current sample value. If the current prediction with fishery data, we created simulations that resem- observation was greater than the data set mean, a new location ble a fleet of vessels fishing over a patchy population. For each was randomly chosen from a fine grid of points within a unit cycle of the simulation, we created a population distributed over square centered over the current location with a spacing of 0.001 space according to a prespecified pattern of spatial autocorrela- units (Figure 1c). If the current observation was lower than the tion that defined the nature of the patches. An initial site selection mean, another location was chosen similarly but from a wider was based on selection criteria that reflected prior knowledge square three units to a side with the same spacing. If the square of good fishing spots. Further sample selection proceeded by window extended beyond the domain, locations were restricted a random walk, with subsequent samples being chosen from to the domain. A total of 9 additional samples were taken for within a short search radius if the current CPUE is high and a each of the initial 20 locations to give a total of 200 observations wide search radius if the current CPUE is low. per simulated surface (Figure 1d). We refer to this sample set as A simulation cycle was defined as the creation of a simulated the “Bias 200” samples. Since the surface was simulated at 400 surface, sampling from it, and then variogram estimation and initial grid nodes, to obtain the sample values for points selected kriging of the sample data sets (Figure 1). The spatially simu- at the finer scale we kriged the simulated data set and predicted Downloaded by [Department Of Fisheries] at 19:14 20 July 2015 lated surfaces were created with known mean (µ), variance (σ2), a value at each location. and pattern of autocorrelation defined by a spherical variogram: To compare kriging with a traditional design-based sampling strategy of the same sample size, we also performed random C 3h h 3 sampling from the kriged surface. We randomly selected 200 γ(h) = C + − for h < a locations from a fine-scale grid with a spacing of 0.3 units. We 0 2 a a generated CPUE values for the 200 locations in a manner similar γ(h) = C0 + C for h ≥ a, (8) to that for the adaptive samples in the Bias 200 data set. We call this sample the “Rand 200” sample. For each simulation and sample design, we estimated the where a is the range of autocorrelation in the data, C0 is the variograms using a common set of 20 lag distances spanning 2 nugget, C is the partial sill (C0 + C = σ ), h is the distance, distances between 1.5 and 21 units (one-half the maximum and γ(h) is the value of the variogram at a given distance. We distance). We fit spherical variograms (equation 6) using the obtained simulated spatially autocorrelated surfaces on a 20 × Cressie (1993) weighted least-squares function and the SPLUS 20 grid with a distance of 30 units on a side using a variogram spatial statistics function variogram.fit (Kaluzny et al. 1997), with a range of 10, a total sill of 1, and a nugget of 0 by first which uses the fitting criterion of the nonlinear bounded mini- generating a random normal (µ = 30, σ = 10) vector of size mization function, nlminb (Insightful). We obtained variograms GEOSTATISTICAL PREDICTION I 1099

a. initial data b. intial 20 catches

20 20 20 20 40 20 40 20 60 40 60 20 80 60 40 60 20 80 40 60 40 60 80 40 80 40 80 40 80 40 80 80 80 20 20 80 20 20 20 0 20 0

20 20 20 20 40 0 40 0 40 40 60 20 60 20 40 40 20 20

0 5 1020 15 20 2540 30 60 20 0 5 1020 15 20 2540 30 60 20

0 5 10 15 20 25 30 0 5 10 15 20 25 30

c. search windows d. resulting 200 catches

20 20 20 20 40 20 40 20 60 40 60 20 80 60 40 60 20 80 40 60 40 60 80 40 80 40 y direction (arbitrary units) (arbitrary direction y 80 40 80 40 80 80 80 20 20 80 20 20 20 0 20 0

20 20 20 20 40 0 40 0 40 40 60 20 60 20 40 40 20 20 0 5 10 15 20 25 30 0 5 1020 15 20 2540 30 60 20 20 40 60 20

0 5 10 15 20 25 30 0 5 10 15 20 25 30 x direction (arbitrary units)

FIGURE 1. (a) Shaded contour plot of simulated spatially autocorrelated data, (b) 20 initial samples selected with a bias towards high values, (c) selection windows for subsequent samples, and (d) subsequent 9 samples for each of the original 20 giving a Bias 200 set.

from the Rand 200 samples and predicted on the same low values), increasing the potential for selection bias, while 20 × 20 grid used for the Bias 200 samples to provide a direct low CVs produce flat (in a three-dimensional profile) data sets comparison between the two sample sets. The simulations were with low potential for bias. repeated 1,000 times. One thousand simulations achieved con- Finally, we considered augmenting the Bias 200 samples to vergence in the estimated values (variogram parameters, mean determine whether additional observations could be effective square prediction error [MSPE], and kriged means) as deter- in reducing bias and variance estimation. To that end we com- Downloaded by [Department Of Fisheries] at 19:14 20 July 2015 mined by the fact that plots of the cumulative medians for these bined the Bias 200 data with 20 random samples (Rand 20), 20 metrics reached stable values. systematic grid samples (Grid 20), and 20 samples chosen to To explore the effects of the pattern of spatial autocorrelation iteratively minimize the total kriging variance (Optimal 20). We upon the kriged and arithmetic means obtained under a regime of obtained the Optimal 20 samples through a sequential process sample selection bias, we varied the variogram range between producing the greatest reduction in the total prediction variance 10% and 60% of the maximum distance (42.4 units) and the (Burgess et al. 1981). This proceeds by sequentially adding a CV between 0.01 and 2.0. We altered the CV by changing the point from a fine grid of locations, calculating the prediction variance of the random normal vector prior to multiplication by variance when this point is added to the original sample, and the Cholesky decomposition of the covariance matrix and adding selecting the location which minimizes the sum of prediction a scalar to change the mean. Increasing the range increases variances. Assuming that this sample has been taken but without the patch size and the potential for selection bias. Data sets needing to have actually taken it, a new location is then chosen, with ranges approaching zero converge on being completely and so on. This results in a sequential increase in sample cov- random (i.e., with zero autocorrelation) and have low potential erage that minimizes the overall kriging variance. The benefit for selection bias. Increasing the CV makes the data set more of this approach is that it does not depend upon knowledge of heterogenous (meaning that it has higher high values and lower the sample values but only on the variogram and the assumption 1100 WALTER ET AL.

that the pattern of spatial autocorrelation remains the same over TABLE 1. Estimated variogram parameters from the Bias 200 samples. The time, so that it can be performed prior to collecting additional upper (UL) and lower (LL) limits represent the upper and lower 2.5% of the data. samples with the variogram from the initial observed samples with a variogram obtained using prior knowledge. Parameter Median LL UL True value Comparing sampling methods.—We used the percent bias to compare the estimation methods, i.e., Range 9.7 9.3 10.0 10.0 Sill 200.9 191.3 210.5 440.0 (predicted mean − true mean) Nugget 0 0 0 0 percent bias = median , (9) true mean In the few instances (9% and 6%, respectively) in which the due to the nugget. When the degree of extrapolation is 100%, Bias 200 or Rand 200 samples estimated a variogram with a the prediction is based upon data beyond the range of autocor- range beyond the maximum possible in the data set (42.4), relation. When the sample selection does not permit inferences we removed that iteration from the analysis as the variogram for the unsampled area, as most design-based estimators do in estimation failed. expectation (Thompson 2002), such an extrapolation is likely To assess prediction accuracy we used the MSPE: to be biased. To illustrate the trade-off between prediction bias N and the spatial extent of prediction, we convert the kriging 2 prediction error into a degree of maximum extrapolation and = [Sˆ (xi ) − S(xi )] ∗, MSPE N (10) i=1 plot it against the bias in the kriged means.

where S(xi) is the true value of the simulated data set at the point RESULTS xi and Sˆ (xi ) is the value predicted by each sample (Bias 200, Rand 200, etc.) for each xi in the set of N* = 400 grid points. Effects of Sampling Bias upon the Variogram Note that the initial 20 samples as well as subsequent samples Estimating the variogram from the sample data is the ini- were obtained from a very fine grid of 10,000 points so as to be tial step in geostatistical prediction. Comparison of empirical very unlikely to co-occur at the same location of the original 400 variogram values obtained from the Bias 200 samples with em- original simulated data points or the prediction locations. As a pirical variograms estimated from the true data sets indicates baseline measure of improvement over using the data set mean underestimation of the sill but generally accurate estimation of as the prediction at each location, we calculated the ratio of the the range and nugget (Table 1; Figure 2, insets a–c). Each real- MSPE for a kriged sample to the MSPE for the arithmetic mean: ization from the superpopulation possessed a slightly different variogram due to the introduction of random variability. Though MSPE(kriged sample) the superpopulation variogram had a zero nugget, the random % improvement = mean 1− . MSPE(arithmetic mean) data creation coupled with estimation of the variogram from (11) fine-scale spatial information afforded by the 400 created data Exploring the prediction variance versus prediction area points often induced a nugget, although this effect was never trade-off .—The kriging prediction variance provides a means estimated to be different than zero in the Rand 200 or Bias 200 for determining how much adjacent information is used to data sets (Figure 2c). obtain a value. With many adjacent samples within the range of The semivariance values (R in Figure 2) for the Rand 200 Downloaded by [Department Of Fisheries] at 19:14 20 July 2015 autocorrelation, a prediction location has low variance. At zero sample from each realization fluctuated around the superpopu- distance, the finest scale of variability is the variogram nugget. lation variogram used to create each data set (solid black line). As predictions extend beyond the range of autocorrelation The Bias 200 sample semivariance values (B) and intervals that from adjacent samples, the prediction variance reaches an include 95% of these values at each lag distance generally fell asymptote at the variogram sill. The difference between these below the true variogram, resulting in underestimation of the two levels, is the partial sill, which represents the variance sill though the range was correctly estimated. “explained” by spatial autocorrelation. Conversely, for a given point, the fraction of unexplained variability or the degree of extrapolation used to obtain the prediction can be obtained by Kriged Means and Prediction Accuracy dividing the variance of the kriging prediction at a location Figure 3 depicts the kriging predictions and kriging standard minus the nugget by the partial sill: error for a single simulation. The arithmetic means of the Bias 200 samples overestimated the true mean (Table 2; Figure 4A), kriging variance − nugget with a median percent bias of 68%. Kriging the Bias 200 sample degree of extrapolation = (12) partial sill set reduced the overestimation to 42%, indicating that kriging improved predictions by 26 percentage points over the entire At the finest spatial scale, extrapolation is zero and the spatial domain. As would be expected for design-based estima- prediction is the sample observation, plus or minus variability tors, the arithmetic means for the Rand 200, Rand 20, and Grid Downloaded by [Department Of Fisheries] at 19:14 20 July 2015

TABLE 2. Comparison of the performance of geostatistical and arithmetic estimators; NE = not estimated or not estimable. See the text for explanations of the sample sets.

% Improvement Percent of times Median 5th 95th 5th 95th in MSPE over ﬁtted range Sample set Estimator Sample size percent error percentile percentile MSPE percentile percentile arithmetic mean > = 42.4 Rand 200 Arithmetic mean 200 0 −19 14 416 278 624 NE NE Rand 200 Kriging 200 0 −14 9 105 53 403 69 6% Bias 200 Arithmetic mean 200 68 45 110 NE NE NE NE NE Bias 200 Kriging 200 43 19 70 425 203 827 −29% Bias 200 + Rand 20 Kriging 220 13 1 30 166 108 306 56 9% Bias 200 + Grid 20 Kriging 220 9 −1 22 145 101 234 60 8% Optimal allocation Kriging 220 7 1 17 128 90 186 64 NE Rand 20 Kriging 20 0 −21 24 291 189 488 28 14% Grid 20 Kriging 20 0 −16 14 238 160 375 40 9% 1101 1102 WALTER ET AL.

a. range range 01020 True Bias200 Rand200 b. sill

R sill R R R R 0 400 1000 R R True Bias200 Rand200 R R R R c. nugget R R R R R R R R R B RB RR R R R R RR R R RB R R nugget R R R R R 01025 R R R True Bias200 Rand200 R RB B R R R R R B R R R R R R R R R R B B R R R R R R R R semivariance R R R RR R R RR R RR R R R R R R R R R R R R R RRR R R R RR RR R R R R R RR RBB R B R R RR R B B R R B B B B B B B BB RRR B B B B B B BB B R RR BB BB BB B B R BR BB BBB B B BB BBB BBB BB RRRR B BBB B B B BB B B B BB B R BB B B B BB BBB B B B BB BB RR BBBBR BBB BB BB BB B B B B RRR BB BB B B B B B B B BB BBR BB BB B BBBBB B 0 200 400 600 800 1000

2468101214 distance (arbitrary units)

FIGURE 2. Sample variograms obtained from the random (R) and biased (B) data sets. The individual letters correspond to the jittered semivariance values at each distance interval, and the dark and light lines represent the upper and lower 95% limits for the two data sets. The variogram used to create the data sets is shown as a dashed line. The insets show the simulation results comparing the “true” and estimated (a) range, (b) sill, and (c) nugget.

20 samples sets were unbiased for the overall mean. Combin- variance, or the percent of extrapolation, increases as predictions ing the Bias 200 data with the Rand 20 and Grid 20 samples extend away from the observed data. Locations falling within the improved estimation of the mean, reducing the bias to 13% and 80% contour were still within the range of autocorrelation, but 9%, respectively. these predictions were based upon a high level of extrapolation Downloaded by [Department Of Fisheries] at 19:14 20 July 2015 In the absence of spatial autocorrelation or without assuming and consequently had high prediction variance. a predictive model, the arithmetic mean would be the default For the data set with selection bias (Bias 200), this trade-off prediction at each location, but it would be fairly inaccurate became one of increasing bias in the predicted mean as predic- (MSPE = 416; Table 2). In contrast, kriging with the Rand tion locations with a greater kriging variance or greater fraction 200 data set was quite accurate (MSPE = 105), representing of unexplained variability were included (Figure 5). By parti- a substantial improvement, as would be expected. Despite the tioning the prediction locations according to the kriging variance substantial reduction in the bias for the overall mean, kriging the or the percent of the unexplained variance (both contours being Bias 200 sample slightly reduced prediction accuracy (MSPE = almost the same in Figure 3c and 3d), it was possible to de- 425). The design-based Rand 20 (291) and Grid 20 (238) sam- termine a percent area coverage at various levels of precision ples moderately improved prediction precision. (numbers at the bottom of Figure 5). Predictions exhibited relatively low bias with kriging variances up to the nugget (zero in almost all realizations) plus 50% of the partial sill. This covered Exploring the Prediction Variance versus Prediction Area 28% of the area with predictions that had a relatively low level of Trade-Off bias of less than 5%. Prediction locations with greater levels of The trade-off between the prediction variance and prediction kriging variance exhibited substantially increasing bias, though area is illustrated in Figure 3c and 3d. The kriging prediction the predictions encompassed greater spatial area. GEOSTATISTICAL PREDICTION I 1103

a. Initial data b. Kriging predictions simulated predicted value value

80 60 60 40 40 20 20 0 0 0 5 10 15 20 25 30 0 5 10 15 20 25 30

0 5 10 15 20 25 30 0 5 10 15 20 25 30

c. Kriging standard error d. Degree of extrapolation

kriging 80 60 40 kriging var. stand.error as % of 80 partial sill 20 y direction (arbitrary units) (arbitrary direction y 60 80 15 40 60 10 20 40 5 20 20 0 40 0

0 5 10 15 20 25 30 0 5 10 15 20 2520 30 20

0 5 10 15 20 25 30 0 5 10 15 20 25 30 x direction (arbitrary units)

FIGURE 3. Single prediction set of (a) true data and Bias 200 sample data points, (b) kriging predictions from the Bias 200 sample, (c) contours of the kriging prediction error, and (d) contours of the percent of unexplained variability, which is measured as the kriging variance divided by the partial sill and ranges between 0% and 100%.

Varying the Range and CV and 60%, respectively (Table 2; Figure 4A, B). Optimal alloca- Increasing the range of autocorrelation of the created data sets tion of 20 additional samples reduced the overall bias to only increased the error of both the arithmetic mean and kriged means 7%, improved the prediction precision by 69% over the arith- of the Bias 200 samples (Figure 6a), though the arithmetic mean metic mean, and filled in gaps in spatial uncertainty (Table 2; showed both the greatest increase and the greatest absolute bias. Figure 7). While both the Rand 20 and Grid 20 samples filled in Downloaded by [Department Of Fisheries] at 19:14 20 July 2015 Increasing the range of autocorrelation increased the patch size gaps, the optimal allocation provided the greatest improvement and allowed the targeted sampling strategy to remain on these in prediction accuracy. large patches with few excursions to low-abundance regions. Similarly, increasing the CV while maintaining the same range DISCUSSION of autocorrelation increased the bias in both means by broaden- Fisheries CPUE presents numerous problems with interpreta- ing the spread of the data relative to the its mean (Figure 6b), tion caused by the absence of control over the sample design and while the kriging estimator always outperformed the arithmetic collection. While numerous methods of standardization exist to mean. control for the various factors affecting catch rates, few have been able to deal with the influence of targeted, repeated, and Augmenting a Fishery-Dependent Sample and Optimal clustered observations upon CPUE estimates or provide guid- Allocation ance as to how to fill in spatial gaps in samples (Walters 2003; Combining the Bias 200 data set with a set of 20 random Carruthers et al. 2011). This paper examined the potential to use samples (Rand 20) or a systematic grid (Grid 20) reduced the samples collected in a manner resembling a fishing process in bias in the kriged means to only 13% and 9%, respectively, and geostatistical prediction of CPUE and showed that (1) fishery- improved prediction accuracy as measured by the MSPE by 60% dependent data will likely provide an adequate variogram for 1104 WALTER ET AL.

A. percent bias in mean CPUE percent bias 0 20406080100

Arith Krige Arith Krige Krige Krige Krige Krige Krige Rand Rand Bias Bias Bias Bias Bias200 Rand Grid 200 200 200 200 Rand20 Grid20 OptAlloc 20 20

B. mean squared prediction error mean square prediction error 0 100 200 300 Arith Krige Krige Krige Krige Krige Krige Krige Rand Rand Bias Bias Bias Bias200 Rand Grid 200 200 200 Rand20 Grid20 OptAlloc 20 20

FIGURE 4. (A) Percent bias in CPUE and (B) mean square prediction error for various sampling set–estimator combinations.

spatial interpolation; (2) geostatistical prediction partially ame- liorates the bias that clustered and targeted sampling imposes upon catch rate data; and (3) the kriging variance can define the spatial limits of data-informed prediction. We elaborate on these three findings below. Downloaded by [Department Of Fisheries] at 19:14 20 July 2015 In our simulations and those of Silliman and Berkowitz (2000), underestimation of the sill from the biased data was a result of the sample variance’s underestimating the population variance. Since the expected variogram total sill is the population variance (Barnes 1991), correct estimation of the sill requires that the sample variance be, on average, unbiased for the unknown population variance. This is unlikely with fishery-dependent catch rate data, though Silliman and Berkowitz (2000) proposed correction factors for the sill provided that one had prior knowledge of the variance. In practice, underestimation of the sill is unlikely to severely FIGURE 5. Error versus the kriging variance as a percent of the partial sill. This measures the degree of extrapolation used to obtain a prediction. The affect the kriged means unless it substantially alters the es- numbers below the points are the percent areas covered by the predictions. The timation of the range. However, since the sill determines the values are jittered for clarity. absolute magnitude of the kriging variance, underestimation of the sill will lower kriging prediction variances proportionally. GEOSTATISTICAL PREDICTION I 1105

a. bias vs range b. bias vs CV 1998) and the two are not really comparable, this is of little practical issue. Rather, the real utility of the kriging variance is

A A A as a measure of spatial sampling sufﬁciency, in which case only A A A A the relative values are of interest. As the degree of extrapolation

A A is independent of the absolute value of the sill, changes in the K A K A K A K K magnitude of the sill have no effect. K K A A K K K percent bias A K percent bias A Variograms obtained from the biased sample data usually K K K A K A K provided correct estimates of the range and nugget. In contrast K A K K A to the sill, the range can have a substantial effect upon kriged

0 50 100 150 0K 100 200 300 400 estimates both locally and over the entire extent of the prediction 10 20 30 40 50 60 0.0 0.5 1.0 1.5 2.0 range (percent of maximum distance) coefficient of variation surface (Stein and Handcock 1989). Overestimation of the range causes samples to exert an undue influence upon prediction FIGURE 6. Percent bias in kriged (K) and arithmetic mean (A) CPUE with locations beyond the actual range of autocorrelation, but this data sets created with (a) an increasing range of autocorrelation (measured as a occurred in only 9% of the simulations. In practical situations in percentage of the maximum distance in the created data sets [42.4 units]) and an expected CV of 0.67 and (b) a range of 10 units and different coefficients of which the estimated range appears to be substantially different variation. than intuition would suggest, it may be necessary to employ prior knowledge of a similar spatial process to assist in variogram This, coupled with the tendency for fishery-dependent data to modeling (Myers 1997); alternatively, if one could characterize have lower variability due to repeated sampling in the same the overestimation of the range for a given sample configuration, location, will produce kriging variances that are lower than a a correction could be obtained (Rufino et al. 2006). comparable design-based variance. As the kriging variance is The simulation results indicate that the kriged predictions fundamentally different from a design-based variance (Warren using fishery-dependent samples will be no worse than the

Iteratively adding a sample to minimize kriging SE 0 5 10 15 20 25 30 0 5 10 15 20 25 30 0 5 10 15 20 25 30

0 5 10 15 20 25 30 0 5 10 15 20 25 30 0 5 10 15 20 25 30 Ydirecon Downloaded by [Department Of Fisheries] at 19:14 20 July 2015 0 5 10 15 20 25 30 0 5 10 15 20 25 30 0 5 10 15 20 25 30

0 5 10 15 20 25 30 0 5 10 15 20 25 30 0 5 10 15 20 25 30

12 10 kri g i ng SE 8 Kriging 6 4 2 Std error 0 0 5 10 15 20 25 30 0 5 10 15 20 25 30

0 5 10 15 20 25 30 0 5 10 15 20 25 30 X direcon

FIGURE 7. Demonstration of the method of iteratively adding a sample to the location of the highest kriging standard error. 1106 WALTER ET AL.

arithmetic sample mean and, within the range of spatial au- pattern of spatial autocorrelation does provide some guidance tocorrelation, can be an improvement though bias still remains. as to the potential strength of selection bias. With relatively Kriging achieves this through a de facto spatial weighting of random (short range; Figure 6a) or relatively flat (very little samples, whereby spatially replicated observations receive less difference between the high and low values and low CV; Figure weight and isolated observations receive greater weight. The 6b) spatial processes, there is less potential to impose bias upon performance of the kriging predictions declined as a result of either the arithmetic or the kriged mean. For low-CV processes extrapolating predictions into areas with few samples. As the the difference between high and low patches is slight, so there Bias 200 samples were clustered in areas of high CPUE, ex- is little potential for the selection process to impose a bias. trapolating these values into areas with few samples (and pre- Similarly, for spatially random phenomena there is less potential sumably low values) overestimated the mean and provided poor to impose a bias. It may also be possible to evaluate the potential prediction accuracy. Kriged predictions for the entire area of- strength of selection bias through analysis of fishing practices by ten extended beyond the range of autocorrelation, where the determining whether there is substantial exploration or a limited predictions reverted towards the arithmetic mean of the biased amount of prior knowledge of high- and low-abundance areas. sample (Isaaks and Srivastava 1989), which was higher than the The practical implications of varying the CV and the range of true mean. Predictions beyond the range come close to but do autocorrelation can be illustrated by considering the spatial dis- not always become the arithmetic sample mean because they re- tribution of a species. If there are strong hotspots of abundance, main a spatially weighted combination of the samples within the there is likely a high CV and strong spatial autocorrelation, thus kriging equations. This reversion towards the sample mean as creating high potential for bias. In contrast, if a species is rela- prediction extends towards and beyond the range of the spatial tively rare or almost random in its distribution (such as a bycatch process is both a desirable property of kriging in the absence of species), the potential for bias is lower. proximate samples and a problem when the sample mean does For most spatial patterns, the strength of the selection bias, not represent the overall mean. or the skill with which fishers target or avoid certain locations, The kriging variance is valuable in making explicit the spatial remains the greatest factor in determining how representative limitations of prediction. Data points in close proximity differ the sampled locations are of the unsampled ones. Assessing the according to the fine-scale or nugget variability, while at dis- strength of selection bias would require an evaluation of fishing tances beyond the range data points differ according to the total practices, such as how often fishermen explore new areas, how variance (nugget + partial sill). The difference in variance be- much prior knowledge of the distribution do they have, or how tween the nugget and the total sill is the partial sill or the variance strong the incentive might be for avoiding certain areas. The explained by the distance between data points. This explained selection bias might be extreme, such that vessels with observers variance provides a means of limiting predictions to areas of actively avoid an area of high bycatch. But in any case, the high precision by contouring the prediction points according strength of selection bias will largely be unknown. The benefit the unexplained variability. that geostatistics offers in the face of this uncertainty is that, Often analysts are faced with a set of fishery data similar to regardless of the strength of the selection bias, geostatistical those in Figure 3a, where the observations only cover part of the predictions will reduce this bias and, more importantly, will area. In this situation, the kriging variance provides guidance identify the areas of greatest uncertainty. as to how far the predicted abundance can be extrapolated and When there are substantial gaps in sample data beyond the at what level of error. Contours of the kriging standard error or range of autocorrelation, additional sampling that provides un- percent extrapolation provide a means of determining the infor- biased inference over the unsampled space is required (Chang

Downloaded by [Department Of Fisheries] at 19:14 20 July 2015 mation content used in a prediction. A prediction at the edges et al. 2006). The simulation performance of the augmented sam- of Figure 3d but far outside the observed locations would still pling approaches indicates that such samples could be collected be within the range of autocorrelation (10 units), but 60–90% with a few additional design-based samples and could be most of this prediction is based upon extrapolation of data from the efficiently obtained by targeting the areas of greatest uncer- fringes of the range of autocorrelation. For a prediction point tainty, as was demonstrated by the optimal sampling strategy. beyond this, 100% of its value is obtained under the assumption Additional samples could be taken by contracted fishing ves- that the observations permit unbiased inference for the unob- sels; alternatively, when large gaps in sample coverage exist, served location, an assumption not usually met with fishery fishing vessels could be encouraged to fill in the gaps in spatial data. In practice an analyst could a priori choose a percentage of coverage. acceptable extrapolation and then restrict inference to that area. Low sample size can impose further complications on Unfortunately, while the kriging variance provides guidance variogram estimation and kriging; however, this is a well- as to how far predictions could be extrapolated before they revert documented result (Rufino et al. 2006) for which several possi- towards the arithmetic sample mean, how far predictions should ble solutions exist (Walter et al. 2007). We chose to use rather be extrapolated is a function of the strength of selection bias, large sample sizes (200) to focus on more insidious and perva- which will likely be unknown. While it may not be possible to sive problems created by biased sample selection. In our expe- specify an objective limit to extrapolation to minimize bias, the rience, fishery-dependent data rarely lack for absolute sample GEOSTATISTICAL PREDICTION I 1107

size. Rather, the problem is that the samples are pseudorepli- Campbell, R. A. 2004. CPUE standardization and the construction of indices of cates, may only cover limited parts of the region of interest, and stock abundance in a spatially varying fishery using general linear models. may have a severe selection bias. Similarly, the design-based Fisheries Research 70:209–227. Carruthers, T. R., R. N. M. Ahrens, M. K. McAllister, and C. J. Walters. 2011. samples that we chose to use (Rand 20, Grid 20) reflect our Integrating imputation and standardization of catch rate data in the calculation experience that survey data usually are design-based and unbi- of relative abundance indices. Fisheries Research 109:157–167. ased but suffer from low sample sizes due to cost and logistical Chang, C.-R., P.-F. Lee, M.-L. Bai, and T.-T. Lin. 2006. Identifying the scale considerations. thresholds for field-data extrapolation via spatial analysis of landscape gra- Another problem that might occur in practice is the imposi- dients. Ecosystems 9:200–214. Cressie, N. A. C. 1993. Statistics for spatial data. Wiley, New York. tion of spatial structure in CPUE data by the fishing behavior Davis, M. W. 1987. Production of conditional simulations via the LU triangular (Stelzenmuller¨ et al. 2006). In this case the spatial structure im- decomposition of the covariance matrix. Mathematical Geology 19:91–98. posed by fishing behavior does not correspond to actual abun- Gregoire, T. G. 1998 Design-based and model-based survey sampling: appreci- dance patterns. Ideally, conventional methods of standardization ating the differences. Canadian Journal of Forest Research 28:1429–1447. could account for such factors and remove these effects upon the Isaaks, E. H., and R. M. Srivastava. 1989. An introduction to applied geostatistics. Oxford University Press, Oxford, UK. observed spatial pattern of CPUE. The residuals following such Kaluzny, S. P., S. C. Vega, T. P. Cardosa, and A. A. Shelly. 1997. S + Spatial analysis would then reflect the underlying spatial structure and Stats user’s manual for Windows and UNIX. Springer-Verlag, New York. could form the basis for geostatistical prediction (Stelzenmuller¨ Matheron, G. 1963. Principles of geostatistics. Economic Geology 58:1246– et al. 2007). 1266. Walters (2003) posed the task of obtaining catch rates from Maunder, M. N., and A. E. Punt. 2004. Standardizing catch and effort: a review of recent approaches. Fisheries Research 70:141–160. commercial CPUE as an exercise in predicting catch rates in Maunder, M. N., J. R. Sibert, A. Fonteneau, J. Hampton, P. Kleiber, and S. J. unfished areas, a process which requires backfilling a spatial– Harley. 2006. Interpreting catch-per-unit-effort data to assess the status of temporal grid of catch rates. In very limited cases it may be pos- individual stocks and communities. International Council for the Exploration sible to augment existing samples, but generally this is not pos- of the Sea Journal of Marine Science 63:1373–1385. sible and clearly is impossible back in time. Conventional CPUE Myers, J. 1997. Geostatistical error management: quantifying uncertainty for environmental sampling and mapping. Wiley, New York. standardization methods may implicitly fill in gaps, largely un- Paloheimo, J. E., and L. M. Dickie. 1964. Abundance and fishing suc- der the assumption that the unsampled locations represent the cess. Rapports et Proces-Verbaux` des Reunions´ Conseil International pour sampled locations without bias, an assumption which is often l’Exploration de la Mer 155:152–163. untrue. Geostatistical prediction methods partially ameliorate Petitgas, P. 2001. Geostatistics in fisheries survey design and stock assessment: the effects of selection bias and allow for statistically-based models, variances, and applications. Fish and Fisheries 2:231–249. Rivoirard, J., J. Simmonds, K. G. Foote, P. Fernandes, and N. Bez. 2000. Geo- prediction into unsampled locations, at least to the extent of statistics for estimating fish abundance. Blackwell Scientific Publications, spatial autocorrelation. Furthermore, these methods force an Oxford, UK. explicit examination of the spatial limits to extrapolation and Rufino, M. M., V. Stelzenmuller,¨ F. Maynou, and G.-P. Zauke. 2006. Assessing allow the analyst to determine the level of extrapolation used the performance of linear geostatistical methods applied to artificial fisheries to make a prediction. This mapping of spatial uncertainty is data. Fisheries Research 82:263–279. Sarndal,¨ C.-E., B. Swensson, and J. Wretman. 1992. Model-assisted survey critical for understanding the limitations of fishery data and sampling. Springer-Verlag, New York. identifying gaps in knowledge. In conjunction with conven- Silliman, S. E., and B. Berkowitz. 2000. The impact of biased sampling on the tional standardization methods that account for factors affecting estimation of the semivariogram within fractured media containing multiple catchability, geostatistical prediction provides an additional tool fracture sets. Mathematical Geology 32:543–560. that addresses sample selection biases and the need to predict Stein, M. L., and M. S. Handcock. 1989. Some asymptotic properties of kriging when the covariance function is misspecified. Mathematical Geology 21:171– Downloaded by [Department Of Fisheries] at 19:14 20 July 2015 CPUE in unsampled areas. 190. Stelzenmuller,¨ V., S. Ehrich, and G.-P. Zauke. 2006. Analysis of mesoscaled spatial distribution of the Dab in the German Bight: Does the type of fishing ACKNOWLEDGMENTS gear employed matter? Fisheries Science 72:95–104. We thank Courtney Harris, William DuPaul, Roger Mann, Stelzenmuller,¨ V.,F. Maynou, and P.Mart´ın. 2007. Spatial assessment of benefits Paul Rago, and Doug Vaughan for reviewing this manuscript. of a coastal Mediterranean Marine Protected Area. Biological Conservation This research was conducted by J.F.W. while he held a NOAA– 136:571–583. Thompson, S. K. 2002. Sampling. Wiley, New York. Sea Grant Service Joint Graduate Fellowship in population dy- Walter, J. F., M. Christman, J. Hoenig, and R. Mann. 2007. Increasing the namics. This is VIMS contribution 3398. number of data points used to estimate the variogram by including data from repeated sampling stanzas. Envirometrics 18:583–598. Walters, C. 2003. Folly and fantasy in the analysis of spatial catch rate data. REFERENCES Canadian Journal of Fisheries and Aquatic Sciences 60:1433–1436. Barnes, R. 1991. The variogram sill and the sample variance. Mathematical Warren, W. G. 1998. Spatial analysis for marine populations: factors to be Geology 23:673–678. considered. Canadian Special Publications in Fisheries and Aquatic Sciences Burgess, T. M., R. Webster, and A. B. McBratney. 1981. Optimal interpolation 125. and isarithmic mapping of soil properties, IV. Sampling strategy. Journal of Webster, R., and M. Oliver. 2001. Geostatistics for environmental scientists. Soil Science 22:643–659. Wiley, Chichester, England. This article was downloaded by: [Department Of Fisheries] On: 20 July 2015, At: 19:14 Publisher: Taylor & Francis Informa Ltd Registered in England and Wales Registered Number: 1072954 Registered office: 5 Howick Place, London, SW1P 1WG

North American Journal of Fisheries Management Publication details, including instructions for authors and subscription information: http://www.tandfonline.com/loi/ujfm20 Reducing Bias and Filling in Spatial Gaps in Fishery- Dependent Catch-per-Unit-Effort Data by Geostatistical Prediction, II. Application to a Scallop Fishery John F. Waltera, John M. Hoenigb & Mary C. Christmanc a National Oceanic and Atmospheric Administration–Fisheries, Southeast Fisheries Science Center, 75 Virginia Beach Drive, Miami, Florida 33149, USA b Virginia Institute of Marine Science, College of William and Mary, Post Office Box 1364, Gloucester Point, Virginia 23062, USA c MCC Statistical Consulting LLC, 2219 Northwest 23rd Terrace, Gainesville, Florida 32605, USA Click for updates Published online: 30 Oct 2014.

To cite this article: John F. Walter, John M. Hoenig & Mary C. Christman (2014) Reducing Bias and Filling in Spatial Gaps in Fishery-Dependent Catch-per-Unit-Effort Data by Geostatistical Prediction, II. Application to a Scallop Fishery, North American Journal of Fisheries Management, 34:6, 1108-1118, DOI: 10.1080/02755947.2014.932866 To link to this article: http://dx.doi.org/10.1080/02755947.2014.932866

PLEASE SCROLL DOWN FOR ARTICLE

Taylor & Francis makes every effort to ensure the accuracy of all the information (the “Content”) contained in the publications on our platform. However, Taylor & Francis, our agents, and our licensors make no representations or warranties whatsoever as to the accuracy, completeness, or suitability for any purpose of the Content. Any opinions and views expressed in this publication are the opinions and views of the authors, and are not the views of or endorsed by Taylor & Francis. The accuracy of the Content should not be relied upon and should be independently verified with primary sources of information. Taylor and Francis shall not be liable for any losses, actions, claims, proceedings, demands, costs, expenses, damages, and other liabilities whatsoever or howsoever caused arising directly or indirectly in connection with, in relation to or arising out of the use of the Content. This article may be used for research, teaching, and private study purposes. Any substantial or systematic reproduction, redistribution, reselling, loan, sub-licensing, systematic supply, or distribution in any form to anyone is expressly forbidden. Terms & Conditions of access and use can be found at http:// www.tandfonline.com/page/terms-and-conditions North American Journal of Fisheries Management 34:1108–1118, 2014 C American Fisheries Society 2014 ISSN: 0275-5947 print / 1548-8675 online DOI: 10.1080/02755947.2014.932866

ARTICLE

Reducing Bias and Filling in Spatial Gaps in Fishery- Dependent Catch-per-Unit-Effort Data by Geostatistical Prediction, II. Application to a Scallop Fishery

Abstract Fishery-dependent catch per unit effort (CPUE) comprises critical input for many stock assessments. Construction of CPUE indices usually employs some method of data standardization. However, conventional methods based on linear models do not effectively deal with the fact that samples are collected with a selection bias or with the problem of filling spatial gaps. Geostatistical interpolation methods can ameliorate some of the biases caused by both of these problems while remaining complementary to traditional linear model-based CPUE standardization. In this paper we present geostatistical estimates of sea scallop Placopecten magellanicus CPUE from tows recorded by onboard observers during an opening of Georges Bank Closed Area II in 1999. By selecting tows for which there was little prior effort (on the basis of accumulated effort measured by vessel monitoring systems), we obtained tows that reflected initial abundance as closely as possible. These tows were used to obtain a variogram which was used in geostatistical prediction of sea scallop CPUE. The kriged mean was substantially lower than the arithmetic sample mean, indicating that a geostatistical approach reduced the influence of repeated sampling in locations of extremely high CPUE and increased the weight of isolated observations in areas of low CPUE. The results produced a map that was qualitatively similar to that obtained from a preseason fishery-independent survey. Overall differences between the two approaches were driven by the extension of predictions into areas at the edges of spatial autocorrelation where kriging predictions approached the grand mean of the data set. Downloaded by [Department Of Fisheries] at 19:14 20 July 2015

In a companion paper, Walter et al. (2014, this issue) used autocorrelation-driven method for spatial interpolation, and the simulation to explore how geostatistical methods address the associated geostatistical prediction errors provide a means of de- problems inherent in interpreting catch-per-unit-effort (CPUE) termining the prediction uncertainty according to the amount of data consisting of size-biased and clustered observations and adjacent spatial information. In conjunction with conventional the need to interpolate between or extrapolate into areas lacking linear model–based standardization methods, which account for samples (Walters 2003). The simulation results demonstrate that factors that affect catchability such as gear and season (Maunder the spatial weighting properties of the geostatistical mean reduce and Punt 2004), geostatistical methods harness spatial autocor- the inﬂuence of sample clustering and the targeting of higher- relation for prediction and have the ability to ameliorate the abundance areas on estimated CPUE. Geostatistics provides an biases created by unequal sample selection probabilities.

*Corresponding author: [email protected] Received July 26, 2013; accepted June 2, 2014

1108 GEOSTATISTICAL PREDICTION II 1109

The widespread utilization of fishery observers and vessel the geostatistical modeling process used here can be found in a monitoring systems has created interest in using such data for companion paper (Walter et al. 2014). stock assessment purposes (Deng et al. 2005; Mills et al. 2007; Fishery and data description.—The North Atlantic sea scal- Palmer and Wigley 2009). Data from onboard observers con- lop fishery occurs on the continental shelf from Cape Hatteras stitute critical information for estimating bycatch and discards north to Canada. A major fishery occurs on Georges Bank, where in many fisheries, yet substantial logistical and theoretical is- several large areas were closed in 1994 to protect groundfish sues exist with respect to the use of these data. First, one has (Figure 1). During these closures, sea scallop biomass accumu- little ability to control the spatial and temporal allocation of lated to high levels (Murawski et al. 2000), permitting a limited observations. Observed samples tend to be clustered, as they opening of Closed Area II to scalloping from June through are repeated observations from the same vessels in similar lo- November 1999. cations, and they may not cover the entire spatial and tempo- During the opening, observers hired by the National Marine ral extent of a fishery. Vessel monitoring systems can address Fisheries Service’s Northeast Fisheries Science Center recorded the problem of repeated observations in the same location by the catch per tow of sea scallops and bycatch species as well placing an observed sample in the context of the amount of as tow location and duration, sea condition, and vessel and gear prior effort that has occurred at that location (Gedamke et al. characteristics for 2,755 tows from 35 vessels on a total of 40 2004). trips. On all vessels, New Bedford style scallop dredges were Incomplete observer coverage and limited sampling re- used, with most vessels towing similar configurations consisting sources invariably lead to gaps in sample coverage. These are of paired 15-foot dredges with frame heights varying from 12 to most problematic when the spatial gaps are likely to have 21 in, chain bags with inside ring diameters of 3.5 in, and twine nonignorable sampling bias (Little and Rubin 2002). One tops with 10-in meshes. method of overcoming such selection bias is to use spatial To obtain usable catch rates, we eliminated tows when it autocorrelation—the tendency of objects to be more alike at could not be determined whether the catch came from the port small distances than at large ones—as a spatial bridge to fill in dredge, the starboard dredge, or both and when the tows were gaps and, to the extent that there is spatial autocorrelation, to of poor quality (such as those involving flipped, tangled, or lost extrapolate to unobserved locations. dredges). Tows with durations greater than 2 h or tow distances In this paper we address the question whether geostatistical greater than 9 nautical miles (based on the distance between methods allow us to use observer data alone to obtain the same the start and stop positions) were also eliminated, as these tows insights into the spatial distribution and relative abundance of could not be rectified to spatial locations and may have entailed sea scallops Placopecten magellanicus as were derived from an the dredge being used as an anchor while the vessel processed the intensive fishery-independent survey. We use vessel monitoring catch. Catch was recorded in either whole weight or meat weight system (VMS) effort data to address the issue of prior depletion of kept and discarded scallops or as total bushel baskets of kept by identifying a set of initial tows that occurred with very little or discarded scallops. Tows in which these two measures could prior fishing effort from a larger set of 2,755 scallop dredge not be reconciled (the weight landed should be a multiple of the tows observed in a 1999 fishery on Georges Bank Closed Area number of bushels) were removed. Two additional tows were II (CAII) off Massachusetts into. From these tows we obtained a removed as the scallop weights were misreported as bushels, variogram and used it to predict relative abundance at the start of resulting in an impossibly high bushel count. the fishery. We then evaluated the performance of geostatistical In 1999, vessel speeds were not available in the observer predictions by comparing variograms and predicted CPUE with database, so we assumed an average vessel speed of 5 knots

Downloaded by [Department Of Fisheries] at 19:14 20 July 2015 estimates obtained from an intensive survey conducted 1 year (Gedamke et al. 2004). To georeference each tow we used the prior to the opening of the fishery. midpoint of the start and stop positions, and to obtain a tow distance we used the product of the tow duration and average vessel speed. We converted latitude and longitude to an equidis- METHODS tant scale in nautical miles with a Universal Transverse Mercator General geostatistical methods.—Geostatistics is a model- projection in North American Datum 83, Zone 19. based method of predicting the values of a spatial process from We used bushels (1 bushel = 35.239 L) of kept and discarded a set of samples (Webster and Oliver 2001). As with any model- whole sea scallops per nautical mile to measure abundance. In based estimation method, the first step is to obtain a model of the few (<5%) cases in which only one dredge was observed, we the process which, in this case, is a measure of spatial autocor- doubled the catch to approximate the catch had two tows been relation, the variogram. The second step is prediction, where the observed based upon the significant correlation between catch in model is used as a means of weighting samples to predict val- port and starboard dredges observed in the fishery-independent ues in unsampled locations. Geostatistics has found widespread survey described below (r2 = 0.9734, P ≤ 0.001, df = 528). We application in fisheries, particularly as an alternative to design- did not correct for selectivity because the mean shell heights based estimators and for interpolating and mapping (Warren were greater than 4.33 in in 95% of the tows, well above the 1998; Rivoirard et al. 2000; Petitgas 2001). Details regarding shell height (3.3 in) of 100% selectivity for commercial dredge 1110 WALTER ET AL.

FIGURE 1. Sea scallop area closures off the East Coast of North America. The study area is the cross-hatched section of Closed Area II, which was opened for ﬁshing in 1999. The contour line is the 50-fathom curve.

gear with 3.5-in rings used in the 1999 fishery (Serchuk and (0.8%) of records the elapsed times were greater than 3 h, and Smolowitz 1980). we removed these anomalous records because it was impossible Processing VMS data.—The U.S. North Atlantic scallop fish- to spatially assign effort to a particular location. Downloaded by [Department Of Fisheries] at 19:14 20 July 2015 ery has a universal VMS that records and transmits vessel posi- Geostatistical prediction of fishing effort.—It is critical that tions to satellite receivers. We obtained 90,944 individual VMS CPUE indices reflect abundance at a recognizable time with records representing the total fishing effort from June 16 through respect to the processes of biomass addition and depletion. With November 13, 1999, in the exemption area of CAII (Figure 1). fishery-dependent data there is little ability to control the timing Each record consisted of the latitude and longitude, the time and of data collection with respect to removals. VMS systems can date, and the time differential between the record and the previ- be used to assign a prior level of fishing effort to each observed ously recorded position for the vessel. The average straight-line catch. If there are enough catches at a common level of effort, velocity of the vessel (representing a minimum velocity because they can be used to generate a CPUE index with a similar—and the path may not have been straight) can be inferred from the known—level of prior effort. distance between the current and previous positions divided by For each observed catch rate we obtained two distinct geo- the elapsed time. We assumed that vessels with a velocity less statistical predictions of fishing effort from the VMS data. than 5.5 knots were fishing and used this time at speed to re- The first represents prior fishing effort at the location of the flect fishing effort (Gedamke et al. 2004). For most records the tow. The second represents the total fishing effort that occurred elapsed time was close to 1 h, as the VMS systems were pro- over the course of the fishery at the location of the tow. Because grammed to report vessel position every hour. In a small fraction the trajectories of the tows were not known exactly, it was not GEOSTATISTICAL PREDICTION II 1111

possible to determine precisely how much effort had occurred north–south and east–west directions and found little evidence previously over the exact path of a tow. Note that the goal of of nonstationarity. We evaluated geometric anisotropy through this method was to provide a measure of relative effort at the lo- plots of the autocorrelation in various angles of rotation and var- cation of an observed tow rather than to determine the absolute ious axis scaling ratios and concluded that corrective measures area swept and bottom contact time, as was the goal of Mills were unwarranted. et al. (2007). After selection of the fishery tows we obtained an empiri- To obtain the predicted prior and total fishing effort, we cal variogram of sea scallop catch per nautical mile using the summed VMS fishing effort by assigning each record to the robust method of Cressie and Hawkins (1980). To fit theoreti- nearest node on a 1-nautical mile × 1-nautical mile grid cov- cal variograms we tested spherical, exponential, and Gaussian ering CAII. Grid nodes with no effort received a value of zero. model forms, but the spherical model fit best and is the only one This provided summed fishing effort on a spatial grid. These presented: point estimates of fishing effort were then used to obtain a variogram so that effort could be predicted anywhere within the C 3h h 3 fished area. γ(h) = C0 + − for h < a, 2 a a For each observed tow all prior effort over the entire fishery was accumulated over the spatial grid. A kriged prediction of γ(h) = C0 + C for h ≥ a, (1) fishing effort was then obtained at the midpoint of the tow to provide a smoothed estimate of the prior effort that had occurred where a is the range of autocorrelation in the data, C0 is the in its general proximity. To determine how much effort would nugget, C is the partial sill (C0 + C = the total sill), h is have subsequently occurred around the location of the tow, all the distance between the points, and γ(h)isthevalueofthe effort was summed over the season and the process repeated. semivariogram at a given distance. Hence, for each observed tow, it was possible to determine how To fit each of these models to the empirical semivariogram, much effort had occurred prior to that tow and how much effort we employed a weighted least-squares function (Cressie 1993): was going to occur after it. For example, if a tow had a prior fishing time of 5 h and a total fishing time of 100 h, this tow K γˆ(h( j)) 2 occurred within the first 5% of the total effort that would occur |N(h( j))| − 1 , (4) γˆ(h( j); ) at that location. J=1 Obtaining initial tows.—In this relatively short fishery opening, during which depletion was expected to be substantial, we where K is the number of variogram distance bins or lags, wanted to obtain catch rates that reflected abundance prior to the |N(h( j))| is the number of pairs of points in each lag, θ is fishery opening as closely as possible. Given the extremely high the set of variogram parameters (nugget, partial sill, and range), level of observer coverage (30% of vessels) in the opening of γ(h( j)) is the modeled semivariance value at distance bin h(j), CAII, we felt that this could be accomplished by selecting tows and γˆ(h( j)) is the observed semivariance value at distance bin that had occurred when there was very little prior accumulated h(j)). Fits were obtained in SPLUS 2000 (Mathsoft, Inc.) using effort. We used both a relative-effort criterion (which selected nonlinear bounded minimization. Spherical theoretical models tows with less than 5% of the accumulated effort at the location were selected on the basis of the smallest objective function for of the tow) and an absolute-effort criterion (which selected tows both the fishery and survey data sets (described below; Table 1; with less than 12.5 h of accumulated effort). The absolute-effort Figure 2).

Downloaded by [Department Of Fisheries] at 19:14 20 July 2015 criterion was necessary because the second tow in a particular Comparison with fishery-independent survey.—Kriged pre- location with little effort would have been excluded according dictions of sea scallop CPUE were obtained with the best-fitting to the 5% criterion, even though there had been very little total variogram model and predicted on a 30-nautical mile × 30- prior effort. The absolute criterion of 12.5 h represented the 5th nautical mile grid over the exemption area of CAII. We compare percentile of effort accumulated over the entire fishing area as these estimates with kriged estimates from a survey conducted well as a low level of total effort covering an area ≤9.8% of the over 6 weeks in August–October 1998, prior to the fishery open- area over which VMS data were accumulated (a circle with a ing. This survey was a collaborative effort between the scallop radius of 1 nautical mile). industry and the scientific community and consisted of 547 ob- Meeting assumptions and estimation of sea scallop CPUE servations from random and systematic grid locations. These variograms.—Geostatistical analysis assumes stationarity, i.e., observations were collected on six commercial scallop vessels that the spatial process has a constant expected mean and vari- using gear similar to that used in the 1999 fishery except that ance over the entire region (Cressie 1993). An additional as- the tows were 10-min straight-line tows at 5 knots. sumption is that of isotropy, or that the correlation between For comparison with the fishery data in 1999, we converted observations at locations is the same in all directions (Webster the numbers of sea scallops in 1998 to the numbers in 1999 and Oliver 2001). We examined the data sets for evidence of by incrementing them for growth and decrementing them nonstationarity (trend) by plotting the data values against the for mortality. Scallops were assigned to 0.196-in shell-height 1112 WALTER ET AL.

TABLE 1. Objective function value, estimated range, sill and nugget, kriged mean CPUE, and pointwise correlations for three variogram models for the ﬁshery data set.

Variable Spherical variogram Exponential variogram Gaussian variogram Objective function value 2,317.13 2,888.44 2,615.4 Estimated range (nautical mile) 9.176 3.331 4.465 Sill 6.93 8.30 5.93 Nugget 3.28 2.14 4.29 Kriged mean (bushels per nautical mile) 4.78 4.786 4.812 Pointwise correlation with spherical model predictions 1 0.980 0.982

bins, and the numbers were adjusted for size-selectivity using shell height per bushel (number of scallops per bushel = the selectivity curve of DuPaul et al. (1989). The size at the 9214.3·(shell height)−2.803; r2 = 0.77, df = 1,034). Hereafter midpoint of each bin was converted to age using a von Berta- these converted tow observations will be referred to as the lanffy growth model obtained from National Marine Fisheries survey tows. Identical conversions have been used with similar Service surveys of Georges Bank in 1998, namely, shell height data (Gedamke et al. 2004) and to project future abundance for (in) = 6.378 ·{1 − exp(−0.3374 · [age − 1)]}. The age was setting quotas (NEFSC 2007). advanced 1 year and a new shell height after 1 year of growth Prediction area versus the degree of extrapolation—As de- was obtained. The numbers of scallops were decremented for scribed in the companion simulation paper (Walter et al. 2014), 1 year of natural mortality ( = 0.1/year; Merrill and Posgay the kriging variance provides a means of determining the de- 1964) and then expanded using the selectivity curve to reﬂect gree to which a prediction is a product of adjacent spatial in- the expected catch of scallops at length in 1999. We converted formation or extrapolation of distant data points. This degree of these numbers into bushels per tow using a regression obtained extrapolation (equation 9 in Walter et al. 2014) is a function of from the 1998 survey of the number of scallops of a given the strength of autocorrelation and the number and proximity Downloaded by [Department Of Fisheries] at 19:14 20 July 2015

FIGURE 2. Empirical and ﬁtted variograms for the 1998 survey data and the ﬁshery catch rates; nm = nautical miles. GEOSTATISTICAL PREDICTION II 1113

FIGURE 3. (a) Kriged sea scallop abundance (bushels per nautical mile) for the fishery data set (contours of relative abundance are shown); (b) kriged survey abundance; (c) fishery minus survey predictions at all prediction points; and (d) total fishing effort as obtained from VMS records (shading) and 2,755 fishery tows (points). The symbols in panels (a), (b) and (d) represent individual dredge tows.

of adjacent samples and varies from 0% to 100%. We plotted RESULTS contours of the degree of extrapolation for the fishery CPUE. Extrapolation is important because as the degree of extrapola- Obtaining Initial Tows tion approaches 100% the kriged prediction converges on the Of the 2,755 total tows, 289 occurred with ≤5% of the total mean of the observed data. Thus, in the absence of data within fishing time at that location. The union of the two criteria (ei- the range of autocorrelation, kriging predicts the sample mean ther less than 5% locally or less than 12.5 h total) gave 1,076 abundance in areas beyond the range of autocorrelation. observations, which we will refer to as the fishery tows in the Downloaded by [Department Of Fisheries] at 19:14 20 July 2015 Some areas of high effort were not observed until substantial remainder of this paper. It is important to note that these tows effort had occurred. Due to our desire to assign observed tows did not necessarily correlate with the timing of the fishery, i.e., to a common level of prior effort, some important fishing areas an initial tow at a location could occur at the end of the season were left with no observations. Unlike gaps in coverage created if it took that long until the area was visited by a vessel. Tow by the absence of catch and effort, these are problematic because locations were recorded only to the nearest minute of latitude effort did occur but the observed CPUE may only reflect abun- and longitude, so some tows shared a midpoint position. In these dance after substantial depletion. The kriging variance coupled cases we used only the first (in time) tow at a location. Because with VMS data provides a means of identifying these areas by our criteria for determining initial tows required approximation plotting contours of the sum of the relative effort (on a scale of and were prone to bias from some level of prior removal, we 0% to100%) and the degree of extrapolation. The values range explored the sensitivity of the results to various cutoff levels of between 0 and 200 but are useful in a relative sense to identify prior effort. These tows covered much of the survey area, though areas that stand out as locations of critical uncertainty. They selection according to prior effort left some gaps in observed are uncertain because of the absence of adjacent information samples in areas where there was high fishing effort (Figure 3a). (an uncertainty shared with unfished areas) but are critical to Also, no tows were observed in the far western region, where interpreting overall catch rates because they are heavily fished. little to no fishing effort occurred. 1114 WALTER ET AL.

TABLE 2. Kriged means, prediction errors, and arithmetic means and standard errors for the survey and ﬁshery data.

Data Kriged mean Kriging prediction error Arithmetic mean Arithmetic standard error Fishery 4.78 2.40 5.95 4.07 Survey 5.11 2.28 5.15 5.53

Variogram Fits of autocorrelation where the geostatistical predictions reverted The variogram of the total effort captured the spatial distribu- to the arithmetic mean. tion of fishing cumulated over the entire season. An exponential variogram model was chosen over a spherical model on the ba- Sensitivity of the Kriged Means to Effort Cutoff sis of objective fitting criteria described in greater detail below. Because the effort level cutoff was used to approximate the The parameter estimates for the fitted model comprised a sill unfished conditions, we explored the sensitivity of the kriged of 14,150,000, a nugget of 470,000, and a range of 4 nautical and arithmetic means to varying the percentage of prior effort miles. allowed on the location of a tow (Figure 4). Increasing the level Spherical variogram models provided the best fit to the both of prior effort reduced both the kriged and arithmetic means, as fishery and survey data sets. However, the parameters differed greater effort brought in observations with greater levels of de- for the two data sets, notably in the magnitude of the sill and the pletion. Nevertheless, the kriged means were consistently lower range of autocorrelation. The fishery CPUE had an estimated than the arithmetic means, reflecting the differential weighting range of 9.147 nautical miles, shorter than the estimated survey of sample observations. range (11.453 nautical miles) and a total sill (partial sill + Sensitivity of the Kriged Means to Variogram nugget) approximately 55% of the survey variogram total sill Functional Form (Figure 2). Since the empirical variogram was estimated from data with error, there is no certainty that the best-fitting theoretical model Geostatistical Prediction truly represents the spatial autocorrelation of the process. For We obtain kriged abundances from the survey data on the this reason we obtained geostatistical abundance estimates with same 30 × 30 prediction grid as for fishery tows (Figure 3b). each of three fitted theoretical models and found minor differ- Geostatistical predictions of sea scallop CPUE measured in ences in the overall predicted mean for fishery CPUE (Table 1), bushels per nautical mile for the entire study area for the fishery indicating that the functional form of the variogram had little and survey data differed by only 6.5% (4.78 and 5.11, respec- substantive impact upon predicted CPUE. tively; Table 2). However, the arithmetic mean values for the two data sets showed a divergence of 15.5%, indicating that the Prediction Area versus the Degree of Extrapolation spatial weighting of kriging for the fishery data gave an estimate As observed in the plot of VMS coverage, the fishery did more similar to that of the survey. not fish in the far western area of CAII and consequently no To compare the two spatial maps, we subtracted kriged from predicted values on each grid node and plotted the differences (Figure 3c). The survey data provided more extensive cover-

Downloaded by [Department Of Fisheries] at 19:14 20 July 2015 age of the entire region and showed well-defined areas of high CPUE in the central area and on the eastern edge. The areas of highest discordance were at the far eastern edge, where the fishery data had gaps in observations. Though observer coverage eventually extended to almost the entirety of the fished area, there was substantial prior effort for the tows in some locations even within the 5% cutoff. In particular, for the tows in two areas of extremely high effort (dark shaded areas on the far right and at the top center of Figure 3d) between 156 and 164 h of effort occurred before a tow had been observed. Thus, some observations would have been influenced by prior depletion and may not have reflected initial abundance. The effects of this can be seen in the lower predicted abundances from the fishery data in the far right and the upper area (Figure 3c). In contrast, for the far western unfished region, the fishery predictions were higher FIGURE 4. Sensitivity analysis of kriged (K) and arithmetic mean (A) CPUE than the survey values, as prediction extended beyond the range as a function of the percentage of total prior VMS fishing effort allowed. GEOSTATISTICAL PREDICTION II 1115

FIGURE 5. Contours of percent extrapolation, deﬁned as 100 times the kriging variance divided by the sum of the variogram nugget and sill. Locations arethe 1,076 initial ﬁshery tows.

tows were observed from this area (Figure 3d). The contours fishery observations without substantial prior effort, such that of the degree of extrapolation for the fishery data (Figure 5) they could not have reasonably been attributed to initial unfished indicate that areas in the far upper left were beyond the range conditions. of autocorrelation. Predictions at these locations reverted to the arithmetic mean and represented 100% extrapolation of catch rates in the observed locations. Where the degree of extrapolation was relatively low (<50%), the mean kriged fishery

Downloaded by [Department Of Fisheries] at 19:14 20 July 2015 and mean kriged survey values were correlated (Figure 6). Be- yond this level of extrapolation, however, the kriged survey values declined as predictions largely extended into the unfished areas, whereas the kriged fishery predictions increasingly became a product of extrapolation and converged on the arithmetic sample mean. Given that the biomass of sea scallops was not evenly distributed, 86% and 96%, respectively, of the abundance estimated from the survey occurred within areas bounded by the 35% and 50% contours in Figure 5, indicating that almost all of the available biomass could be contained within an area where little extrapolation of the fishery data would be necessary. The contours of the sum of the degree of extrapolation and FIGURE 6. Comparison of the kriged means obtained from the fishery and relative fishing effort (Figure 7) identify two locations of highest survey data sets as they are increasingly a product of extrapolation. Predictions critical uncertainty, one in the top center and the other at the far are averaged for bins partitioned according to the percent extrapolation for the right. These areas had extremely high effort (Figure 3d) but few fishery predictions. 1116 WALTER ET AL.

FIGURE 7. Contours of the degree of extrapolation plus the percent of total VMS effort, which identify areas of critical uncertainty due to high effort and a high level of extrapolation.

DISCUSSION These results mirror simulation results (Walter et al. 2014) We obtained several important results regarding the perfor- that suggest that the geostatistical mean will be less biased than mance of geostatistical methods for obtaining an index of abun- a sample mean whereby all data points receive equal weight dance using fishery data. First, we found that the sea scallop under conditions of sample selection similar to those of fishery- observer data provided a variogram that was different from the dependent catch rate data. Furthermore, when changes in the survey data due to the lower sill but similar in the range of auto- area fished over time represent a substantial problem (Walters correlation, thus providing a variogram that generally reflected 2003; Carruthers et al. 2011), the ability to fill in spatial gaps will provide a more robust treatment of CPUE data. Geostatistical Downloaded by [Department Of Fisheries] at 19:14 20 July 2015 the most critical component of spatial autocorrelation necessary for geostatistical modeling. Next, the geostatistical esti- treatment of fishery data also readily facilitated mapping, which, mator produced a much lower mean abundance than the arith- in this case, was quite similar to the survey map of sea scallop metic sample mean, indicating that geostatistical methodology abundance. Lastly, the methodology of partitioning observed reduced the influence of the clustered and repeated observations catch rates using VMS data elucidated some of the possibili- common in fishery-dependent data. Furthermore, while the geo- ties and limitations of obtaining relative abundance information statistical mean abundance from the fishery data was lower than from fishery-dependent data that are valuable considerations for that from the survey data, it closely tracked abundance in areas future applications. with low levels of data extrapolation. It is rare to use fishery- As with any model-based predictive method, geostatistics dependent CPUE to estimate absolute abundance; it is much requires obtaining a model for prediction, in this case the vari- more common to use it as a relative abundance measure and ogram. While geostatistical prediction does not require design- then to estimate fishery catchability within a stock assessment based or random sampling, obtaining the variogram requires that model. In these situations a geostatistical treatment of fishery- samples be at least representative of the spatial process of independent catch rates is likely to better track abundance than terest (Webster and Oliver 2001), which fishery-dependent data CPUE derived from a generalized linear model (in essence an may not ensure (Paloheimo and Dickie 1964). In this applica- arithmetic mean). tion, the observer data provided a variogram with a similar range GEOSTATISTICAL PREDICTION II 1117

but a reduced sill. The range is the most critical variogram pa- This reversion to the mean is generally a convenient property of rameter for kriging because it affects the absolute distance over kriging in most situations; in this case, however, the mean of the which prediction can be made and it has the greatest potential fished areas is not representative of much of the unfished areas. to alter the predicted abundance. Conversely, in the central areas of extremely high fishing ef- In contrast, the sill and nugget parameters do not influence fort, the fishery data predicted lower abundance in the central the geostatistical mean other than through their effect upon the ridge than the survey. This result was incongruous with the ex- range when all three parameters are jointly estimated. The sill pected results but was due to two factors: allowing some prior and nugget scale the absolute value of the kriging prediction depletion to occur and the gaps in sample coverage in areas of variance. As the kriging variance is only used in a relative man- high sea scallop densities, fishing effort, and exploitation rates. ner to identify areas of uncertainty, this relative scaling has neg- Particularly in the high-abundance area on the far eastern edge, ligible impact upon the kriged means. The reduced sill for the few observed samples were included in the set of initial catches fishery data matched the simulation results that predict lower and low abundance was predicted on the basis of adjacent sam- sills in the biased data due to underestimation of variability ples. Though observer coverage eventually extended to some of (Silliman and Berkowitz 2000). Most tows occurred in areas of these areas of high effort, there were few observed tows before higher abundance, reducing the frequency of low tows. Also, 60–80% of the effort had been expended, so these observations commercial tow lengths (8 km, but rarely straight-line) were likely suffered from substantial prior depletion and would not longer than the survey (1.8 km straight-line), so they integrate have reflected initial abundance. the patchy sea scallop distribution over a greater area. In practice, it may be unlikely that geostatistics (or any other The variogram functional form with the best fit was a spher- model-based methodology) will fill in very large gaps in sample ical model for both the survey and the fishery. Kriged means coverage. It is in this situation that the geostatistical prediction were relatively insensitive to variogram functional form, largely variance becomes valuable in identifying the areas of greatest because the different models fitted to the same data provided uncertainty. Though less critical for assessing fishery removals similar spatial patterns with only slight differences at short spa- because of low effort, areas where the extrapolation level was tial scales. The potential for substantial bias occurs at longer high (the far western edges of Figure 7) are important to include spatial scales approaching or exceeding the true range of auto- for developing time series of CPUE (i.e., the spatial gap-filling correlation. For each of the three variogram models, it appears requirement of Walters [2003]). For bycatch monitoring, catch the sea scallop observer data provided ranges of autocorrelation rates obtained under the assumption that the observed areas similar to the survey, even if the sill was underestimated. represent unobserved locations could be misleading (Lewison If we assume that the 1998 survey extrapolated to 1999 abun- and Crowder 2003) when substantial fishing effort occurs in dances represented our best estimate of true relative abundance unobserved locations. In a bycatch monitoring situation, identi- (or at least the relative abundance upon which the 1999 quota fication of the areas of “critical uncertainty” (Figure 7), where was set), then the geostatistical estimator provided an overall both the degree of extrapolation and the level of fishing effort CPUE only 6% lower than the survey. In contrast, the arith- are high, may be critical for evaluating the adequacy of bycatch metic sample mean of the fishery data was 15.5% higher than estimates and observer sample coverage. the arithmetic mean of the survey data, indicating that the com- While there are numerous methods of standardizing fishery- bination of kriging and using only the first observation at a dependent catch rate data to account for lack of control of the location reduced the influence of clustered observations in lo- sampling process (Maunder and Punt 2004), most rely upon cations of high abundance. Furthermore, the map of relative the assumptions of linear modeling theory, i.e., that samples

Downloaded by [Department Of Fisheries] at 19:14 20 July 2015 abundance qualitatively resembled the survey map. In the hypo- are independent, identically distributed, and collected without thetical situation in which there was no survey, the geostatistical a selection bias. Geostatistics provides a means of spatially prediction obtained with observations partitioned by prior effort weighting observations, thus partially reducing the influence represents a substantially improved treatment of the data versus of selection bias and providing a means of extrapolation into alternatives which might assume independent observations. unsampled areas, at least to the extent of the range of autocorre- Nevertheless, some substantial differences remain between lation. Furthermore, geostatistical prediction can be combined the two geostatistical predictions involving extrapolation of the with traditional CPUE treatments based upon linear models by arithmetic mean into unfished areas beyond the range of auto- standardizing to account for factors such as vessel type, gear correlation and lower predicted fishery abundances in a heavily configuration, and seasonal effects prior to modeling the var- fished central region. At the far western edges of the open area iogram and making geostatistical predictions (Petitgas 2001). of CAII there was very little effort, presumably because there Furthermore, geostatistics provides a map of relative abundance were few sea scallops there, an observation corroborated by the that is useful in understanding the dynamics of fishery catch survey (Figure 3b). In these areas predictions were increasingly and bycatch (Kulka et al. 1996) and allows extrapolation of in- based on extrapolation—to the point that at the range of autocor- formation into unsampled locations, a desirable property when relation they reverted to the arithmetic mean, which was much fishery-dependent data leave gaps in sample coverage. Even for higher than the abundance in this area based on the 1998 survey. fishery-independent survey data, a similar type of data-informed 1118 WALTER ET AL.

extrapolation may be useful when there are gaps in sample cov- Merrill, A. S., and J. A. Posgay. 1964. Estimating the natural mortality rate of erage (NEFSC 2007). sea scallop. Research Bulletin of the International Commission of Northwest Atlantic Fisheries 1:88–106. Mills, C. M., S. E. Townsend, S. Jennings, P. D. Eastwood, and C. A. Houghton. ACKNOWLEDGMENTS 2007. Estimating high-resolution trawl fishing effort from satellite-based ves- We thank William DuPaul, Courtney Harris, Roger Mann, sel monitoring system data. ICES Journal of Marine Science 64:248–255. Murawski, S. A., R. Brown, H.-L. Lai, P. J. Rago, and L. Hendrickson. 2000. Paul Rago, and Doug Vaughan for reviewing this manuscript. Large-scale closed areas as a fishery-management tool in temperate ma- This research was conducted by J.F.W. while he held a NOAA– rine ecosystems: the Georges Bank experience. Bulletin of Marine Science Sea Grant Joint Graduate Fellowship in population dynamics. 66:775–798. This is VIMS contribution 3399. NEFSC (Northeast Fisheries Science Center). 2007. 45th Northeast Regional Stock Assessment Workshop (45th SAW) report. Northeast Fisheries Science Center, Reference Document 07-16, Woods Hole, Massachusetts. REFERENCES Palmer, M. C., and S. E. Wigely. 2009. Using positional data from vessel mon- Carruthers, T. R., R. N. M. Ahrens, M. K. McAllister, and C. J. Walters. 2011. itoring systems to validate the logbook-reported area fished and the stock Integrating imputation and standardization of catch rate data in the calculation allocation of commercial fisheries landings. North American Journal of Fish- of relative abundance indices. Fisheries Research 109:157–167. eries Management 29:928–942. Cressie, N. A. C. 1993. Statistics for spatial data. Wiley, New York. Paloheimo, J. E., and L. M. Dickie. 1964. Abundance and fishing suc- Cressie, N., and D. M. Hawkins. 1980. Robust estimation of the variogram. cess. Rapports et Proces-Verbaux` des Reunions,´ Conseil International pour Mathematical Geology 12:115–125. l’Exploration de la Mer 155:152–163. Deng, R., C. Dichmont, D. Milton, M. Haywood, D. Vance, N. Hall, and D. Petitgas, P. 2001. Geostatistics in fisheries survey design and stock assessment: Die. 2005. Can vessel monitoring system data also be used to study trawl- models, variances, and applications. Fish and Fisheries 2:231–249. ing intensity and population depletion? The example of Australia’s north- Rivoirard, J., J. Simmonds, K. G. Foote, P. Fernandes, and N. Bez. 2000. Geo- ern prawn fishery. Canadian Journal of Fisheries and Aquatic Sciences 62: statistics for estimating fish abundance. Blackwell Scientific Publications, 611–622. New York. DuPaul, W. D., E. J. Heist, and J. E. Kirkley. 1989. Comparative analysis of Serchuk, F. M., and R. J. Smolowitz. 1980. Size selection of sea scallops by an sea scallop escapement/retention and resulting economic impacts. Virginia offshore scallop survey dredge. International Council for the Exploration of Institute of Marine Science, College of William and Mary, Contract S-K, NA the Sea, C.M. 1980/K:24, Copenhagen. 88EA-H-00011, Gloucester Point, Virginia. Silliman, S. E., and B. Berkowitz. 2000. The impact of biased sampling on the Gedamke, T., W. DuPaul, and J. M. Hoenig. 2004. A spatially explicit open- estimation of the semivariogram within fractured media containing multiple ocean DeLury analysis to estimate gear efficiency in the dredge fishery fracture sets. Mathematical Geology 32:543–560. for sea scallop. North American Journal of Fisheries Management 24: Walter, J. F., J. M. Hoenig, and M. C. Christman. 2014. Reducing bias and filling 335–351. in spatial gaps in fishery-dependent catch-per-unit-effort data by geostatis- Kulka, D. W., A. T. Pinhorn, R. G. Halliday, D. Pitcher, and D. Stansbury. 1996. tical prediction, I. Methodology and simulation. North American Journal of Accounting for changes in spatial distribution when estimating abundance Fisheries Management 34:1095–1107. from commercial fishing data. Fisheries Research 28:321–342. Walters, C. 2003. Folly and fantasy in the analysis of spatial catch rate data. Lewison, R. L., and L. Crowder. 2003. Estimating fishery bycatch and effects Canadian Journal of Fisheries and Aquatic Sciences 60:1433–1436. on a vulnerable seabird population. Ecological Applications 13:743–753. Warren, W. G. 1998. Spatial analysis for marine populations: factors to be Little, R. J. A., and D. B. Rubin. 2002. Statistical analysis with missing data, considered. Canadian Special Publications in Fisheries and Aquatic Sciences 2nd edition. Wiley, New York. 125:21–28. Maunder, M. N., and A. E. Punt. 2004. Standardizing catch and effort: a review Webster, R., and M. Oliver. 2001. Geostatistics for environmental scientists. of recent approaches. Fisheries Research 70:141–160. Wiley, Chichester, UK. Downloaded by [Department Of Fisheries] at 19:14 20 July 2015 This article was downloaded by: [Department Of Fisheries] On: 20 July 2015, At: 19:14 Publisher: Taylor & Francis Informa Ltd Registered in England and Wales Registered Number: 1072954 Registered office: 5 Howick Place, London, SW1P 1WG

North American Journal of Fisheries Management Publication details, including instructions for authors and subscription information: http://www.tandfonline.com/loi/ujfm20 Optimizing Fishing Quotas to Meet Target Fishing Fractions of an Internationally Exploited Stock of Pacific Sardine David A. Demera & Juan P. Zwolinskib a National Oceanic and Atmospheric Administration, National Marine Fisheries Service, Fisheries Resources Division, Southwest Fisheries Science Center, 8901 La Jolla Shores Drive, La Jolla, California 92037, USA b Institute of Marine Sciences, University of California, Santa Cruz, Earth and Marine Sciences Building, Room A317, Santa Cruz, California 95064, USA Published online: 04 Nov 2014. Click for updates

To cite this article: David A. Demer & Juan P. Zwolinski (2014) Optimizing Fishing Quotas to Meet Target Fishing Fractions of an Internationally Exploited Stock of Pacific Sardine, North American Journal of Fisheries Management, 34:6, 1119-1130, DOI: 10.1080/02755947.2014.951802 To link to this article: http://dx.doi.org/10.1080/02755947.2014.951802

PLEASE SCROLL DOWN FOR ARTICLE

Taylor & Francis makes every effort to ensure the accuracy of all the information (the “Content”) contained in the publications on our platform. However, Taylor & Francis, our agents, and our licensors make no representations or warranties whatsoever as to the accuracy, completeness, or suitability for any purpose of the Content. Any opinions and views expressed in this publication are the opinions and views of the authors, and are not the views of or endorsed by Taylor & Francis. The accuracy of the Content should not be relied upon and should be independently verified with primary sources of information. Taylor and Francis shall not be liable for any losses, actions, claims, proceedings, demands, costs, expenses, damages, and other liabilities whatsoever or howsoever caused arising directly or indirectly in connection with, in relation to or arising out of the use of the Content. This article may be used for research, teaching, and private study purposes. Any substantial or systematic reproduction, redistribution, reselling, loan, sub-licensing, systematic supply, or distribution in any form to anyone is expressly forbidden. Terms & Conditions of access and use can be found at http:// www.tandfonline.com/page/terms-and-conditions North American Journal of Fisheries Management 34:1119–1130, 2014 C American Fisheries Society 2014 ISSN: 0275-5947 print / 1548-8675 online DOI: 10.1080/02755947.2014.951802

ARTICLE

Optimizing Fishing Quotas to Meet Target Fishing Fractions of an Internationally Exploited Stock of Paciﬁc Sardine

David A. Demer* National Oceanic and Atmospheric Administration, National Marine Fisheries Service, Fisheries Resources Division, Southwest Fisheries Science Center, 8901 La Jolla Shores Drive, La Jolla, California 92037, USA Juan P. Zwolinski Institute of Marine Sciences, University of California, Santa Cruz, Earth and Marine Sciences Building, Room A317, Santa Cruz, California 95064, USA

Abstract Two stocks of Pacific Sardine Sardinops sagax migrate seasonally and synchronously along the west coasts of Mex- ico, the USA, and Canada. Landings from the two stocks are currently combined in U.S. assessments of the northern stock, but the stocks may be differentiated by their associated seawater habitats, which are predominantly characterized by different ranges of sea surface temperature. We compared the combined and temperature-differentiated landings of the two stocks in each country for the period 1993–2011, demonstrating how different attributions of the landings affected the estimated annual fishing fraction (F) for the northern stock. Using combined or stock- differentiated landings and assessed biomasses, we found that the current harvest control rule (HCR) for Pacific Sardine has not consistently maintained a total F below the U.S. target value because the “distribution” parameter (used to account for the northern stock’s proportion in the U.S. Exclusive Economic Zone [EEZ]), has not adequately accounted for northern stock landings in Mexico and Canada. We propose a refinement to the HCR, giving explicit consideration to the summed landings in Mexico and Canada, to more optimally set the annual U.S. quota. The performance of our method was compared with (1) the values of F that would have been achieved during the federal management period (2000–2011) if the U.S. quotas had always been met and (2) the generally lower actual values of F that were calculated using the default HCR formulation (1993–2011). We demonstrate that application of our method would permit more U.S. fishing for Pacific Sardine when the northern stock is large and predominantly located in the U.S. EEZ and would curtail U.S. fishing when a large proportion of the stock is present and fished in the Mexican EEZ, Canadian EEZ, or both. Downloaded by [Department Of Fisheries] at 19:14 20 July 2015

In the California Current, there are two migrating stocks of (∼90 nautical miles south of Ensenada), northward to south- adult Pacific Sardine Sardinops sagax (Clark and Janssen 1945; ern Alaska (northwest of Vancouver Island, British Columbia, Felix-Uraga´ et al. 2004; Smith 2005) that exhibit large fluc- Canada; e.g., Vrooman 1964; Hargreaves et al. 1994; Smith tuations in abundance (e.g., MacCall 1979; Hill et al. 2012) 2005). and distribution (e.g., Lo et al. 2011a; Demer et al. 2012; The seasonal north–south migrations of the two Pacific Sar- Garc´ıa-Morales et al. 2012). The southern stock spans sea- dine stocks are approximately synchronous within their respec- sonally from southern Baja California, Mexico (∼850 nauti- tive domains, resulting in somewhat segregated spatial distri- cal miles south of Ensenada; 1 nautical mile = 1.852 km), to butions and different spawning and life history characteristics Point Conception, California (∼100 nautical miles north of San for these stocks (Murphy 1966; Felix-Uraga´ et al. 2005; Smith Pedro); the northern stock extends from San Quint´ın, Mexico 2005). The latitudinal extent of the seasonal migration appears

*Corresponding author: [email protected] Received December 12, 2013; accepted July 16, 2014 1119 1120 DEMER AND ZWOLINSKI

to depend on the stock size (Zwolinski and Demer 2012) and of the stock-size-dependent migration, potentially little to all of is constrained by the dynamics of the stock’s potential oceano- the northern stock may be in or pass through U.S. waters. For graphic habitat. example, during the 2006–2011 spring surveys, the large majority of the northern stock was located off the U.S. west coast. Habitats of the Northern Stock A small proportion of the stock may have been present in wa- To predict the potential habitat and associated seasonal dy- ters off Mexico during spring (Zwolinski et al. 2011), whereas namics for the northern stock of Pacific Sardine, Zwolinski et al. fishery landings indicated that few to no northern stock fish (2011) developed a model parameterized with satellite-sensed were off Canada at that time (Zwolinski et al. 2011; Demer and sea surface temperature (SST), chlorophyll-a concentration, and Zwolinski 2014). In contrast, during the 2008 summer survey, gradient of sea surface height. The model showed that dur- Pacific Sardine migrated north (Demer et al. 2012), and few to ing spring, the potential habitat for Pacific Sardine is located no individuals from the northern stock were present off Mexico offshore of central and southern California (Figure 1). During (Zwolinski et al. 2011; Demer and Zwolinski 2014). Thus, the summer, the potential habitat is compressed along the coast, proportion of the northern stock that occurs within any U.S. extending northward to Vancouver Island. During winter, the management area is not constant because this stock migrates habitat distribution, coupled with the virtual absence of land- and potentially extends seasonally into Mexican and Canadian ings in the north, suggests that the Pacific Sardine are offshore waters. Furthermore, the proportions of Pacific Sardine catches as they return south. associated with each country do not equate to the potential habitat present in the respective regions, or the proportions of the Differentiated Landings stock in the respective exclusive economic zones (EEZs). The Oregon, Washington, and Vancouver Island fisheries are seasonal, generally occurring between May and November de- U.S. Harvest Control Rule pending on the migration, region, and allocation. During this pe- The Pacific Fishery Management Council (PFMC) currently riod, the potential habitat of the northern stock develops along manages the northern stock of Pacific Sardine. The biomass the northeastern Pacific coast from the south and progresses of the southern stock is neither managed nor assessed, and its northward. In the north, the seasonal habitat and migration of exploitation rate is unknown. Since January 2000, the PFMC has Pacific Sardine, as well as regulations and weather, cause fish- set the U.S. annual catch level for the northern stock by using a ing for this species to begin and end first off Oregon, then off harvest control rule (HCR), optionally formulated as a harvest Washington, and lastly off Vancouver Island (Zwolinski et al. guideline (HG), an overfishing level (OFL), or an allowable 2011). In contrast, to the south, fisheries off Ensenada, Mex- biological catch (PFMC 2011). To date, only the default HG ico, and San Pedro, California, can span all months (Demer and formulation has been used; HG is calculated as Zwolinski 2014). Although the U.S. currently only assesses the northern stock and the models include all of the Pacific Sar- HG = (B − C) · F · D, (1) dine landings from Ensenada to Vancouver Island (Hill et al. 2010, 2011, 2012), accurate apportionment of the landings data where B is the assessed biomass of age-1 and older (age-1 + ) from the northern and southern stocks is critical for successful fish of the northern stock as estimated for July 1 of the previous assessment and management of both stocks (Smith 2005). year; C is the cutoff (currently 150,000 metric tons), defined as Refining the method proposed by Felix-Uraga´ et al. (2004, the lowest level of estimated biomass at which harvest is allowed 2005), we (Demer and Zwolinski 2014) used indices of potential (i.e., C is used as a buffer to protect the spawning stock); F is the Downloaded by [Department Of Fisheries] at 19:14 20 July 2015 habitat and associated SST ranges to partition and attribute catch target fishing fraction (0.05–0.15; currently 0.15), which is the data from each fishing zone to each respective Pacific Sardine target proportion of northern stock biomass above the C to be stock. We showed that fishery landings from central California harvested by the fisheries; and D is the distribution parameter and northward are consistently from habitat associated with the (currently 0.87), defined as the designated proportion of B in migrating northern stock, whereas the landings at Ensenada and U.S. waters (PFMC 2011). By mathematical complement, 1 – San Pedro may represent either the southern stock or northern D (currently 0.13) is the designated proportion of B outside of stock depending on the local presence of suitable potential habi- the U.S. EEZ. tat. The catches off southern California were from the northern The OFL equation is stock during winter and spring and from the southern stock during summer, transitioning during the months of June–July and = · · , November–December. OFL B FMSY D (2)

Distribution of the Northern Stock where FMSY (currently 0.18) is the target F that allows for the The U.S. ﬁshing areas span both the spring spawning and maximum sustainable yield (MSY; PFMC 2011). The allowable summer feeding areas used by the northern stock during migra- biological catch equals the OFL multiplied by a fraction to tion. Depending on the time of the year and the latitudinal extent account for scientiﬁc uncertainty and accommodate risk policy FISHING QUOTAS FOR PACIFIC SARDINE 1121

FIGURE 1. Pacific Sardine biomass densities (metric tons per square nautical mile; 1 nautical mile = 1.852 km), averaged over 2 km intervals, off the west coast of the USA, as estimated from acoustic–trawl surveys conducted during spring 2006, 2008, 2010, and 2011 and summer 2008 (Demer et al. 2012, 2013; Zwolinski et al. 2012). Confined by their dynamic potential habitat (dashed lines), the fish were located offshore of central and southern California during spring and were close to the coast in the northeast Pacific during summer. The fish did not span their entire potential habitat, but they did fill more of the potential habitat when their population was larger. More specifically, both the area occupied ( = 41,567 km2 × biomass + 18,341, where biomass is in metric tons; R2 = 0.9788) and the proportion of their habitat occupied ( = 0.0896 × biomass + 0.0593; R2 = 0.9998) in high density (Lo et al. 2011b) increased as estimated biomass increased (Demer and Zwolinski 2014).

to protect against overfishing (i.e., when landings exceed the than 150 mm (Morales 1993; Castaneda˜ 2012). The Pacific OFL or when the fishing rate exceeds FMSY). Sardine fishery in Canada is restricted by a 15% harvest rate In each of the HCR formulations, it is currently assumed that for a 3-year running average of the estimated proportion of 87% of the northern stock of Pacific Sardine is resident in the the U.S.-assessed Pacific Sardine biomass that migrates into U.S. EEZ. This value was derived from an analysis of aerial Canadian waters (Ware 1999). In 2010, for example, the U.S. spotter logbook data for the period 1963–1992 (PFMC 1998). assumed that 87% of the northern stock resided within its EEZ, During the period analyzed, the residual portion of the northern Canada assumed that 27.2% of the northern stock migrated into stock’s biomass (i.e., 13%) was ascribed entirely to the Mexican its EEZ (DFO 2011), and Mexico made no assumption in this

Downloaded by [Department Of Fisheries] at 19:14 20 July 2015 EEZ because the northern stock had not yet resumed its seasonal regard. Therefore, under the current HCR, it is quite possible migration into the Canadian EEZ (PFMC 2011). If the assumed that the tri-national landings of the northern stock can exceed D-value is now incorrectly large, the HCR may be less conser- the biomass corresponding to the U.S. target F. vative than expected. Conversely, because U.S. assessment and management of Pacific Sardine assume that all of the landings Study Objectives at ports spanning from Ensenada to Vancouver Island are from In the present analysis, our principal assumption was that the the northern stock, and given that some of the U.S. landings are total tri-national landings (L) for the northern stock of Pacific probably from the southern stock (Demer and Zwolinski 2014), Sardine should be less than or equal to the U.S. target F biomass, the HCR may be more conservative than expected. Therefore, irrespective of the HCR formulation (e.g., L ≤ [B − C] · F or without explicitly accounting for the actions of the tri-national L ≤ B · FMSY ). The value of L is the sum of annual landings Pacific Sardine fishery and differentiation of the landings, the at Ensenada, Mexico (L Mexico); multiple U.S. ports (LUSA); total F for the northern stock is probably different than the U.S. and Vancouver Island, Canada (LCanada). First, we explored target. If so, the stock biomass and yield may be reduced due to whether the distribution parameter D has been sufficient to too much fishing (PFMC 1998) or the stock may not be exploited keep LUSA and L Mexico + LCanada below the proportional val- to the intended extent. ues of F indicated (or implied) by the HG and OFL formula- The Pacific Sardine fishery in Mexico is not regulated by tions (i.e., if and when LUSA > [B − C] · F · D or LUSA > quotas but is technically restricted to fish with SL values greater B · FMSY · D;orL Mexico + LCanada > [B − C] · F [1 − D] or 1122 DEMER AND ZWOLINSKI

L Mexico + LCanada > B · FMSY. [1 − D], respectively). If not, The maximum value of D (Dmax) was calculated, via mathe- and if the target F-values were exceeded during 1993–2011, we matical complement, by substituting LMexico + LCanada for LUSA aim to show that future U.S. quotas could be optimized (max- and 1 − D for D in equation (3) and rearranging: imized under our principal assumption) using any of the HCR + formulations by explicitly accounting for the “undifferentiated” LMexico LCanada Dmax ≤ 1 − . (6) (combined) northern and southern stock landings or the “differ- (B − C) · F entiated” northern stock landings at Ensenada and Vancouver + Island. We consider both the undifferentiated and differentiated If Dmax is less than D, then LMexico LCanada exceeds the landings because future consideration may be given to the pro- proportional F implied by the HCR. If Dmax is less than zero, + portions of Pacific Sardine landings at Ensenada and San Pedro then LMexico LCanada alone exceeds the target total F. that are attributed to the southern stock. Finally, we aim to show Similarly, Dmin and Dmax were calculated using the OFL for- that application of our method could not only reduce discrep- mulation (equation 4): ancies between the actual F and the target F but could also (1) L optimally increase U.S. quotas when the northern stock is large D ≥ USA min · (7) and primarily located within the U.S. EEZ and (2) inherently B FMSY reduce U.S. exploitation of the stock when large proportions of and the landings are at Mexico, Canada, or both. LMexico + LCanada Dmax ≤ 1 − . (8) METHODS B · FMSY Differentiation of fishery landings.—Because the annual landings of Pacific Sardine at Ensenada and San Pedro may For either HCR formulation (HG: equations 5 and 6; or OFL: include fish from the southern and northern stocks, a monthly equations 7 and 8), if Dmin exceeds Dmax then the total landings SST index associated with the potential habitat for the northern for the northern stock exceed the target total F biomass (i.e., > − · > · stock was used to ascribe the landings from these ports to either L [B C] F or L B FMSY , respectively). the northern stock or southern stock (see Demer and Zwolinski Determining the fishing fraction.—Based on the HG formu- 2014 for details). The undifferentiated or differentiated landings lation used by the PFMC from 2000 to 2011 (equation 1) and at Ensenada, U.S. ports, and Vancouver Island were summed to assuming that the northern stock is located entirely within the + − = estimate the total tri-national landings. EEZs of the USA, Mexico, and Canada (i.e., D [1 D] Exploring the effectiveness of the distribution parameter.— 1.0), values of the historic total F for 1993–2011 were calculated The undifferentiated and differentiated landings were used to by explore whether D in the HCR has effectively kept L and USA L + L + L L + L below the proportional target values of F. F = Mexico USA Canada . Mexico Canada − (9) First, we note that since the PFMC implemented the HCR in B C 2000, Pacific Sardine landings at U.S. ports have not exceeded Likewise, the annual proportions of F associated with land- the annual quota (PFMC 2011; Hill et al. 2012). Therefore, after ings at Mexico (F ), the USA (F ), and Canada (F ) ≤ Mexico USA Canada substituting LUSA HG for HG in equation (1) and rearranging, were calculated by we have LMexico Downloaded by [Department Of Fisheries] at 19:14 20 July 2015 ≤ ≤ − · · . FMexico = , (10a) LUSA HG (B C) F D (3) B − C

Similarly, after substituting LUSA ≤ OFL in equation (2) and rearranging, we obtain L F = USA , (10b) USA B − C LUSA ≤ OFL ≤ B · FMSY · D. (4) and The minimum value of D (Dmin) is calculated by rearranging equation (3): LCanada F = . (10c) Canada B − C LUSA Dmin ≥ . (5) (B − C) · F We did not also calculate the historic values of FMSY because to date, the PFMC has only used HG (i.e., the default HCR If Dmin exceeds D, then LUSA exceeds the proportional F formulation). indicated by the HCR. If Dmin exceeds 1.0, then LUSA alone Optimizing U.S. quotas.—For the future, assuming that the exceeds the target total F. total tri-national F for the northern stock of Paciﬁc Sardine FISHING QUOTAS FOR PACIFIC SARDINE 1123

should not exceed the target F (i.e., LMexico + LUSA + LCanada ≤ landings comprise an appreciable proportion (11–15%) of the F · [B − C]), optimal values for LUSA could be calculated by total landings. On average, during 1993–2011, the U.S. fishery rearranging equation (9) as accounted for 53% (SD = 12%) of the undifferentiated landings of Pacific Sardine. LUSA ≤ F · (B − C) − (LMexico + LCanada) . (11) Differentiated Fishery Landings of the Northern Stock Differentiated landings of Pacific Sardine in Mexico and in Similarly, optimal values for LUSA could be calculated based on the OFL formulation: the USA were reduced from the aforementioned estimates, and the proportions of the catch landed in Canada were consequently increased (Table 1). For the northern stock, peak annual land- LUSA ≤ FMSY · B − (LMexico + LCanada) . (12) ings in Mexico were less than 40,000 metric tons and constituted In equations (11) and (12), the summed landings at Ensenada less than 50% of the total landings—and often, since 2001, less and Vancouver Island serve to limit the total F to the target value. than 20% of the total. The peak U.S. landings of the northern The practical challenge is to forecast this sum. If the annual val- stock were approximately 107,000 metric tons in 2007, but even ues for L + L are just stochastic estimates of the smaller landings of the northern stock comprised larger propor- Mexico Canada > historic mean landings (i.e., no autocorrelation or trend), then a tions ( 80%) of the total catch in 1998, 2002, and 2004–2007. mean (e.g., long-term, short-term, arithmetic, or geometric) or Stock differentiation did not affect the landings of the north- median value should serve to stabilize the total F about the tar- ern stock in Canada, but the proportion of Canadian landings get value. However, Pacific Sardine biomass and migration are among the total catch did increase to between 15% and 23% dur- highly variable and exhibit trends. Consequently, for the period ing 2009–2011. On average, during 1993–2011, the U.S. fishery accounted for 70% (SD = 13%) of the differentiated northern 1993–2011, the annual values for LMexico + LCanada (Table 1) had positive autocorrelation coefficients (undifferentiated: r = 0.59, stock landings. 2 = = 2 = r 0.35; differentiated: r 0.69, r 0.47). For the PFMC Exploring the Effectiveness of the Distribution Parameter management period 2000–2011, the values were somewhat dif- Values of D and D were calculated by using the HG for- = 2 = = min max ferent (undifferentiated: r 0.46, r 0.21; differentiated: r mulation (equations 5 and 6), the annual assessment-estimated 2 = 0.73, r 0.53). Because the annual summed landings during B of age-1 + fish, a C-value of 150,000 metric tons, and an F these periods included some short-term trend information (i.e., of 0.15 (Table 1). During 1994–2011, Dmax ranged from 0.35 21–53% of the year-to-year variance is explained), the prior in 2011 to 0.79 in 2007 and was always less than the D of year’s value is the best alternative predictor tested (Figure 2). 0.87. During the entire study period, the mean Dmax was 0.53 For 1993–2011, the annual HG was retrospectively opti- (SD = 0.21). That is, LMexico + LCanada consistently exceeded mized by solving equation (11) with the target F (e.g., 0.15), the proportional target F biomass implied by the HG. In 1993, + the assessment-estimated B of age-1 fish, the current C (e.g., D was negative because the total landings at Ensenada and + max 150,000 metric tons), and LMexico LCanada from the prior year. Vancouver Island alone exceeded F (Figure 3a). With stock dif- The OFL was similarly optimized by solving equation (12) with ferentiation, Dmax ranged from 0.67 in 2011 to 0.94 in 1997 and the target FMSY (e.g., 0.18), the assessment-estimated B of age- 1998, and the mean Dmax was 0.85 (SD = 0.07; Table 1; Figure 1 + fish, and LMexico + LCanada from the prior year. Because the 3b). In other words, with stock differentiation, LMexico + LCanada historic landings did not necessarily reach the respective har- sometimes exceeded the proportional target F biomass implied vest quotas (Hill et al. 2012), the performance of our method by the HG and also exceeded the target F biomass on average. Downloaded by [Department Of Fisheries] at 19:14 20 July 2015 was evaluated against (1) the allowed values of F, substituting The Dmin and Dmax values were also calculated using the the actual U.S. quotas (Q; Table 1) for LUSA (equation 9; 2000– OFL formulation (equations 7 and 8), the annual assessment- 2011); and (2) the actual values of F calculated using estimates estimated B of age-1 + fish, and the FMSY of 0.18 (Table 1). of B from the 2012 assessment (equation 9; 1993–2011; Hill During 1993–2011, Dmax ranged from 0.45 in 1993 to 0.84 et al. 2012). in 2007, with a mean of 0.69 (SD = 0.09), and was always less than the D of 0.87 (Table 1; Figure 3c). In other words, RESULTS LMexico + LCanada consistently exceeded the proportional target F biomass implied by the OFL. With stock differentiation, Dmax Undifferentiated Fishery Landings ranged from 0.73 in 2011 to 0.95 in 1997 and 1998, and the Without differentiating the landings from just the northern mean was 0.88 (SD = 0.06; Table 1; Figure 3d). On average, stock of Pacific Sardine, landings in Mexico peaked at about with stock differentiation, LMexico + LCanada did not exceed the 70,000 metric tons in 1997, 2000, 2008, and 2011, comprising proportional target F biomass implied by the OFL. 41–60% of the total annual landings in those years (Table 1). Landings at U.S. ports exceeded 100,000 metric tons in 2002 Determining the U.S. Fishing Fraction and 2007, comprising about 68% and 77% of the total landings, The total F and the proportions that were associated with respectively. Only between 2009 and 2011 did the Canadian each country (FMexico, FUSA, and FCanada) were estimated 1124 DEMER AND ZWOLINSKI

TABLE 1. Estimated prior-year (July 1) biomass (B; metric tons) of age-1 + northern-stock Pacific Sardine from the 2012 assessment (Hill et al. 2012); the actual U.S. harvest quota (Q; metric tons); undifferentiated and stock-differentiated biomass landed (metric tons) at Ensenada, Mexico (LMexico), at U.S. ports (LUSA), and at Vancouver Island, British Columbia, Canada (LCanada), and their respective proportions of the total annual catch (Prop.); and the fishing fractions allocated to the USA (Dmin) and Mexico plus Canada (Dmax), determined using harvest guideline (HG) and overfishing limit (OFL) harvest control rule formulas (equations 1 and 2; PFMC 2011). For 1992, only the values needed to calculate the autocorrelation of LMexico + LCanada are shown. Mexico USA Canada HG OFL

Year BQLMexico Prop. LUSA Prop. LCanada Prop. Dmax Dmin Dmax Dmin Biomass, quota, and undifferentiated landings 1992 34,568 0 1993 324,550 18,144 32,045 0.66 16,192 0.34 0 0.00 −0.23 0.62 0.45 0.28 1994 507,320 9,072 20,877 0.62 12,701 0.38 0 0.00 0.61 0.24 0.77 0.14 1995 691,760 47,305 35,396 0.46 41,489 0.54 23 0.00 0.56 0.51 0.72 0.33 1996 915,256 34,791 39,065 0.53 34,056 0.47 0 0.00 0.66 0.30 0.76 0.21 1997 977,035 48,988 68,439 0.60 46,198 0.40 71 0.00 0.45 0.37 0.61 0.26 1998 998,922 43,545 47,812 0.54 41,056 0.46 488 0.01 0.62 0.32 0.73 0.23 1999 1,134,060 120,474 58,569 0.50 57,522 0.50 25 0.00 0.60 0.39 0.71 0.28 2000 1,333,310 186,791 67,845 0.48 72,497 0.51 1,721 0.01 0.61 0.41 0.71 0.30 2001 1,246,290 134,737 46,071 0.37 78,520 0.62 1,266 0.01 0.71 0.48 0.79 0.35 2002 1,032,760 118,442 46,845 0.31 101,367 0.68 739 0.00 0.64 0.77 0.74 0.55 2003 868,532 110,908 41,342 0.35 74,599 0.64 978 0.01 0.61 0.69 0.73 0.48 2004 634,081 122,747 41,897 0.30 92,613 0.67 4,438 0.03 0.36 1.28 0.59 0.81 2005 976,986 136,179 55,323 0.37 90,130 0.61 3,232 0.02 0.53 0.73 0.67 0.51 2006 1,107,780 118,937 57,237 0.38 90,776 0.61 1,575 0.01 0.59 0.63 0.71 0.46 2007 1,365,980 152,564 36,847 0.22 127,695 0.77 1,522 0.01 0.79 0.70 0.84 0.52 2008 1,356,860 89,093 66,866 0.41 87,175 0.53 10,425 0.06 0.57 0.48 0.68 0.36 2009 1,286,760 66,932 55,911 0.40 67,083 0.48 15,334 0.11 0.58 0.39 0.69 0.29 2010 1,106,180 72,039 56,821 0.39 66,891 0.46 22,223 0.15 0.45 0.47 0.60 0.34 2011 1,077,220 50,526 70,337 0.51 46,745 0.34 20,719 0.15 0.35 0.34 0.53 0.24 Mean 996,906 88,538 49,766 0.44 65,543 0.53 4,462 0.03 0.53 0.53 0.69 0.36 SD 290,304 49,608 13,821 0.12 29,889 0.12 7,188 0.05 0.21 0.24 0.09 0.16 Differentiated northern stock landings 1992 1,051 0 1993 9,469 0.43 12,400 0.57 0 0.00 0.80 0.47 0.84 0.21 1994 9,074 0.45 11,111 0.71 0 0.00 0.88 0.21 0.90 0.12 1995 11,184 0.24 34,558 0.76 23 0.00 0.89 0.43 0.91 0.28 1996 16,038 0.42 22,445 0.58 0 0.00 0.88 0.20 0.90 0.14 1997 8,911 0.28 23,266 0.72 71 0.00 0.94 0.19 0.95 0.13 Downloaded by [Department Of Fisheries] at 19:14 20 July 2015 1998 8,773 0.20 34,681 0.84 488 0.01 0.94 0.27 0.95 0.19 1999 37,861 0.48 41,189 0.52 25 0.00 0.77 0.28 0.81 0.20 2000 29,243 0.30 67,086 0.68 1,721 0.02 0.84 0.38 0.87 0.28 2001 13,547 0.18 56,800 0.79 1,266 0.02 0.92 0.35 0.93 0.25 2002 13,552 0.14 82,052 0.85 739 0.01 0.91 0.62 0.92 0.44 2003 18,487 0.22 64,846 0.77 978 0.01 0.85 0.60 0.88 0.41 2004 6,297 0.07 76,964 0.88 4,438 0.05 0.88 1.06 0.91 0.67 2005 15,583 0.16 75,334 0.80 3,232 0.03 0.87 0.61 0.89 0.43 2006 11,215 0.12 78,905 0.86 1,575 0.02 0.92 0.55 0.94 0.40 2007 19,919 0.16 106,748 0.83 1,522 0.01 0.89 0.59 0.91 0.43 2008 21,548 0.19 80,222 0.73 10,425 0.09 0.84 0.44 0.87 0.33 2009 24,718 0.24 63,773 0.64 15,334 0.15 0.79 0.37 0.83 0.28 2010 17,076 0.18 57,529 0.59 22,223 0.23 0.76 0.40 0.80 0.29 2011 31,751 0.34 41,642 0.44 20,719 0.22 0.67 0.30 0.73 0.21 Mean 17,066 0.25 54,291 0.70 4,462 0.05 0.85 0.44 0.88 0.30 SD 8,676 0.12 26,710 0.13 7,188 0.07 0.07 0.21 0.06 0.14 FISHING QUOTAS FOR PACIFIC SARDINE 1125 Downloaded by [Department Of Fisheries] at 19:14 20 July 2015

FIGURE 2. Alternative predictors for the sum of Paciﬁc Sardine landings at Ensenada, Mexico (LMexico), and Vancouver Island, British Columbia (LCanada), including the (a) prior year’s value, (b) all prior years’ arithmetic mean, (c) prior 3 years’ arithmetic mean, (d) all prior years’ geometric mean, (e) prior 3 years’ geometric mean, and (f) all prior years’ median. The 1:1 line (solid black line) indicates perfect prediction. Prediction bias is indicated by the difference between the slope of the least-squares ﬁt to the data and 1.0. The dashed lines indicate the 1:1 line ± 50%. The smallest root mean square deviation (RMS) = n − 2 / (RMS t=1 [yt yˆt ] n,whereyt is the actual value during year t, yˆt is the predicted value, and n is the number of annual predictions) indicates deviation from perfect prediction (i.e., imprecision). The prior year’s value had the best combination of low bias (slope ∼ 1.0) and high precision (low RMS) and was therefore the best candidate predictor. 1126 DEMER AND ZWOLINSKI

FIGURE 3. Annual estimates of the Pacific Sardine fishing fractions allocated to the USA (Dmin) and Mexico plus Canada (Dmax), calculated using the prior year’s (July 1) annual assessment-estimated biomass (B)ofage-1+ fish and (a) the harvest guideline (HG) formulation with annual undifferentiated landings; (b) the HG formulation with differentiated northern stock landings; (c) the overfishing limit (OFL) formulation with annual undifferentiated landings; and (d) the OFL formulation with differentiated northern stock landings. The Dmin values (dashed lines) were estimated using landings at U.S. ports. The Dmax values (solid lines) were estimated using landings at Ensenada, Mexico, and Vancouver Island, British Columbia, Canada. The value of Dmin exceeds Dmax when the total fishing fraction (F) exceeds the target F based on 2012 assessment-estimated B (Hill et al. 2012). In 1993, Dmax was less than zero (panel a) because the combined landings in Mexico and Canada exceeded the target F of 0.15. In 2004, Dmin exceeded 1.0 (panels a and b) because the U.S. landings alone exceeded the target F of 0.15 and the FMSY of 0.18.

annually using equations (9) through (10c), which were pa- Optimizing the U.S. Fishing Fraction Downloaded by [Department Of Fisheries] at 19:14 20 July 2015 rameterized with undifferentiated and differentiated landings Equation (11) was solved for optimized FUSA using prior-year from each country (LMexico, LUSA, and LCanada) and assessment values for LMexico and LCanada, the assessed B of age-1 + Pacific estimates (Hill et al. 2012) of northern stock B (Figure 4). Ir- Sardine from July 1 of the previous year, and an F of 0.15 (Figure respective of stock differentiation, F trended upward during 4c, d). When applied to the HG, our method stabilized the total 1994–2011 and peaked during (1) 2004 due to a peak in the F within ( ± 95% confidence interval) approximately ± 23.4% U.S. landings and (2) 2010–2011 due to increased landings in (undifferentiated stocks) or ± 20.7% (differentiated stocks) of Mexico and Canada (Figure 4). The undifferentiated F (Figure the target F of 0.15 (Figure 4a, c). For the management period 4a) exceeded the HG target F-value (0.15) during 1993, 2002– 2000–2011, compared to the potential tri-national fishing frac- 2006, and 2010. In 2002–2006, the rise in F was driven mostly tion calculated using equation 9 and substituting the U.S. harvest by LUSA. In contrast, the more recent increase in F resulted from quotas (Q; Table 1) for LUSA, our method reduced the root increases in LMexico and LCanada. With stock differentiation (Fig- mean square deviation (RMS) by 80% (RMS = 0.0142 versus ure 4b), F only exceeded 0.15 in 2004, and this was due to 0.0704) for the presently undifferentiated landings (Figure 4c). LUSA. Since then, because FUSA has decreased while FMexico and For the period 1993–2011, compared to the actual tri-national FCanada have increased, the F for the northern stock has remained fishing fraction (equation 9, Figures 4a, b), our method reduced relatively stable between approximately 0.10 and 0.13. the RMS by 77% (0.0153 versus 0.0663) for the presently FISHING QUOTAS FOR PACIFIC SARDINE 1127

FIGURE 4. Total fishing fraction (total F; solid black line from 1993 to 2011) and the values of F for Mexico (dashed line), the USA (dotted–dashed line), and Canada (dotted line), defined as the respective annual landings of Pacific Sardine divided by the prior year’s (July 1) assessment-estimated biomass ofage-1+ fish (Hill et al. 2012) minus the cutoff value (150,000 metric tons), and calculated based on (a) the actual harvest guideline (HG) quota and undifferentiated landings, (b) the actual HG quota and differentiated northern stock landings, (c) the optimized HG value and undifferentiated landings, and (d) the optimized HG value and differentiated northern stock landings. For comparison with the current harvest control rule, panels a and c show the total F (solid black line from 2000 to 2011) that would have occurred if the U.S. landings equaled the HG quota. The optimized HG values panels c and d (dotted–dashed line; calculated using equation 9) would ensure that the total F (solid black line) approximates the target F (0.15; gray horizontal line) of biomass above the cutoff.

undifferentiated landings (Figure 4c) and reduced the RMS by U.S. EEZ (0.57), as estimated using the aerial spotter data along 64% (0.0182 versus 0.0508) for the potentially differentiated with ﬁsh egg and larva data for 1951–1985 (PFMC 1998). When Downloaded by [Department Of Fisheries] at 19:14 20 July 2015 landings (Figure 4d). These results were also plotted in terms using the OFL, the average Dmax was 0.69 (SD = 0.09). of landed B computed with and without stock differentiation Calculated with the differentiated landings and the HG, Dmax (Figure 5). was less than 0.87 during 8 of the 19 years in the study. The Between 1993 and 2011, without differentiation, LUSA was average Dmax was 0.85 (SD = 0.07), and the lowest value (0.67) on average 10,239 metric tons (SD = 29,457 metric tons) below occurred most recently, in 2011 (Table 1). When calculated by the target F (Figures 4c, 5c); during 1993 and 2002–2007, F use of the OFL, Dmax was less than 0.87 during 5 of the 19 years; exceeded the target. With differentiation, LUSA was on average the average Dmax was 0.88 (SD = 0.06), and the lowest value 53,834 metric tons (SD = 30,111 metric tons) below the target (0.73) also occurred in 2011. Therefore, the constant value of D F (Figures 4d, 5d); during 2004, F exceeded the target. in the HCR did not sufﬁciently account for landings at Ensenada When Dmax was calculated with undifferentiated landings and and Vancouver Island, thereby creating the potential for the either the HG or the OFL (Figure 3a, c), Dmax for 1993–2011 target F or FMSY to be exceeded. was always less than the D-value of 0.87 derived from the aerial When calculated with undifferentiated landings and the HG, spotter data (1963–1992 data; PFMC 1998). Incidentally, when the target F was indeed exceeded in 1993, 2002–2006, 2010, using the HG, the average Dmax of 0.53 (SD = 0.12; Table 1) and 2011 (Figures 3a, 4a). When calculated with undifferenti- was slightly less than the average proportion of the stock in the ated landings and the OFL, the target FMSY was exceeded during 1128 DEMER AND ZWOLINSKI

FIGURE 5. Total harvested biomass (metric tons) of Pacific Sardine (solid black line) and harvested biomasses (landings L) for Mexico (dashed line), the USA (dotted–dashed line), and Canada (dotted line) based on (a) the actual harvest guideline (HG) quota and undifferentiated landings, (b) the actual HG quota and differentiated northern stock landings, (c) the optimized HG value and undifferentiated landings, and (d) the optimized HG value and differentiated northern stock landings. The biomass that is equivalent to the target fishing fraction (target F = 0.15) of the prior year’s (July 1) assessment-estimated (Hill et al. 2012) biomass of age-1 + fish minus the cutoff (150,000 metric tons) is shown (solid gray line). For comparison with the current harvest control rule, panels a and c show the total fished biomass (solid black line from 2000 to 2011) that would have occurred if the U.S. landings equaled the HG quota. The optimized U.S. landings (panels c and d; dotted–dashed line; calculated using equation 9) ensure that the total landings (solid black line) approximate the target F (0.15) of biomass above the cutoff value.

2004 (Figure 3c). These results would have been different if the mammals, ﬁshes, and seabirds) and other causes of natural at-

Downloaded by [Department Of Fisheries] at 19:14 20 July 2015 values used for B (i.e., from the 2012 assessment; Hill et al. trition. The performance of our method depends on the variance 2012) had been taken from another assessment (e.g., the 2011 and trend of the summed landings at the ports in Mexico and or 2010 assessment). Based on calculations with differentiated Canada (LMexico + LCanada). Although the summed annual land- landings and the HG, the target F was exceeded only during 2004 ings are positively autocorrelated, harvest quotas for Mexico (Figures 3b, 4b). Again, these results would have been different and Canada could be even better predictors of the landings there if B-values had been obtained from a different assessment, par- if such quotas existed and were known prior to setting the U.S. ticularly if the assessment had considered stock-differentiated quota. In that case, the quotas could be substituted for LMexico landings. and LCanada in equation (11) or (12). We recognize that the potential precedent of allocating the U.S. landings based on landings or quotas in Mexico and Canada DISCUSSION may draw some concern. This is because landings are influenced To set the U.S. quotas for Pacific Sardine such that the actual by the fish stock, the environment, markets, geopolitics, and F more consistently matches the target F or FMSY, we proposed other human factors; without cooperative management, the USA a practical method. In this method, the exploitation of the north- has no control over the international landings (PFMC 1998) and ern stock by Mexico and Canada, uncontrolled by the USA, is could be subjected to lower or unstable F. Nevertheless, due to essentially added to the mortality from predators (e.g., marine the variety of common goals (e.g., stabilizing socio-economic FISHING QUOTAS FOR PACIFIC SARDINE 1129

returns, sustaining exploitation rates, and preserving biodiver- sagax) in the California Current. ICES Journal of Marine Science 71:328– sity), international managers may be motivated to cooperate in 335. optimizing quotas with respect to target total F. Demer, D. A., J. P. Zwolinski, K. A. Byers, G. R. Cutter, J. S. Renfree, T. S. Sessions, and B. J. Macewicz. 2012. Prediction and confirmation of seasonal The abundance, distribution, and migration of the northern migration of Pacific Sardine (Sardinops sagax) in the California Current stock of Pacific Sardine in the California Current are highly ecosystem. U.S. National Marine Fisheries Service Fishery Bulletin 110:52– variable. Depending on the size and demographic structure of 70. the stock, Pacific Sardine migrate seasonally, spawning offshore Demer, D. A., J. P. Zwolinski, G. R. Cutter Jr., K. A. Byers, B. J. Macewicz, of southern and central California during spring and foraging and K. T. Hill. 2013. Sampling selectivity in acoustic-trawl surveys of Pacific Sardine (Sardinops sagax) biomass and length distribution. ICES Journal of off central and northern California, Oregon, Washington, and Marine Science 70:1369–1377. Vancouver Island during summer. The seasonal migration DFO (Department of Fisheries and Oceans Canada). 2011. Evaluation of Pacific spans the waters off Mexico, the USA, and Canada. To set the Sardine (Sardinops sagax) stock assessment and harvest guidelines in British annual U.S. quota for the northern stock, the HCR designates Columbia. Canadian Science Advisory Secretariat Science Advisory Report a proportion of the stock to be present in the U.S. EEZ. 2011/016. Felix-Uraga,´ R., V. M. Gomez-Mu´ noz,˜ C. Quinõnez-Vel´ azquez,´ and W. Garc´ıa- However, prorating a coastal pelagic fish stock’s target F based Franco. 2004. On the existence of Pacific Sardine groups off the west coast on the designated proportion in U.S. waters will not protect of Baja California and southern California. California Cooperative Oceanic the stock against high combined exploitation (PFMC 1998). Fisheries Investigations Reports 45:146–151. We developed a practical method to optimize the U.S. harvest Felix-Uraga,´ R., V. M. Gomez-Mu´ noz,˜ C. Quinõnez-Vel´ azquez,´ F. N. Melo- quotas irrespective of the HCR and its parameterization and to Barrera, K. T. Hill, and W. Garc´ıa-Franco. 2005. Pacific Sardine (Sardinops sagax) stock discrimination off the west coast of Baja California and south- approximate the target total F of the internationally exploited ern California using otolith morphometry. California Cooperative Oceanic northern stock of Pacific Sardine. Fisheries Investigations Reports 46:113–121. We demonstrated that our method, if used, could serve to Garc´ıa-Morales, R., B. Shirasago-German, R. Felix´ -Uraga, and E. L. Perez-´ stabilize the actual values of F about the target value, permit Lezama. 2012. Conceptual models of Pacific Sardine distribution in the Cal- more U.S. fishing of Pacific Sardine during most periods when ifornia Current system. Current Development in Oceanography 5:1–47. Hargreaves, N. B., D. M. Ware, and G. A. McFarlane. 1994. Return of the Pacific the northern stock is large and predominantly located in the Sardine (Sardinops sagax) to the British Columbia coast in 1992. Canadian U.S. EEZ, and curtail U.S. fishing during periods when a large Journal of Fisheries and Aquatic Sciences 51:460–463. proportion of the northern stock is present and fished in the Hill, K. T., P. R. Crone, N. C. H. Lo, D. A. Demer, J. P. Zwolinski, and B. J. Mexican EEZ, Canadian EEZ, or both. Although the USA does Macewicz. 2012. Assessment of the Pacific Sardine resource in 2012 for U.S. not currently assess or manage the southern stock, these methods management in 2013. NOAA Technical Memorandum NMFS-SWFSC-501. Hill, K. T., P. R. Crone, N. C. H. Lo, B. J. Macewicz, E. Dorval, J. D. McDaniel, could also be applied to that stock if it is present and exploited and Y. Gu. 2011. Assessment of the Pacific Sardine resource in 2011 for U.S. in U.S. waters. management in 2012. NOAA Technical Memorandum NMFS-SWFSC-487. Hill, K. T., N. C. H. Lo, B. J. Macewicz, P. R. Crone, and R. Felix-Uraga. 2010. Assessment of the Pacific Sardine resource in 2010 for U.S. management in ACKNOWLEDGMENTS 2011. NOAA Technical Memorandum NMFS-SWFSC-469. Lo, N. C. H, B. J. Macewicz, and D. A. Griffith. 2011a. Migration of Pacific We are grateful to Randy Cutter, Roger Hewitt, Sam Her- Sardine (Sardinops sagax) off the west coast of United States in 2003–2005. rick, Dale Sweetnam, Russ Vetter, Bill Perrin, Alec MacCall, Bulletin of Marine Science 87:395–412. and others for reviewing drafts of this paper as part of the Na- Lo, N., B. J. Macewicz, and D. A. Griffith. 2011b. Spawning biomass of Pacific tional Marine Fisheries Service (NMFS) Southwest Fisheries Sardine (Sardinops sagax) off U.S. in 2011. NOAA Technical Memorandum Science Center’s internal review process. We also thank Car- NMFS-SWFSC-486. MacCall, A. D. 1979. Population estimates for the warming years of the Pa- Downloaded by [Department Of Fisheries] at 19:14 20 July 2015 ryn de Moor, Sarah Kraak, and three anonymous reviewers for cific Sardine fishery. California Cooperative Oceanic Fisheries Investigations their thorough and constructive critiques. This article does not Report 20:72–82. necessarily reflect the official views or policies of the NMFS, Morales, G. J. 1993. [Mexican Official Norm 003-PESC-1993 to regulate the the National Oceanic and Atmospheric Administration, or the use of the following sardine species: monterrey, pina, crinuda, bocona, Department of Commerce. and japonesa, and the species of anchovy and mackerel using purse-seine methods in Pacific waters under federal jurisdiction, including the Gulf of California.] Mexico Distrito Federal, Fisheries Secretary, Mexico City. (In Spanish.) Available: www.conapesca.sagarpa.gob.mx/work/sites/cona/ REFERENCES resources/LocalContent/8739/21/003pesc1993SARDINA.pdf. (September Castaneda,˜ F. J. M. 2012. Fishery management plan for the fish- 2014). ery of small pelagics (sardines, anchovies, mackerel, and similar) off Murphy, G. I. 1966. Population biology of the Pacific Sardine (Sardinops northwestern Mexico. Available: http://www.dof.gob.mx/nota detalle popup. caerulea). Proceedings of the California Academy of Sciences 34:1–84. php?codigo=5276945. (September 2014). PFMC (Pacific Fishery Management Council). 1998. Amendment 8, appendix Clark, F. N., and J. F. Janssen Jr. 1945. Movements and abundance of the sardine B, options and analyses for the coastal pelagic species fishery management as measured by tag returns. California Department of Fish and Game Fishery plan. PFMC, Portland, Oregon. Bulletin 61:7–42. PFMC (Pacific Fishery Management Council). 2011. Coastal pelagic species Demer, D. A., and J. P. Zwolinski. 2014. Corroboration and refinement of a fishery management plan as amended through amendment 13. PFMC, Port- method to differentiate landings from two stocks of Pacific Sardine (Sardinops land, Oregon. 1130 DEMER AND ZWOLINSKI

Smith, P. E. 2005. A history of proposals for subpopulation structure in the Sardine stock. Proceedings of the National Academy of Sciences of the USA Pacific Sardine (Sardinops sagax) population off western North America. 109:4175–4180. California Cooperative Oceanic Fisheries Investigations Report 46:75–82. Zwolinski, J. P., D. A. Demer, K. A. Byers, G. R. Cutter, J. S. Renfree, S. T. Vrooman, A. M. 1964. Serologically differentiated stocks of the Pacific Sardine, Sessions, and B. J. Macewicz. 2012. Distributions and abundances of Pacific Sardinops caerulea. Journal of Fisheries Research Board of Canada 21:661– Sardine (Sardinops sagax) and other pelagic fishes in the California Current 680. ecosystem during spring 2006, 2008, and 2010, estimated from acoustic-trawl Ware, D. M. 1999. Life history of Pacific Sardine and a suggested framework surveys. U.S. National Marine Fisheries Service Fishery Bulletin 110:110– for determining a B.C. catch quota. Canadian Science Advisory Secretariat 122. Research Document 1999/204. Zwolinski, J. P., R. L. Emmett, and D. A. Demer. 2011. Predicting habitat to Zwolinski, J. P., and D. A. Demer. 2012. A cold oceanographic regime with optimize sampling of Pacific Sardine (Sardinops sagax). ICES Journal of high exploitation rates in the Northeast Pacific forecasts a collapse of the Marine Science 68:867–879. Downloaded by [Department Of Fisheries] at 19:14 20 July 2015 This article was downloaded by: [Department Of Fisheries] On: 20 July 2015, At: 19:14 Publisher: Taylor & Francis Informa Ltd Registered in England and Wales Registered Number: 1072954 Registered office: 5 Howick Place, London, SW1P 1WG

North American Journal of Fisheries Management Publication details, including instructions for authors and subscription information: http://www.tandfonline.com/loi/ujfm20 Growth and Mortality of Hatchery-Reared Striped Bass Stocked into Nonnatal Systems Jody L. Callihana, Charlton H. Godwinb, Kevin J. Dockendorfc & Jeffrey A. Buckeld a Department of Applied Ecology, North Carolina State University, Campus Box 7617, Raleigh, North Carolina 27695, USA b North Carolina Division of Marine Fisheries, Northern District Office, 1367 U.S. 17 South, Elizabeth City, North Carolina 27909, USA c North Carolina Wildlife Resources Commission, 1721 Mail Service Center, Raleigh, North Carolina 27699, USA d Department of Applied Ecology, North Carolina State University, Center for Marine Sciences Click for updates and Technology, 303 College Circle, Morehead City, North Carolina 28557, USA Published online: 30 Oct 2014.

To cite this article: Jody L. Callihan, Charlton H. Godwin, Kevin J. Dockendorf & Jeffrey A. Buckel (2014) Growth and Mortality of Hatchery-Reared Striped Bass Stocked into Nonnatal Systems, North American Journal of Fisheries Management, 34:6, 1131-1139, DOI: 10.1080/02755947.2014.951805 To link to this article: http://dx.doi.org/10.1080/02755947.2014.951805

PLEASE SCROLL DOWN FOR ARTICLE

Taylor & Francis makes every effort to ensure the accuracy of all the information (the “Content”) contained in the publications on our platform. However, Taylor & Francis, our agents, and our licensors make no representations or warranties whatsoever as to the accuracy, completeness, or suitability for any purpose of the Content. Any opinions and views expressed in this publication are the opinions and views of the authors, and are not the views of or endorsed by Taylor & Francis. The accuracy of the Content should not be relied upon and should be independently verified with primary sources of information. Taylor and Francis shall not be liable for any losses, actions, claims, proceedings, demands, costs, expenses, damages, and other liabilities whatsoever or howsoever caused arising directly or indirectly in connection with, in relation to or arising out of the use of the Content. This article may be used for research, teaching, and private study purposes. Any substantial or systematic reproduction, redistribution, reselling, loan, sub-licensing, systematic supply, or distribution in any form to anyone is expressly forbidden. Terms & Conditions of access and use can be found at http:// www.tandfonline.com/page/terms-and-conditions North American Journal of Fisheries Management 34:1131–1139, 2014 C American Fisheries Society 2014 ISSN: 0275-5947 print / 1548-8675 online DOI: 10.1080/02755947.2014.951805

ARTICLE

Growth and Mortality of Hatchery-Reared Striped Bass Stocked into Nonnatal Systems

Jody L. Callihan* Department of Applied Ecology, North Carolina State University, Campus Box 7617, Raleigh, North Carolina 27695, USA Charlton H. Godwin North Carolina Division of Marine Fisheries, Northern District Ofﬁce, 1367 U.S. 17 South, Elizabeth City, North Carolina 27909, USA Kevin J. Dockendorf North Carolina Wildlife Resources Commission, 1721 Mail Service Center, Raleigh, North Carolina 27699, USA Jeffrey A. Buckel Department of Applied Ecology, North Carolina State University, Center for Marine Sciences and Technology, 303 College Circle, Morehead City, North Carolina 28557, USA

Abstract Cross-stocking involves the use of fish from nonnatal sources to augment populations. This practice may not be effective, especially if fish from different populations are not well adapted to the environmental conditions of the areas intended for enhancement. Yet, the ecological consequences of cross-stocking have received little attention, particularly in coastal environments. We used tag return data (1990–2010) from an ongoing stock enhancement program to compare the growth and mortality of hatchery-reared Striped Bass Morone saxatilis of Roanoke River origin between their natal (Albemarle Sound estuary) and two nonnatal systems (Tar-Pamlico and Neuse rivers) in North Carolina. Despite their Roanoke River origin, stocked juveniles exhibited high fidelity (>90%) to nonnatal systems and similarly high growth as in their natal habitat (von Bertalanffy K values were statistically similar among systems and ranged from 0.54 to 0.61). However, time-at-liberty estimators of total mortality (Z) indicated stocked Striped Bass experienced significantly higher mortality in nonnatal (Z values, 0.48–0.51) versus natal (Z = Downloaded by [Department Of Fisheries] at 19:14 20 July 2015 0.33) systems. Therefore, while cross-stocking may not contribute to stock rebuilding, it appeared to be an effective management tool for supporting local put-and-take fisheries for this recreationally and commercially important species.

Due to practical constraints such as costs, hatchery proximity, 1990; Rulifson and Laney 1999; Norton et al. 2000). While there or difficulties obtaining natal broodstock (for imperiled popu- are obvious genetic concerns with this approach (e.g., outbreed- lations), stocking programs may use fish from nonnatal sources ing depression: Blankenship and Leber 1995; Bulak et al. 2004; for population enhancement (Lichatowich et al. 1999; Rulifson Ward 2006), few studies have evaluated the ecological conse- and Laney 1999; St. Pierre 1999). This practice of stocking fish quences of cross-stocking in coastal environments. For example, in nonnatal systems is referred to as “cross-stocking” (Murphy fish populations often develop beneficial local adaptations to

*Corresponding author: [email protected] Received March 11, 2014; accepted July 16, 2014 1131 1132 CALLIHAN ET AL.

their natal habitat such as unique bioenergetic properties that al- to multiple inlets and lower river flows, salinities are higher in low them to maximize growth in their natal area (Conover 1990; the Tar-Pamlico and Neuse rivers, especially near the mouths of Conover et al. 1997). Therefore, individuals of those populations these coastal rivers, where salinities can reach 15–20‰ (Epperly may not grow and survive as well when placed in nonnatal sys- and Ross 1986; Orlando et al. 1994). tems. Clearly, for any stocking program to be successful and Stocking program.—While hatchery-reared Striped Bass of contribute to fisheries and/or rebuild populations, stocked fish Roanoke River origin have been stocked as juveniles into the must exhibit favorable growth and survival and also demonstrate coastal waters of North Carolina for decades (Rulifson and high fidelity to the areas targeted for enhancement (Rimmer and Laney 1999), only since 1980 have fish been tagged by the Russell 1998; Jenkins et al. 2004; Hervas et al. 2010). North Carolina Division of Marine Fisheries (NCDMF) to per- Striped Bass Morone saxatilis have been extensively cross- mit an evaluation of stocking efforts. During this period (i.e., stocked in North America during the past century (Harrell et al. post-1980), broodstock were collected on the Roanoke River 1990; Woodroffe 2011). The first hatchery for this species was spawning grounds in spring (April–May), transported to the built in the 1880s along the Roanoke River in North Carolina, Watha State Fish Hatchery (Figure 1), and spawned in indoor which comprises the largest spawning run of Striped Bass south tanks. One-week-old larvae were then transported to the Eden- of the Chesapeake Bay (Worth 1884). Accordingly, Roanoke ton National Fish Hatchery (Figure 1) and placed in outdoor River broodstock have served as a major source of hatchery- (∼1 acre [0.4 ha]) ponds. This grow-out phase lasted through reared Striped Bass stocked into North Carolina waters and late fall (November–December), at which time juveniles were along the U.S. eastern seaboard. The current paradigm sug- harvested from the ponds for stocking purposes. Striped Bass gests that Roanoke River Striped Bass are a resident population were stocked in the Tar-Pamlico and Neuse rivers during the that generally remains in their natal estuary (Albemarle Sound) following 1.5 months, and ∼10,000 fish were released during throughout their lives, making only local spawning migrations each stocking event (total numbers stocked varied across years, during spring to their freshwater spawning grounds on the upper see Table 1). Tagged fish (∼3,000 per system) were released Roanoke River (Hassler et al. 1981). However, a recent study with ∼7,000 untagged counterparts in early to mid-December; by Callihan et al. (2014) revealed that some members of this the mean release date across years was December 11 and ranged population, mainly larger and older fish (TL > 850 mm, age from December 6 to 19. One stocking event occurred in Albe- > 10), are indeed migratory and engage in seasonal coastwide marle Sound each year from 1990 to 1996, and all stocked fish migrations (north in the summer, south in the fall–winter). were tagged (Table 1). Striped Bass were stocked at the same In our study, we used tag return data from a stock enhance- location in each system (Figure 1) across years. The external ment program to estimate and compare the growth and mor- tags indicated that a “reward” (US$5 or a baseball cap) would tality of hatchery-reared Striped Bass of Roanoke River origin be offered for reporting information on recaptured Striped Bass between their natal (Albemarle Sound estuary, including the to the NCDMF; contact information for NCDMF was printed Roanoke River) and two nonnatal systems (the Tar-Pamlico and on the tag. Neuse rivers) in North Carolina. Roanoke River adults served For the purposes of this study, we analyzed tag returns from as the sole source of broodstock from 1980 to 2010 to produce Striped Bass stocked during the years 1990–2003 (Table 1). juvenile Striped Bass that were stocked into North Carolina Multiple tag types have been used since the inception of the tag- coastal waters. Therefore, this data set of tag returns afforded ging program and included Carlin disk, cinch-up spaghetti, and a unique opportunity to compare the growth and mortality of a model FM-84 internal anchor tags. Sprankle et al. (1996) found recreationally and commercially important fish species between that internal anchor tags had much higher annual retention rates

Downloaded by [Department Of Fisheries] at 19:14 20 July 2015 its natal and nonnatal habitats. As such, our study provides an in Striped Bass (85%) than spaghetti tags (62%). Therefore, to important example of the effectiveness of cross-stocking as a maintain consistency and not bias growth and mortality esti- management tool for coastal fish populations. mates (the latter of which were based on mean time at liberty) we only used data from fish tagged with internal anchor tags: 1990 and onwards for the Albemarle Sound, and after 1992 and METHODS 1993 for the Neuse and Tar-Pamlico rivers, respectively. In ad- Study area.—Our study focused on three systems in coastal dition, due to a change in hatchery procedures (fish densities North Carolina: (1) the Albemarle Sound estuary, (2) the Tar- in grow-out ponds were decreased and food rations increased), Pamlico River, and (3) the Neuse River (Figure 1). These sys- juveniles produced after 2003 were larger (mean TL = 164 mm, tems exhibit similar temperatures (annual range, 5–30◦C) and range = 145–183 mm) at the time of stocking than those from bathymetry (mean depth, ∼4 m at midtide: NOAA 1990), but earlier years (mean TL = 127 mm, range = 122–132 mm). Ac- their salinity regimes differ markedly. The Albemarle Sound is cordingly, we excluded tag returns from fish released after 2003 fresh to oligohaline (salinities < 5‰) as high flows from the to ensure that size at release was similar among systems. Roanoke and Chowan rivers dominate the system relative to Growth estimation.—Striped Bass exhibited high fidelity to marine waters derived from a single, distant ocean inlet (Ore- their stocking system, which allowed us to estimate system- gon Inlet) (Figure 1). Meanwhile, due to their close proximity specific growth and mortality. In other studies, tagged Striped GROWTH AND MORTALITY OF STOCKED STRIPED BASS 1133 Downloaded by [Department Of Fisheries] at 19:14 20 July 2015

FIGURE 1. Coastal North Carolina study area for assessment of growth and mortality of hatchery-reared Striped Bass, showing the Albemarle Sound estuary (shaded in black, which includes the Roanoke and Chowan rivers), Tar-Pamlico River (shaded in gray), and Neuse River (shaded in black). Stars indicate stocking locations in each system. ENFH = the Edenton National Fish Hatchery, WSFH = the Watha State Fish Hatchery.

Bass have been reported by fishers in coastal waters from North system; values for the Albemarle Sound and Neuse River were Carolina to Maine (Dorazio et al. 1994; Welsh et al. 2007; Cal- similarly high at 94% (442 of 472) and 96% (582 of 605), respec- lihan et al. 2014); thus, Striped Bass fisheries (and potential tively (the boundaries for each system are shown in Figure 1). tag recovery locations) occur along the entire eastern seaboard Therefore, growth and mortality estimates for fish released into of North America. Yet, the majority (>90%) of tag returns of a given system should incorporate and reflect any effects of stocked fish was from the same river system in which fish were system-specific conditions. initially released. Specifically, 91% of tag returns (479 of 527) System-specific growth was estimated using the von Berta- from fish stocked in the Tar-Pamlico River occurred in that lanffy (VB) growth model. We were able to estimate growth 1134 CALLIHAN ET AL.

TABLE 1. Annual number of juvenile Striped Bass stocked and tagged in each system. Fish were released into the Tar-Pamlico and Neuse rivers during multiple stocking events (∼10,000 ﬁsh per event) in late fall (November–December); one stocking event per year occurred in Albemarle Sound and all stocked ﬁsh were tagged. The total number (and percent) of tag returns from each release year (through the end of the study reporting period on December 31, 2010) are also provided. Data are only shown for those release years included in statistical analyses (1990–2003).

Albemarle Sound Tar-Pamlico River Neuse River Release Number Number Number of Number Number Number of Number Number Number of year stocked tagged returns stocked tagged returns stocked tagged returns 1990 2,000 2,000 60 (3.0%) 1991 2,994 2,994 292 (9.8%) 1992 2,465 2,465 83 (3.4%) 116,820 2,527 129 (5.1%) 1993 2,180 2,180 23 (1.1%) 118,600 2,204 37 (1.7%) 1994 2,481 2,481 2 (0.1%) 183,254 2,320 27 (1.1%) 79,933 2,212 7 (0.3%) 1995 2,498 2,498 14 (0.6%) 140,972 2,497 51 (2.0%) 1996 2,490 2,490 2 (0.1%) 100,760 4,998 122 (2.4%) 1997 24,031 4,865 114 (2.3%) 1998 83,195 2,500 73 (2.9%) 1999 17,954 2,750 118 (4.3%) 2000 108,000 2,900 37 (1.3%) 2001 37,000 3,000 30 (1.0%) 2002 147,654 2,960 18 (0.6%) 2003 159,996 3,000 21 (0.7%) Total 17,108 17,108 476 (2.8%) 681,807 20,636 398 (1.9%) 636,362 18,097 386 (2.1%)

parameters (K: the Brody growth coefficient, and L∞:the the low abundance levels and truncated age structure of those asymptotic length; Jennings et al. 2001) using size-at-age data populations (NCDMF and NCWRC 2013). Consequently, most because the age of recaptured fish was known. Fish age was cal- Striped Bass at liberty during the later years of this study (2000s) culated as the time between the spawn date in the hatchery and were those stocked in the Tar-Pamlico and Neuse rivers. It is pos- the recapture date reported by fishers. Because the exact spawn sible that large-scale (i.e., statewide) environmental conditions date was known for most, but not all release batches (79%, or (e.g., temperature) differed between earlier years (1990s) and 11 of 14), we used the mean spawn date across years (April 28; later years (2000s), which could confound growth comparisons range, April 18 to May 10) to calculate fish age. While most re- among systems (because fish in Albemarle Sound were only ex- capture lengths (97%, 584 of 600) were reported by recreational posed to earlier-year conditions). To explore this potential issue, or commercial fishers, some returns (n = 16) were made during we examined whether growth trends based on all data (release scientific surveys by the NCDMF and North Carolina Wildlife years, 1990–2003) persisted when analyses were restricted to a Resources Commission (NCWRC), which allowed us to assess common time period over which releases occurred in all systems the reliability of our growth data. In the case of multiple re- (1992–1996) (Table 1). Downloaded by [Department Of Fisheries] at 19:14 20 July 2015 captures (1.4% of all returns), only lengths from the terminal Mortality estimation.—Because the fate of recaptured fish recapture were used as in Patterson et al. (2001) to ensure statis- was not known in our study (i.e., whether they were harvested tical independence such that data from a given fish was included or released by fishers), we did not estimate mortality using only once in analyses. We modeled length at recapture as a func- traditional approaches such as the Brownie model (Brownie tion of fish age. In these models, the mean size (127 mm TL) et al. 1985), which require this information and estimate sur- and age (227 d) at stocking was used as a data point and the vival based on dead recoveries. The assumption that all re- t0 parameter was constrained to zero. We used nonlinear least covered fish are dead, when in fact many are released alive squares to fit VB models to the observed data; residuals were after being caught, can result in overestimates of the mortality normally distributed and not heteroscedastic. Likelihood ratio rate (Smith et. al 2000), especially when catch and release is (LR) tests were used to compare growth curves and parameters common, as is the case in Striped Bass fisheries. Instead, we among systems (Kimura 1980). estimated the total instantaneous mortality rate (Z) of Striped The Albemarle–Roanoke stock of Striped Bass was declared Bass using time-at-liberty (TAL) estimators that were origi- to be fully recovered in 1997 (ASMFC 1998). Therefore, stock- nally developed by Gulland (1955) and Chapman (1961), and ing was discontinued in Albemarle Sound in 1997, but continued largely ignored until McGarvey (2009) recently renewed this in other coastal systems (Tar-Pamlico and Neuse rivers) due to approach. GROWTH AND MORTALITY OF STOCKED STRIPED BASS 1135

The basic premise of the TAL approach is that, in a fish mean comparisons (among systems) were assessed using Tukey- population experiencing higher mortality, there should be fewer adjusted P-values. Statistical tests were performed in Statistical tagged individuals at liberty for long periods than in a popu- Analysis System (SAS, version 9.1.3) using α = 0.05. lation experiencing lower mortality (i.e., mortality is inversely proportional to TAL) (McGarvey 2009). Using this rationale, RESULTS Chapman (1961) developed the following equations (for a finite sample of tag returns) to estimate Z (assumed to be constant Growth over time) and its SE: On average, Striped Bass were more than 400 mm at age 2 and 500 mm at age 3 across systems, and therefore exceeded ˆ nr − 1 1 ZChapman the minimum size limit for recreational and commercial harvest Zˆ = , Zˆ = √ , Chapman τ SE Chapman nr ¯ nr − 2 (457 mm TL, 1991–present) during their second year of life (Figure 2). Growth curves for Striped Bass from the Tar-Pamlico where nr = the number of tag returns from fish released in a and Neuse rivers were statistically indistinguishable from one given system and τ¯ = the mean TAL of fish released in a given another (LR test: P = 0.50), but each differed from the Albe- system. marle Sound fish (LR tests: P-values < 0.0001). This difference We used the above equations to estimate Z of Striped Bass in growth curves was due to the slightly higher L∞ values in the stocked into each system. Estimates of Z were made for each Tar-Pamlico (657 mm) and Neuse (636 mm) rivers versus the annual stocking by system, with the exception of 1994 and 1996 Albemarle Sound (582 mm) (LR tests: P-values < 0.0001) as K in the Albemarle Sound and 1994 in the Neuse River due to low values were similar (0.54–0.61) and did not significantly differ sample sizes (Table 1). As recommended by McGarvey et al. among systems (LR test: P = 0.12) (Figure 2). (2009), we excluded multiple recaptures from our analyses. In Growth parameters and trends were similar when data were addition, we only included TAL data from hook-and-line re- restricted to 1992–1996 releases only. That is, K values were captures (not gill nets and pound nets) so that gear type did not similar among systems (0.52–0.62), and L∞ values were slightly confound mortality comparisons. For example, many recaptures higher in the Tar-Pamlico (595 mm) and Neuse (665 mm) rivers from Albemarle Sound were from commercial gill nets (140- than in Albemarle Sound (565 mm). mm stretched mesh) that are inefficient at capturing old, large fish. Using these data could bias mortality estimates (shorter Mortality TALs, higher Z values) relative to other systems (Tar-Pamlico Stocked Striped Bass were at liberty for much shorter periods and Neuse rivers) where gill-net recaptures were rare. Further- in the Tar-Pamlico and Neuse rivers than in Albemarle Sound. more, because the TAL approach requires a sufficiently long For instance, only 6% of returns in the Tar-Pamlico and Neuse period to estimate Z such that most tagged fish would have rivers occurred beyond 4 years at liberty, whereas returns beyond died (McGarvey 2009), or their tags would have fallen out, we 4 years were fairly common (24%) in the Albemarle Sound only included TAL data for Striped Bass released from 1990 (Figure 3). The ANOVA confirmed this trend as mean TALs to 2000 because the maximum TAL observed in the database were significantly lower (P-values < 0.0001) in the Tar-Pamlico was 10 years (the database included returns reported through (2.0 years) and Neuse (2.1 years) rivers than in the Albemarle December 31, 2010, up to 20 years after tagging). We assumed Sound (3.0 years). Mean TALS did not differ between the Tar- tag loss (which is a component of Z estimated via the TAL Pamlico and Neuse rivers (ANOVA: P = 0.34). approach: McGarvey 2009) was constant across systems and Total mortality of Striped Bass was consistently higher in

Downloaded by [Department Of Fisheries] at 19:14 20 July 2015 did not confound among-system comparisons of mortality. Fi- the Tar-Pamlico and Neuse rivers than in the Albemarle Sound nally, the TAL approach does not require information on report- regardless of release year. In the Albemarle Sound, Z val- ing rates and is therefore insensitive to differences in reporting ues ranged from 0.27 to 0.39 across release years 1990–1995 rates among systems; this is a major advantage of the approach (Figure 4). In the Tar-Pamlico and Neuse rivers, Z values ex- (McGarvey 2009; McGarvey et al. 2009). ceeded 0.4 and ranged from 0.42 to 0.65, with the exception To simulate differences in mortality, we used the standard ex- of one release year (1994 in the Tar-Pamlico River) (Figure 4). ponential decay model for fish mortality (Jennings et al. 2001) During 3 years (1992, 1993, and 1995), fish from the same batch to project the decline of 100,000 stocked fish over time in each (raised together in the hatchery) were stocked into two different system based on data pooled across years within each system. systems 1 d apart. These fish showed the same trend in mortal- We pooled data because Z values were consistent across release ity among systems; that is, higher Z values in the Tar-Pamlico years within each system (see Results). To test statistical signif- (1993, 1995) and Neuse (1992) rivers than in the Albemarle icance, we compared mean TALs among systems (data pooled Sound (Figure 4). within each system across years) using a one-way ANOVA. Overall, system-specific Z values, calculated from data For this analysis, a data transformation (ln transformed) was pooled across release years, were higher in nonnatal systems, necessary to meet the assumptions of homogeneity of variance the Tar-Pamlico (Z = 0.51, SE = 0.03) and Neuse (Z = 0.48, and normality of residuals. The significance levels of pairwise SE = 0.03) rivers, than in the natal system, Albemarle Sound 1136 CALLIHAN ET AL.

FIGURE 2. (a–c) System-specific growth of hatchery-reared Striped Bass released into the wild. Growth parameters (K, L∞) were obtained by fitting the von Bertalanffy (VB) growth model to size-at-age data from tag returns of fish stocked into each system. Black dots show lengths reported from fishers, white dots show lengths of recaptured fish measured in scientific surveys, black stars indicate the mean size and age at release, and the thick black line is the VB growth Downloaded by [Department Of Fisheries] at 19:14 20 July 2015 model fitted to observed data. (d) Mean growth curves estimated for each system in relation to the current minimum size limit and known age at maturity (age 4: Olsen and Rulifson 1992; Boyd 2011). Alb Sound = the Albemarle Sound estuary, Tar-Pam = the Tar-Pamlico River, Neuse = theNeuseRiver(seeFigure1for system boundaries).

(Z = 0.33, SE = 0.02). Based on these Z estimates, the exponen- Tar-Pamlico and Neuse rivers) and contributed to local fish- tial decay model indicated that twice as many stocked Striped eries, because they grew just as quickly as they did in their natal Bass would survive to maturity (age 4) in the Albemarle Sound system, reaching legal catchable size by age 2, and were caught (26,317 fish) than would in the Tar-Pamlico and Neuse rivers: by fishers. However, total mortality was significantly higher in 13,252 and 14,517 fish, respectively. nonnatal versus natal systems, the causes of which are unknown. These results will help guide future research initiatives and management efforts to enhance Striped Bass populations in North DISCUSSION Carolina’s southern coastal rivers. Our study demonstrated that cross-stocking can be an effec- Fishery-dependent tag return data are not always reli- tive management tool, at least in supporting put-and-take type able for the purpose of growth estimation. For example, fisheries. Despite their Roanoke River origin, stocked Striped Støttrup et al. (2002) found that return data from commer- Bass exhibited high fidelity (>90%) to nonnatal systems (the cial fishers suggested a lack of growth during the summer GROWTH AND MORTALITY OF STOCKED STRIPED BASS 1137

during scientific surveys (Støttrup et al. 2002). In our study, the VB growth model provided a good fit to recapture lengths at age. Moreover, the lengths of Striped Bass recaptured during scientific surveys agreed with the mean growth curves for each system (which were largely derived from fishery-dependent data) (Figure 2a–c), providing evidence that size information from fishers was reliable. Although our analyses indicated L∞ was significantly higher in nonnatal versus natal systems, this result may be a statistical artifact of low sample sizes. Because Striped Bass experienced higher mortality in the Tar-Pamlico and Neuse rivers, there were few tag returns from fish older than age 6 in these systems. The few data points for fish > age 6 in the Tar-Pamlico and Neuse rivers lie above the mean growth curve for those systems (Fig- ure 2). Therefore, L∞ values were probably overestimated for the Tar-Pamlico and Neuse systems and in actuality were more FIGURE 3. System-specific times at liberty for Striped Bass tagged and re- similar to those for the Albemarle Sound, which would agree leased during 1990–2000. Sample sizes refer to the number of tag returns with size-at-age data for Striped Bass < 6 years old (size at age reported through December 31, 2010, from fish released in each system. See was similar among systems for those ages). It is also possible the Figure 1 for system boundaries and Figure 2 for definitions of abbreviations. higher L∞ values for the Tar-Pamlico and Neuse systems were partly due to the inherent negative correlation between L∞ and K growing season for juvenile Turbot Psetta maxima stocked along (Pilling et al. 2002). Although not significantly different, K val- the coast of Denmark (there was no relationship between re- ues were lower (and L∞ values higher) for the Tar-Pamlico and ported recapture lengths and TAL). Yet, growth was positive, as Neuse systems relative to that of Albemarle Sound (Figure 2). would be expected, based on recaptures of stocked fish made While we can only speculate why the growth of Striped Bass was similar among systems (possibly due to similar water temperatures), our growth results have important management implications. Stocked Striped Bass grew rapidly and exceeded the current minimum size limit (457 mm TL) between the ages of 2 and 3 across systems. Most Roanoke River Striped Bass do not mature until age 4. By age 4, more than 93% of Roanoke River females were found to be mature compared with only 29– 44% at age 3 (Olsen and Rulifson 1992; Boyd 2011). Therefore, fishing pressure on immature fish could be limiting the ability of the stocking program to enhance Striped Bass populations in southern coastal rivers within North Carolina, at least in the Tar-Pamlico and Neuse rivers where harvest is still permitted, especially if the higher mortality we observed in those systems

Downloaded by [Department Of Fisheries] at 19:14 20 July 2015 was due to fishing mortality. As with growth patterns, we cannot discern the underlying reasons for the higher total mortality of Striped Bass in nonnatal relative to natal systems. Roanoke River Striped Bass may exhibit a more exploratory foraging behavior (a herita- ble trait in fish: Bell 2009; Wisenden et al. 2011) if there are fewer predators in their oligohaline natal estuary (Albemarle Sound). If so, Roanoke River fish may suffer greater predation mortality when they are transplanted into more saline systems (Tar-Pamlico and Neuse rivers) that contain more marine preda- FIGURE 4. Mortality rate estimates of tagged Striped Bass by release year. tors. There are several other biotic and/or abiotic explanations Note that in some years (1992, 1993, and 1995), Striped Bass were stocked for differences in mortality, but we do not speculate further. into multiple systems. Symbols (triangles, filled circles, and squares) represent Our mortality results provide important baseline data for North the mean estimates of Z for each release year by system combination. Mean estimates of Z and their SEs (illustrated as vertical error bars) were determined Carolina’s collaborative (NCWRC and NCDMF) Striped Bass via the time-at-liberty approach (McGarvey 2009). See Figure 2 for definitions stock enhancement program. As of 2012, this program began us- of abbreviations. ing local (river-specific) broodstock to produce juveniles, which 1138 CALLIHAN ET AL.

were then stocked into the same system from which their par- stricted to the common time period of release across all systems ents (broodstock) were obtained. Therefore, our data will permit (1992–1996), indicating the differing release years among sys- comparisons between these two alternative stocking methods tems did not confound among-system comparisons of growth. (cross-stocking versus using local fish) to determine which is In conclusion, our study demonstrated that cross-stocking the most successful and cost-effective approach. For instance, can be an effective management tool for Striped Bass popula- the additional costs and logistics of raising and maintaining tions in coastal North Carolina, at least in terms of supporting locally sourced fish may be justifiable if juveniles from local put-and-take type fisheries and possibly for stock rebuilding. (natal) broodstock exhibit higher survival in the Tar-Pamlico Current research, which is being conducted on the degree of and Neuse rivers (e.g., due to local adaptations). Striped Bass natural reproduction in the Tar-Pamlico and Neuse In addition to providing important baseline data to evaluate rivers (i.e., the relative contributions of stocked versus wild the efficiency of different stocking methods, our results will al- fish), will provide further insight into the type of fishery that low mangers to more precisely estimate fishing mortality levels. can be achieved with stocking. For example, if spawning suc- The current status of Striped Bass stocks in the Tar-Pamlico and cess is poor due to a lack of suitable habitat, stocked fish may Neuse rivers is considered to be “unknown, but of concern” due not successfully reproduce in these systems and therefore only to the lack of precise stock assessment data from these two sys- support a put-and-take type fishery. In a broader context, as one tems (NCDMF and NCWRC 2013). Cohort-based catch curves of the few studies investigating the ecological consequences of have been used to estimate Z values for each river system (stock) cross-stocking in coastal fish populations, our findings provide in prior assessments, but the results were deemed too imprecise an important reference for managers considering this approach, for management purposes (NCDMF and NCWRC 2013). The which may be the only stock enhancement option available SEs of Z values estimated via catch curves approached or ex- in some cases, especially for severely depleted populations in ceeded the mean values of Z, as proportional standard errors which adult broodstock are rare and difficult to obtain. Still, it (PSEs) were, on average, 98% and 72% in the Tar-Pamlico and is important to bear in mind that the effectiveness (and poten- Neuse rivers, respectively, for the same release years (cohorts) tial genetic effects) of cross-stocking may vary widely among analyzed in our study (1992–2000) (NCDMF and NCWRC species and largely depend on population structure and specif- 2013). These imprecise estimates of Z were likely due to a ically where within a species’ range source fish are procured truncated age structure (few age-classes on the descending limb from. For instance, juveniles from more distant source popula- of the catch curve), indicative of high total mortality, which lim- tions may perform poorly relative to those from nearby (adja- ited the effectiveness of the catch-curve method. The approach cent) systems whose environmental conditions are more similar we used resulted in more precise estimates of Z: mean PSEs for to the natal habitat. the Tar-Pamlico and Neuse rivers were 12% and 17%, respectively, based on year-specific Z values and 6% for both systems based upon data pooled across release years. Accordingly, our Z ACKNOWLEDGMENTS estimates and approach should provide a better (more precise) We thank the many individuals from the NCDMF and evaluation of stock status of Striped Bass in the Tar-Pamlico NCWRC who were involved in the hatchery-production and tag- and Neuse rivers. For example, an independent estimate of the ging of juvenile Striped Bass. We thank Jeff Evans (NCWRC) instantaneous natural mortality rate (e.g., M=0.15: NCDMF for providing data on spawning dates and Stephen Jackson (U.S. and NCWRC 2013) could be subtracted from our Z estimates Fish and Wildlife Service) for length-at-release data. We are also to obtain fishing mortality rates (F) for each river system. In grateful to the many fishers who provided tag return informa- Downloaded by [Department Of Fisheries] at 19:14 20 July 2015 doing so, it should be noted that our Z values are overestimates tion and the NCDMF personnel who processed and compiled in the absolute sense due to tag loss (McGarvey et al. 2009). these data. Analyses for this study were completed while J. L. However, this bias is minimal and at most, the values for Z in Callihan was a Marine Fisheries Management Fellow supported our study were overestimated by 0.02–0.16 based on available by NCDMF (Coastal Recreational Fisheries License fund award estimates of annual tag loss rates in Striped Bass (2% to 15%) 3210) and North Carolina Sea Grant award E/GS-6. from double-tagging studies (Waldman et al. 1991; Sprankle et al. 1996). Although we used data from multiple release batches (release years), this did not bias our study inferences on mortality REFERENCES and growth. Across release years, Z values were consistently ASMFC (Atlantic States Marine Fisheries Commission). 1998. Addendum III to amendment 5 interstate fishery management plan for Atlantic Striped Bass: higher in nonnatal (Tar-Pamlico and Neuse rivers) versus natal 1999–2000 fisheries, Albemarle/Roanoke stock recovery, Delaware River systems (Albemarle Sound), clearly demonstrating that “batch stock recovery. ASMFC, Washington, D.C. effects” did not confound among-system comparisons of mor- Bell, A. M. 2009. Approaching the genomics of risk-taking behavior. Advances tality. Nor did “batch effects” confound growth comparisons. in Genetics 68:83–104. Striped Bass were only released in nonnatal systems after 1996. Blankenship, H. L., and K. M. Leber. 1995. A responsible approach to marine stock enhancement. Pages 167–175 in H. L. Schramm Jr. and R. G. Piper, However, growth trends calculated from the entire data set editors. Uses and effects of cultured fishes in aquatic ecosystems. American (release years 1990–2003) persisted when analyses were re- Fisheries Society, Symposium 15, Bethesda, Maryland. GROWTH AND MORTALITY OF STOCKED STRIPED BASS 1139

Boyd, J. B. 2011. Maturation, fecundity, and spawning frequency of the Albe- NCDMF (North Carolina Division of Marine Fisheries) and NCWRC marle/Roanoke Striped Bass stock. Master’s thesis. East Carolina University, (North Carolina Wildlife Resources Commission). 2013. Amendment Greenville, North Carolina. I to the North Carolina estuarine Striped Bass fishery management Brownie, C., D. R. Anderson, K. P.Burnham, and D. S. Robson. 1985. Statistical plan. NCDMF and NCWRC, Morehead City. Available: http://portal. inference from band recovery data–a handbook, 2nd edition. U.S. Fish and ncdenr.org/c/document library/get file?uuid=d3fdf967–82d5–4653–8b79– Wildlife Services Resource Publication 156. 20247c5ed5ad&groupId = 38337. (September 2014). Bulak, J. S., C. S. Thomason, K. Han, and B. Ely. 2004. Genetic variation NOAA (National Oceanic and Atmospheric Administration). 1990. Estuaries and management of Striped Bass populations in the coastal rivers of South of the United States: vital statistics of a national resource base. NOAA, Carolina. North American Journal of Fisheries Management 24:1322–1329. Strategic Environmental Assessments Division, Office of Ocean Resources Callihan, J. L., C. H. Godwin, and J. A. Buckel. 2014. Effect of demogra- Conservation and Assessment, Silver Spring, Maryland. phy on spatial distribution: movement patterns of the Albemarle Sound– Norton, S. F., Z. A. Eppley, and B. D. Sidell. 2000. Allometric scaling of Roanoke River stock of Striped Bass (Morone saxatilis) in relation to their maximal enzyme activities in the axial musculature of Striped Bass, Morone recovery. U.S. National Marine Fisheries Service Fishery Bulletin 112: saxatilis. Physiological and Biochemical Zoology 73:819–828. 131–143. Olsen, E. J., and R. A. Rulifson. 1992. Maturation and fecundity of Roanoke Chapman, D. G. 1961. Statistical problems in dynamics of exploited fisheries River–Albemarle Sound Striped Bass. Transactions of the American Fisheries populations. Proceedings of the Berkeley Symposium on Mathematical Statis- Society 121:524–537. tics and Probability 4:153–168. Orlando, S. P. Jr., P. H. Wendt, C. J. Klein, M. E. Pattillo, K. C. Dennis, and Conover, D. O. 1990. The relation between capacity for growth and length of G. H. Ward. 1994. Salinity characteristics of South Atlantic estuaries. Na- growing season: evidence for an implications of countergradient variation. tional Oceanic and Atmospheric Administration, Office of Ocean Resources Transactions of the American Fisheries Society 119:416–430. Conservation and Assessment, Silver Spring, Maryland. Conover, D. O., J. J. Brown, and A. Ehtisham. 1997. Countergradient variation Patterson, W. F., J. C. Watterson, R. L. Shipp, and J. H. Cowan Jr. 2001. Move- in growth of young Striped Bass (Morone saxatilis) from different latitudes. ment of tagged Red Snapper in the northern Gulf of Mexico. Transactions of Canadian Journal of Fisheries and Aquatic Sciences 54:2401–2409. the American Fisheries Society 130:533–545. Dorazio, R. M., K. A. Hattala, C. B. McCollough, and J. E. Skjeveland. 1994. Tag Pilling, G. M., G. P. Kirkwood, and S. G. Walker. 2002. An improved method recovery estimates of migration of Striped Bass from spawning areas of the for estimating individual growth variability in fish, and the correlation be- Chesapeake Bay. Transactions of the American Fisheries Society 123:950– tween von Bertalanffy growth parameters. Canadian Journal of Fisheries and 963. Aquatic Sciences 59:424–432. Epperly, S. P., and S. W. Ross. 1986. Characterization of the North Car- Rimmer, M. A., and D. J. Russell. 1998. Survival of stocked Barramundi, Lates olina Pamlico–Albemarle estuarine complex. NOAA Technical Memoran- calcarifer (Bloch), in a coastal river system in far northern Queensland, dum NMFS-SEFC-175. Australia. Bulletin of Marine Science 62:325–335. Gulland, J. A. 1955. On the estimation of population parameters from marked Rulifson, R. A., and R. W. Laney. 1999. Striped Bass stocking programs in the members. Biometrika 42:269–270. United States: ecological and resource management issues. Canadian Stock Harrell, R. M., J. H. Kerby, and R. V. Minton. 1990. Culture and propaga- Assessment Secretariat Research Document 99/007. tion of Striped Bass and its hybrids. American Fisheries Society, Bethesda, Smith, D. R., K. P. Burnham, D. M. Kahn, X. He, C. J. Goshorn, K. H. Hattala, Maryland. and A. W. Kahnle. 2000. Bias in survival estimates from tag-recovery models Hassler, W. W., N. L. Hill, and J. T. Brown. 1981. The status and abundance of where catch-and-release is common, with an example from Atlantic Striped Striped Bass, Morone saxatilis, in the Roanoke River and Albemarle Sound, Bass (Morone saxatilis). Canadian Journal of Fisheries and Aquatic Sciences North Carolina, 1956–1980. North Carolina Department of Natural Resources 57:886–897. and Community Development, Division of Marine Fisheries, Special Scien- Sprankle, K., J. Boreman, and J. B. Hestbeck. 1996. Loss rates for dorsal loop tific Report 38, Morehead City. and internal anchor tags applied to Striped Bass. North American Journal of Hervas, S., K. Lorenzen, M. A. Shane, and M. A. Drawbridge. 2010. Quantitative Fisheries Management 16:461–464. assessment of a White Seabass (Atractoscion nobilis) stock enhancement St. Pierre, R. A. 1999. Restoration of Atlantic Sturgeon in the northeastern program in California: post-release dispersal, growth, and survival. Fisheries USA with special emphasis on culture and restocking. Journal of Applied Research 105:237–243. Ichthyology 15:180–182. Jenkins, W. E., M. R. Denson, C. B. Bridgham, M. R. Collins, and T. J. Smith. Støttrup, J. G., C. R. Sparrevohn, J. Modin, and K. Lehmann. 2002. The use of 2004. Year-class component, growth, and movement of juvenile Red Drum releases of reared fish to enhance natural populations a case study on Turbot Downloaded by [Department Of Fisheries] at 19:14 20 July 2015 stocked seasonally in a South Carolina estuary. North American Journal of Psetta maxima (Linne,´ 1758). Fisheries Research 59:161–180. Fisheries Management 24:636–647. Waldman, J. R., D. J. Dunning, and M. T. Mattson. 1991. Long-term retention Jennings, S., M. J. Kaiser, and J. D. Reynolds. 2001. Marine fisheries ecology. of anchor tags and internal anchor tags by Striped Bass. North American Blackwell Scientific Publications, Oxford. Journal of Fisheries Management 11:232–234. Kimura, D. K. 1980. Likelihood methods for the von Bertalanffy growth curve. Ward, R. D. 2006. The importance of identifying spatial population structure U.S. National Marine Fisheries Service Fishery Bulletin 77:765–776. in restocking and stock enhancement programmes. Fisheries Research 80:9– Lichatowich, J., L. Mobrand, and L. Lestelle. 1999. Depletion and extinction of 18. Pacific salmon (Oncorhynchus spp.): a different perspective. ICES Journal of Welsh, S. A., D. R. Smith, R. W. Laney, and R. C. Tipton. 2007. Tag-based Marine Science 56:467–472. estimates of annual fishing mortality of a mixed Atlantic coastal stock McGarvey, R. 2009. Methods of estimating mortality and movement rates from of Striped Bass. Transactions of the American Fisheries Society 136:34– single-tag recovery data that are unbiased by tag non-reporting. Reviews in 42. Fisheries Science 17:291–304. Wisenden, B. D., C. D. Sailer, S. J. Radenic, and R. Sutrisno. 2011. Mater- McGarvey, R., J. M. Matthews, and J. E. Feenstra. 2009. Estimating mortality nal and exploratory-boldness behavioural syndrome in Zebrafish. Behaviour from times-at-large: testing accuracy and precision using simulated single 148:1443–1456. tag-recovery data. ICES Journal of Marine Science 66:573–581. Woodroffe, J. 2011. Historical ecology of striped bass stocking in the south- Murphy, B. R. 1990. Evidence for a genetically unique Walleye population eastern United States. Master’s thesis. East Carolina University, Greenville, in the upper Tombigbee River system of northeastern Mississippi. Pro- North Carolina. ceedings of the Southeastern Fishes Council 22:14–16. Available: http:// Worth, S. G. 1884. Report upon the propagation of Striped Bass at Weldon, www.sefishescouncil.org/proceedings/sfcpro22.pdf. (September 2014). N.C., in the spring of 1884. U.S. Fish Commission Bulletin 4:225–230. This article was downloaded by: [Department Of Fisheries] On: 20 July 2015, At: 19:14 Publisher: Taylor & Francis Informa Ltd Registered in England and Wales Registered Number: 1072954 Registered office: 5 Howick Place, London, SW1P 1WG

North American Journal of Fisheries Management Publication details, including instructions for authors and subscription information: http://www.tandfonline.com/loi/ujfm20 Validation of Daily Ring Deposition in the Otoliths of Age-0 Alligator Gar Peter C. Sakarisac, David L. Buckmeierb & Nathan G. Smithb a Department of Biology and Chemistry, Southern Polytechnic State University, 1100 South Marietta Parkway, Marietta, Georgia 30060, USA b Texas Parks and Wildlife Department, Heart of the Hills Fisheries Science Center, 5103 Junction Highway, Mountain Home, Texas 78058, USA c Present address: School of Science and Technology, Georgia Gwinnett College, 1000 University Center Lane, Lawrenceville, Georgia 30043, USA. Published online: 03 Nov 2014. Click for updates

To cite this article: Peter C. Sakaris, David L. Buckmeier & Nathan G. Smith (2014) Validation of Daily Ring Deposition in the Otoliths of Age-0 Alligator Gar, North American Journal of Fisheries Management, 34:6, 1140-1144, DOI: 10.1080/02755947.2014.951806 To link to this article: http://dx.doi.org/10.1080/02755947.2014.951806

PLEASE SCROLL DOWN FOR ARTICLE

Taylor & Francis makes every effort to ensure the accuracy of all the information (the “Content”) contained in the publications on our platform. However, Taylor & Francis, our agents, and our licensors make no representations or warranties whatsoever as to the accuracy, completeness, or suitability for any purpose of the Content. Any opinions and views expressed in this publication are the opinions and views of the authors, and are not the views of or endorsed by Taylor & Francis. The accuracy of the Content should not be relied upon and should be independently verified with primary sources of information. Taylor and Francis shall not be liable for any losses, actions, claims, proceedings, demands, costs, expenses, damages, and other liabilities whatsoever or howsoever caused arising directly or indirectly in connection with, in relation to or arising out of the use of the Content. This article may be used for research, teaching, and private study purposes. Any substantial or systematic reproduction, redistribution, reselling, loan, sub-licensing, systematic supply, or distribution in any form to anyone is expressly forbidden. Terms & Conditions of access and use can be found at http:// www.tandfonline.com/page/terms-and-conditions North American Journal of Fisheries Management 34:1140–1144, 2014 C American Fisheries Society 2014 ISSN: 0275-5947 print / 1548-8675 online DOI: 10.1080/02755947.2014.951806

MANAGEMENT BRIEF

Validation of Daily Ring Deposition in the Otoliths of Age-0 Alligator Gar

Peter C. Sakaris*1 Department of Biology and Chemistry, Southern Polytechnic State University, 1100 South Marietta Parkway, Marietta, Georgia 30060, USA David L. Buckmeier and Nathan G. Smith Texas Parks and Wildlife Department, Heart of the Hills Fisheries Science Center, 5103 Junction Highway, Mountain Home, Texas 78058, USA

however, little is known about the population dynamics of this Abstract species. Interest in Alligator Gar management probably stems We validated techniques for estimating the daily age of young- from their international popularity among anglers pursuing fish of-the-year Alligator Gar, Atractosteus spatula. Alligator Gar eggs that can weigh more than 100 kg. Locally, gar fillets are also were hatched and larvae were grown in aquaria and then trans- $ ferred to an outdoor artificial stream. Fish were sampled from 21 d important to commercial fishers, valued up to US 6.61 per kilo- to 118 d posthatch. Sagittal otoliths were extracted from each fish, gram (DiBenedetto 2009). Recent assessments have found that and each otolith was embedded in a clear epoxy resin. Otoliths were the historic range of Alligator Gar has declined, and that re- sectioned along a transverse plane, sanded with 600-grit sandpaper maining populations appear vulnerable to habitat loss and over- to reveal daily rings, and then polished with 1500-grit sandpaper. fishing (Etnier and Starnes 1993; Pflieger 1997; Ferrara 2001; We estimated the daily age of each otolith three times, making each estimate in random order without reference to known ages. The Jelks et al. 2008). Reasons for this range reduction are unknown, mean of the three counts for each otolith was considered the final but loss of connectivity to historic floodplain spawning areas is age estimate. Daily age estimates of 50 Alligator Gar (21–118 d old, suspected as a major cause (Inebnit 2009). Alligator Gar are 43–458 mm total length) were closely related to known fish ages, long-lived and late to mature, and their recruitment is often validating daily ring deposition. Absolute error in age estimates episodic and infrequent (Ferrara 2001; Buckmeier et al. 2012). increased after 62 d posthatch; 68% of the estimated ages were within ± 3 d of the known age through 62 d, whereas only 26% As a result, sustainability of remaining Alligator Gar fisheries of the estimated ages were within ± 3 d of the known age from and reestablishment of extirpated populations will require a bet- 76 to 118 d posthatch. However, mean estimates for each cohort ter understanding of the recruitment dynamics of this species. were within 5 d of known age through 104 d posthatch. Daily incre- Downloaded by [Department Of Fisheries] at 19:14 20 July 2015 Alligator Gar are known to spawn over inundated terrestrial ments near the core were difficult to identify, but later increments vegetation during flood pulses (Brinkman 2008; Inebnit 2009). were more discernable and deposited in a regular pattern. A significant relationship was also identified between fish total length However, data regarding the timing and frequency of spawning, and otolith radius (P < 0.01), indicating that estimates of growth the environmental conditions required for successful reproduc- histories should be reliable. We encourage researchers to utilize tion, and conditions of early growth and survival are mostly our aging technique to estimate hatch dates, the timing and fre- lacking. Alligator Gar are known to spawn in the spring (April quency of hatching, and early growth rates of Alligator Gar in wild to June) when water temperatures exceed 20◦C (Ferrara 2001; populations. This early life history information will be valuable in enhancing management and conservation of this species. DiBenedetto 2009; Inebnit 2009). Inebnit (2009) speculated that a minimum of 7 d of flooding was needed for successful spawn- In recent years, interest in management and conservation ing and larval development, and an additional 14 d of flooding of Alligator Gar Atractosteus spatula has greatly increased; was needed to provide suitable nursery habitat for developing

*Corresponding author: [email protected] 1Present address: School of Science and Technology, Georgia Gwinnett College, 1000 University Center Lane, Lawrenceville, Georgia 30043, USA. Received March 21, 2014; accepted July 16, 2014

1140 MANAGEMENT BRIEF 1141

FIGURE 1. (A) Left sagittal otolith from a 48-d-old Alligator Gar, depicting section location along a transverse plane (black dashed line = approximate sectioning plane) and orientation (A = anterior, P = posterior). (B) Multiple nuclei (black arrows) often revealed in the core region of otolith sections. (C) Otolith section of a 48-d-old Alligator Gar, with inset images displaying daily increment deposition around the core and away from the core. After about 20 d posthatch, daily

Downloaded by [Department Of Fisheries] at 19:14 20 July 2015 increments appeared more distinct and showed a more regular pattern of deposition. The white dotted line depicts the measuring transect for the otolith radius.

Alligator Gar in the Fourche LaFave River, Arkansas. Growth difficult. The development of a viable technique would also al- of Alligator Gar in captivity can be rapid, but may vary substan- low fish managers to assess the effects of environmental factors tially in the wild, presumably depending upon environmental (e.g., the flood pulse and other hydrologic variables) on spawn- conditions and availability of food. ing and hatching success and the early growth and survival of For many species, researchers have validated daily age es- this species. In this study, we validated a technique for using timates from larval otoliths to allow estimation of vital rates, otoliths to estimate the daily age of young-of-the-year Alligator timing and success of spawning events, and correlation with Gar. Specifically, we determined that deposition of daily incre- biotic and abiotic factors (Miller and Storck 1982; Davis et al. ments occurred in sagittal otoliths and evaluated the accuracy 1985; Graham and Orth 1987; Sweatman and Kohler 1991; Par- of age estimates through 118 d. rish et al. 1994; DiCenzo and Bettoli 1995; Sakaris and Irwin 2008; Sakaris et al. 2011). However, no validation studies of daily increment deposition in the otoliths of lepisosteids have METHODS been published. For Alligator Gar, the ability to estimate daily Known-age fish collection.—Alligator Gar eggs were collected age and back-calculate hatch dates would be useful in evaluating from a captive stock of wild-caught fish that spawned in research early life history characteristics, because direct observations are ponds on June 1 and 2, 2010, at Heart of the Hills Fisheries 1142 SAKARIS ET AL.

Science Center near Mountain Home, Texas. Water tempera- TABLE 1. Aging statistics for 10 cohorts of known-age young-of-year Alli- tures during spawning were 28–29◦C. Each day, eggs were trans- gator Gar (total n = 50) including the mean, SD, CV, and the range. ferred to indoor aquaria, where we attempted to maintain water ◦ Mean Estimated temperatures near 28 C. Eggs hatched on June 4 and 5, and yolk Known estimated SD CV age sacs were absorbed by both groups on June 7. Fish were initially age (d) N age (d) (d) (%) range (d) fed starter trout feed (Silver Cup Granulated Salmon/Trout 1; Nelson and Sons, Inc., Murray, Utah), zooplankton, and blood- 21 3 22 2.3 10.3 21–25 worms (Chironomidae). By June 14, fish were piscivorous and 29 3 31 3.2 10.5 27–33 were fed Fathead Minnow Pimephales promelas fry. Between 30 3 27 4.4 16.1 22–30 June 23 and 28 (21–27 d posthatch), larval Alligator Gar were 34 9 34 3.3 9.8 30–41 moved to an outdoor artificial stream where they were raised to 48 6 53 6.6 12.4 47–66 age 118 d. In the artificial stream, Alligator Gar were fed a va- 62 7 64 3.5 5.5 60–69 riety of fish species, including Western Mosquitofish Gambusia 76 5 77 11.0 14.3 62–90 affinis,KoiCarpCyprinus carpio, and Sunfish Lepomis spp. As 90 5 91 9.8 10.7 83–107 Alligator Gar developed, 5 to 10 fish were killed about every 104 5 105 9.0 8.5 92–117 2 weeks from 21 d through 118 d posthatch. In addition, fish 118 4 111 5.1 4.6 103–114 that died from other causes throughout the study were saved if date of death could be determined. Age (d) and total length (TL; mm) were recorded and sagittal otoliths were removed for age analyses. for each otolith was considered the estimated age. Otolith radii (mm) were measured for each transverse section along a tran- Otolith preparation and aging.—We processed otoliths to yield sect line from the center of the core to the ventral tip of each a thin section in the transverse plane (Figure 1A), similar to otolith (Figure 1C). Buckmeier et al. (2012). Sagittal otoliths of Alligator Gar larvae (N = 60) were embedded in a clear epoxy resin (105 Epoxy Statistical analyses.—Linear regression analysis was used to Resin, 206 Slow Hardener; WestSystem, Bay City, Michigan). evaluate relations between (1) the estimated age and the known For fish from 21 to 34 d posthatch, we used a Buehler Isomet age of the fish and (2) the fish TL and otolith radius. We tested high-precision sectioning saw (Buehler, Ltd., Lake Bluff, Illi- null hypotheses (t-tests) that the slope of the regression between nois) to remove a smaller section of the resin block that con- estimated age and known age equaled 1 (i.e., 1:1 ratio) and tained the otolith. Otoliths embedded in resin were then mounted that the intercept equaled zero (i.e., intercept was at the origin). with crystal bond to microscope slides, positioned perpendic- Significance was set at α = 0.05. Aging accuracy and precision ularly to the plane of the slide. Otoliths were incrementally were evaluated by calculating the SD, coefficient of variation ground wet with 600-grit sandpaper and routinely viewed with a [CV = 100 × (SD/mean)], and range of estimated ages for compound microscope (under 100 × and 400 × magnification) each cohort (i.e., known-age group). The deviance of each age until the core and daily increments were visible in a transverse estimate from known age was also calculated. plane (Figure 1B). The otolith was then inverted and ground incrementally on the other side until a thin section (<0.1-mm thick) was obtained. For fish ≥48 d, we used the Isomet saw

Downloaded by [Department Of Fisheries] at 19:14 20 July 2015 to section each otolith along a transverse plane. We used a thin RESULTS wafering blade (0.15 mm thick; Buehler Diamond Wafering We successfully estimated the ages of 50 of the 60 known- Blade, No. 11-4283) to maximize our ability to produce sec- age Alligator Gar, ranging from 21 to 118 d posthatch (Table 1). tions with clearly visible cores. Transverse otolith sections were The other 10 otolith sections were considered unreadable, be- fixed with crystal bond to microscope slides and lightly sanded cause no core was recognizable after processing. Total lengths with 600-grit sandpaper. All otolith sections were polished sev- of the age-0 Alligator Gar ranged from 43 to 458 mm, and eral times with 1500-grit sandpaper until otolith cores and rings mean ± SD growth rate was 3.56 ± 0.76 mm/d. Fish length were clearly visible for age estimation. was strongly related to otolith radius (r2 = 0.982, P < 0.01, Otoliths were read three times, each time in a different ran- Figure 2). dom order, and no two counts of the same otolith were made Daily increments around the core were typically difficult to consecutively (Miller and Storck 1982; DiCenzo and Bettoli distinguish up to approximately 20 d posthatch, but these rings 1995; Sakaris and Irwin 2008; Sakaris et al. 2011). All otolith were readable upon close inspection (Figure 1C). We frequently sections were viewed using an image analysis system at 400 × observed multiple nuclei, or origins of growth, in the core re- magnification (ProgRes-CF, Jenoptik, Germany). Daily incre- gion of otolith sections (Figure 1B). After 20 d posthatch, daily ments were counted from the inner core to the outer edge of increments appeared to be more regularly deposited and were each otolith section (Figure 1C). The mean of the three counts easier to discern and enumerate (Figure 1C). MANAGEMENT BRIEF 1143

500 through 62 d posthatch and that cohorts can be assigned ages 450 through 104 d. As a consequence of constant relative errors 400 (CVs) among age groups, we expected an increase in absolute 350 aging errors in older fish. Similar results have been reported for 300 other freshwater fishes (Miller and Storck 1982; Sweatman and 250 Kohler 1991; DiCenzo and Bettoli 1995; Roberts et al. 2004; 200 Sakaris and Irwin 2008; Sakaris et al. 2011). 150 We partly attribute the increased absolute aging error af- 100 Total Length (mm TL) ter 62 d to our otolith processing methods. Cutting otoliths 50 with the Isomet saw often produced undesirable scratches on 0 the surfaces of otolith sections. Isomet sectioning also pro- 0.00 0.50 1.00 1.50 2.00 2.50 3.00 3.50 4.00 duced some transverse otolith sections with faint cores. All Otolith Radius (mm) 10 otolith sections that we considered unreadable were sec- FIGURE 2. Relation between total length and otolith radius for age-0 Alligator tioned with the Isomet saw, and 9 of these 10 unreadable Gar (regression model: TL = 134.71 × (otolith radius) – 14.109, r2 = 0.982, otolith sections were from older fish (76–118-d posthatch). All P < 0.01). hand-sanded otoliths yielded readable cores that were assessed an estimated age. Although sectioning with the saw signifi- Increments on sagittal otoliths of Alligator Gar corre- cantly reduced our processing time, we recommend that all gar sponded to daily age. Mean daily increment counts were closely otoliths be sanded by hand to enhance the clarity of otolith 2 related to known age of Alligator Gar (r = 0.953, P < 0.01; sections. Figure 3). The slope (0.965) of the regression was not sig- Daily increments around the core were also often difficult nificantly different from one (t =−1.13, P > 0.20), and the to distinguish. However, daily increments were more distinct intercept was not significantly different from zero (t = 1.33, and showed a more regular pattern of deposition after about 20 P = 0.19), indicating that increment deposition began at hatch. d posthatch. This transition from less to more distinguishable Absolute error in age estimates increased after 62 d posthatch, daily increments after 20 d may have resulted from Alliga- with 68% of the estimated ages (21 of 31) being ± 3dof tor Gar being transferred from aquaria to an outdoor artificial the known age through 62 d and only 26% (5 of 19) of the stream. Alligator Gar were held at a constant temperature (28◦C) estimated ages being ± 3 d of the known age from 76 to 118 while in aquaria, but were subject to a more natural variation d posthatch. However, mean ages of cohorts (i.e., known-age in temperature in the artificial stream. Other studies have indi- groups) were ± 5 d of known ages through 104 d posthatch, and cated that daily fluctuations in water temperature can result in CVs indicated that variability was relatively consistent across increased otolith clarity and overall reading accuracy of daily all age-groups (Table 1). increments. For example, otolith increments were clearer and counts of increments were more accurate for juvenile Colorado DISCUSSION Pikeminnow Ptychocheilus lucius reared at fluctuating temper- Our study demonstrates that daily increments are deposited atures (Bestgen and Bundy 1998). Bundy and Bestgen (2001) in the sagittal otoliths of larval Alligator Gar. In addition, we reported similar results for the Razorback Sucker Xyrauchem determined that individual Alligator Gar can be reliably aged texanus. Ages of Razorback Suckers raised in constant temperatures were underestimated by 4–7 d, while otolith increment

Downloaded by [Department Of Fisheries] at 19:14 20 July 2015 140 counts for fish raised in fluctuating temperatures were similar to their known ages (Bundy and Bestgen 2001). Future studies 120 should test the effects of water temperature regime on otolith 100 clarity and reading accuracy. 80 Sagittal otoliths appear to grow proportionally to the Alli- gator Gar TL, based on a strong relation with otolith radius. 60 In our study, only 2.8% of the variation in fish length could 40 Dashed line = 1:1 not be explained by otolith radius. This result indicates that Estimated Age (d) 20 otolith radius is a robust predictor of total length and that back- calculation methods for determining growth histories can be ap- 0 plied to young-of-the-year Alligator Gar for at least 3–4 months 0 20 40 60 80 100 120 140 after hatch. Although other environmental factors (e.g., water Known Age (d) temperature and water chemistry) may influence the growth of FIGURE 3. Relation between estimated age and known age for age-0 Alligator otoliths (Secor and Dean 1992; Bestgen and Bundy 1998), our Gar (N = 50; regression model: y = 0.965x + 2.838, r2 = 0.953, P < 0.01; data suggest that these factors may not be as influential for solid line = regression). young-of-the-year Alligator Gar. However, we recommend that 1144 SAKARIS ET AL.

future studies explore how environmental factors may affect DiCenzo, V. J., and P. W. Bettoli. 1995. Verification of daily ring deposition in otolith growth in lepisosteids. the otoliths of age-0 Spotted Bass. Transactions of the American Fisheries We encourage fisheries managers to utilize our daily ag- Society 124:633–636. Etnier, D. A., and W. C. Starnes. 1993. The fishes of Tennessee. University of ing technique to study the early life history of Alligator Gar. Tennessee Press, Knoxville. Estimation of daily age and subsequent hatch dates will allow Ferrara, A. M. 2001. Life-history strategy of Lepisosteidae: implications for the evaluation of abiotic and biotic factors in regards to the the conservation and management of Alligator Gar. Doctoral dissertation. timing of spawning and hatching and to the early growth and Auburn University, Auburn, Alabama. survival of Alligator Gar. Identifying critical environmental re- Graham, R. J., and D. J. Orth. 1987. Otolith aging of young-of-the-year Small- mouth Bass. Pages 483–491 in R. C. Summerfelt and G. E. Hall, editors. Age quirements (e.g., annual flood pulse) for spawning, hatching, and growth of fish. Iowa State University Press, Ames. and early growth of Alligator Gar is vital to the conservation Inebnit, T. E. III. 2009. Aspects of the reproductive and juvenile ecology of Alli- and management of this valuable species. gator Gar in the Fourche LaFave River, Arkansas. Master’s thesis. University of Central Arkansas, Conway. Jelks, H. L., S. J. Walsh, N. M. Burkhead, S. Contreras-Balderas, E. Diaz- ACKNOWLEDGMENTS Pardo, D. A. Hendrickson, J. Lyons, N. E. Mandrak, F. McCormick, J. We thank the staff at Texas Parks and Wildlife Department S. Nelson, S. P. Platania, B. A. Porter, C. B. Renaud, J. J. Schmitter- Heart of the Hills Fisheries Science Center, especially R. Wie- Soto, E. B. Taylor, and M. L. Warren Jr. 2008. Conservation status of necke for assistance with fish rearing. We also thank Southern imperiled North American freshwater and diadromous fishes. Fisheries 33: Polytechnic State University for providing equipment and facil- 372–407. Miller, S. J., and T. Storck. 1982. Daily growth rings in otoliths of young-of- ities for otolith processing and analysis. Funding was provided, the-year Largemouth Bass. Transactions of the American Fisheries Society in part, by the U.S. Fish and Wildlife Service through the State 111:527–530. Wildlife Grants Program, grant T-53-1, and through the Federal Parrish, D. L., B. Vondracek, and W. J. Eckmayer. 1994. Accuracy and Aid in Sport Fish Restoration, grant F-231-R, to the Texas Parks precision of daily age estimates for Walleyes from Ohio hatcheries and Wildlife Department. and Lake Erie. Transactions of the American Fisheries Society 123: 601–605. Pflieger, W. 1997. The fishes of Missouri. Missouri Department of Conservation, REFERENCES Jefferson City. Bestgen, K. R., and J. M. Bundy. 1998. Environmental factors affect daily in- Roberts, M. E., J. E. Wetzel II, R. C. Brooks, and J. E. Garvey. 2004. Daily crement deposition and otolith growth in young Colorado Squawfish. Trans- increment formation in otoliths of the Redspotted Sunfish. North American actions of the American Fisheries Society 127:105–117. Journal of Fisheries Management 24:270–274. Brinkman, E. L. 2008. Contributions to the life history of Alligator Gar Atrac- Sakaris, P. C., D. J. Daugherty, and D. L. Buckmeier. 2011. Validation of daily tosteus spatula, in Oklahoma. Master’s thesis. Oklahoma State University, ring deposition in the otoliths of age-0 Blue Catfish and Flathead Catfish. Stillwater. Pages 689–697 in P.H. Michaletz and V.H. Travnichek, editors. Conservation, Buckmeier, D. L., N. G. Smith, and K. R. Reeves. 2012. Utility of Alligator Gar ecology, and management of catfish: the second international symposium. age estimates from otoliths, pectoral fin rays, and scales. Transactions of the American Fisheries Society, Symposium 77, Bethesda, Maryland. American Fisheries Society 141:1510–1519. Sakaris, P. C., and E. R. Irwin. 2008. Validation of daily ring deposition in Bundy, J. M, and K. R. Bestgen. 2001. Confirmation of daily increment de- the otoliths of age-0 Channel Catfish. North American Journal of Fisheries position in otoliths of young Razorback Suckers. Southwestern Naturalist Management 28:212–218. 46:34–40. Secor, D. H., and J. M. Dean. 1992. Comparison of otolith-based back- DiBenedetto, K. 2009. Life history characteristics of Alligator Gar, Atracosteus calculation methods to determine individual growth histories of larval Striped spatula, in the Bayou Dularge area of south central Louisiana. Master’s thesis. Bass, Morone saxatilis. Canadian Journal of Fisheries and Aquatic Sciences Louisiana State University, Baton Rouge. 49:1439–1454. Davis, R. D., T. W. Storck, and S. J. Miller. 1985. Daily growth increments in Sweatman, J. J., and C. C. Kohler. 1991. Validation of daily otolith increments the otoliths of young-of-the-year Gizzard Shad. Transactions of the American for young-of-the-year White Crappie. North American Journal of Fisheries Downloaded by [Department Of Fisheries] at 19:14 20 July 2015 Fisheries Society 114:304–306. Management 11:499–503. This article was downloaded by: [Department Of Fisheries] On: 20 July 2015, At: 19:14 Publisher: Taylor & Francis Informa Ltd Registered in England and Wales Registered Number: 1072954 Registered office: 5 Howick Place, London, SW1P 1WG

North American Journal of Fisheries Management Publication details, including instructions for authors and subscription information: http://www.tandfonline.com/loi/ujfm20 Use of a Statewide Angler Tag Reporting System to Estimate Rates of Exploitation and Total Mortality for Idaho Sport Fisheries Kevin A. Meyera & Daniel J. Schilla a Idaho Department of Fish and Game, 1414 East Locust Lane, Nampa, Idaho 83686, USA Published online: 10 Nov 2014.

Click for updates

To cite this article: Kevin A. Meyer & Daniel J. Schill (2014) Use of a Statewide Angler Tag Reporting System to Estimate Rates of Exploitation and Total Mortality for Idaho Sport Fisheries, North American Journal of Fisheries Management, 34:6, 1145-1158, DOI: 10.1080/02755947.2014.951803 To link to this article: http://dx.doi.org/10.1080/02755947.2014.951803

PLEASE SCROLL DOWN FOR ARTICLE

Taylor & Francis makes every effort to ensure the accuracy of all the information (the “Content”) contained in the publications on our platform. However, Taylor & Francis, our agents, and our licensors make no representations or warranties whatsoever as to the accuracy, completeness, or suitability for any purpose of the Content. Any opinions and views expressed in this publication are the opinions and views of the authors, and are not the views of or endorsed by Taylor & Francis. The accuracy of the Content should not be relied upon and should be independently verified with primary sources of information. Taylor and Francis shall not be liable for any losses, actions, claims, proceedings, demands, costs, expenses, damages, and other liabilities whatsoever or howsoever caused arising directly or indirectly in connection with, in relation to or arising out of the use of the Content. This article may be used for research, teaching, and private study purposes. Any substantial or systematic reproduction, redistribution, reselling, loan, sub-licensing, systematic supply, or distribution in any form to anyone is expressly forbidden. Terms & Conditions of access and use can be found at http:// www.tandfonline.com/page/terms-and-conditions North American Journal of Fisheries Management 34:1145–1158, 2014 C American Fisheries Society 2014 ISSN: 0275-5947 print / 1548-8675 online DOI: 10.1080/02755947.2014.951803

ARTICLE

Use of a Statewide Angler Tag Reporting System to Estimate Rates of Exploitation and Total Mortality for Idaho Sport Fisheries

Kevin A. Meyer* and Daniel J. Schill Idaho Department of Fish and Game, 1414 East Locust Lane, Nampa, Idaho 83686, USA

Abstract From 2006 to 2009, 18,712 fish were tagged and released in 45 tagging events using nonreward and high-reward T- bar anchor tags to estimate the rate of exploitation (u) by anglers in various Idaho fisheries. In total, 3,100 nonreward tags and 592 high-reward tags were reported by anglers. Annual u was adjusted for tag loss, tagging mortality, and angler tag reporting rate. Tag loss, estimated by double-tagging a subsample of fish, varied greatly among species; tag loss was lowest for Yellow Perch Perca flavescens (1.1% in year 1 and 4.5% in year 2) and crappies Pomoxis spp. (2.9% and 4.8%) and was highest for Largemouth Bass Micropterus salmoides (14.8% and 30.3%), Walleyes Sander vitreus (11.4% and 43.2%), and Smallmouth Bass M. dolomieu (10.5% and 41.6%). Short-term (7–33-d) mortality averaged about 1% for both hatchery and wild fish. The nonreward-tag reporting rate averaged 54.5% across all species and years. Adjusted u averaged 19.4% (range = 2.0–44.3%) and was generally highest for crappies (mean = 28.7%) and Smallmouth Bass (22.0%) and lowest for wild trout (9.5%). Estimates of total annual mortality (A), based on the difference in tag returns between years 1 and 2, were plausible for some species but were unusually high for other species, especially Smallmouth Bass and wild trout. The implausibility of some estimates of A probably resulted from a combination of factors, including the reduced vulnerability of larger, older fish to angling, which would have caused a reduction in tag returns in year 2, likely due to a shift in fish behavior or habitat preference as tagged fish grew in size. Our results demonstrate the utility of using the high-reward tagging method to estimate u for fisheries under a variety of circumstances, but fisheries managers should use caution in attempting to simultaneously estimate A from the tag returns.

Mortality assessment is an essential part of ﬁsh population however, except in small populations, difﬁculties in measuring

Downloaded by [Department Of Fisheries] at 19:14 20 July 2015 management (Allen and Hightower 2010) because mortality population size often preclude this approach (Miranda et al. rates can have important influences on the abundance, size struc- 2002). A more commonly employed technique for estimating ture, recruitment, and growth of fish populations (Ricker 1975). u consists of releasing a known number of marked fish with However, estimating the various components of mortality in fish tags, relying on anglers to return the tags, and calculating the populations is often challenging due to sampling limitations and proportion harvested (Pollock et al. 2001). an inability to fully meet the assumptions of various estimation Several authors have provided a sizeable list of assumptions procedures (Miranda and Bettoli 2007). associated with this approach (e.g., Ricker 1975; Pollock et al. Estimation of exploitation rate (u; i.e., the proportion of 2001; Miranda et al. 2002), and violation of any of these as- harvestable-sized fish that are removed from a population annu- sumptions can render the estimates of u unreliable unless the ally via fishing) has been a hallmark of fishery management for resulting bias is accounted for. In particular, violations of the over 50 years. Perhaps the most straightforward technique for tag loss, tagging mortality, and tag reporting rate assumptions estimating u is to determine the ratio of fish harvested during are sometimes evaluated empirically and corrected for directly the fishing season to population size at the start of the season; in exploitation tagging studies (Pollock et al. 2001; Miranda

*Corresponding author: [email protected] Received December 23, 2013; accepted July 21, 2014 1145 1146 MEYER AND SCHILL

et al. 2002). For example, tag loss can be directly estimated may be difficult to substantiate or correct (for example, the by releasing double-tagged fish. Immediate or delayed tagging assumption that the catch is not size selective), and substan- mortality can be minimized by gentle handling and capture tech- tial effort is needed to collect the necessary data (Walters and niques, instituting favorable holding and tagging environments, Martell 2004). Alternatively, if u is estimated with angler tag and choosing the appropriate tag for the size and species of fish returns and if the tag returns are monitored for two full years, (Nielsen 1992). Even if these procedures are followed, tagging then total annual mortality (A) can be simply estimated by com- may still cause some mortality, usually in the short term, and paring the tag returns in year 2 to those in year 1 (McCammon this can be estimated with holding experiments wherein tagged and LaFaunce 1961; Ricker 1975) after adjusting for any ad- fish as well as untagged control fish are held in an ambient envi- ditional tag loss in the second year relative to the first. This ronment (such as a net-pen) for one to several days, after which method also relies on meeting several important assumptions, their relative survival is compared (Nielsen 1992; Pollock et al. some of which can be evaluated empirically. For example, vul- 2001). Tag reporting rate can be evaluated by releasing both nerability to angling is assumed to be constant over all ages and nonreward and high-reward tags and comparing their relative sizes of fish in the harvestable population. However, if vulner- returns (e.g., Nichols et al. 1991; Meyer et al. 2012a). ability to angling declines as the fish grow to larger sizes and Once the above corrections are accounted for, annual rates of become more difficult to catch, then tag returns in year 2 would exploitation by anglers can be estimated. However, because of likely be biased low and estimates of total mortality would be difficulties in (1) meeting the necessary assumptions to reliably inflated. Another important assumption is that the tag report- estimate u from tag returns or (2) estimating the departure from ing rate does not differ between years. A secondary objective these assumptions, it has been suggested that biologists avoid of this study was to calculate tag-derived estimates of A and directly estimating u altogether in favor of indirect population determine whether the assumptions of this method (particularly monitoring methods, such as analyzing angler catch rates, fish the assumption of equal vulnerability to catch by anglers) were size, fish growth, and recruitment variability (Miranda et al. adequately supported by the data. 2002). This recommendation was also based on the perception that a large financial investment would typically be required for field tagging and reward tag payments, although others have METHODS argued that this perception is based on little evidence and that Fish tagging.—From 2006 to 2009, Idaho Department of high-reward tagging studies should not be dismissed out of hand Fish and Game (IDFG) personnel tagged 18,712 fish distributed as being too expensive (Pollock et al. 2001; also see Walters and across 45 tagging events (Table 1). This study included a vari- Martell 2004). ety of species: White Crappie Pomoxis annularis, Black Crap- Estimates of u that incorporate corrections for tag loss, tag- pie Pomoxis nigromaculatus, Largemouth Bass Micropterus ging mortality, and tag reporting rate have error that is propa- salmoides, Smallmouth Bass Micropterus dolomieu, Yellow gated with each applied correction, perhaps accounting for the Perch Perca flavescens, Walleye Sander vitreus, and several observation that confidence bounds for such estimates are sel- species of trout, including Rainbow Trout Oncorhynchus mykiss, dom reported in the literature (Miranda et al. 2002). Although Brook Trout Salvelinus fontinalis, Cutthroat Trout O. clarkii, simulation or resampling techniques can be used to develop such and Rainbow Trout × Cutthroat Trout hybrids. White Crap- confidence bounds (e.g., Miranda et al. 2002; Isermann et al. pies and Black Crappies were combined in this study because 2005), a simpler approach for most fishery managers would be they often occur in sympatry in Idaho waters and anglers do not to use straightforward mathematical formulas that account for distinguish between species.

Downloaded by [Department Of Fisheries] at 19:14 20 July 2015 the error propagation (McFadden 1961; Yates 1980). The pri- Wild fish were generally collected by using a boat-mounted mary aim of the present study was therefore to evaluate whether electrofisher (settings were 300–600-V DC, 60 Hz, and 4–7-ms results from a statewide angler tag reporting system in Idaho pulse width, producing a 25–40% duty cycle and an average (presented by Meyer et al. 2012a) could produce precise esti- output of 1–5 A). During electrofishing, fish were captured and mates of u (corrected for tag loss, tagging mortality, and tag placed in a live well in small quantities until they were tagged reporting rate) for a variety of Idaho sport fisheries. and released near the point of capture. Wild trout were also By itself, an estimate of annual u—even with precise error captured at weirs. Hatchery trout that were used in this study bounds—may or may not be adequate for use in evaluating a were raised to catchable size (i.e., about 250 mm TL), netted fishery management program. Unless the evaluation was ini- out of the raceway, tagged, and held in a pen within the raceway tially designed only to quantify the return rates of stocked fish until stocking (usually 1–3 d later). (e.g., Koenig and Meyer 2011) or unless an estimated u for a All fish were anesthetized with spearmint oil (Danner et al. wild stock is clearly low, additional mortality components (e.g., 2011) and measured for TL (nearest mm); fish were then tagged total or natural mortality) are often needed for a more refined with T-bar anchor tags that were fluorescent orange, 70 mm fishery assessment. Often, this is accomplished by estimating in total length (51 mm of tubing), and treated with algaecide. total mortality using catch curves (Miranda and Bettoli 2007). Tags were labeled on two sides, with one side stating the agency However, catch curves have their own set of assumptions that and phone number and the other side listing a tag number and Downloaded by [Department Of Fisheries] at 19:14 20 July 2015

TABLE 1. Summary of nonreward and high-reward tags released from 2006 to 2009 and returned by anglers for each of 45 ﬁsh tagging events in Idaho.

Nonreward tags Reporteda High-reward tags Harvested Angler nonreward-tag Year Water body Species Origin Released Total Year 1 Year 2 in year 1 Released Reported reporting rate ( ± 90% CI) 2006 Brownlee Reservoir Crappies Wild 449 111 78 28 62 41 14 72.4 ± 33.8 2007 Brownlee Reservoir Crappies Wild 399 108 97 9 67 42 20 56.8 ± 22.8 2008 Brownlee Reservoir Crappies Wild 379 73 63 9 48 40 19 40.5 ± 17.2 2009 Brownlee Reservoir Crappies Wild 398 85 79 5 69 42 16 56.1 ± 25.1 2006 C.J. Strike Reservoir Crappies Wild 210 54 42 7 26 18 6 77.1 ± 54.6 2007 C.J. Strike Reservoir Crappies Wild 366 113 71 34 46 40 12 102.9 ± 51.4 2008 C.J. Strike Reservoir Crappies Wild 382 110 79 25 57 40 18 64.0 ± 26.8 2009 C.J. Strike Reservoir Crappies Wild 380 112 98 13 84 38 26 43.1 ± 15.4 2006 Mann Lake Crappies Wild 252 111 84 12 64 25 18 61.2 ± 25.6 2006 Ben Ross Reservoir Largemouth Bass Wild 108 13 9 2 5 16 5 38.5 ± 33.3 2007 Ben Ross Reservoir Largemouth Bass Wild 227 47 36 8 6 24 13 38.2 ± 19.7 2006 Pend Oreille River Largemouth Bass Wild 332 94 75 12 33 33 25 37.4 ± 13.8 2006 Brownlee Reservoir Smallmouth Bass Wild 392 92 89 1 35 38 22 40.5 ± 15.8 2007 Brownlee Reservoir Smallmouth Bass Wild 399 80 79 1 37 42 15 56.1 ± 26.0 2008 Brownlee Reservoir Smallmouth Bass Wild 382 109 107 2 57 40 18 63.4 ± 26.5 2009 Brownlee Reservoir Smallmouth Bass Wild 342 51 50 1 25 36 14 38.3 ± 19.0 2006 Cascade Reservoir Smallmouth Bass Wild 106 12 10 2 6 2 1 22.6 ± 38.8 2006 C.J. Strike Reservoir Smallmouth Bass Wild 292 90 87 1 29 29 14 63.8 ± 30.2 2007 C.J. Strike Reservoir Smallmouth Bass Wild 379 144 142 1 61 40 19 80.0 ± 32.1 2008 C.J. Strike Reservoir Smallmouth Bass Wild 381 92 91 1 41 40 19 50.8 ± 21.1 2009 C.J. Strike Reservoir Smallmouth Bass Wild 381 104 104 1 61 42 22 52.1 ± 20.1 2007 Dworshak Reservoir Smallmouth Bass Wild 383 96 87 8 40 40 22 45.6 ± 17.7 2006 Milner Reservoir Smallmouth Bass Wild 401 134 132 2 26 40 22 60.8 ± 23.0 2009 Oakley Reservoir Walleye Wild 224 39 32 7 24 24 6 69.6 ± 50.2 2007 Salmon Falls Creek Walleye Wild 559 123 72 41 45 42 14 66.0 ± 30.6 Reservoir 2009 Cascade Reservoir Yellow Perch Wild 379 61 44 18 37 40 11 58.5 ± 31.5 2006 Coeur d’Alene River Rainbow Trout, Wild 78 11 10 1 4 10 4 35.3 ± 33.9 Cutthroat Trout, and hybrids 2007 Henry’s Lake Cutthroat Trout and Wild 669 18 12 5 6 75 4 50.4 ± 45.9 hybrids 2006 Moyie River Rainbow Trout and Wild 374 16 14 2 7 29 3 41.4 ± 42.8 Brook Trout 1147 Downloaded by [Department Of Fisheries] at 19:14 20 July 2015 1148

TABLE 1. Continued.

Nonreward tags Reporteda High-reward tags Harvested Angler nonreward-tag Year Water body Species Origin Released Total Year 1 Year 2 in year 1 Released Reported reporting rate ( ± 90% CI) 2006 South Fork Snake Rainbow Trout Wild 243 48 35 8 10 25 4 123.5 ± 105.7 River 2007 South Fork Snake Rainbow Trout and Wild 521 72 55 9 29 48 12 55.3 ± 28.4 River hybrids 2006 Williams Lake Rainbow Trout Wild 226 40 27 10 21 24 11 38.6 ± 21.6 2007 Williams Lake Rainbow Trout Wild 228 20 15 2 10 31 6 45.3 ± 34.7 2008 Anderson Ranch Rainbow Trout Hatchery 606 26 23 3 20 63 6 45.0 ± 33.6 Reservoir 2007 Boise River Rainbow Trout Hatchery 380 89 89 0 45 39 18 50.7 ± 21.6 2006 Cascade Reservoir Rainbow Trout Hatchery 755 27 25 2 18 80 4 71.5 ± 63.0 2006 Chesterﬁeld Reservoir Rainbow Trout Hatchery 378 38 29 9 17 40 12 33.5 ± 18.3 2007 Glendale Reservoir Rainbow Trout × Hatchery 379 87 86 1 64 39 22 40.7 ± 16.0 Cutthroat Trout hybrid 2006 Lake Walcott Rainbow Trout Hatchery 698 86 53 23 44 95 22 53.2 ± 20.9 2007 Little Wood Reservoir Rainbow Trout × Hatchery 378 6 5 0 5 40 3 21.2 ± 24.6 Cutthroat Trout hybrid 2006 Lucky Peak Reservoir Rainbow Trout Hatchery 380 42 39 2 30 40 4 110.5 ± 95.1 2007 Mann Creek Rainbow Trout Hatchery 380 76 70 5 66 40 15 53.3 ± 24.8 Reservoir 2006 Mann Lake Rainbow Trout Hatchery 343 58 56 0 36 40 9 75.2 ± 44.3 2007 North Fork Payette Rainbow Trout Hatchery 670 53 53 0 31 72 14 40.7 ± 20.1 River 2008 Ririe Reservoir Cutthroat Trout Hatchery 380 29 18 10 10 40 8 38.2 ± 25.1 Total 16,948 3,100 2,651 343 1,564 1,764 592 54.5 ± 4.0

a For reported tags, year 1 + year 2 does not always equal the total because some of the reported tags could not be assigned to a year. ANGLER TAG REPORTING SYSTEM 1149

reward amount if applicable. Tags were either nonreward or high also pooled by species across all tagging events to create mean reward. High-reward tags were equally split between values of λ for each species (see Meyer et al. 2012a). US$100 and $200, both of which have recently been shown Estimates of nonreward-tag reporting rate theoretically to elicit 100% tag reporting by anglers (Meyer et al. 2012a). should range between 0% and 100%. However, individual es- High-reward tags made up about 10% of the total number of timates of nonreward-tag reporting rates can exceed 100% by tags released for each tagging event. All species were tagged chance due to small numbers of tags for individual tagging just below the dorsal fin as recommended by Guy et al. (1996). events and the random nature of anglers’ encounters with tagged To reduce the rate of tagging mortality, individual fish were fish. As a hypothetical example, if 100 nonreward tags and 10 evaluated (up to the point of release) in terms of whether they high-reward tags are released in a water body and if anglers were unfit for this study due to visible signs of stress from report 11 nonreward tags and 2 high-reward tags, then the esti- capture and handling procedures (Nielsen 1992). Because we mated nonreward-tag reporting rate for this tagging event would were primarily interested in estimating exploitation, fish that be (11/100)/(2/10) = 0.55, or 55%. However, because of small were selected for tagging were always of harvestable size based sample sizes for individual tagging events, anglers might—by on angler interest and angling regulations, except at one study chance alone—encounter high-reward tags a bit less often than location (i.e., Largemouth Bass at Ben Ross Reservoir) where they probabilistically should have. In this example, if anglers angler catch of sub-harvestable-sized fish was also of interest. had encountered and reported only one high-reward tag, the re- Meyer et al. (2012a) provided a more detailed summary of sulting nonreward-tag reporting rate would be 110%. Clearly, the tag reporting system used in this study. Briefly, a website and this result does not mean that anglers were more willing to report toll-free automated telephone hotline were established through nonreward tags than high-reward tags in this instance, but rather which anglers could voluntarily report tags, although some tags it reflects chance variation in tag encounters by anglers and the were mailed to or dropped off at IDFG offices. In addition, small sample sizes on which reporting rates were based for in- informational posters and stickers were distributed to IDFG dividual tagging events. Consequently, estimates of u based on license vendors, regional offices, and sporting goods stores to λ from individual tagging events were considered less reliable, publicize tagging efforts, explain how the information was being and herein we only present estimates based on mean λ for each used, and provide tag return instructions. No other information species (also see Meyer et al. 2012a). was provided to anglers, and signs were not posted at individual Estimating tag loss.—To estimate the rate of tag loss, more water bodies so that site-specific estimates of u in the future than half of the tagged fish (nonreward and high-reward tags) would not require labor-intensive sign maintenance activity at were double-tagged with a “secondary” nonreward tag. Tag loss each water body where tags are released. rate (Tag l) was estimated as Estimating tag reporting rate.—The angler tag reporting rate (λ) was estimated using the reporting rate of nonreward tags 2 = n DT1 × n DT1 , relative to that of high-reward tags (Pollock et al. 2001): Tagl 2 × n DT2 2 × n DT2

where nDT1 is the number of double-tagged fish for which anglers λ = Rr ÷ Nr , reported that only a single tag was present, and nDT2 is the total Rt Nt number of double-tagged fish reported, whether one tag or both tags were present. The second part of the equation accounts for where Rt and Rr are the numbers of nonreward tags released fish that lost both tags and therefore had no chance of being

Downloaded by [Department Of Fisheries] at 19:14 20 July 2015 and reported, respectively; and Nt and Nr are the numbers of reported (Miranda et al. 2002). Sample size was not adequate to high-reward tags released and reported. Although in some in- estimate Tag l at each water body, so data were pooled to develop stances anglers did release ﬁsh with tags intact, this usually only one estimate of Tag l for each species. Tag loss was estimated occurred for nonreward tags because anglers were required to separately for year 1 (i.e., tags returned within 1 year of the turn in (not just report) high-reward tags before payment could tagging event) and year 2 (i.e., tags returned after year 1 but be issued. Consequently, we only counted a reported tag once before 2 years had expired since the tagging event). Variance λ so that would not be biased by the fact that nonreward tags was calculated according to the formula given by Fleiss (1981): were much more likely to be reported multiple times. Variance in λ was calculated according to Henny and Burnham (1976) as PQ Var( Tagl ) = , n λ 2 2 2 1 Rt Var(λ) = λ × + × × Nr . where P is Tag , Q is 1 − P, and n is the number of double-tagged R R N l r r t ﬁsh that were reported by anglers (multiplied by 2). From the estimate of variance, we calculated 90% CIs. The secondary From the estimate of variance, we calculated 90% CIs. Report- nonreward tags were returned by anglers at the same rate as ing rates were estimated for each tagging event, but data were the primary nonreward tags; thus, to simplify analyses and to 1150 MEYER AND SCHILL

help meet the assumption of independence in tag reporting, and secondary tags were used only to estimate Tag l and were not considered when estimating λ. For each species, we calculated x 2 Var (x) Var (y) Var( x/y) = × + , the year-1 Tag l for larger-sized fish (i.e., those above the median y x2 y2 TL) and smaller-sized fish (below the median), and we used a paired t-test (α = 0.10, with each species treated as the sampling where x and y are independent components of the formula for unit) to evaluate whether tag loss appeared to be related to fish u (each with their own variance, as established in earlier equa- size. tions) being multiplied or divided by one another. These variance Estimating tagging mortality.—To estimate tagging mortal- estimators account for uncertainty due to the number of tags at ity (Tag m), we used the same electrofishing methodology as large, such that precision is improved at higher sample sizes. above to capture wild Smallmouth Bass, Largemouth Bass, crap- From these results, we derived 90% CIs for u. Estimates of u pies, and trout at several water bodies. Fish were placed in 1-m3, were compared across species and years by using ANOVA with wire-mesh cages at a density of 28–80 fish/m3 and were lowered α = 0.10. to the bottom in 3–10 m of water (n = 9 holding trials with wild The rate of catch-and-release angling occurring in these fish- fish). Half of the fish were tagged and the other half were un- eries was estimated in the same way as for u except that in tagged; tagged and untagged fish were held in the same cage for the numerator (i.e., u), the number of nonreward-tagged fish each trial. Short-term mortality of tagged and untagged fish was reported as harvested within 1 year of the tagging event was estimated as the proportion of fish alive at 1 d or 7 d after initial replaced with the number reported as not harvested. As with capture. For hatchery trout, short-term mortality was estimated estimates of u, for each tagging event we estimated the rate of by tagging fish in raceways and holding them in 6.8-m3 pens (at catch and release by using only the mean λ for the appropriate a density of 74 fish/m3) for 22–33 d (n = 5 holding trials with species. hatchery fish). Estimating total annual mortality.—Total annual mortality Estimating angler exploitation.—Unadjusted u was calcu- rate (A) was estimated as lated as the number of nonreward-tagged fish that were reported as harvested within 1 year of the tagging event, divided by the Rr(y2) A = 1 − , number of fish that were released with nonreward tags for the Rr(y1) tagging event. All anglers that reported a tag were asked whether they had intentionally planned to harvest the fish they were re- where Rr(y2) is the number of nonreward tags reported by anglers porting or whether they only harvested the fish because it had a in year 2 and Rr(y1) is the number of nonreward tags reported in tag. We asked this question because a small proportion of anglers year 1 (Ricker 1975; Miranda and Bettoli 2007). The estimation indicated confusion as to whether tag reporting was mandatory of A had several important assumptions that we could evaluate or voluntary and whether the tag was removable without killing empirically. One assumption was that λ did not differ between the fish. Few anglers reported that they had only harvested their years, and our results support this assumption (see below; also fish because the fish was tagged; however, assuming these data see Meyer et al. 2012a). Another assumption was that λ did not were accurate, a small proportion of the total number of tagged differ between larger-sized and smaller-sized fish because as fish were harvested “unintentionally,” and those fish were not fish grew in size from year 1 to year 2, a change in λ caused by included in the calculation of u. fish growth would cause the likelihood of tags being returned by Adjusted u (u ) incorporated λ, Tag l, and Tag m and was esti- anglers in year 2 to differ from the likelihood in year 1, thereby

Downloaded by [Department Of Fisheries] at 19:14 20 July 2015 mated using the formula invalidating the estimate of A. This assumption was evaluated by splitting fish of each species into groups of larger-sized fish (i.e., those above the median) and smaller-sized fish (those below the median) and comparing λ between the two groups with a = u , u paired t-test (α = 0.10, treating each species as the sampling λ(1 − Tagl )(1 − Tagm) unit). Hatchery trout were excluded from these analyses since they were all stocked at essentially the same size. If Tag l was different between larger- and smaller-sized fish, this would have where u is the unadjusted annual exploitation rate (Allen and been corrected for by the additional tag loss adjustment we had Hightower 2010). For each tagging event, we estimated u using already made for year-2 tag returns, but as noted above we did the mean λ for the appropriate species. Variance for u was evaluate whether fish size affected tag loss. calculated using the approximate formulas for the variance of Another important assumption for estimating A was that vul- products and the variance of ratios (Yates 1980): nerability to angling was not size dependent. To assess whether estimates of A were potentially impaired by size dependence in fish vulnerability to anglers, each fish with a nonreward tag Var( x × y) = x2 × Var( y) + y2 × Var( x) was considered the sampling unit; fish TL at tagging was binned ANGLER TAG REPORTING SYSTEM 1151

into 50-mm size-groups, and dummy variables were used to define whether a fish was caught and reported by an angler (1) or not (0). We used ANOVA (α = 0.10) to evaluate whether the proportion of tagged fish caught by anglers varied between size-groups. To minimize any bias that fish growth might have on this analysis, we limited returns to those occurring within 1 year of tagging. A final assumption was that Tag l in year 2 was not higher than that in year 1. We expected that this assumption would be violated because regardless of whether or not Tag l was size dependent, tags were more likely to have fallen off by the end of year 2 than by the end of year 1. To account for this, we adjusted our estimates of A by enlarging the total number of year-2 tag returns by the amount of additional tag loss in year 2 relative to FIGURE 1. Estimates of mean ( ± 90% CI) annual rates (%) of exploitation by year 1, which standardized the effect of tag loss on tag returns anglers (adjusted), catch-and-release angling, and total mortality summarized by species. between years. A second adjustment to A was made to account for those anglers who had (1) “unintentionally” harvested a fish only because it had a tag, and otherwise would have released the fish; or (2) removed and reported the tag but released the fish. (mean = 8.6%) and smaller-sized fish (mean = 7.8%; t = 1.47, These tagged fish were not incorporated into estimates of u,but df = 5, P = 0.20). they were also unavailable to anglers the next year, which could Tagging mortality (Tag m) was low (i.e., < 5%) for all trials have lowered tag returns in year 2. We derived 90% CIs for A (Table 3). Mean short-term (i.e., 7-d) mortality (using each trial by using the above-mentioned formulas from Yates (1980). as the sampling unit) of tagged wild fish was about 1% higher We used an α value of 0.10 for all statistical significance tests than that of untagged wild fish. For hatchery trout, Tag m av- and for calculating CIs. This less-stringent significance level eraged 0% after 1 d and 0.8% after 22–33 d. Based on these (compared to the more standard use of α = 0.05) was adopted findings, Tag m was assumed to be very low but not zero; there- to balance type I and type II errors in our statistical tests (Cohen fore, 1% was used as an estimate. 1990; Stephens et al. 2005) and because IDFG fish managers Taking into account the angler tag reporting rate for nonre- were content with the tradeoff of having more precision in the ward tags, Tag l, and Tag m, estimates of u for individual tagging estimates of u and A at the expense of less confidence in the events averaged 19.4% but varied widely from a low of 2.0% estimates. to a high of 44.3% (Table 4). Based on ANOVA (using each tagging event as a sampling unit, or point estimate, of u; n = 45), exploitation differed among species (F = 2.80, df = 6, P = RESULTS 0.02) but not between years (F = 2.10, df = 3, P = 0.12). Ex- In total, 16,948 fish with nonreward tags and 1,764 fish with ploitation was generally highest for crappies (mean u ± SE = high-reward tags were released, and anglers reported 3,100 non- 28.7 ± 2.7%) and Smallmouth Bass (mean u = 22.0 ± 2.3%) reward tags and 592 reward tags (Table 1). Nonreward-tag re- and lowest for wild trout (mean u = 9.5 ± 9.5%; Figure 1). porting rate averaged 54.5 ± 4.0% (weighted mean ± 90% CI) The 90% CIs around the u estimates were (on average) 34% Downloaded by [Department Of Fisheries] at 19:14 20 July 2015 across all species and years, with individual estimates varying of the point estimate, and nearly one-half of the u estimates from 21.2% to over 100% (Table 1). There was no apparent vari- had 90% CIs that were less than or equal to 25% of the point ation in tag reporting rates over time, with weighted means of estimate (Table 4). 53, 56, 50, and 56% from 2006 to 2009. A paired t-test indicated The rate of catch-and-release angling in these fisheries aver- that the tag reporting rate did not differ between the larger-sized aged 14.8% but also varied widely among species and among fish (mean = 55.8%) and smaller-sized fish (mean = 53.5%; tagging events (Table 4). Rates of catch-and-release angling t = 1.14, df = 5, P = 0.31). were lowest for the three most harvest-oriented species, includ- Average Tag l (weighted mean ± 90% CI) across all species ing Yellow Perch (3.2%), Walleyes (mean = 7.1%), and crappies was 8.2 ± 0.7% in year 1 and 17.8 ± 2.3% in year 2 (Ta- (mean = 9.6%). Catch-and-release rates were highest for Large- ble 2). Tag loss varied among species and was lowest for Yellow mouth Bass (mean = 30.6%) and Smallmouth Bass (mean = Perch (1.1 ± 1.9% in year 1; 4.5 ± 3.7% in year 2) and 29.1%). crappies (2.9 ± 1.1%; 4.8 ± 3.1%) and was highest for Large- Based on tag returns from one full year and two full years mouth Bass (14.8 ± 5.1%; 30.3 ± 11.9%), Walleyes (11.4 ± after release, A averaged 81.7% and varied from 28.3% to 5.2%; 43.2 ± 12.8%), and Smallmouth Bass (10.5 ± 1.5%; 100% (Table 4; Figure 1). However, this mean value was in- 41.6 ± 14.3%). A paired t-test indicated that Tag l in the first fluenced by the extremely high estimates for Smallmouth Bass year after release did not differ between the larger-sized fish (11 estimates; mean = 95.2%) and hatchery trout (12 estimates; 1152 MEYER AND SCHILL

TABLE 2. Estimates of tag loss within 1 and 2 years of tagging for various ﬁsh species in Idaho.

Year-1 tag reporting Year-2 tag reporting Double-tagged fish Double-tagged fish reported as: reported as: Number of double- Single- Double- Year-1 tag loss Single- Double- Year-2 tag loss Species tagged fish released tagged tagged (%) ( ± 90% CI) tagged tagged (%) ( ± 90% CI) Crappies 2,530 39 650 2.9 ± 1.1 15 147 4.8 ± 3.1 Hatchery trout 3,080 108 553 8.8 ± 2.0 34 41 27.8 ± 13.0 Wild trout 1,079 31 143 9.7 ± 3.7 19 54 14.7 ± 7.0 Largemouth Bass 458 32 90 14.8 ± 5.1 20 21 30.3 ± 11.9 Smallmouth Bass 3,103 198 838 10.5 ± 1.5 24 14 41.6 ± 14.3 Walleye 633 20 77 11.4 ± 5.2 30 16 43.2 ± 12.8 Yellow Perch 309 1 43 1.1 ± 1.9 3 32 4.5 ± 3.7 Total 11,192 429 2,394 8.2 ± 0.7 145 325 17.8 ± 2.3

TABLE 3. Estimates of short-term mortality for tagged and untagged wild fish that were held in cages and tagged hatchery fish that were held in raceway net pens (no untagged hatchery fish were held). For each holding experiment, all tagged and untagged fish were held in one cage.

Tagged ﬁsh Untagged ﬁsh Percent Percent Fish TL (mm) mortality Fish TL (mm) mortality Water body Species Origin n Mean Range 1 d 7 d n Mean Range 1 d 7 d Brownlee Crappies Wild 20 193 185–225 0 5 20 193 180–215 0 0 Reservoir Brownlee Crappies Wild 20 217 200–228 0 0a 20 213 200–228 0 0a Reservoir C.J. Strike Crappies Wild 40 200 190–239 7.5a 40 199 195–215 5a Reservoir Brownlee Smallmouth Bass Wild 14 395 205–465 0 0 14 381 310–470 0 0 Reservoir Brownlee Smallmouth Bass Wild 20 418 308–518 0 0a 20 390 306–475 0 0a Reservoir Downloaded by [Department Of Fisheries] at 19:14 20 July 2015 C.J. Strike Smallmouth Bass Wild 15 318 305–347 6.7 6.7 15 329 306–395 0 6.7 Reservoir Lake Lowell Largemouth Bass Wild 16 344 308–388 0 0 15 353 305–419 0 0 South Fork Rainbow Trout Wild 20 477 380–605 0 20 459 368–520 0 Snake River South Fork Rainbow Trout Wild 20 425 320–570 0 0 20 449 285–583 0 0 Boise River Ririe Cutthroat Trout Hatchery 500 285 0 Reservoir Hatchery Rainbow Trout Hatchery 500 280 0 1.0b Hatchery Rainbow Trout Hatchery 500 250 0 0.4b Hatchery Rainbow Trout Hatchery 500 260 0 1.8b Hatchery Rainbow Trout Hatchery 100 240 0 0b

aSampling occurred 8 d after release. bFish were held in the hatchery for 22–33 d. ANGLER TAG REPORTING SYSTEM 1153

TABLE 4. Annual rates of exploitation by anglers (adjusted), catch-and-release angling, and total mortality ( ± 90% CI) for each of 45 ﬁsh tagging events in Idaho.

Exploitation Catch-and-release Total mortality Year Water body Species Origin (%) angling (%) (%) 2006 Brownlee Reservoir Crappies Wild 24.1 ± 4.7 6.2 ± 5.1 62.8 ± 8.9 2007 Brownlee Reservoir Crappies Wild 29.3 ± 5.4 13.1 ± 5.4 90.1 ± 4.9 2008 Brownlee Reservoir Crappies Wild 22.1 ± 4.9 6.9 ± 4.9 85.2 ± 7.2 2009 Brownlee Reservoir Crappies Wild 30.3 ± 5.5 4.4 ± 5.5 93.5 ± 4.5 2006 C.J. Strike Reservoir Crappies Wild 21.6 ± 6.5 13.3 ± 6.5 82.4 ± 9.5 2007 C.J. Strike Reservoir Crappies Wild 21.9 ± 5.0 11.9 ± 5.0 50.1 ± 9.6 2008 C.J. Strike Reservoir Crappies Wild 26.0 ± 5.2 10 ± 5.2 66.9 ± 8.6 2009 C.J. Strike Reservoir Crappies Wild 38.6 ± 6.1 6.4 ± 6.1 86.3 ± 5.6 2006 Mann Lake Crappies Wild 44.3 ± 7.9 13.8 ± 7.9 85.2 ± 6.3 2006 Ben Ross Reservoir Largemouth Bass Wild 14.4 ± 10.4 11.5 ± 10.4 75.5 ± 21.8 2007 Ben Ross Reservoir Largemouth Bass Wild 8.2 ± 5.5 41.1 ± 6.1 75.3 ± 10.9 2006 Pend Oreille River Largemouth Bass Wild 30.9 ± 8.7 39.3 ± 8.8 82.1 ± 6.7 2006 Brownlee Reservoir Smallmouth Bass Wild 18.7 ± 5.0 28.8 ± 5.0 98.5 ± 2.0 2007 Brownlee Reservoir Smallmouth Bass Wild 19.4 ± 5.0 22.0 ± 5.0 98.3 ± 2.2 2008 Brownlee Reservoir Smallmouth Bass Wild 31.2 ± 6.3 27.4 ± 6.3 97.6 ± 2.3 2009 Brownlee Reservoir Smallmouth Bass Wild 15.3 ± 4.9 15.3 ± 4.9 97.5 ± 3.5 2006 Cascade Reservoir Smallmouth Bass Wild 11.8 ± 7.7 7.9 ± 7.7 74.3 ± 21.5 2006 C.J. Strike Reservoir Smallmouth Bass Wild 20.8 ± 6.0 41.6 ± 6.1 98.5 ± 2.0 2007 C.J. Strike Reservoir Smallmouth Bass Wild 33.7 ± 6.5 44.7 ± 6.6 99.1 ± 1.3 2008 C.J. Strike Reservoir Smallmouth Bass Wild 22.5 ± 5.5 27.5 ± 5.5 98.6 ± 1.9 2009 C.J. Strike Reservoir Smallmouth Bass Wild 33.5 ± 6.5 23.6 ± 6.5 98.8 ± 1.7 2007 Dworshak Reservoir Smallmouth Bass Wild 21.9 ± 5.4 25.7 ± 5.4 88.0 ± 5.4 2006 Milner Reservoir Smallmouth Bass Wild 13.6 ± 4.2 55.3 ± 4.4 98.0 ± 1.9 2009 Oakley Reservoir Walleye Wild 18.1 ± 5.9 6.0 ± 5.8 71.7 ± 12.4 2007 Salmon Falls Creek Walleye Wild 13.6 ± 3.3 8.2 ± 3.2 26.4 ± 8.1 Reservoir 2009 Cascade Reservoir Yellow Perch Wild 17.0 ± 4.4 3.2 ± 4.4 57.3 ± 12.2 2006 Coeur d’Alene River Rainbow Trout, Wild 11.0 ± 8.8 16.4 ± 8.8 89.4 ± 15.3 Cutthroat Trout, and hybrids 2007 Henry’s Lake Cutthroat Trout and Wild 1.9 ± 1.3 1.9 ± 1.3 56.9 ± 22.4 hybrids 2006 Moyie River Rainbow Trout and Wild 4.0 ± 2.5 4.0 ± 2.5 85.1 ± 14.9 Downloaded by [Department Of Fisheries] at 19:14 20 July 2015 Brook Trout 2006 South Fork Snake River Rainbow Trout Wild 8.8 ± 4.5 22.0 ± 4.6 75.9 ± 11.3 2007 South Fork Snake River Rainbow Trout and Wild 11.9 ± 3.6 10.7 ± 3.6 82.9 ± 7.9 hybrids 2006 Williams Lake Rainbow Trout Wild 19.8 ± 6.8 5.7 ± 6.8 61.4 ± 14.7 2007 Williams Lake Rainbow Trout Wild 9.4 ± 4.8 4.7 ± 4.8 86.2 ± 13.9 2008 Anderson Ranch Rainbow Trout Hatchery 7.5 ± 2.7 1.1 ± 2.7 84.8 ± 11.8 Reservoir 2007 Boise River Rainbow Trout Hatchery 26.9 ± 6.2 26.3 ± 6.2 100 2006 Cascade Reservoir Rainbow Trout Hatchery 5.4 ± 2.1 2.1 ± 2.1 90.7 ± 9.2 2006 Chesterﬁeld Reservoir Rainbow Trout Hatchery 10.2 ± 4.0 7.2 ± 4.0 63.5 ± 14.1 2007 Glendale Reservoir Rainbow Trout × Hatchery 38.4 ± 7.3 13.2 ± 7.2 98.6 ± 2.0 Cutthroat Trout hybrids 2006 Lake Walcott Rainbow Trout Hatchery 14.3 ± 3.5 2.9 ± 3.4 49.3 ± 10.8 1154 MEYER AND SCHILL

TABLE 4. Continued.

Exploitation Catch-and-release Total mortality Year Water body Species Origin (%) angling (%) (%) 2007 Little Wood Reservoir Rainbow Trout × Hatchery 3.0 ± 2.2 0 100 Cutthroat Trout hybrids 2006 Lucky Peak Reservoir Rainbow Trout Hatchery 17.9 ± 5.2 5.4 ± 5.2 93.9 ± 6.0 2007 Mann Creek Reservoir Rainbow Trout Hatchery 39.4 ± 7.3 2.4 ± 7.3 91.6 ± 5.2 2006 Mann Lake Rainbow Trout Hatchery 23.8 ± 6.2 13.2 ± 6.2 100 2007 North Fork Payette River Rainbow Trout Hatchery 10.5 ± 3.0 7.5 ± 3.0 100 2008 Ririe Reservoir Cutthroat Trout Hatchery 6.0 ± 3.1 4.8 ± 3.1 34.9 ± 17.7

mean = 84.3%). Mean A for wild fish populations not including expenditure was not prohibitive. Moreover, a large percentage Smallmouth Bass was 73.6%. of the reward payments were needed to determine the value that Several species showed graphical evidence of size-selective would elicit a λ of 100% (see Meyer et al. 2012a). With this vulnerability to anglers (Figure 2). Vulnerability to anglers gen- information now at hand, future studies will not need to release erally increased at the smallest sizes, reached an apex at about so many reward tags, and $200 tags will not be necessary. We 250–350 mm (depending on the species), and declined beyond believe that in most instances, a nonreward : high-reward tag- the apex. Based on ANOVA results, this pattern of higher vul- ging ratio of about 10:1 will provide adequate estimates of λ nerability for intermediate-sized fish and lower vulnerability for at individual water bodies. If a statewide or provincial program smaller and/or larger fish was statistically significant for wild is established, reward tags will only be required in some of the trout (F = 8.61, df = 6, P < 0.0001), crappies (F = 2.73, df water bodies, and the release of fewer than 100 high-reward = 2, P = 0.07), and Largemouth Bass (F = 2.16, df = 4, P = tags per year would probably provide an adequate estimate of 0.07). The pattern was similar but not statistically significant for λ across the area of inference. It should be noted that values Smallmouth Bass (F = 1.17, df = 4, P = 0.15), and there was of λ are likely to be higher for species that anglers are more no apparent relationship for Yellow Perch (F = 0.40, df = 1, likely to harvest and lower for species that anglers are more P = 0.75) or Walleyes (F = 0.90, df = 4, P = 0.48). No such likely to catch and release (Meyer et al. 2012a), so we rec- analyses could be conducted for hatchery trout because all fish ommend estimating exploitation with λ values that are species were stocked at relatively the same size (i.e., 250 mm). specific or group specific (e.g., wild trout). Following these sug- gestions would yield a substantial reduction in reward payments compared with those employed in the current study and would DISCUSSION likely involve little sacrifice in confidence for actual estimates of The statewide angler tag reporting system allowed us to pro- λ or exploitation. Certainly, the total costs associated with con- duce a multitude of estimates of u with reasonably tight conducting statewide evaluations of angler exploitation using tag fidence bounds, which Idaho fisheries managers have subse- returns would be less than the costs of using creel surveys and quently found to be useful in understanding and managing these population estimates at the same waters to obtain exploitation Downloaded by [Department Of Fisheries] at 19:14 20 July 2015 sport fish populations. Using this system, IDFG biologists now estimates (see Miranda and Bettoli 2007). tag 30,000–40,000 fish/year to evaluate angler exploitation in Our results indicate that rates of exploitation by anglers in a variety of Idaho’s resident fisheries, especially those target- Idaho were highest for harvest-oriented coolwater and warmwa- ing the roughly 2 million catchable-sized (i.e., TL = 250 mm) ter fisheries and were similar to rates of exploitation typical of hatchery Rainbow Trout that are stocked annually in put-and- other North American fisheries. For example, Allen et al. (1998) take fisheries across the state (Cassinelli and Koenig 2013); reviewed 18 estimates of crappie exploitation, finding an average stocking rates and locations have been adjusted accordingly. of 48% and a range of 0–84%, slightly higher than the average The study design and statistical analyses presented herein were (29%) and range (22–44%) in this study. Baccante and Colby relatively simple yet practical and provided a straightforward (1996) similarly reviewed 46 estimates of Walleye exploitation method of producing accurate and precise confidence bounds in North America, and their average (21%) was similar to that around estimates of u , relative to more complicated methods, identified for Walleyes in the present study (16%). In reviewing such as those using simulation or resampling procedures (e.g., 32 studies of Largemouth Bass exploitation in North America, Miranda et al. 2002; Isermann et al. 2005). Allen et al. (2008) found that harvest from 1953 to 2003 was A costly aspect of the program was paying anglers a total parabolic shaped due to voluntary catch-and-release behavior of about $90,000 for high-reward tag returns. However, when since 1990; mean exploitation from 1990 to 2003 was 18%, considering the cost per estimate of u produced ($2,000), the identical to the rate we found for Largemouth Bass. Finally, ANGLER TAG REPORTING SYSTEM 1155

0.30 540 Crappie Largemouth bass 0.30 57 2085 165 200 135 0.25 235 0.20 0.20 76 0.10 0.15

0.10 0.00 200 250 300 350 400 450 500 200 250 300 350 400 450 500 Smallmouth bass 0.20 977 325 Wild trout 80 470 98 435 1935 0.15 117 0.25 67

0.10 294 400 309 0.15 0.05 173

0.00 0.05 200 250 300 350 400 450 500

Return rate (% ± 90% CIs) 200 0.20 0.25 Walleye 186 Yellow perch 136 242 62 154 0.20 199 0.15 0.15 82

0.10 0.10 0.05 0.00 0.05 200 250 300 350 400 450 500 200 250 300 350 400 450 500 Fish length (mm)

FIGURE 2. Percentage of tag returns ( ± 90% CI) in various size-at-tagging bins (50-mm TL-groups) for ﬁsh species in Idaho. Numbers above bars indicate sample sizes within each size bin.

Marinac-Sanders and Coble (1981) estimated exploitation for the mean A of 76% for wild trout in this study (mostly Rainbow Downloaded by [Department Of Fisheries] at 19:14 20 July 2015 Smallmouth Bass and summarized estimates for seven other Trout and Cutthroat Trout) was much higher than the average studies around North America, ﬁnding an average of 41%; of 47% reported by Meyer et al. (2012b) for 24 Rainbow Trout although our average for Smallmouth Bass (22%) was much populations in Idaho. Estimates of A for the remaining species lower, it was nevertheless within the range of estimates com- in our study appeared to be more realistic. For example, the piled by Marinac-Sanders and Coble (1981). This difference for mean for crappies (78%) was similar to the mean (75%) of 33 Smallmouth Bass can be partly explained by the fact that the North American estimates summarized by Allen et al. (1998); previous estimates were all made prior to the recent adoption estimates for Yellow Perch and Walleyes in this study were of the voluntary catch-and-release ethic by many bass anglers similarly in accordance with other estimates in the literature (Slipke et al. 1998; Myers et al. 2008). (e.g., Quist et al. 2004; Schoenebeck and Brown 2011). Total In contrast to the plausibility of estimates of u in the present annual mortality for stocked catchable-sized trout in Idaho often study, estimates of A were dubiously high for some species, approaches 100% in both lotic (High and Meyer 2009) and lentic especially Smallmouth Bass and wild trout. For example, the (Koenig and Meyer 2011) environments, and estimates from mean A for Smallmouth Bass populations in this study was the present study concur with those prior ﬁndings. Finally, the 95%—much higher than the average of 54% across eight North mean for Largemouth Bass (77%) in our study was similar to but American studies (Marinac-Sanders and Coble 1981). Similarly, slightly higher than the mean (64%) from 34 North American 1156 MEYER AND SCHILL

studies reviewed by Allen et al. (1998), although our estimate mortality associated with anchor tags is generally thought to be was based on only three tagging events. so low that biologists often assume zero mortality without any The implausibility of estimates of A for some species could evaluation (e.g., Knapp et al. 1991; Muoneke 1994; Schultz and have stemmed from violations of a number of important as- Robinson 2002). If tagging and handling in our study resulted in sumptions. At least one of the potentially violated assumptions more severe mortality that was delayed for several months, this was that all sizes of fish have equal vulnerability to anglers could have reduced year-2 tag returns and perhaps partly ex- (Figure 2). For crappies, basses, and wild trout, the smallest fish plained our implausibly high estimates of A. However, because became more vulnerable as they grew; this could have resulted tagging with anchor tags is generally believed to cause little in higher-than-expected tag returns by anglers in year 2, which initial mortality and even less delayed mortality (Nielsen 1992; would have led to underestimation of total mortality. However, Pollock et al. 2001), delayed tagging mortality was unlikely to the vast majority of fish tagged in our study were at or beyond have had an appreciable impact on our estimates of A. the apex of the relationship between fish size and vulnerability Another possible explanation for the dearth of tag returns to angling, especially for Smallmouth Bass (96%), Largemouth in year 2 is violation of the assumption that catch-and-release Bass (88%), and wild trout (72%). In contrast, for crappies, a mortality was negligible. Violation of this assumption would greater percentage of tagged fish (73%) were at or before the have had the largest impacts on estimates of A for Smallmouth apex. Thus, for most of the tagged bass and wild trout in our Bass and Largemouth Bass, as they had the highest rates of study, vulnerability to anglers was lower in year 2 than in year catch-and-release angling. However, although catch-and-release 1, which likely resulted in reduced tag returns in year 2 for these mortality can be high for black bass in some situations (reviewed species and consequently led to an overestimation of their to- by Cooke et al. 2002), it is generally considered to be relatively tal mortality. Statistical significance in this relationship was not low (Siepker et al. 2007); thus, high rates of catch-and-release evident for Smallmouth Bass, but graphically the assumption of mortality seem unlikely to be the primary cause of the inflated equal vulnerability across all sizes of fish nevertheless appeared estimates of A. Our estimates of A also involved the assumption tenuous (Figure 2). Size-selective angler catch is not uncommon that if an angler caught a tagged fish but chose to not report the in freshwater sport fisheries (e.g., Serns and Kempinger 1981; tag, the angler released the fish with the tags intact. However, it Miranda and Dorr 2000), but accounts are often anecdotal and is possible that some anglers were disgruntled because the fish poorly documented (e.g., Miranda and Dorr 2000), and the direc- they caught had tags attached to them, and uncooperative anglers tion of change is not necessarily consistent across species. For may have chosen to pull all of the tags off before releasing the example, larger White Bass Morone chrysops were more vul- fish and to avoid reporting the tags. This would have inflated the nerable to anglers in Kansas reservoirs (Schultz 2004), whereas actual Tag l, and the inflation would not be detectable since our vulnerability of crappies declined for larger fish in southeastern estimates of Tag l were derived from double-tagged fish that had North America (Miranda and Dorr 2000). lost only one of the two tags. A number of possible mechanisms are likely responsible for Taking all of these assumptions into consideration, the in- what we are terming “unequal vulnerability in angler catch,” flated estimates of A we observed for some species, particularly including selective angling gear, fish morphology, fish behav- wild trout and Smallmouth Bass, were perhaps caused by the ior, habitat preferences, and seasonality of angler effort (e.g., cumulative effects of slight violations of several assumptions. Miranda and Dorr 2000; Cooke et al. 2005; Heermann et al. Often, little or no effort is made to evaluate many of the assump- 2013). Perhaps the most likely explanation is a probable shift in tions associated with estimating A from two consecutive years of behavior or habitat preference, with larger fish moving to deeper angler tag returns (e.g., Rawstron and Hashagen 1972; Rieman

Downloaded by [Department Of Fisheries] at 19:14 20 July 2015 water, making them less vulnerable to anglers. Such a behav- 1987; Muoneke 1994; Schultz and Robinson 2002). Based on ioral shift of larger fish to deeper water has been observed for the present results, caution should be used by biologists who are many of the species in our study, including crappies (Markham considering an angler tagging study to estimate A for fish pop- et al. 1991), black basses (Probst et al. 1984; Cole and Moring ulations unless it can be shown that the assumptions necessary 1997), and wild trout (Baltz et al. 1991; Jakober et al. 2000). to estimate A are not being violated. Estimates of Tag m in our study were based on only 14 hold- Tag loss rates for the species in this study were generally ing trials that were of short duration. Nevertheless, our results lower than those reported in most other studies using T-bar concur with literature findings that suggest minimal tagging anchor tags. For example, year-1 tag loss for crappies was lower mortality when anchor tags are used. For example, mortality in this study (mean = 2.9%) than in studies by Larson et al. was 3% over 2–4 d for Yellow Perch in South Dakota (Scholten (1991; mean = 18%) and Miranda et al. (2002; 47%). Tag loss et al. 2002), 11% over 2 d for crappies in southeastern North was similarly low for Largemouth Bass in our study (mean America (Miranda et al. 2002), 0% over 191 d for Largemouth = 14.8% in year 1) compared with studies by Tranquilli and Bass in an Illinois pond (Tranquilli and Childers 1982), 0% Childers (1982; 56% after 6 months) and Hartman and Janney over 300 d for Brook Trout in a Wisconsin pond (Carline and (2006; 57% after 13 months). The estimate of 8.8% annual Brynildson 1972), and 0% over 160 d for Arctic Char Salvelinus tag loss for hatchery trout was similar to estimates reported by alpinus in a Canadian hatchery (Rikardsen et al. 2002). Tagging Carline and Brynildson (1972; 4% after 8 months), Mourning ANGLER TAG REPORTING SYSTEM 1157

et al. (1994; 11% after 4 months), and Walsh and Winkelman REFERENCES (2004; 9% after 6 months). Often, the cumulative tag loss after Allen, M. S., and J. E. Hightower. 2010. Fish population dynamics: mortality, 2 years in this study (e.g., 5% for crappies) was still lower than growth, and recruitment. Pages 43–80 in W. A. Hubert and M. C. Quist, published rates of tag loss within 1 year. editors. Inland fisheries management in North America, 3rd edition. American Fisheries Society, Bethesda, Maryland. Allen, M. S., L. E. Miranda, and R. E. Brock. 1998. Implications of compensatory and additive mortality to the management of selected sport fish CONCLUSIONS populations. Lakes and Reservoirs: Research and Management 3:67–79. Despite the well-reasoned concerns of Miranda et al. (2002) Allen, M. S., C. J. Walters, and R. Myers. 2008. Temporal trends in Largemouth regarding cost and imprecision, we believe our results demon- Bass mortality, with fishery implications. North American Journal of Fisheries Management 28:418–427. strate that using the high-reward tagging method to estimate u Baccante, D. A., and P.J. Colby. 1996. Harvest, density and reproductive charac- can be an effective tool for inland fisheries management. As teristics of North American Walleye populations. Annales Zoologici Fennici noted above, nearly half of the estimates of u herein had 90% 33:601–615. CIs that were within 25% of the point estimate (Table 4), and Baltz, D. M., B. Vondracek, L. R. Brown, and P. B. Moyle. 1991. Seasonal many more estimates were low enough that a fairly strong sense changes in microhabitat selection by Rainbow Trout in a small stream. Trans- actions of the American Fisheries Society 120:166–176. of angler harvest impact was discernible despite imprecision in Carline, R. F., and O. M. Brynildson. 1972. Effects of the Floy anchor tag on the estimates. From a management perspective, perhaps only the growth and survival of Brook Trout (Salvelinus fontinalis). Journal of the 3 of the 45 estimates of u in Table 4 had error bounds that Fisheries Research Board of Canada 29:458–460. were too wide to be informative. The straightforward concept Cassinelli, J., and M. Koenig. 2013. Hatchery trout evaluations. Idaho Depart- of angler exploitation estimates makes them easily understood ment of Fish and Game, Annual Performance Report 13–04, Boise. Cohen, J. 1990. Things I have learned (so far). American Psychologist 45:1304– and thus perhaps more readily accepted by biologists, anglers, 1312. and policymakers alike. Fish managers routinely use such in- Cole, M. B., and J. R Moring. 1997. Relation of adult size to movements and formation in public meetings to (1) inform the angling public distribution of Smallmouth Bass in a central Maine Lake. Transactions of the of the harvest rates for specific fish populations or (2) defend American Fisheries Society 126:815–821. or modify harvest regulations as needed (Allen and Hightower Cooke, S. J., B. L. Barthel, C. D. Suski, M. J. Siepker, and D. P. Philipp. 2005. Influence of circle hook size on hooking efficiency, injury, and size 2010). As long as the assumptions of the high-reward tagging selectivity of Bluegill with comments on circle hook conservation benefits in method can be met and as long as correction parameters can be recreational fisheries. North American Journal of Fisheries Management 25: estimated if necessary, this method is likely to be cost effective 211–219. for biologists attempting to evaluate the effects of angler harvest Cooke, S. J., J. F. Schreer, D. H. Wahl, and D. P. Philipp. 2002. Physiologi- on the fisheries they manage. cal impacts of catch-and-release angling practices on Largemouth Bass and Smallmouth Bass. Pages 489–512 in D. P. Philipp and M. S. Ridgway, edi- Conversely, our study demonstrates the potential incompat- tors. Black bass: ecology, conservation, and management. American Fisheries ibility of the tag return method in estimating A. When vulner- Society, Symposium 31, Bethesda, Maryland. ability to anglers is not equal across all size-classes of fish, Danner, G. R., K. W. Muto, A. M. Zieba, C. M. Stillman, J. A. Seggio, and S. T. estimates of A may prove spurious. Simultaneous monitoring of Ahmad. 2011. Spearmint (l-carvone) oil and wintergreen (methyl salicylate) growth in the fish population by using other field sampling meth- oil emulsion is an effective immersion anesthetic of fishes. Journal of Fish and Wildlife Management 2:146–155. ods may allow biologists to model and correct size-selectivity Fleiss, J. L. 1981. Statistical methods for rates and proportions, 2nd edition. issues before estimating A; future research along these lines Wiley, New York. would be useful. Alternatively, other more commonly employed Guy, C. S., H. L. Blankenship, and L. A. Nielsen. 1996. Tagging and mark- approaches for estimating A, such as building catch curves, may ing. Pages 353–383 in B. R. Murphy and D. W. Willis, editors. Fisheries techniques, 2nd edition. American Fisheries Society, Bethesda, Maryland. Downloaded by [Department Of Fisheries] at 19:14 20 July 2015 be needed in conjunction with tag return estimates of u if a de- Hartman, K. J., and E. C. Janney. 2006. Visual implant elastomer and anchor tailed investigation of population mortality sources is desired. tag retention in Largemouth Bass. North American Journal of Fisheries Man- agement 26:665–669. Heermann, L., M. Emmrich, M. Heynen, M. Dorow, U. Konig,¨ J. Borcherd- ACKNOWLEDGMENTS ing, and R. Arlinghaus. 2013. Explaining recreational angling catch rates We thank the IDFG staff members who assisted with tagging of Eurasian Perch, Perca fluviatilis: the role of natural and fishing- efforts, especially Art Butts, Patrick Kennedy, Tony Lamansky, related environmental factors. Fisheries Management and Ecology 20: 187–200. and Liz Mamer for tagging assistance and data management. We Henny, C. J., and K. P. Burnham. 1976. A reward band study of mallards to are grateful to Rick Alsager and the Nampa Fish Hatchery crew estimate reporting rates. Journal of Wildlife Management 40:1–14. and to Joe Chapman and the Hagerman State Fish Hatchery crew High, B., and K. A. Meyer. 2009. Survival and dispersal of hatchery triploid for their assistance with the hatchery trout used in this study. Rainbow Trout in an Idaho river. North American Journal of Fisheries Man- Special thanks are extended to the U.S. Bureau of Reclamation agement 29:1797–1805. Isermann, D. A., D. W. Willis, D. O. Lucchesi, and B. G. Blackwell. 2005. Sea- for providing a grant to cover tag reward costs. Mike Quist, sonal harvest, exploitation, size selectivity, and catch preferences associated Robert Arlinghaus, Mike Allen, and Jacob Hughes provided with winter Yellow Perch anglers on South Dakota lakes. North American helpful critiques on earlier versions of the paper. Journal of Fisheries Management 25:827–840. 1158 MEYER AND SCHILL

Jakober, M. J., T. E. McMahon, and R. F. Thurow. 2000. Diel habitat partitioning Probst, W. E., C. F. Rabeni, W. G. Covington, and R. F. Marteney. 1984. Re- by Bull Charr and Cutthroat Trout during fall and winter in Rocky Mountain source use by stream-dwelling Rock Bass and Smallmouth Bass. Transactions streams. Environmental Biology of Fishes 59:79–89. of the American Fisheries Society 113:283–294. Knapp, M., L. Talent, and E. Maughan. 1991. A comparison of two exploitation Quist, M. C., J. L. Stephen, C. S. Guy, and R. D. Schultz. 2004. Age structure rate estimators for crappie in an Oklahoma reservoir. U.S. Forest Service and mortality of Walleyes in Kansas reservoirs: use of mortality caps to es- General Technical Report RM 207:283–286. tablish realistic management objectives. North American Journal of Fisheries Koenig, M. K., and K. A. Meyer. 2011. Relative performance of diploid and Management 24:990–1002. triploid catchable Rainbow Trout stocked in Idaho lakes and reservoirs. North Rawstron, R. R., and K. A. Hashagen. 1972. Mortality and survival rates of American Journal of Fisheries Management 31:605–613. tagged Largemouth Bass Micropterus salmoides at Merle Collins Reservoir. Larson, S. C., B. Saul, and S. Schleiger. 1991. Exploitation and survival of Black California Fish and Game 58:221–230. Crappies in three Georgia reservoirs. North American Journal of Fisheries Rikardsen, A. H., M. Woodgate, and D. A. Thompson. 2002. A comparison Management 11:604–613. of Floy and soft VI alpha tags on hatchery Arctic Charr, with emphasis Marinac-Sanders, P., and D. W. Coble. 1981. The Smallmouth Bass population on tag retention, growth and survival. Environmental Biology of Fishes 64: and fishery in a northern Wisconsin lake, with implications for other waters. 269–273. North American Journal of Fisheries Management 1:15–20. Ricker, W. E. 1975. Computation and interpretation of biological statistics of Markham, J. L., D. L. Johnson, and R. W. Petering. 1991. White Crappie summer fish populations. Fisheries Research Board of Canada Bulletin 191. movements and habitat use in Delaware Reservoir, Ohio. North American Rieman, B. E. 1987. Fishing and population dynamics of Largemouth Bass Journal of Fisheries Management 11:504–512. (Micropterus salmoides) in select northern Idaho lakes. Doctoral dissertation. McCammon, G. W., and D. A. LaFaunce. 1961. Mortality rates and movement University of Idaho, Moscow. in the Channel Catfish population of the Sacramento Valley. California Fish Schoenebeck, C. W., and M. L. Brown. 2011. Gender- and year-specific mortal- and Game 47:5–26. ity of Yellow Perch with evidence of compensatory mortality. North American McFadden, J. T. 1961. A study of the Brook Trout (Salvelinus fontinalis). Journal of Fisheries Management 31:474–482. Wildlife Monographs 7:1–73. Scholten, G. W., D. A. Isermann, and D. W. Willis. 2002. Retention and survival Meyer, K. A., F. S. Elle, J. A. Lamansky, E. R. J. M. Mamer, and A. E. Butts. rates associated with the use of t-bar anchor tags in marking Yellow Perch. 2012a. A reward fish tagging study to estimate angler tag reporting rates in Proceedings of the South Dakota Academy of Science 81:35–38. Idaho. North American Journal of Fisheries Management 32:696–703. Schultz, R. D. 2004. Size-specific angling return rates of White Bass in Kansas Meyer, K. A., B. High, and F. S. Elle. 2012b. Effects of stocking catchable-size reservoirs. North American Journal of Fisheries Management 24:719–723. hatchery Rainbow Trout on wild Rainbow Trout abundance, survival, growth, Schultz, R. D., and D. A. Robinson Jr. 2002. Exploitation and mortality rates and recruitment. Transactions of the American Fisheries Society 141:224– of White Bass in Kansas reservoirs. North American Journal of Fisheries 237. Management 22:652–658. Miranda, L. E., and P. W. Bettoli. 2007. Mortality. Pages 229–277 in C. S. Guy Serns, S. L., and J. J. Kempinger. 1981. Relationship of angler exploitation to and M. L. Brown, editors. Analysis and interpretation of freshwater fisheries the size, age, and sex of Walleyes in Escanaba Lake, Wisconsin. Transactions data. American Fisheries Society, Bethesda, Maryland. of the American Fisheries Society 110:216–220. Miranda, L. E., R. E. Brock, and B. S. Dorr. 2002. Uncertainty of exploita- Siepker, M. J., K. G. Ostrand, S. J. Cooke, D. P. Philipp, and D. H. Wahl. 2007. A tion estimates made from tag returns. North American Journal of Fisheries review of the effects of catch-and-release angling on black bass, Micropterus Management 22:1358–1363. spp.: implications for conservation and management of populations. Fisheries Miranda, L. E., and B. S. Dorr. 2000. Size selectivity of crappie angling. North Management and Ecology 14:91–101. American Journal of Fisheries Management 20:706–710. Slipke, J. W., M. J. Maceina, V. H. Travnichek, and K. C. Weathers. 1998. Mourning, T. E., K. D. Fausch, and C. Gowan. 1994. Comparison of visible Effects of a 356-mm minimum length limit on the population characteristics implant tags and floy anchor tags on hatchery Rainbow Trout. North American and sport fishery of Smallmouth Bass in the Shoals Reach of the Tennessee Journal of Fisheries Management 14:636–642. River, Alabama. North American Journal of Fisheries Management 18:76–84. Muoneke, M. I. 1994. Dynamics of a heavily exploited Texas White Bass Stephens, P. A., S. W. Buskirk, G. D. Hayward, and C. Martinez Del Rio. 2005. population. North American Journal of Fisheries Management 14:415–422. Information theory and hypothesis testing: a call for pluralism. Journal of Myers, R. A., J. Taylor, M. Allen, and T. Bonvechio. 2008. Temporal trends in Applied Ecology 42:4–12. voluntary release of Largemouth Bass. North American Journal of Fisheries Tranquilli, J. A., and W. F. Childers. 1982. Growth and survival of Largemouth Management 28:428–433. Bass tagged with floy anchor tags. North American Journal of Fisheries Downloaded by [Department Of Fisheries] at 19:14 20 July 2015 Nichols, J. D., R. J. Blohm, R. E. Reynolds, R. E. Trost, J. E. Hines, and J. Management 2:184–187. P. Bladen. 1991. Band reporting rates for mallards with reward bands of Walsh, M. G., and D. L. Winkelman. 2004. Anchor and visible implant elastomer different dollar values. Journal of Wildlife Management 55:119–126. tag retention by hatchery Rainbow Trout stocked into an Ozark stream. North Nielsen, L. A. 1992. Methods of marking fish and shellfish. American Fisheries American Journal of Fisheries Management 24:1435–1439. Society, Special Publication 23, Bethesda, Maryland. Walters, C. J., and S. J. D. Martell. 2004. Fisheries ecology and management. Pollock, K. H, J. M. Hoenig, W. S. Hearn, and B. Calingaert. 2001. Tag reporting Princeton University Press, Princeton, New Jersey. rate estimation 1: an evaluation of the high-reward tagging method. North Yates, F. 1980. Sampling methods for censuses and surveys, 4th edition. Charles American Journal of Fisheries Management 21:521–532. Griffin, London. This article was downloaded by: [Department Of Fisheries] On: 20 July 2015, At: 19:14 Publisher: Taylor & Francis Informa Ltd Registered in England and Wales Registered Number: 1072954 Registered office: 5 Howick Place, London, SW1P 1WG

North American Journal of Fisheries Management Publication details, including instructions for authors and subscription information: http://www.tandfonline.com/loi/ujfm20 Sampling Characteristics and Calibration of Snorkel Counts to Estimate Stream Fish Populations Daniel M. Weaverad, Thomas J. Kwakb & Kenneth H. Pollockc a North Carolina Cooperative Fish and Wildlife Research Unit, Department of Applied Ecology, North Carolina State University, Campus Box 7617, Raleigh, North Carolina 27695, USA b U.S. Geological Survey, North Carolina Cooperative Fish and Wildlife Research Unit, Department of Applied Ecology, North Carolina State University, Campus Box 7617, Raleigh, North Carolina 27695, USA c Department of Applied Ecology, North Carolina State University, Campus Box 7617, Raleigh, Click for updates North Carolina 27695, USA d Present address: Maine Cooperative Fish and Wildlife Research Unit, Department of Wildlife, Fisheries, and Conservation Biology, University of Maine, 5755 Nutting Hall, Orono, Maine 04469, USA Published online: 10 Nov 2014.

To cite this article: Daniel M. Weaver, Thomas J. Kwak & Kenneth H. Pollock (2014) Sampling Characteristics and Calibration of Snorkel Counts to Estimate Stream Fish Populations, North American Journal of Fisheries Management, 34:6, 1159-1166, DOI: 10.1080/02755947.2014.951808 To link to this article: http://dx.doi.org/10.1080/02755947.2014.951808

PLEASE SCROLL DOWN FOR ARTICLE

Taylor & Francis makes every effort to ensure the accuracy of all the information (the “Content”) contained in the publications on our platform. However, Taylor & Francis, our agents, and our licensors make no representations or warranties whatsoever as to the accuracy, completeness, or suitability for any purpose of the Content. Any opinions and views expressed in this publication are the opinions and views of the authors, and are not the views of or endorsed by Taylor & Francis. The accuracy of the Content should not be relied upon and should be independently verified with primary sources of information. Taylor and Francis shall not be liable for any losses, actions, claims, proceedings, demands, costs, expenses, damages, and other liabilities whatsoever or howsoever caused arising directly or indirectly in connection with, in relation to or arising out of the use of the Content. This article may be used for research, teaching, and private study purposes. Any substantial or systematic reproduction, redistribution, reselling, loan, sub-licensing, systematic supply, or distribution in any form to anyone is expressly forbidden. Terms & Conditions of access and use can be found at http:// www.tandfonline.com/page/terms-and-conditions North American Journal of Fisheries Management 34:1159–1166, 2014 C American Fisheries Society 2014 ISSN: 0275-5947 print / 1548-8675 online DOI: 10.1080/02755947.2014.951808

ARTICLE

Sampling Characteristics and Calibration of Snorkel Counts to Estimate Stream Fish Populations

Daniel M. Weaver1 North Carolina Cooperative Fish and Wildlife Research Unit, Department of Applied Ecology, North Carolina State University, Campus Box 7617, Raleigh, North Carolina 27695, USA Thomas J. Kwak* U.S. Geological Survey, North Carolina Cooperative Fish and Wildlife Research Unit, Department of Applied Ecology, North Carolina State University, Campus Box 7617, Raleigh, North Carolina 27695, USA Kenneth H. Pollock Department of Applied Ecology, North Carolina State University, Campus Box 7617, Raleigh, North Carolina 27695, USA

Abstract Snorkeling is a versatile technique for estimating lotic fish population characteristics; however, few investigators have evaluated its accuracy at population or assemblage levels. We evaluated the accuracy of snorkeling using prepositioned areal electrofishing (PAE) for estimating fish populations in a medium-sized Appalachian Mountain river during fall 2008 and summer 2009. Strip-transect snorkel counts were calibrated with PAE counts in identical locations among macrohabitats, fish species or taxa, and seasons. Mean snorkeling efficiency (i.e., the proportion of individuals counted from the true population) among all taxa and seasons was 14.7% (SE, 2.5%), and the highest efficiencies were for River Chub Nocomis micropogon at 21.1% (SE, 5.9%), Central Stoneroller Campostoma anomalum at 20.3% (SE, 9.6%), and darters (Percidae) at 17.1% (SE, 3.7%), whereas efficiencies were lower for shiners (Notropis spp., Cyprinella spp., Luxilus spp.) at 8.2% (SE, 2.2%) and suckers (Catostomidae) at 6.6% (SE, 3.2%). Macrohabitat type, fish taxon, or sampling season did not significantly explain variance in snorkeling efficiency. Mean snorkeling detection probability (i.e., probability of detecting at least one individual of a taxon) among fish taxa and seasons was 58.4% (SE, 6.1%). We applied the efficiencies from our calibration study to adjust snorkel counts from an intensive snorkeling survey conducted in a nearby reach. Total fish density estimates from strip-transect counts

Downloaded by [Department Of Fisheries] at 19:15 20 July 2015 adjusted for snorkeling efficiency were 7,288 fish/ha (SE, 1,564) during summer and 15,805 fish/ha (SE, 4,947) during fall. Precision of fish density estimates is influenced by variation in snorkeling efficiency and sample size and may be increased with additional sampling effort. These results demonstrate the sampling properties and utility of snorkeling to characterize lotic fish assemblages with acceptable efficiency and detection probability, less effort, and no mortality, compared with traditional sampling methods.

Fish-sampling gear bias and quantitative gear comparison form the scientific basis to assess and interpret fish population and calibration are important considerations in fisheries research density, condition, and vital rates, as well as assemblage struc- and management. Sampling techniques and data collection ture. Examination of the efficiency of sampling techniques

*Corresponding author: [email protected] 1Present address: Maine Cooperative Fish and Wildlife Research Unit, Department of Wildlife, Fisheries, and Conservation Biology, University of Maine, 5755 Nutting Hall, Orono, Maine 04469, USA. Received April 29, 2014; accepted July 23, 2014 1159 1160 WEAVER ET AL.

allows for enhanced understanding of sampling properties and assumed that PAE provided the most accurate, but not perfect, bias and may provide insight to reduce and quantitatively adjust estimate of the true fish population. techniques for future application (Peterson and Paukert 2009). Our objectives in this assessment were to (1) determine Visual estimation techniques have been used for decades whether PAE can be effectively applied as a comparative gear to quantify animal populations; however, many uncertainties for calibrating snorkel counts, (2) estimate snorkeling efficiency limit their accuracy and efficiency (Emlen 1971). Among vi- among taxa in the fish assemblage both spatially and temporally, sual techniques, snorkeling is an inexpensive, nonlethal, and and (3) apply taxon-specific estimates of snorkeling efficiency versatile method for quantitatively sampling fish populations from our empirical calibration to adjust strip-transect snorkel (Dolloff et al. 1996; Dunham et al. 2009), but observer effi- count data collected from a nearby study reach. ciency is presumably influenced by environmental conditions and fish behavior. Water turbidity, flow rate, and temperature STUDY AREA change temporally within and among streams and rivers, which We evaluated the accuracy of snorkel counts for estimating may affect observer efficiency (Hillman et al. 1992). Observer nongame fish densities in a southern Appalachian Mountain training, experience, and fish identification skills may also affect river, the North Toe River, in Mitchell County, North Carolina, efficiency of counts and the accuracy of characterizing assem- during fall 2008 and summer 2009. The North Toe River drains blages (Dunham et al. 2009). Furthermore, changes in the com- approximately 474 km2 and is a high-gradient, medium-sized position of local fish assemblages in response to environmental river that joins the Cane River to become the Nolichucky River, conditions may be influenced by a variety of factors, includ- a tributary to the French Broad River, which ultimately flows ing variable species behavior, habitat affinity, and life history west to the Tennessee River. We conducted our gear calibration (Kwak 1991; Lobb and Orth 1991; Grossman and Ratajczak study in a 0.5-km reach containing a diversity of pool, riffle, and 1998), which may affect snorkeling efficiency among seasons run habitats. (Ensign et al. 1995; Korman et al. 2010). We used the snorkeling efficiency estimates obtained from Calibrating snorkel counts is important when the primary our calibration study to adjust snorkel counts from a more inten- objective is estimating fish density, and a typical approach is to sive survey we conducted 11 km downstream, during summer apply two gears concurrently and compare the relative catch. and fall, in the town of Spruce Pine, North Carolina, over a 1-km Other investigators have calibrated snorkel counts by compar- reach (Weaver 2010; Weaver and Kwak 2013). The native fish ing them with a standardized population-estimating procedure, assemblage of the river is typical of the area and is composed usually multiple-pass electrofishing (Hankin and Reeves 1988; of Smallmouth Bass Micropterus dolomieu and sunfishes Lep- Mullner et al. 1998; Roni and Fayram 2000; Thurow et al. 2006), omis spp. as sport fishes that coexist with multiple species and or mark–recapture methods (Slaney and Martin 1987; Orell and families of coolwater and warmwater nongame fish. The river Erkinaro 2007; Pink et al. 2007). Multiple-pass electrofishing is also stocked seasonally by the North Carolina Wildlife Re- methods are most appropriate for application in small streams sources Commission with stream-dwelling salmonids to support where block nets can be effectively deployed, but are less suit- popular recreational trout fisheries (NCWRC 2009; Weaver and able for large streams or rivers. Mark–recapture methods have Kwak 2013). We examined the entire fish assemblage, stratified primarily been applied to game fish species (e.g., salmonids) that into taxa, based on their habitat use, behavior, and morpho- can survive capture, handling, marking, and recapture. Smaller, logical traits. These fish taxa included (1) Central Stoneroller more sensitive fish species, such as small-bodied cyprinids, may Campostoma anomalum, (2) River Chub Nocomis micropogon, be subject to variably high rates of sampling mortality or injury, (3) shiners Notropis spp., Cyprinella spp., and Luxilus spp., (4) Downloaded by [Department Of Fisheries] at 19:15 20 July 2015 especially when common electrofishing techniques are used suckers Catostomidae, and (5) darters Percidae. (Holliman et al. 2003a, 2003b; Snyder 2003). Faced with the challenge of working in a medium-sized river and with the objective to estimate sampling efficiency and METHODS density of nongame fish populations, we chose prepositioned Prepositioned areal electrofisher.—The PAE grid consisted areal electrofishing (PAE), a passive form of electrofishing, of two 15-m lengths of 0.65-mm-diameter stainless steel cable to calibrate snorkel counts. Prepositioned areal electrofishing that were spaced 1 m apart using two 7-cm-diameter PVC pipes has been used to effectively sample small- and medium-sized to form an effective sampling area of 15 m2. The grid was rivers (Bowen and Freeman 1998) and discrete macrohabitats, powered by AC from a 3,500-W generator and converter that such as pools, riffles, and runs (Bain et al. 1985; Walsh et al. regulated amperage output. The grid was connected to the power 2002; Schwartz and Herricks 2004), as well as diverse fish as- source by a 30-m electrical cord. We adjusted electric output semblages (Larimore and Garrels 1985; Grabowski and Isely prior to sampling to ensure fish immobilization. 2007). Furthermore, based on its high rate of fish immobiliza- Snorkeling calibration procedure.—During fall 2008 and tion (98%: Bain et al. 1985), high efficiency (100%: Janac and summer 2009, we used standard macrohabitat classifications Jurajda 2010), and low sampling bias (Ensign et al. 2002), we (Arend 1999) to select 12 transect sampling locations: four SAMPLING AND SNORKEL COUNTS TO ESTIMATE STREAM FISH 1161

pools, four riffles, and four runs. One PAE sample and one density estimate obtained by snorkeling (Dˆ js) by the corre- snorkeling survey were conducted at each transect during both sponding density estimate obtained from PAE (Dˆ je)as seasons. We sampled all transects within a 2-d period. For each transect, a homogenous area of macrohabitat was selected to Dˆ js encompass the entire grid sampling area. The grid was placed in qˆ = . (2) j ˆ the water parallel to flow, and a 15-min waiting period ensued D je to allow the sampling area to return to predisturbed conditions to minimize fright bias (Bain et al. 1985; Bowen and Freeman These snorkeling efficiency estimates represent the probability 1998). A single observer entered the water 5 m downstream of detecting an individual from a particular fish taxon as a pro- from the grid, snorkeled slowly upstream through the center portion of fish of that taxon present. Efficiency was calculated of the grid, and identified and counted all fish visible within separately for fish taxa and macrohabitat types in a similar man- and outside of the grid. A weighted marker was placed at each ner. The SE of efficiency estimates was calculated as a measure location of fish observed outside of the grid. The observer min- of precision following the equation for SE of a sample mean imized counting fish twice by moving constantly through the (Sokal and Rohlf 1981). transect keeping recorded individuals within peripheral vision. Statistical analysis.—We compared snorkel counts and PAE After passing through the grid, the observer exited the water, catch during fall and summer using Spearman’s rank correlation. and another 15-min waiting period lapsed. The same observer We assumed a positive monotonic correlation between snorkel snorkeled all transects during both seasons. counts and PAE catch. Our snorkeling efficiency estimates did After the second 15-min waiting period, two dipnetters posi- not conform to a normal distribution (Shapiro–Wilk W-test: P < tioned themselves 5 m downstream from the grid. The grid was 0.05; Zar 1999), so we applied statistical procedures that did then electrified for 60 s, and all stunned fish within the grid were not assume a normal distribution. We fit a generalized linear collected as both dipnetters moved upstream. Power to the grid model to test for biotic and abiotic influences on snorkeling was shut off, and dipnetters searched downstream to collect any efficiency (JMP Pro version 11.0, SAS, Cary, North Carolina). remaining stunned fish. Captured fish were identified, counted, We modeled snorkeling count as a function of macrohabitat and released back into the river. After each transect was sam- type, fish taxon, season, and loge(PAE count) using a binomial pled by both techniques, the distance of the farthest identifiable distribution and logit link function. fish was measured from the corresponding marker to the center Fish density.—We applied our snorkeling efficiency esti- of the grid, serving as an estimate of snorkeling sampling area mates for each fish taxon to strip-transect counts from an in- and a sampling covariate for stream visibility for each transect. tensive snorkeling survey in the same river that was designed We measured water temperature (◦C), dissolved oxygen (mg/L), to characterize the fish populations stratified among all macro- specific conductivity (µS/cm), pH, and turbidity (NTU) using habitat types. The intensive survey consisted of 10 transects, a Hydrolab model MS5 multiprobe datasonde and Surveyor 4a each 30-m long, stratified by macrohabitat with transect widths display unit. based on the farthest fish observed during that sampling period Snorkeling efficiency.—The length of each snorkel transect (Weaver 2010). Adjusted density estimates were calculated as was identical to the length of the PAE grid; however, because ˆ observers could generally see farther than the width of the PAE D j Nˆ j = (3) grid, the total area snorkeled was consistently larger than the qˆ j area sampled by the PAE. Therefore, for each of the 12 tran-

Downloaded by [Department Of Fisheries] at 19:15 20 July 2015 sects sampled during each season, a strip-transect count density (Thompson and Seber 1994), where Nˆ is the adjusted density estimate was calculated from the count and respective total area estimate of a fish taxon j, Dˆ is the strip-transect count density sampled from each gear-type as estimate (equation 1), and qˆ is the snorkeling efficiency estimate for that taxon expressed as a proportion from equation (2). We ˆ derived variance approximations of the adjusted density esti- ˆ = C j D j (1) mates (Nˆ ) for a fish taxon j using the delta method (Williams (w × l × t) i et al. 2002) as (Keast and Harker 1977), where density Dˆ for a particular fish 2 Dˆ var Dˆ j var qˆ j species or taxon j is calculated as the count Cˆ divided by the var Nˆ = j + , (4) j qˆ ˆ 2 2 product of the width wi, length l, and number of transects t.The j D j qˆ j width of each snorkeling transect was two times the farthest fish observed, and that for the prepositioned areal grid remained and then calculated the associated SE of the adjusted density constant at 1.0 m. estimate as the square root of the variance. We then com- Snorkeling efficiency estimates qˆ of a particular taxon j were pared strip-transect count density estimates with adjusted strip- calculated as a proportion by dividing the strip-transect count transect density estimates that were corrected for snorkeling 1162 WEAVER ET AL.

TABLE 1. Total fish counts by snorkeling and prepositioned areal electrofishing (number of fish) among fish taxa and macrohabitat types during fall 2008 and summer 2009 in 12 transects in the North Toe River, Mitchell County, North Carolina.

Summer Fall Variable Snorkeling Electrofishing Snorkeling Electrofishing Fish taxon Shiners 28 100 42 153 Central Stoneroller 34 53 93 128 River Chub 24 29 19 43 Suckers 7 31 3 17 Darters17224057 Total 110 235 197 398 Macrohabitat type Pool 23 70 76 158 Riffle 75 119 100 171 Run12462169 Total 110 235 197 398

efficiency through empirical calibration at the taxon and assem- Snorkeling Efficiency blage levels. Spearman’s rank correlation indicated that snorkel counts and electrofishing catch among fish taxa were significantly correlated within seasons for both seasons (summer: rs = 0.725, P < 0.0001, N = 60; fall: r = 0.657, P < 0.0001, N = 60; RESULTS s Figure 1). During gear calibration, the total area sampled by snorkeling Mean snorkeling efficiency among all taxa and seasons was 12 transects was 504 m2 during the summer and the farthest fish 14.7% (SE, 2.5%) and ranged from 4.0% to 30.3% (Table 3). observed was at 1.40 m (2.80 m total width). During the fall the Between seasons, mean snorkeling efficiency among taxa was total area sampled was 900 m2 and the farthest fish observed 11.2% (SE, 3.1%) during the fall, ranging from 4.0% to 19.5%, was at 2.50 m (5.00 m total width). The total area we sampled and was 18.1% (SE, 3.9%) during the summer, ranging from by PAE was constant between seasons at 180 m2 (15 m2 × 9.1% to 30.3%. We observed relatively low mean efficiencies 12 samples each season). We observed no stunned fish outside for suckers and shiners (6.6% and 8.2%, respectively) and of the grid area while the grid was electrified, confirming that higher efficiencies for Central Stoneroller and River Chub the voltage gradient was concentrated within the grid, a char- (20.3% and 21.1%, respectively). Among macrohabitats, runs acteristic also observed by Bowen and Freeman (1998). During had the lowest efficiency (9.4%) relative to pool (13.7%) our intensive survey, the total area snorkeled was 1,116 m2 (10 and riffle (17.7%) habitats. However, a generalized linear transects, each 30 m long, and the farthest fish was observed at 2 model did not significantly explain the variance in snorkeling Downloaded by [Department Of Fisheries] at 19:15 20 July 2015 1.86 m) during the summer and 1,470 m (10 transects, each 30 m long, and the farthest fish was observed at 2.45 m) during the fall. Among environmental variables measured during the TABLE 2. Snorkeling detection probability (%) among fish taxa and mean ◦ calibration procedure, we found higher temperatures (21.5 C detection probability (SE in parentheses) between seasons, assuming that prepo- versus 17.2◦C), higher turbidity (6.1 versus 2.4 NTU), lower sitioned areal electrofishing detected all available taxa in each transect and dissolved oxygen (9.2 versus 9.7 mg/L), and higher pH (8.0 season sampled (N = 12 each season) in the North Toe River, North Carolina. versus 7.7) during summer sampling than in the fall. Detection probability (%) We obtained consistently lower total fish counts from snorkeling than from those obtained by PAE for all fish taxa, macrohab- Fish guild Summer Fall itats, and seasons (Table 1). In 4 of 24 total transects sampled, we identified one fish taxon during snorkeling that was not captured Shiners 72.7 55.6 by electrofishing. Excluding those exceptions, if we assume that Central Stoneroller 66.7 42.9 electrofishing detects all fish taxa present, then snorkeling detec- River Chub 62.5 75.0 tion probability averaged 58.4% (SE, 6.1%) among all taxa and Suckers 33.3 28.6 seasons and was equivalent between seasons (summer: 58.5% Darters 57.1 90.0 [SE, 6.8%]; fall: 58.4% [SE, 11.0%]; Table 2). Mean 58.5 (6.8) 58.4 (11.0) SAMPLING AND SNORKEL COUNTS TO ESTIMATE STREAM FISH 1163

60 Summer gear comparison during two seasons and then adjusted strip- Fall transect snorkeling counts into density estimates using taxon- 50 specific efficiency estimates. No covariate that we examined 40 significantly explained the variance in snorkeling efficiency, which suggests there are distinct differences in fish behavior. 30 Our mean estimate of snorkeling efficiency (14.7%) is similar 20 to those for trout and salmon (16.0%: Roni and Fayram 2000;

Snorkel count(number) 12.5%: Thurow et al. 2006); however, among ﬁsh taxa that we 10 sampled, efﬁciency ranged from 4% to 30%, further emphasiz-

0 ing the need to calibrate gears at the population or taxon level 0 102030405060708090for accuracy of density estimates at the community level. We Electrofishing catch (number) suggest that calibrating snorkeling efficiency with PAE is a practical approach to estimate fish density in medium-sized rivers FIGURE 1. Total snorkel count versus total prepositioned areal electrofishing catch for five taxa in 12 transects during each season, fall 2008 (N = 60) and where conventional sampling techniques may not be possible. summer 2009 (N = 60), in the North Toe River, North Carolina. Our estimates of snorkeling efficiency (mean, 14.7%; range, 4–30% among taxa) were generally lower than those of other efficiency among macrohabitat type, fish taxon, or sampling stream fish sampling techniques. Peterson and Rabeni (2001) 2 season (P > 0.05). found that sampling efficiency of a 1-m quadrat varied widely among fish families and ranged from 31% to 84%. Multiple-pass Fish Density removal sampling using backpack electrofishing was 63–97% Strip-transect count density estimates for combined fish taxa efficient (Wiley and Tsai 1983; Rodgers et al. 1992; Meyer and were 1,030 fish/ha during the summer and 1,769 fish/ha dur- High 2011), but efficiency modeled for single-pass backpack ing the fall (Table 4). Adjusted strip-transect count density es- electrofishing ranged from 7% to 29% among taxa (Price and timates were 7,288 fish/ha (SE, 1,564) during the summer and Peterson 2010). Multiple seine-haul efficiency estimates in Pied- 15,805 fish/ha (SE, 4,947) during the fall. Overall, strip-transect mont and coastal streams were 70–79% (Weinstein and Davis count density estimates and adjusted strip-transect count den- 1980; Wiley and Tsai 1983). The utility of these other tech- sity estimates were lower for all fish taxa during summer than niques in a medium-sized river, such as the North Toe River that during fall. Shiner and Central Stoneroller estimates were most we studied, is usually logistically problematic and hindered by abundant followed by darters and River Chub, and suckers were water depth (e.g., for seining, backpack electrofishing) and ma- the least abundant. neuverability (e.g., for boat electrofishing: Curry et al. 2009). Thus, despite generally lower efficiency, the snorkeling technique provides a practical method for characterizing lotic fish DISCUSSION assemblages when other methods are inappropriate. We quantified snorkeling efficiency for the fish assemblage Our snorkeling efficiency estimates among fish taxa and cor- of a medium-sized southern Appalachian Mountain river using responding adjusted strip-transect count density estimates may

TABLE 3. Mean snorkeling efﬁciency estimates (percent of individuals sampled) and SE among ﬁsh taxa and macrohabitat types during fall 2008 and summer

Downloaded by [Department Of Fisheries] at 19:15 20 July 2015 2009 seasons in the North Toe River, North Carolina.

Summer Fall Variable Efficiency (%) SE N Efficiency (%) SE N Fish taxon Shiners 9.82 2.69 11 6.54 3.54 9 Central Stoneroller 21.10 13.22 3 19.49 13.88 7 River Chub 30.32 11.01 8 11.87 4.17 8 Suckers 9.14 5.67 7 4.00 2.90 7 Darters 20.09 6.89 6 14.16 2.51 10 Macrohabitat type Pool 15.32 7.85 12 12.04 2.66 17 Riffle 20.88 3.59 14 14.54 7.06 14 Run 13.59 6.56 9 5.17 2.33 10 1164 WEAVER ET AL.

TABLE 4. Comparison of strip-transect count and adjusted strip-transect count density estimate (mean ± SE) among ﬁsh taxa for 10 transects during intensive snorkeling surveys conducted during fall 2008 and summer 2009 in the North Toe River, North Carolina.

Strip-transect count (ﬁsh/ha) Adjusted strip-transect count (ﬁsh/ha) Fish taxon Summer mean SE Fall mean SE Summer mean SE Fall mean SE Shiners 439.1 29.8 455.8 18.3 4,469.2 1,259.7 6,969.2 3,775.4 Central Stoneroller 277.8 26.8 816.3 52.3 1,316.2 834.3 4,188.4 2,995.9 River Chub 98.6 5.4 210.9 11.1 325.1 119.4 1,775.1 630.7 Suckers 17.9 1.8 47.6 4.1 196.1 187.2 1,190.5 867.7 Darters 197.1 13.0 238.1 9.5 981.3 336.8 1,681.5 305.7 Total 1,030.5 60.2 1,768.7 81.2 7,287.9 1,563.9 15,804.7 4,947.0

have been influenced by several environmental and behavioral ever, the measurement value of the farthest fish serves as an variables. Some riverine fish exhibit seasonal migration in re- index of observer visibility, which may be an important factor sponse to declining temperature (Jonsson 1991), as well as in- in future studies that examine spatial and temporal variability in creases in cryptic behavior and a greater tendency to seek cover visual estimates of fish density. Alternative indices for observer in reaction to lower water temperatures and higher water veloc- visibility could include water turbidity or Secchi disk depth (En- ities (Gibson 1978; Taylor 1988). We found reduced efficiency sign et al. 1995; Mullner et al. 1998), but our index integrates of snorkel counts with lower water temperature, suggesting that multiple instream factors that influence visibility in situ (e.g., fish can increase cover association and cryptic behavior dur- geomorphology, substrate composition, instream cover, as well ing fall compared with summer (Hillman et al. 1992; Rodgers as water turbidity, color, and light conditions) and provides an et al. 1992). Reduced efficiency may have also been related to exact transect width measurement. an increase in fish density, as we estimated more than twice as Our fish density estimates are within the ranges of those ob- many fish per area during the fall compared with the summer, tained for other riverine fishes (Ensign et al. 1995; Hewitt et al. which may have caused the observer to overlook some fish while 2009), although notable differences exist. Hewitt et al. (2009) snorkeling. estimated density of the endangered Cape Fear Shiner Notropis The portion of unexplained variance in our correlations be- mekistocholas in the Cape Fear River basin, North Carolina, to tween snorkeling counts and electrofishing catch (rs = 0.66– range from 795 to 1,393 fish/ha using the strip-transect count 0.73) is likely due to differences in fish behavior among taxa method. Our strip-transect counts (unadjusted for efficiency) and sizes and associated variability in detection probability be- for all shiners were 439.1 and 455.8 fish/ha for the summer tween snorkeling and electrofishing. Other investigators have and fall, respectively. The difference in density estimates be- reported high positive correlations (r2 > 0.90) between vi- tween our sampling and that of Hewitt et al. (2009) may be due sual counts and multiple-pass electrofishing removal estimates to study site differences, as our sampling was conducted in a (Hankin and Reeves 1988; Mullner et al. 1998; Wildman and high-gradient mountain river (lower productivity) versus their Neumann 2003); however, those studies focused exclusively on sampling in a Piedmont river (higher productivity), reflecting salmonids. We demonstrated clear differences in detection prob- actual population differences rather than variation in sampling

Downloaded by [Department Of Fisheries] at 19:15 20 July 2015 ability among stream fish taxa and among individual fish within efficiency. Ensign et al. (1995) estimated mean densities of 5– a taxon (i.e., efficiency), highlighting the importance of exam- 272 fish/ha for Black Jumprock Moxostom cervinum, 3–163 ining fish sampling efficiency and detection probability of sam- fish/ha for Roanoke Logperch Percina rex, and 79–2,360 fish/ha pling techniques at multiple organizational scales, especially for Roanoke Darter Etheostoma roanoka over various sampling if research or management objectives are at the assemblage sites using line-transect distance sampling. Our density esti- level. mates derived from strip-transect counts adjusted for efficiency A challenge in implementing the strip-transect count method are much higher than those estimates based on actual counts, is determining the boundaries of the transect width. Our strip- and we consider the adjusted estimates to be more accurate es- transect count estimate relied upon measurements of the fish timates of actual fish density than those proportional estimates seen farthest from the center line of observation (0.5wi, equa- based on unadjusted counts (Hayes et al. 2007; Peterson and tion 1) to standardize counts per area. We assumed that (1) this Paukert 2009). index correctly measures the maximum visibility distance of The use of PAE to calibrate snorkel counts appears to be a the observer, (2) all fish taxa have an equal probability of being viable approach when other sampling techniques are inappropri- observed at this distance, and (3) distances are measured ac- ate. Fish mortality from electrofishing, especially if AC power is curately. We acknowledge that violations to these assumptions employed, may pose a concern, especially when studying rare may result in under- or overestimation of fish density. How- or endangered species. An initial calibration study, however, SAMPLING AND SNORKEL COUNTS TO ESTIMATE STREAM FISH 1165

may provide sufficient efficiency estimates to apply to more in- Bowen, Z. H., and M. C. Freeman. 1998. Sampling effort and estimates of tensive or extensive visual surveys to quantify fish populations species richness based on prepositioned area electrofisher samples. North over time without introducing the acute or cumulative effects American Journal of Fisheries Management 18:144–153. Curry, R. A., R. M. Hughes, M. E. McMaster, and D. J. Zafft. 2009. Coldwater of electrofishing mortality. This is a highly applicable approach fish in rivers. Pages 139–158 in S. A. Bonar, W. A. Hubert, and D. W. Willis, if sampling mortality could bias results of research or manage- editors. Standard methods for sampling North American freshwater fishes. ment assessments. Precision of fish density estimates adjusted American Fisheries Society, Bethesda, Maryland. for snorkeling efficiency (expressed as SE here) is influenced Dolloff, C. A., J. Kershner, and R. F. Thurow. 1996. Underwater observation. by spatial and temporal variation in efficiency and the effect Pages 533–554 in B. R. Murphy and D. W. Willis, editors. Fisheries techniques, 2nd edition. American Fisheries Society, Bethesda, Maryland. of sample size. Thus, the precision of future fish density esti- Dunham, J. B., A. E. Rosenberger, R. F. Thurow, C. A. Dolloff, and P. J. Howell. mates may be increased by larger sample sizes in developing 2009. Coldwater fish in wadeable streams. Pages 119–138 in S. A. Bonar, calibration relationships. W. A. Hubert, and D. W. Willis, editors. Standard methods for sampling Our study further highlights the utility of gear calibration North American freshwater fishes. American Fisheries Society, Bethesda, when conducting underwater visual surveys (Dolloff et al. 1996; Maryland. Emlen, J. T. 1971. Population densities of birds derived from transect counts. Dunham et al. 2009; Peterson and Paukert 2009). Our specific Auk 88:323–342. snorkeling efficiency estimates may be appropriately applied to Ensign, W. E., P. L. Angermeier, and C. A. Dolloff. 1995. Use of line-transect related studies of similar fish assemblages under comparable methods to estimate abundance of benthic stream fishes. Canadian Journal of environmental conditions or for imprecise approximations of Fisheries and Aquatic Sciences 52:213–222. fish abundance. Spatial, temporal, and observer variation in ef- Ensign, W. E., A. J. Temple, and R. J. Neves. 2002. Effects of fright bias on sampling efficiency of stream fish assemblages. Journal of Freshwater ficiency, however, may compel integration of gear comparison Ecology 17:127–139. methods or environmental covariates into site- or water-body- Gibson, R. J. 1978. The behavior of juvenile Atlantic Salmon (Salmo salar)and specific efficiency models to more accurately estimate fish abun- Brook Trout (Salvelinus fontinalis) with regard to temperature and to water dance, depending on sampling objectives. Our findings support velocity. Transactions of the American Fisheries Society 107:703–712. the conclusion that stream fish assessments over large spatial Grabowski, T. B., and J. J. Isely. 2007. Spatial and temporal segregation of spawning habitat by catostomids in the Savannah River, Georgia and South areas can be conducted with acceptable efficiency and detection Carolina, U.S.A. Journal of Fish Biology 70:782–798. probability, less effort, and no mortality with visual sampling Grossman, G. D., and R. E. Ratajczak. 1998. Long-term patterns of microhabitat methods compared with traditional sampling methods such as use by fish in a southern Appalachian stream from 1983 to 1992: effects of electrofishing. hydrologic period, season and fish length. Ecology of Freshwater Fish 7:108– 131. Hankin, D. G., and G. H. Reeves. 1988. Estimating total fish abundance and total ACKNOWLEDGMENTS habitat area in small streams based on visual estimation methods. Canadian We thank Ben Wallace, Justin Dycus, Patrick Cooney, Scott Journal of Fisheries and Aquatic Sciences 45:834–844. Favrot, Brad Garner, Christen Brown, Michael Fisk, Josh Raabe, Hayes, D. B., J. R. Bence, T. J. Kwak, and B. E. Thompson. 2007. Abundance, Stephen Poland, Kyle Rachels, and Jeremy Remmington for as- biomass, and production estimation. Pages 327–374 in C. S. Guy and M. L. Brown, editors. Analysis and interpretation of freshwater fisheries data. sistance in the field. Bill Halteman provided statistical guidance, American Fisheries Society, Bethesda, Maryland. and comments from Rob Dunn and Ryan Heise improved earlier Hewitt, A. H., T. J. Kwak, W. G. Cope, and K. H. Pollock. 2009. Population versions of this manuscript. This project was funded with Sport density and instream habitat suitability of the endangered Cape Fear Shiner. Fish Restoration Funds through the North Carolina Wildlife Re- Transactions of the American Fisheries Society 138:1439–1457. sources Commission (Project F-68, Study 10). Jake Rash, Doug Hillman, T. W., J. W. Mullan, and J. S. Griffith. 1992. Accuracy of underwater counts of juvenile Chinook Salmon, Coho Salmon, and steelhead. North Besler, Steve Fraley, Mallory Martin, and Kent Nelson admin- American Journal of Fisheries Management 12:598–603. Downloaded by [Department Of Fisheries] at 19:15 20 July 2015 istered funding and offered helpful guidance and insight. The Holliman, F. M., J. B. Reynolds, and T. J. Kwak. 2003a. A predictive risk model North Carolina Cooperative Fish and Wildlife Research Unit for electroshock-induced mortality of the endangered Cape Fear Shiner. North is jointly supported by North Carolina State University, North American Journal of Fisheries Management 23:905–912. Carolina Wildlife Resources Commission, U.S. Geological Sur- Holliman, F. M., J. B. Reynolds, and T. J. Kwak. 2003b. Electroshock-induced injury and mortality in the Spotfin Chub, a threatened minnow. North Amer- vey, U.S. Fish and Wildlife Service, and Wildlife Management ican Journal of Fisheries Management 23:962–966. Institute. Any use of trade, product, or firm names is for descrip- Janac, M., and P. Jurajda. 2010. Modified sampling design for age-0 fish elec- tive purposes only and does not imply endorsement by the U.S. trofishing at beach habitats. North American Journal of Fisheries Management Government. 30:1210–1220. Jonsson, N. 1991. Influence of water flow, water temperature and light on fish migration in rivers. Nordic Journal of Freshwater Research 66:20–35. REFERENCES Keast, A., and J. Harker. 1977. Strip counts as a means of determining densities Arend, K. K. 1999. Macrohabitat identification. Pages 75–93 in M. B. Bain and habitat utilization patterns in lake fishes. Environmental Biology of Fishes and N. J. Stevenson, editors. Aquatic habitat assessment: common methods. 1:181–188. American Fisheries Society, Bethesda, Maryland. Korman, J., A. S. Decker, B. Mossop, and J. Hagen. 2010. Comparison of elec- Bain, M. B., J. T. Finn, and H. E. Book. 1985. A quantitative method for trofishing and snorkeling mark-recapture estimation of detection probability sampling riverine microhabitats by electrofishing. North American Journal and abundance of juvenile steelhead in a medium-sized river. North American of Fisheries Management 5:489–493. Journal of Fisheries Management 30:1280–1302. 1166 WEAVER ET AL.

Kwak, T. J. 1991. Ecological characteristics of a northern population of Schwartz, J. S., and E. E. Herricks. 2004. Use of prepositioned areal electrofish- Pallid Shiner. Transactions of the American Fisheries Society 120:106– ing devices with rod electrodes in small streams. North American Journal of 115. Fisheries Management 24:1330–1340. Larimore, R. W., and D. D. Garrels. 1985. Assessing habitats used by warmwater Slaney, P. A., and A. D. Martin. 1987. Accuracy of underwater census of trout stream fishes. Fisheries 10(2):10–16. populations in a large stream in British Columbia. North American Journal Lobb, M. D., and D. J. Orth. 1991. Habitat use by an assemblage of fish in of Fisheries Management 7:117–122. a large warmwater stream. Transactions of the American Fisheries Society Snyder, D. E. 2003. Electrofishing and its harmful effects on fish. U.S. Govern- 120:65–78. ment Printing Office, Information and Technology Report USGS/BRD/ITR- Meyer, K. A., and B. High. 2011. Accuracy of removal electrofishing estimates 2003-0002, Denver. of trout abundance in Rocky Mountain streams. North American Journal of Sokal, R. R., and F. J. Rohlf. 1981. Biometry, 2nd edition. Freeman, New York. Fisheries Management 31:923–933. Taylor, E. B. 1988. Water temperature and velocity as determinants of micro- Mullner, S. A., W. A. Hubert, and T. A. Wesche. 1998. Snorkeling as an alter- habitats of juvenile Chinook and Coho salmon in a laboratory stream channel. native to depletion electrofishing for estimating abundance and length-class Transactions of the American Fisheries Society 117:22–28. frequencies of trout in small streams. North American Journal of Fisheries Thompson, S. K., and G. A. Seber. 1994. Detectability in conventional and Management 18:947–953. adaptive sampling. Biometrics 50:712–724. NCWRC (North Carolina Wildlife Resources Commission). 2009. The eco- Thurow, R. F., J. T. Peterson, and J. W. Guzevich. 2006. Utility and validation nomic impact of mountain trout fishing in North Carolina. NCWRC, Federal of day and night snorkel counts for estimating Bull Trout abundance in first- Aid in Fish Restoration, Project F-86, Raleigh. to third-order streams. North American Journal of Fisheries Management Orell, P., and J. Erkinaro. 2007. Snorkeling as a method for assessing spawning 26:217–232. stock of Atlantic Salmon, Salmo salar. Fisheries Management and Ecology Walsh, M. G., D. B. Fenner, and D. L. Winkelman. 2002. Comparison of an 14:199–208. electric seine and prepositioned area electrofishers for sampling stream fish Peterson, J. T., and C. P. Paukert. 2009. Converting nonstandard fish sam- communities. North American Journal of Fisheries Management 22:77–85. pling data to standardized data. Pages 195–212 in S. A. Bonar, W. Weaver, D. M. 2010. Effects of stocked trout on native nongame riverine fishes. A. Hubert, and D. W. Willis, editors. Standard methods for sampling Master’s thesis. North Carolina State University, Raleigh. North American freshwater fishes. American Fisheries Society, Bethesda, Weaver, D. M., and T. J. Kwak. 2013. Assessing effects of stocked trout on Maryland. nongame fish assemblages in southern Appalachian Mountain streams. Trans- Peterson, J. T., and C. F. Rabeni. 2001. Evaluating the efficiency of a one-square- actions of the American Fisheries Society 142:1495–1507. meter quadrat sampler for riffle-dwelling fish. North American Journal of Weinstein, M. P., and R. W. Davis. 1980. Collection efficiency of seine and Fisheries Management 21:76–85. rotenone samples from tidal creeks, Cape Fear River, North Carolina. Estu- Pink, M., T. C. Pratt, and M. G. Fox. 2007. Use of underwater visual distance aries 3:98–105. sampling for estimating habitat-specific population density. North American Wildman, T. L., and R. M. Neumann. 2003. Comparison of snorkeling and Journal of Fisheries Management 27:246–255. electrofishing for estimating abundance and size structure of Brook Trout Price, A. L., and J. T. Peterson. 2010. Estimation and modeling of electrofishing and Brown Trout in two southern New England streams. Fisheries Research capture efficiency for fishes in wadeable warmwater streams. North American 60:131–139. Journal of Fisheries Management 30:481–498. Wiley, M. L., and C. Tsai. 1983. The relative efficiencies of electrofishing vs. Rodgers, J. D., M. F. Solazzi, S. L. Johnson, and M. A. Buckman. 1992. Com- seines in piedmont streams of Maryland. North American Journal of Fisheries parison of three techniques to estimate juvenile Coho Salmon populations Management 3:243–253. in small streams. North American Journal of Fisheries Management 12: Williams, B. K., J. D. Nichols, and M. J. Conroy. 2002. Analysis and man- 79–86. agement of animal populations: modeling, estimation, and decision making. Roni, P., and A. Fayram. 2000. Estimating winter salmonid abundance in small Academic Press, San Diego, California. western Washington streams: a comparison of three techniques. North Amer- Zar, J. H. 1999. Biostatistical analysis, 4th edition. Prentice Hall, Englewood ican Journal of Fisheries Management 20:683–692. Cliffs, New Jersey. Downloaded by [Department Of Fisheries] at 19:15 20 July 2015 This article was downloaded by: [Department Of Fisheries] On: 20 July 2015, At: 19:21 Publisher: Taylor & Francis Informa Ltd Registered in England and Wales Registered Number: 1072954 Registered office: 5 Howick Place, London, SW1P 1WG

North American Journal of Fisheries Management Publication details, including instructions for authors and subscription information: http://www.tandfonline.com/loi/ujfm20 Comment: Population Structure and Run Timing of Sockeye Salmon in the Skeena River, British Columbia Michael H. H. Pricea, Andrew G. J. Rosenbergerb, Greg G. Taylorc & Jack A. Stanfordd a Skeena Wild Conservation Trust, 4505 Greig Avenue, Terrace, British Columbia V8G 1M6, Canada b Raincoast Conservation Foundation, Post Office Box 2429, Sidney, British Columbia V8L 3Y3, Canada c Watershed Watch Salmon Society, 1037 Madore Avenue, Coquitlam, British Columbia V3K 3B7, Canada d Flathead Lake Biological Station, University of Montana, 32125 Bio Station Lane, Polson, Click for updates Montana 59860, USA Published online: 10 Nov 2014.

To cite this article: Michael H. H. Price, Andrew G. J. Rosenberger, Greg G. Taylor & Jack A. Stanford (2014) Comment: Population Structure and Run Timing of Sockeye Salmon in the Skeena River, British Columbia, North American Journal of Fisheries Management, 34:6, 1167-1170, DOI: 10.1080/02755947.2014.956162 To link to this article: http://dx.doi.org/10.1080/02755947.2014.956162

PLEASE SCROLL DOWN FOR ARTICLE

Taylor & Francis makes every effort to ensure the accuracy of all the information (the “Content”) contained in the publications on our platform. However, Taylor & Francis, our agents, and our licensors make no representations or warranties whatsoever as to the accuracy, completeness, or suitability for any purpose of the Content. Any opinions and views expressed in this publication are the opinions and views of the authors, and are not the views of or endorsed by Taylor & Francis. The accuracy of the Content should not be relied upon and should be independently verified with primary sources of information. Taylor and Francis shall not be liable for any losses, actions, claims, proceedings, demands, costs, expenses, damages, and other liabilities whatsoever or howsoever caused arising directly or indirectly in connection with, in relation to or arising out of the use of the Content. This article may be used for research, teaching, and private study purposes. Any substantial or systematic reproduction, redistribution, reselling, loan, sub-licensing, systematic supply, or distribution in any form to anyone is expressly forbidden. Terms & Conditions of access and use can be found at http:// www.tandfonline.com/page/terms-and-conditions North American Journal of Fisheries Management 34:1167–1170, 2014 C American Fisheries Society 2014 ISSN: 0275-5947 print / 1548-8675 online DOI: 10.1080/02755947.2014.956162

COMMENT

Comment: Population Structure and Run Timing of Sockeye Salmon in the Skeena River, British Columbia

In their article, Beacham et al. (2014) acknowledge that

[t]he underlying priority in management of the [Skeena River Sockeye Salmon Oncorhynchus nerka] ﬁshery is one of conservation ﬁrst of stocks and species of conservation concern, and secondarily where and when possible, allowing for harvest of more abundant stocks of Sockeye Salmon.

The authors then describe genetic stock identification at a salmon test fishery as a novel and important technique with which to estimate population abundance for fisheries management and protect populations of concern. While we agree that the conservation of salmon is the highest priority for fisheries management, we believe that Beacham et al. fail to consider the appropriate population-level structure required for conservation in Canada, do not emphasize the complexity of the Sockeye Salmon populations in the Skeena River catchment, and overstate the approach’s utility in contemporary salmon management. Ultimately, the proposed approach will not achieve the necessary conservation goals because of the substantial overlap in the return of wild and artificially enhanced Sockeye Salmon. A primary objective of Canada’s Policy for Conservation of Wild Pacific Salmon (WSP) is to sustain the genetic and geographic diversity of wild salmon and preserve their habitats through the protection of conservation units (CUs; DFO 2005,

Downloaded by [Department Of Fisheries] at 19:21 20 July 2015 2009). Despite the importance of this fundamental structure for management decision making, monitoring, and conservation, there was no mention of CUs in Beacham et al. (2014). FIGURE 1. Skeena River catchment lake-type (dark gray [numbers correspond to those in Table 1]) and river-type (medium gray = Skeena River–High Instead, the authors examined 27 populations that in their ge- Interior; light gray = Skeena River) Sockeye Salmon conservation units. Aster- netic analysis segregated into 12 “stock reporting groups.” isks indicate the locations of the Babine artificial spawning channels. In the Skeena River catchment, there are 31 lake-type and 2 river-type Sockeye Salmon CUs (Holtby and Ciruna 2007; our Figure 1); only 9 of the 33 CUs can be identified genetically as a CU with >95% confidence at the Tyee test fish- which is the focus of the conservation objectives under the ery (Beacham et al. 2014; our Table 1). Importantly, more WSP. than one-half of the Skeena River Sockeye Salmon CUs for Beacham et al. (2014) briefly describe the dynamics of the which there is enough information to determine abundance Skeena River commercial fishery, whereby artificial spawn- status are “of concern,” including some that are severely de- ing channels in Babine Lake have annually increased Sockeye pressed (Korman and English 2013; our Table 1). In not fo- Salmon production for fisheries yet resulted in less productive cusing on CUs, Beacham et al. do not fully account for wild populations that have been subject to overexploitation. their inherent genetic, life history, and geographic uniqueness, They state that “[t]he identification of timing of the return of 1167 1168 PRICEETAL.

TABLE 1. The 33 Sockeye Salmon conservation units (CUs) in the Skeena River catchment. Location numbers correspond to those in Figure 1; “Genetic baseline” refers to whether or not a particular CU has a genetic baseline; “Tyee identified” refers to whether or not a CU can be identified in the Tyee test fishery with 95% confidence; “Proportion” is the estimated stock composition identified at the Tyee test fishery; and “Status” is the abundance status estimated by Korman and English (2013). Conservation units without an assigned status are data deficient.

Conservation unit Location Type Genetic baseline Tyee identiﬁed Proportion Status Alastair 1 Lake Yes Yes 2.6 (0.6) Green Aldrich 2 Lake No No Asitika 3 Lake No No Atna 4 Lake No No Azuklotz 5 Lake No No Green Babine 6 Lake Yes No Amber Bear 7 Lake Yes Yes 0.8 (0.4) Red Bulkley 8 Lake No No Club 9 Lake No No Damshilgwit 10 Lake Yes No Green Dennis 11 Lake No No Ecstall, lower 12 Lake No No Johanson 13 Lake No No Johnston 14 Lake No No Kitsumkalum 15 Lake Yes Yes 2.0 (0.5) Green Kitwanga 16 Lake Yes Yes 0.3 (0.2) Red Kluatantan 17 Lake No No Kluayaz 18 Lake No No Lakelse 19 Lake Yes Yes 1.4 (0.4) Amber Maxan 20 Lake No No McDonell 21 Lake Yes Yes 0.9 (0.3) Green Morice 22 Lake Yes Yes 1.2 (0.4) Red Motase 23 Lake Yes Yes 0.1 (0.1) Amber Nilkitkwa 24 Lake Yes No Red Sicintine 25 Lake No No Slamgeesh 26 Lake Yes No 0.9 (0.4) Spawning 27 Lake No No Stephens 28 Lake Yes No Green Sustut 29 Lake Yes Yes 0.5 (0.2) Swan 30 Lake Yes No Red Tahlo–Morrison 31 Lake Yes No Amber Skeena River River No No Skeena River–High Interior River No No Downloaded by [Department Of Fisheries] at 19:21 20 July 2015

specific populations from information gained through fisheries will not be effective because there are large overlaps in run tim- is thus a key component of reducing exploitation on populations ing (Figure 2B), run timing is highly variable between years, of conservation concern.” We agree wholeheartedly. However, and the annual variation is not consistent among CUs (Cox- the coarse resolution of Beacham et al.’s (2014) results, whereby Rogers 2012). Importantly, run times are calculated through the all wild Babine Sockeye Salmon CUs (each known to have a Tyee test fishery after the fish have passed through the marine unique run time; Smith and Jordan 1973; Figure 2A) are grouped mixed-stock fishery; hence, there is no measure of CU compo- genetically with artificially enhanced fish as a “middle run time,” sition prior to the impact of the fishery. While a few CUs with would not prevent the exploitation of these CUs of concern in extremely early run timing might be managed separately, the mixed-stock fisheries and fails to meet the intent of the WSP. majority cannot be (Walters et al. 2008). More broadly, the authors’ proposed approach of using tempo- Beacham et al. (2014) state that “[t]he application of ge- ral differences in run timing to target or protect specific CUs netic stock identification technology to a salmon test fishery COMMENT 1169

abundance in the fishery; half of all stock reporting groups comprise less than 1% of the relative fishery abundance. Beacham et al. (2014) declare that “[d]etermination of spawning escapement is one of the key pieces of information required in the assessment of the status of specific salmon populations.” We agree, and we acknowledge that such escapement estimation is difficult to perform by visual observation. How- ever, we disagree that the Tyee test fishery can provide an alternative and reasonable determination of spawning escapement for Sockeye Salmon CUs as required by the WSP (DFO 2005, 2009), especially for small populations of conservation concern. This is because the majority of Skeena River CUs remain anonymous in the Tyee test fishery; the genetic identification is performed postseason rather than during the fishery; and of the stock reporting groups that can be identified, all but one (Babine Lake) have such low abundance that the stock composition estimates are imprecise. As Walters et al. (2008) confirm, “[t]he [Tyee test] fishery provides at best only a ‘noisy’ estimate of escapement (plus or minus 20% of run size).” Beacham et al. (2014) imply that the detection of blocked spawning grounds for Kwinageese Sockeye Salmon was due only to a Nass River test fishery, which is erroneous; had the test fishery alone (i.e., without spawning ground surveys) been responsible for escapement estimates, the habitat concern and resulting reduction in spawners may never have been known. Im- portantly, ground site visits and overflights provide much more FIGURE 2. Average weekly percentages of Skeena River Sockeye Salmon detailed information (such as changes to habitat quality and passing the Tyee test fishery. Panel (A) shows the average run times for wild Babine conservation units (Babine, Tahlo–Morrison, and Nilkitkwa) and artifi- quantity, and disease events on the spawning grounds) than the cially enhanced Babine Sockeye Salmon. Panel (B) shows the average run times Tyee test fishery. Furthermore, other viable and cost-effective for 11 wild (Lakelse, Alastair, Zymoetz, Morice, Kispiox, Sustut, Slamgeesh, ways to count fish on the spawning grounds (such as camera- Motase, Bear, Kitsumkalum, and Kitwanga) and the combined wild and arti- equipped drones and fixed-position remote cameras) should be ficially enhanced Babine Sockeye Salmon stock reporting groups reported by considered. Beacham et al. (2014). All run time data are for the period 2000–2010 (reported by Cox-Rogers 2012). In summary, the approach proposed by Beacham et al. (2014), whereby stock reporting groups are assigned to one of three broad run times, will not achieve the necessary conservation can provide a powerful and cost-effective method of estimat- goals of the Canadian federal government’s WSP because of the ing relative population abundance and timing for application in substantial overlap in the return of wild and artificially enhanced fishery management.” While we agree that the genetic approach Sockeye Salmon and the lack of concurrent genetic identifica-

Downloaded by [Department Of Fisheries] at 19:21 20 July 2015 proposed by Beacham et al. can provide useful information for tion of the CUs of concern. If the goal of fisheries management complex mixed-stock ocean fisheries, we feel the authors over- for Skeena River Sockeye Salmon is to conserve vulnerable CUs state the approach’s utility in contemporary salmon management while catching abundant artificially enhanced Babine fish, man- in two ways. First, the proposed approach currently is unable agement must move a larger proportion of the fishery in-river and to identify 72% of Skeena River Sockeye Salmon CUs, which to terminal locations so that populations of concern may reach is the appropriate population-level structure required for fish- the spawning grounds. Yet if what we have done in building the eries management in Canada (DFO 2005, 2009; our Table 1). Babine spawning channels is to create a situation in which we Sixteen of 33 CUs currently do not have baseline genetic in- literally cannot manage the Skeena River Sockeye Salmon fish- formation and therefore remain unassigned stocks in the Tyee ery according to the WSP, perhaps it is time for Fisheries and test fishery. The inability to differentiate CUs of concern fails to Oceans Canada to reduce channel production. Indeed, “if con- address the exploitation-related problems that these populations servation objectives of the Wild Salmon Policy are focusing on face. Second, the stock composition estimates are imprecise for concerns about wild Skeena Sockeye, then reduced smolt out- populations that comprise <5% of the relative abundance in put from Babine Lake would help those stocks” (Walters et al. the fishery (Beacham et al. 2014). With the exception of the 2008). Such a reduction would simplify Sockeye Salmon fish- aggregate Babine Lake group, there is no Skeena River stock eries management in the Skeena River and help meet Canada’s reporting group that comprises more than 2.6% of the relative stated goals of conserving vulnerable conservation units. 1170 PRICEETAL.

MICHAEL H. H. PRICE* the Skeena River, British Columbia. North American Journal of Fisheries Management 34:335–348. Skeena Wild Conservation Trust, 4505 Greig Avenue, Cox-Rogers, S. 2012. Skeena Sockeye substock run timing and abundance Terrace, British Columbia V8G 1M6, Canada evaluated using Tyee test fishery DNA. Available: http://skeenawild.org/ images/uploads/docs/Cox-Rogers 2012 Skeena sockeye run-timing.pdf. (June 2014). ANDREW G. J. ROSENBERGER DFO (Fisheries and Oceans Canada). 2005. Canada’s policy for conservation of wild Pacific salmon. DFO, Vancouver. Available: http://www.pac.dfo- Raincoast Conservation Foundation, Post Office Box 2429, mpo.gc.ca/fm-gp/species-especes/salmon-saumon/wsp-pss/docs/wsp-pss- Sidney, British Columbia V8L 3Y3, Canada eng.pdf. (October 2014). DFO (Fisheries and Oceans Canada). 2009. Framework for characterizing con- GREG G. TAYLOR servation units of Pacific salmon (Oncorhynchus spp.) for implementing the Wild Salmon Policy. Canadian Science Advisory Secretariat Science Advi- Watershed Watch Salmon Society, 1037 Madore Avenue, sory Report 2008/052. Holtby, B. L., and K. A. Ciruna. 2007. Conservation units for Pacific salmon un- Coquitlam, British Columbia V3K 3B7, Canada der the Wild Salmon Policy. Canadian Science Advisory Secretariat Science Advisory Report 2007/070. JACK A. STANFORD Korman, J., and K. K. English. 2013. Benchmark analysis for Pacific salmon conservation units in the Skeena watershed. Available: http:// Flathead Lake Biological Station, University of Montana, skeenasalmonprogram.ca/libraryfiles/lib 318.pdf. (June 2014). 32125 Bio Station Lane, Polson, Montana 59860, USA Smith, R. D., and F. P. Jordan. 1973. Timing of Babine Lake Sockeye Salmon stocks in the north coast commercial fishery as shown by several taggings at *Corresponding author: [email protected] the Babine counting fence and rates of travel through the Skeena and Babine rivers. Fisheries Research Board of Canada Technical Report 418. Walters, C. J., J. A. Lichatowich, R. M. Peterman, and J. D. Reynolds. 2008. Report of the Skeena Independent Science Review Panel. Report of the Skeena REFERENCES Independent Science Review Panel to the Canadian Department of Fisheries Beacham, T. D., S. Cox-Rogers, C. MacConnachie, B. McIntosh, and C. G. and Oceans and the British Columbia Ministry of the Environment, Victoria, Wallace. 2014. Population structure and run timing of Sockeye Salmon in British Columbia. Downloaded by [Department Of Fisheries] at 19:21 20 July 2015 This article was downloaded by: [Department Of Fisheries] On: 20 July 2015, At: 19:15 Publisher: Taylor & Francis Informa Ltd Registered in England and Wales Registered Number: 1072954 Registered office: 5 Howick Place, London, SW1P 1WG

North American Journal of Fisheries Management Publication details, including instructions for authors and subscription information: http://www.tandfonline.com/loi/ujfm20 Population Structure and Run Timing of Sockeye Salmon in the Skeena River, British Columbia: Response to Comment Terry D. Beachama, Steven Cox-Rogersb, Cathy MacConnachiea, Brenda McIntosha & Colin G. Wallacea a Fisheries and Oceans Canada, Pacific Biological Station, 3190 Hammond Bay Road, Nanaimo, British Columbia V9T 6N7, Canada b Fisheries and Oceans Canada, North Coast Stock Assessment Unit, 417 2nd Avenue West, Prince Rupert, British Columbia V8J 1G8, Canada Published online: 10 Nov 2014. Click for updates

To cite this article: Terry D. Beacham, Steven Cox-Rogers, Cathy MacConnachie, Brenda McIntosh & Colin G. Wallace (2014) Population Structure and Run Timing of Sockeye Salmon in the Skeena River, British Columbia: Response to Comment, North American Journal of Fisheries Management, 34:6, 1171-1176, DOI: 10.1080/02755947.2014.956163 To link to this article: http://dx.doi.org/10.1080/02755947.2014.956163

PLEASE SCROLL DOWN FOR ARTICLE

Taylor & Francis makes every effort to ensure the accuracy of all the information (the “Content”) contained in the publications on our platform. However, Taylor & Francis, our agents, and our licensors make no representations or warranties whatsoever as to the accuracy, completeness, or suitability for any purpose of the Content. Any opinions and views expressed in this publication are the opinions and views of the authors, and are not the views of or endorsed by Taylor & Francis. The accuracy of the Content should not be relied upon and should be independently verified with primary sources of information. Taylor and Francis shall not be liable for any losses, actions, claims, proceedings, demands, costs, expenses, damages, and other liabilities whatsoever or howsoever caused arising directly or indirectly in connection with, in relation to or arising out of the use of the Content. This article may be used for research, teaching, and private study purposes. Any substantial or systematic reproduction, redistribution, reselling, loan, sub-licensing, systematic supply, or distribution in any form to anyone is expressly forbidden. Terms & Conditions of access and use can be found at http:// www.tandfonline.com/page/terms-and-conditions North American Journal of Fisheries Management 34:1171–1176, 2014 American Fisheries Society 2014 ISSN: 0275-5947 print / 1548-8675 online DOI: 10.1080/02755947.2014.956163

COMMENT

Population Structure and Run Timing of Sockeye Salmon in the Skeena River, British Columbia: Response to Comment

The use of genetic stock identification together with abun- aggregates, resulting in a relatively small number of manage- dances estimated from a test fishery is fairly standard in the ment units. Holt and Irvine (2013) further noted that management of ocean mixed-stock salmon fisheries in British Columbia (Beacham et al. 2004a, 2004b, 2008) as well as pro- [d]ifferences between CUs and management units result in chal- viding information on the abundance and timing of maturing lenges for managers who are responsible to biological status of CUs and fishery objectives for the management unit. If CUs are nested salmon entering freshwater in major river systems (Beacham within management units, reference points on management units and Wood 1999; Parken et al. 2008; Beacham et al. 2012, can be chosen to ensure biological objectives on component CUs are 2014b). Similar uses of genetic variation have been made out- achieved. side of British Columbia, in the management of ocean fisheries (Dann et al. 2013; Satterthwaite et al. 2014) and freshwater as- This was the approach followed by Beacham et al. (2014b), and sessments (Shaklee et al. 1999; Doctor et al. 2010; Flannery it is analogous to the approach followed in the management– et al. 2010). In their comment, Price et al. (2014) suggest that assessment of Fraser River Sockeye Salmon, for which har- Beacham et al. (2014b) overstate the utility of genetic stock vest rate and escapement guidelines are applied to management identification at a salmon test fishery in contemporary salmon units in order to achieve biological objectives for the component management. Price et al. concede that “the genetic approach CUs. For the majority of Sockeye Salmon fisheries in British proposed by Beacham et al. can provide useful information for Columbia, fish from different CUs that have similar run timing complex mixed-stock ocean fisheries,” but they seem reluctant are managed as aggregates in mixed-stock fisheries because the to accept the idea that information useful for both fisheries man- application of different harvest rates to individual CUs is typ- agement and stock assessment can be derived from a test fishery ically not possible. Management of the individual component in the lower portion of a major river drainage. CUs is not practical for Sockeye Salmon, and we do not believe As noted by Price et al. (2014), Canada has a policy for the that it is required under Canada’s WSP given the example out- conservation of wild Pacific salmon Oncorhynchus spp. (the lined by Holt and Irvine (2013). This approach is unlikely to Wild Salmon Policy [WSP]) and the designation of conserva- change for the foreseeable future. tion units (CUs) is an important component of that policy. Price Holtby and Ciruna (2007) designated 33 CUs for Sockeye et al. are critical of the genetic approach proposed by Beacham Salmon within the Skeena River drainage, 31 “lake-type” CUs

Downloaded by [Department Of Fisheries] at 19:15 20 July 2015 et al. (2014b), contending that it does not provide individual and 2 “river-type” CUs, with the abundance of Sockeye Salmon resolution for all Skeena River Sockeye Salmon O. nerka CUs within particular CUs being of no consequence with respect to in a lower-river test fishery and therefore cannot be used as a defining them. The existence of one river-type CU (Skeena– management and assessment tool in meeting the conservation High Interior) was questionable, as Holtby and Ciruna (2007) requirements of the WSP. We disagree and offer some perspec- noted (referring to the river-type CUs) that “[t]wo of the CUs tive for readers not aware of the CU issue being raised by Price consist of single populations of unknown status, and it is pos- et al. Holtby and Ciruna (2007) defined the CUs for Sockeye sible that neither CU is extant.” Price et al. (2014) mistakenly Salmon, generally assuming that the Sockeye Salmon spawning indicated that a middle Skeena River river-type CU was not rep- and rearing in individual lakes constituted individual CUs re- resented in the genetic baseline, but it was in fact represented gardless of the size of the population within the lake. Given the by two populations (Kispiox River and Nangeese River). Price CU approach, Irvine and Fraser (2008) noted that there can be in- et al. suggest that in our advocacy of applying genetic stock consistencies between CUs and management units. In their eval- identification to a Skeena River test fishery we uation of a case study of the WSP, Holt and Irvine (2013) noted that Sockeye Salmon can exist as a large number of isolated fail to consider the appropriate population-level structure required for conservation in Canada, do not emphasize the complex- CUs but that because the fisheries for Sockeye Salmon occur ity of the Sockeye Salmon populations in the Skeena River catch- primarily in marine and estuarine areas where separate popula- ment, and overstate the approach’s utility in contemporary salmon tions comingle management has focused on mixed-population management. 1171 1172 BEACHAM ET AL.

Are these valid comments? Price et al. are correct in pointing out method of estimating relative population abundance and tim- that Beacham et al. (2014b) never discussed the CU concept in ing for application in fishery management.” Price et al. indicate their description of the Tyee test fishery stock identification re- that “[n]umerous small CUs currently do not have baseline ge- sults. Instead, we focused on what information could be derived netic information and therefore remain unassigned stocks in the from genetic stock identification for fishery management appli- Tyee test fishery.” This observation is correct, but we point out cations. Because the abundance of Sockeye Salmon was of no that the populations included in our existing baseline cover the consequence in establishing CUs, some CUs are unlikely ever large majority of core spawning areas of known escapement to be identified within a season at current levels of fishery sam- within the Skeena River drainage based on cross-referencing pling because of their low relative abundance, but we suggest with current and historic spawning ground escapement records that this does not lessen the value of genetic stock identification (Spilsted and Spencer 2009). We are unaware of any application in contemporary salmon fisheries management. of genetic stock identification that covers all of the populations Perhaps some additional insights into the validity of Price potentially contributing to a mixed-stock sample. Tagging stud- et al.’s (2014) comments can be obtained by examining the ies conducted in the 1940s and 1950s for Sockeye Salmon in role that genetic stock identification plays in the management the Skeena River drainage that had been tagged in marine wa- of Sockeye Salmon from the Fraser River, which supports the ters or in the lower river and subsequently recovered within the largest Sockeye Salmon fishery in Canada; is managed un- drainage indicated that all but one of the 2,323 tagged individ- der the terms of the Pacific Salmon Treaty between Canada uals recovered were from populations currently in the baseline and the United States; and has greater population diversity (Aro and McDonald 1968). As Beacham et al. (2014b) indi- within the watershed than the Sockeye Salmon from the Skeena cated, “[n]o populations of significant abundance are known to River drainage, comprising some 37 CUs (Holtby and Ciruna be unrepresented in the current baseline, and thus estimates of 2007). Genetic stock identification is applied to marine and stock composition of actual fishery samples with the current freshwater test fishery catches (Beacham et al. 2004a, 2004b) baseline should be reliable.” The absence of small CUs from the as well as to commercial catches on both sides of the U.S.– baseline will essentially have little effect on the estimated stock Canadian border. The role of genetic stock identification in compositions. For example, the Johnston Lake population—a the management of Fraser River Sockeye Salmon has been de- CU that was added to the baseline of Beacham et al. (2014b)— scribed by Pacific Salmon Commission staff as follows: was estimated to have comprised only 0.5% of the individuals sampled at the test fishery (Table 1). As Fraser Sockeye comprise a diverse array of over 40 popula- Next, Price et al. (2014) indicate that “stock composition tions and management is done by stock aggregate, genetic analyses estimates are imprecise for populations that comprise <5% of conducted by the DFO’s [Fisheries and Oceans Canada] in-season stock ID team have been critical in ensuring that stock strengths the relative abundance in the fishery.” This observation is again are sufficient to provide harvestable surpluses. Furthermore, the ge- correct and was noted by Beacham et al. (2014b). The impre- netic analyses helped to ensure that fisheries were opened at times cision stems from the relatively high coefficients of variation and places targeting abundant stocks while minimizing impacts on associated with estimates of stock composition <5% (see Table weak stocks. The real-time analysis of multiple samples over the 5 in Beacham et al. 2014b for estimates and associated standard course of a three-month period is not an easy task (particularly given uncertain shipping times and variable needs regarding re- deviations), and this is a reflection of the sample size of the sult delivery times), but DFO’s in-season stock ID team contin- mixed-stock sample. However, the accuracy of the estimated ues to deliver high-throughput, high-quality analyses to meet the stock composition is of far greater importance for fishery man- demands of the Fraser River Panel. Pacific Salmon Commission agement applications than is the precision, and there is little staff could not conduct their in-season assessment without this re- Downloaded by [Department Of Fisheries] at 19:15 20 July 2015 doubt that accurate estimates of stock composition were pro- liable service. [M. Lapointe, Pacific Salmon Commission, personal communication] vided, as was outlined for individual identification of Sockeye Salmon by Beacham et al. (2011a, 2014a). The management of Fraser River Sockeye Salmon is more com- Price et al. (2014) suggest that plex than that of Sockeye Salmon from the Skeena River, and genetic stock identification information from test fisheries is the coarse resolution of Beacham et al.’s results, whereby all considered crucial to the management of Fraser River fisheries wild Babine Sockeye Salmon CUs (each known to have a unique run as well as for stock assessment. We conclude that there is no time; Smith and Jordan 1973; Figure 2A) are grouped genetically with artificially enhanced fish as a “middle run time,” would not validity to Price et al.’s (2014) contention that our advocacy of prevent the exploitation of these CUs of concern in mixed-stock genetic stock identification “overstate[s] the approach’s utility fisheries and fails to meet the intent of the WSP. in contemporary salmon management.” Price et al. (2014) further suggest that Beacham et al. (2014b) As previously noted, though, management units containing “overstate the approach’s utility in contemporary salmon man- comigrating CUs can be managed to meet the biological ob- agement in two ways,” the approach in question being “the ap- jectives of the component CUs while fulfilling the intent of the plication of genetic stock identification technology to a salmon WSP (Holt and Irvine 2013). Price et al. are correct in stat- test fishery [which] can provide a powerful and cost-effective ing that Beacham et al. (2014b) could not reliably separate the COMMENT 1173

TABLE 1. Estimated percentage stock composition of Skeena River Sockeye Salmon as obtained from a lower-river test ﬁshery during 2000–2013 and estimated with a 29-population baseline (14 reporting groups) incorporating variation at 14 microsatellites. Fish were analyzed in proportion to weekly run abundance within the year; N is the number of ﬁsh successfully genotyped.

Reporting Mean group 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 (SD) N 986 1,181 794 907 653 724 996 664 736 573 664 595 698 506 762 Johnston Lake 0.1 0.1 0.5 0.8 1.7 1.3 0.1 0.0 0.0 0.6 1.5 0.2 0.1 0.5 0.5 (0.2) Alastair Lake 1.3 1.5 0.9 4.0 4.8 6.4 2.0 3.5 0.4 0.5 2.9 3.5 3.4 4.1 2.8 (0.6) Lakelse Lake 0.8 2.0 0.4 1.1 3.4 2.7 1.3 1.1 0.4 0.8 1.2 2.0 1.0 5.0 1.7 (0.4) Zymoetz River 0.3 0.3 1.2 1.7 0.6 2.2 0.4 0.9 0.0 0.6 0.7 1.3 0.0 1.6 0.8 (0.3) Kitwanga Lake 0.0 0.1 0.0 0.3 0.1 0.0 0.1 0.0 0.3 0.6 2.5 0.0 0.1 1.0 0.4 (0.2) Kitsumkalum Lake 0.5 1.3 1.1 5.7 2.0 1.8 0.2 0.8 1.1 1.4 3.1 1.5 1.5 3.5 1.8 (0.5) Kispiox River 1.6 1.5 3.8 3.4 2.3 3.1 1.4 2.3 4.6 2.3 1.7 1.9 3.7 2.6 2.6 (0.6) Motase Lake 0.1 0.1 0.0 0.3 0.0 0.3 0.0 0.0 0.0 0.5 0.1 0.4 0.2 0.0 0.1 (0.1) Morice Lake 0.5 1.4 2.0 1.7 0.4 0.3 0.2 2.2 1.5 1.7 1.6 1.2 0.3 2.4 1.2 (0.4) Bear Lake 1.2 1.8 0.6 0.8 0.8 0.2 0.6 0.0 0.9 1.5 0.5 0.6 0.5 0.6 0.8 (0.4) Sustut Lake 0.2 0.1 0.3 0.9 0.5 0.3 0.2 0.8 0.1 0.7 0.9 0.7 0.7 1.8 0.6 (0.2) Slamgeesh Lake 0.3 0.0 0.9 0.8 0.8 0.4 0.9 0.1 0.8 5.3 0.9 1.0 1.3 5.6 1.4 (0.4) Tahlo–Morrison lakes 3.4 3.3 2.1 5.9 2.7 0.2 5.4 0.2 4.4 6.3 1.4 6.0 2.4 5.3 3.5 (1.7) Babine Lake, other 89.6 86.5 86.4 72.6 79.8 81.0 87.3 88.3 85.5 77.2 81.1 79.7 84.8 66.0 81.9 (2.2) Babine Lake, totala 93.0 89.8 88.5 78.5 82.5 81.2 92.7 88.5 89.9 83.5 82.5 85.7 87.2 71.3 85.4 (1.3)

aIncludes Tahlo–Morrison and other Babine Lake reporting groups.

populations in this CU from other populations in Babine Lake. rison Lake), but Sockeye Salmon spawning in Morrison Creek However, since the results outlined in Beacham et al. were ob- are believed to produce fry that rear in Morrison Arm of Babine tained, we have acquired access to samples from the Morrison Lake. Lake lake-spawning population as well as enhancing our base- With the wild Tahlo Creek and Morrison Lake CU reli- line by 49 individuals collected in 2013 from the Tahlo Creek ably identified, as well as the Johnston Lake CU, we reesti- population. Tahlo Lake is drained by lower Tahlo Creek, which mated the annual stock compositions at the Tyee test fishery flows into Morrison Lake; that lake, in turn, is drained by Morri- in order to evaluate the contribution of these two CUs to the son Creek, which flows into Babine Lake. Wild salmon are found test fishery. With a 29-population baseline assigned to 14 re- in the Tahlo–Morrison CU (the Morrison Arm of Babine Lake). porting groups (the original baseline enhanced by the addition We evaluated whether the three populations (Tahlo Creek, Mor- of the Morrison Lake and Johnston Lake populations and the rison Lake, and Morrison Creek) could be reliably identified number of reporting groups increased by identification of the to the wild Tahlo–Morrison CU. The evaluation was conducted Tahlo–Morrison and Johnston Lake reporting groups [accuracy as outlined by Beacham et al. (2014b), where each of these of Johnston Lake = 99.9%]). The Tahlo–Morrison lakes com- three populations was tested by estimating the stock compo- ponent was an average 3.5% of escapement past the test fishery

Downloaded by [Department Of Fisheries] at 19:15 20 July 2015 sition of a single-population mixture comprised of 200 simu- during 2000–2013, and the Johnston Lake component was 0.5% lated multilocus genotypes and the entire 29-population Skeena (Table 1). River baseline was used to estimate the stock composition of Smith and Jordan (1973) reported that the timing of the upper each mixture. A target of 90% accuracy was set for remaining Babine River and lower Babine River populations (which spawn as a separate reporting group. This analysis indicated that the at the two ends of Nilkitkwa Lake in the Babine River) was later Tahlo Creek population could be reliably identified to the wild than that of other Babine Lake populations. We estimated the Tahlo–Morrison CU (population: 84.8% accurate; CU: 91.4% mean date through the lower test fishery for other Babine Lake accurate), as could the lake-spawning population in Morrison populations to be between July 17 and July 24 but to be July Lake (96.0%; 97.3%); but the Morrison Creek population could 30 for the upper Babine River population and August 2 for the not be reliably identified to this CU (12.0%; 28.7%). However, lower Babine River population, dates that are in accordance with the Morrison Creek population was reliably identified as part of the results of Smith and Jordan (1973). However, we could not the original Babine Lake complex outlined by Beacham et al. provide reliable estimates of stock composition for this Nilk- (current reporting group accuracy = 92.2%) and was retained itkwa Lake CU (upper and lower Babine River populations con- within that reporting group. Sockeye Salmon spawning in lower sidered a reporting unit; upper Babine River population accu- Tahlo Creek are believed to produce fry that rear in Morrison racy = 56%, regional CU accuracy = 83%; lower Babine River Lake (along with fry from lake-spawning individuals in Mor- population accuracy = 21%, regional CU accuracy = 75%), 1174 BEACHAM ET AL.

so we retained these two populations within the Babine Lake rate of the stock of conservation concern was actually lower reporting group. Beacham et al. (2011b) reported that for less than during the preceding period of substantially reduced total genetically distinct populations larger baseline population sam- Chinook Salmon catch in the fishery. Given this example, that ple sizes are required to achieve a given level of accuracy in of the previously mentioned Fraser River Sockeye Salmon, and estimated stock composition. Thus, it may be that with greater the many examples of genetic stock identification in operation baseline sample sizes for these two populations, sufficient ac- in British Columbia and elsewhere over the past 15 years, we curacy would be attained to classify these populations as an suggest that there is demonstrable evidence that genetic stock additional reporting group for stock identification applications identification technology provides a powerful and cost-effective in the Skeena River drainage. Increased baseline sample sizes method of obtaining relevant information for fishery manage- for existing baseline populations may aid in finer stock discrim- ment applications. ination, such that accurate estimates of stock composition for Price et al. (2014) have misinterpreted our comments on the additional CUs (such as Nilkitkwa Lake and Stevens Lake) may utility of the lower Nass River Sockeye Salmon test fishery. be possible. They state that we The addition of samples to the baseline of Beacham et al. (2014b) illustrates an important point about the baselines used imply that the detection of blocked spawning grounds for Kwinageese Sockeye Salmon was due only to a Nass River test in genetic stock identification applications. Baselines are not fishery, which is erroneous; had the test fishery alone (i.e., without static, tending to grow in both the number of populations repre- spawning ground surveys) been responsible for escapement esti- sented and the number and types of genetic markers used in stock mates, the habitat concern and resulting reduction in spawners may discrimination. The Tahlo–Morrison component of the baseline never have been known. was enhanced owing to considerable local interest in develop- Beacham et al. (2014b) stated the following with respect to ing an application that could separate the wild and enhanced Kwinageese Sockeye Salmon: Sockeye Salmon components from Morrison Arm in Babine Lake. Additional baseline sampling led to the development of In 2009, discrepancies began to occur between estimated this new capability, illustrating the capacity of genetic stock Kwinageese River escapement in the lower Nass River test fishery identification to enable different applications to be undertaken. and abundance observed on the spawning grounds, with far fewer Sockeye Salmon reaching the spawning grounds than were estimated Price et al. (2014) ignored this key aspect of baseline develop- to be in the test fishery. ment. After noting that many small CUs lack baseline genetic information, they state that “the inability to differentiate CUs of Clearly, as indicated by Beacham et al., there were two key concern fails to address the exploitation-related problems that sources of data regarding Kwinageese River Sockeye Salmon these populations currently face.” Perhaps because they were not abundance, namely, the lower Nass River test fishery and obser- completely familiar with developing genetic stock identification vations on the spawning grounds in the Kwinageese River. As applications, Price et al. assumed that the baseline outlined by outlined by Beacham et al., Beacham et al. was static in nature. Their perceived baseline [b]ased upon the estimated stock composition in the lower Nass deficiency would be easily rectified if the appropriate baseline River test fishery, assessment staff had confidence that Sockeye samples were available. In our experience, stakeholders and Salmon originating from the Kwinageese River were actually in the other interested parties provide us with access to appropriate lower Nass River, narrowing the location where the loss of salmon samples that are then incorporated into the baseline and esti- was occurring. mates of stock composition are made. However, in the Skeena Price et al. (2014) may wish to minimize the importance of Downloaded by [Department Of Fisheries] at 19:15 20 July 2015 River Sockeye Salmon example, even if all of the perceived the application of genetic stock identification at a salmon test baseline deficiencies were rectified the estimated stock compo- fishery in contemporary salmon management, but the facts speak sitions are unlikely to change appreciably owing to the fact that for themselves in this particular example. DNA-level variation the missing populations contribute so little to drainage escape- was first applied to Nass River Sockeye Salmon escapement ment. estimation 18 years ago (Beacham and Wood 1999), and it Does the application of genetic stock identification technol- has remained an important source of information for fishery ogy to a salmon test fishery provide a powerful and cost-effective managers. As indicated by Beacham at al. (2014b), method of estimating relative population abundance and timing for application in fishery management? In an application of [t]he test fishery results were important in the management this technology to the management of the Chinook Salmon O. and assessment of all Nass River Sockeye Salmon, as managers tshawytscha troll fishery of Haida Gwaii in British Columbia, had subsequently restricted fishery openings to permit passage of Beacham et al. (2008) concluded that the economic payback to Kwinageese River salmon after spawning escapement had declined dramatically in 2009. the northern troll fleet was over 40 times the cost of generating the in-season stock identification estimates on which manage- Referring to fishery management requirements, Price et al. ment decisions were based. When in-season stock identification (2014) suggest that “the Tyee test fishery provides at best only was applied to fishery management decisions, the exploitation a “noisy” estimate of escapement (plus or minus 20% of run COMMENT 1175

2.5 ported by Beacham et al. (2014b) have been conducted on the catch of steelhead O. mykiss at the Tyee test ﬁshery (Beacham 2.0 et al. 2012), and an independent science panel recommended that genetic stock identiﬁcation be conducted on this catch in

1.5 order to estimate escapement to major systems (Walters et al. 2008)—completely analogous to the Sockeye Salmon Tyee test fishery example reported by Beacham et al. 1.0 Price et al. (2014) state that “run timing is highly variable between years, and the annual variation is not consistent among 0.5 CUs.” With respect to the timing of Sockeye Salmon entry into the Skeena River, we compared the results for specific popula- Babine abundance above Tyee (millions) Babine abundance above Tyee 0.0 tions reported by Aro and McDonald (1968) based on tagging 0 500 1000 1500 2000 2500 3000 with those reported by Beacham et al. (2014b) based on genetic Babine CPUE Index at Tyee (June 10-Aug 25) stock identification at the lower-river test fishery. Aro and Mc- FIGURE 1. Annual seasonal catch per unit effort index (CPUE) of Babine Donald (1968) state that the run returning to Lakelse Lake “is Lake–origin Sockeye Salmon at the Tyee test fishery versus escapement at the apparently early and of short duration”—which is exactly the Babine Lake counting fence plus Babine Lake–origin catch between Tyee and result depicted in Figure 5 of Beacham et al. (2014b). With re- the counting fence, 2000–2013. spect to the return to Alastair Lake, Aro and McDonald (1968) state that “while the bulk of the Alastair Sockeye run is early, a size).” This comment applies to the typical range of in-season segment of the run has been present in the fishing area as late as (i.e., during the fishery) escapement estimation error vis-a-vis` early August,” which again is exactly the result depicted in Fig- postseason estimates generated from spawning ground counts. ure 5 of Beacham et al. With respect to the return to Morice Lake The discrepancy is caused by uncertainty in the expansion fac- (Bulkley River), Aro and McDonald (1968) state that “the data tors (catchability) used to generate the in-season escapement indicate that the Bulkley River Sockeye run has been present in estimates; however, the in-season estimates are still adequate the fishing area from mid-June to late July, but are inadequate for managing the fishery because they are reported to managers to show whether it was present in August,” again consistent as probability distributions, not as point estimates (P. Hall, Fish- with the results presented by Beacham et al. Aro and McDonald eries and Oceans Canada, personal communication). For the (1968) summarized the timing of Skeena River Sockeye Salmon types of stock assessments needed by managers, the Tyee test runs as follows: fishery does in fact provide an accurate index of drainage escape- The earliest runs which entered the river were the Alastair and ment or “abundance.” For example, as discussed by Beacham Lakelse runs. These were followed a short while later by the Babine et al. (2014b), known escapement numbers in the Skeena River Lake and Bulkley River runs. The Lakelse Lake run and the main drainage were available only from the counting fence at the out- part of the Alastair Lake run were present in the fishery during June let to Babine Lake. DNA-derived stock composition data have and early July. The tagging data indicate that the Bulkley River run been available from the Tyee test fishery on an annual basis continued to the end of July and the Babine Lake run into late August. The limited information on the other Skeena runs suggests that they beginning in 2000. By comparing the known Babine Lake– were present off the river mouth in July. specific annual CPUE index at Tyee between 2000 and 2013 with the known Babine Lake–origin abundance at the counting These results are essentially the same as those derived from ge- Downloaded by [Department Of Fisheries] at 19:15 20 July 2015 fence plus the Babine Lake–specific catch between Tyee and the netic stock identification of the lower-river test fishery catches counting fence, a reasonably accurate and precise relationship that were reported by Beacham et al. (2014b), but we were able between Babine Lake CPUE at Tyee and subsequent abundance to provide more detailed information on timing because we typ- was obtained (r2 = 0.78, P < 0.01; Figure 1). Beacham et al. ically observed greater recoveries for a population in the test had noted previously that fishery than were derived directly from tagging. We also indicated previously that the timing for the upper and lower Babine [i]f an absolute measure of abundance is available from a popula- River populations relative to other Babine Lake populations was tion that constitutes a major portion of the return, then the abundance in accordance with the results of Smith and Jordan (1973). We of the major population is divided by its seasonal contribution to the test fishery, providing an estimate of drainage escapement. conclude that Price et al.’s (2014) assertion that the Tyee test fishery does not provide accurate estimates of the timing of This description accurately depicts the Babine Lake component Sockeye Salmon entry into the river is without merit. of the return (85%, as estimated in the test fishery), and because Price et al. (2014) suggested that drones or some other tech- an absolute estimate of the abundance of Babine Lake–origin nology could be used to determine escapement, but it is unlikely Sockeye Salmon is available from the counting fence an esti- that a cost-effective absolute estimate of drainage abundance mate of the abundance of all other populations on the spawning will be possible with drone technology due to weather and tech- grounds can be made. In addition, analyses similar to those re- nical constraints, while this estimate can currently be provided 1176 BEACHAM ET AL.

by genetic stock identification. Genetic stock identification at the stock analysis of Sockeye Salmon (Oncorhynchus nerka) in British Columbia. Tyee test fishery has been used for calibrating annual Sockeye Canadian Journal of Fisheries and Aquatic Sciences 68:550–562. Salmon escapement estimates for the Pacific Salmon Commis- Beacham, T. D., C. G. Wallace, K. D. Le, and M. Beere. 2012. Population structure and run timing of steelhead trout in the Skeena River, British Columbia. sion review process, and it has been useful in assessing the stock North American Journal of Fisheries Management 32:262–275. status of Sockeye Salmon within the Skeena River drainage. It Beacham, T. D., I. Winther, K. L. Jonsen, M. Wetklo, L. Deng, and J. R. Candy. has also been recommended as a way to estimate steelhead 2008. The application of rapid microsatellite-based stock identification to escapement in the drainage (Walters et al. 2008). The current management of a Chinook Salmon troll fishery off the Queen Charlotte Is- baseline can certainly be enhanced with additional populations lands, British Columbia. North American Journal of Fisheries Management 28:849–855. of interest, but as reliable estimates of the abundance and tim- Beacham, T. D., and C. C. Wood. 1999. Application of microsatellite DNA ing of Sockeye Salmon stocks are currently available from the variation to estimation of stock composition and escapement of Nass River Tyee test fishery, we expect that genetic stock identification ap- Sockeye Salmon (Oncorhynchus nerka). Canadian Journal of Fisheries and plications will play an increasing role in the assessment and Aquatic Sciences 56:297–310. management of Skeena River salmon resources. Dann, T. H., C. Habicht, T. T. Baker, and J. E. Seeb. 2013. Exploiting genetic diversity to balance conservation and harvest of migratory salmon. Canadian Journal of Fisheries and Aquatic Sciences 70:785–793. TERRY D. BEACHAM* Doctor, K. K., R. Hilborn, M. Rouse, and T. Quinn. 2010. Spatial and temporal patterns of upriver migration by Sockeye Salmon populations in the Wood Fisheries and Oceans Canada, Pacific Biological Station, River system, Bristol Bay, Alaska. Transactions of the American Fisheries 3190 Hammond Bay Road, Nanaimo, Society 139:80–91. British Columbia V9T 6N7, Canada Flannery, B. G., T. D. Beacham, J. R. Candy, R. R. Holder, G. F. Maschmann, E. J. Kretschmer, and J. K. Wenburg. 2010. Mixed-stock analysis of Yukon River Chum Salmon: application and validation in a complex fishery. North STEVEN COX-ROGERS American Journal of Fisheries Management 30:1324–1338. Fisheries and Oceans Canada, North Coast Stock Assessment Holt, C. A., and J. R. Irvine. 2013. Distinguishing benchmarks of biological Unit, 417 2nd Avenue West, Prince Rupert, British Columbia status from management reference points: a case study on Pacific salmon in Canada. Environmental Conservation 40:345–355. V8J 1G8, Canada Holtby, B. L., and K. A. Ciruna. 2007. Conservation units for Pacific salmon under the Wild Salmon Policy. Canadian Science Advisory Secretariat Science CATHY MACCONNACHIE,BRENDA MCINTOSH, Advisory Report 2007/070. AND COLIN G. WALLACE Irvine, J. R., and G. A. Fraser. 2008. Canada’s Wild Salmon Policy and the maintenance of diversity. Pages 391–398 in J. Nielsen, J. J. Dodson, K. Friedland, Fisheries and Oceans Canada, Pacific Biological Station, T. R. Hamon, J. Musick, and E. Verspoor, editors. Reconciling fisheries with 3190 Hammond Bay Road, Nanaimo, conservation: proceedings of the fourth World Fisheries Congress, volume I. British Columbia V9T 6N7, Canada American Fisheries Society, Symposium 49, Bethesda, Maryland. Parken, C. K., J. R. Candy, J. R. Irvine, and T. D. Beacham. 2008. Genetic and coded wire tag results combine to allow more precise management of *Corresponding author: [email protected] a complex Chinook Salmon aggregate. North American Journal of Fisheries Management 28:328–340. Price, M. H. H., A. G. J. Rosenberger, G. G. Taylor, and J. A. Stanford. 2014. REFERENCES Comment: population structure and run timing of Sockeye Salmon in the Aro, K. V., and J. McDonald. 1968. Times of passage of Skeena River Sockeye Skeena River, British Columbia. North American Journal of Fisheries Man- and Pink salmon through the commercial fishing area. Fisheries Research agement 34:1167–1170. Board of Canada Manuscript Report 984. Satterthwaite, W. H., M. S. Mohr, M. R. O’Farrell, E. C. Anderson, M. A. Banks, Beacham, T. D., R. J. Beamish, J. R. Candy, S. Tucker, J. H. Moss, and M. S. J. Bates, M. R. Bellinger, L. A. Borgerson, E. D. Crandall, J. C. Garza, B. Downloaded by [Department Of Fisheries] at 19:15 20 July 2015 Trudel. 2014a. Stock-specific migration size of juvenile Sockeye Salmon J. Kormos, P. W. Lawson, and M. L. Palmer-Zwahlen. 2014. Use of genetic in British Columbia waters and in the Gulf of Alaska. Transactions of the stock identification data for comparison of the ocean spatial distribution, size American Fisheries Society 143:876–888. at age, and fishery exposure of an untagged stock and its indicator: California Beacham, T. D., J. R. Candy, E. Porszt, S. Sato, and S. Urawa. 2011a. Mi- coastal versus Klamath River Chinook Salmon. Transactions of the American crosatellite identification of Canadian Sockeye Salmon rearing in the Bering Fisheries Society 143:117–133. Sea. Transactions of the American Fisheries Society 140:296–306. Shaklee, J. B., T. D. Beacham, L. Seeb, and B. A. White. 1999. Managing Beacham, T. D., S. Cox-Rogers, C. MacConnachie, B. McIntosh, and C. G. fisheries using genetic data: case studies from four species of Pacific salmon. Wallace. 2014b. Population structure and run timing of Sockeye Salmon in Fisheries Research 43:45–78. the Skeena River, British Columbia. North American Journal of Fisheries Smith, H. D., and F. P. Jordan. 1973. Timing of Babine Lake Sockeye Salmon Management 34:335–348. stocks in the north coast commercial fishery as shown by several taggings at Beacham, T. D., M. Lapointe, J. R. Candy, B. McIntosh, C. MacConnachie, A. the Babine counting fence and rates of travel through the Skeena and Babine Tabata, K. Kaukinen, L. Deng, K. M. Miller, and R. E. Withler. 2004a. Stock rivers. Fisheries Research Board of Canada Technical Report 418. identification of Fraser River Sockeye Salmon using microsatellites and major Spilsted, B., and B. Spencer. 2009. Documentation of north coast (statistical histocompatibility complex variation. Transactions of the American Fisheries areas 1–6) salmon escapement information. Canadian Manuscript Report of Society 133:1117–1137. Fisheries and Aquatic Sciences 2802. Beacham, T. D., M. Lapointe, J. R. Candy, K. M. Miller, and R. E. Withler. Walters, C. J., J. A. Lichatowich, R. M. Peterman, and J. D. Reynolds. 2008. 2004b. DNA in action: rapid application of DNA variation to Sockeye Salmon Report of the Skeena Independent Science Review Panel. Report of the Skeena fisheries management. Conservation Genetics 5:411–416. Independent Science Panel to the Canadian Department of Fisheries and Beacham, T. D., B. McIntosh, and C. Wallace. 2011b. A comparison of poly- Oceans and the British Columbia Ministry of the Environment, Victoria, morphism of genetic markers and population sample sizes required for mixed- British Columbia. This article was downloaded by: [Department Of Fisheries] On: 20 July 2015, At: 19:15 Publisher: Taylor & Francis Informa Ltd Registered in England and Wales Registered Number: 1072954 Registered office: 5 Howick Place, London, SW1P 1WG

North American Journal of Fisheries Management Publication details, including instructions for authors and subscription information: http://www.tandfonline.com/loi/ujfm20 Timing of First Annulus Formation in White Sucker Otoliths Daniel W. Beckmana & John W. Calfeea a Department of Biology, Missouri State University, 901 South National Avenue, Springfield, Missouri 65897, USA Published online: 14 Nov 2014.

Click for updates

To cite this article: Daniel W. Beckman & John W. Calfee (2014) Timing of First Annulus Formation in White Sucker Otoliths, North American Journal of Fisheries Management, 34:6, 1187-1189, DOI: 10.1080/02755947.2014.951809 To link to this article: http://dx.doi.org/10.1080/02755947.2014.951809

PLEASE SCROLL DOWN FOR ARTICLE

Taylor & Francis makes every effort to ensure the accuracy of all the information (the “Content”) contained in the publications on our platform. However, Taylor & Francis, our agents, and our licensors make no representations or warranties whatsoever as to the accuracy, completeness, or suitability for any purpose of the Content. Any opinions and views expressed in this publication are the opinions and views of the authors, and are not the views of or endorsed by Taylor & Francis. The accuracy of the Content should not be relied upon and should be independently verified with primary sources of information. Taylor and Francis shall not be liable for any losses, actions, claims, proceedings, demands, costs, expenses, damages, and other liabilities whatsoever or howsoever caused arising directly or indirectly in connection with, in relation to or arising out of the use of the Content. This article may be used for research, teaching, and private study purposes. Any substantial or systematic reproduction, redistribution, reselling, loan, sub-licensing, systematic supply, or distribution in any form to anyone is expressly forbidden. Terms & Conditions of access and use can be found at http:// www.tandfonline.com/page/terms-and-conditions North American Journal of Fisheries Management 34:1187–1189, 2014 C American Fisheries Society 2014 ISSN: 0275-5947 print / 1548-8675 online DOI: 10.1080/02755947.2014.951809

MANAGEMENT BRIEF

Timing of First Annulus Formation in White Sucker Otoliths

Daniel W. Beckman* and John W. Calfee Department of Biology, Missouri State University, 901 South National Avenue, Springﬁeld, Missouri 65897, USA

population, unrecognized aging errors can affect age-based Abstract conservation and management plans (Beamish and McFarlane The purpose of this study was to document the time of forma- 1983). Yearly aging studies of Catostomus species based on tion for the first two otolith annuli for White Sucker Catostomus otoliths have included mostly mature fish older than age 2 years, commersoni. Young-of-the-year and age-1 individuals were sampled monthly from a population in the White River drainage of the and age validation studies for White Suckers have been done Missouri Ozarks. The formation of a single annual opaque region with populations dominated by fish of age 5 years and older during the months of June–September of the first 2 years of life (Thompson and Beckman 1995). For reliable growth modeling was documented using edge analysis, whereby the progression of and population assessment, it is especially important to doc- otolith annulus formation is followed throughout the year in the ument whether the first annulus is formed in age-0 or age-1 population. The documentation that the first annulus was formed within the first 6 months following birth provides for more accu- fish. For example, White Suckers spawn primarily during spring rate and precise age estimations than previously available for this months (Wakefield and Beckman 2005), and otolith annulus species. form primarily in late summer to fall months (Thompson and Beckman 1995); therefore, the first annulus could form at an White Sucker Catostomus commersoni is a widespread approximate age of either 0.5 or 1.5 years. The current study catostomid found throughout much of Canada and the northern was designed to document the time of the formation for the first United States east of the Rocky Mountains; coolwater streams two otolith annuli by following Missouri Ozarks White Sucker in southern Missouri are near the southern edge of its range populations in the wild from hatch through age 2 years. (Pflieger 1975). In southwest Missouri, White Suckers moves from White River reservoirs into tributaries to spawn over gravel substrates in clear, shallow stream waters, being an important METHODS component of these aquatic ecosystems (Wakefield and Beck- This study was carried out on Bee Creek, a tributary of Lake man 2005). Growth of White Suckers is highly variable through- Taneycomo, a reservoir on the White River of southwest Mis- out their range (Trippel and Harvey 1987), and age and growth souri. Visual observations of White Suckers were made from Downloaded by [Department Of Fisheries] at 19:15 20 July 2015 data can thus provide a valuable measure of environmental con- boat or shore in Bee Creek upstream of its confluence with Lake dition. Obtaining reliable growth data requires valid age estima- Taneycomo beginning 3 March 1994 and approximately weekly tion methods. thereafter. Spawning was documented by observing behaviors Otolith annulus counts have been documented as a reliable associated with reproduction, including aggregation in shallows age estimator for several species of suckers (Catostomidae) in and clearing of nesting areas (Curry and Spacie 1984). Spawn- the genus Catostomus (Sylvester and Berry 2006; Quist et al. ing was verified by collecting eggs from nesting areas follow- 2007; Bettinger and Crane 2011). Aging methods were vali- ing spawning. Following initial spawning activities, sweepnet dated for White Suckers by Thompson and Beckman (1995), sampling was carried out three times per week in pelagic near- using edge analysis (Beckman and Wilson 1995). Although surface waters of shallow backwater areas, where White Sucker such validations have shown the reliability of otolith aging for larvae tend to congregate following swim-up (Clifford 1972). adult fish, immature fish have often been excluded from studies White Suckers were sampled monthly from May 1994 to because sample collection methods typically exclude smaller April 1995, except in December and January, in areas of fish. Unless validation is carried out for all age-groups in a Bee Creek where spawning was documented. Individuals were

*Corresponding author: [email protected] Received May 14, 2014; accepted July 23, 2014 1187 1188 BECKMAN AND CALFEE

recognized as young-of-the-year fish by monitoring growth 100 after observed spawning events, egg collection, and appearance 90 of larvae. Age-1 fish were recognized by size, since early co- 80 70 horts maintain a distinct size distribution. So that all size-classes 60 were collected, the White Suckers were sampled with sweepnets 50 µ Young-of-the-year (750- m mesh), seines, and boat electrofishing (Calfee 1995). 40 Age 1 Electrofishing was carried out by intermittently moving shore- 30 ward while gradually progressing in a downstream direction. 20

An observer at the front of the boat netted all White Suckers Edge with Opaque Otolith Percent 10 0 observed to be affected by the current. May June July Aug Sep Oct Nov Dec Jan Feb Mar Apr The two lapilli otoliths were removed from each fish, one of 38 36 47 28 26 40 20 35 49 16 which was processed for aging by embedding in an epoxy resin. Month The embedded otolith was sectioned in the transverse plane by FIGURE 1. Percent of individuals with opaque zones at the growing edge of using a Struers Minitom low-speed saw with a diamond-tipped otoliths by month of capture for White Suckers in Bee Creek, White River system cutoff wheel (Thompson and Beckman 1995). Sections were of Missouri, as young-of-the-year (n = 335) and age 1 (n = 136). Numbers below sanded to a thickness of 0.50–0.75 mm and polished on a felt months indicate sample sizes for young-of-the-year individuals. µ pad using 0.3- m alumina suspension polishing compound. 2 months later in young-of-the-year fish than in age-1 fish (Fig- Daily growth increments were used to age larval fish up ure 1). These studies, along with other aging studies of Ozarks to 25 d posthatch (Calfee 1995). For juveniles, otolith trans- fishes (Beckman 2002; Beckman and Hutson 2012; Simmons verse sections were assessed using a compound microscope and Beckman 2012; Beckman and Howlett 2013), suggest that × with transmitted light at 40 magnification to identify concen- the predominant time of formation of the otolith opaque growth tric opaque (dark) and translucent (light) regions. Opaque re- zone (annulus) falls within the months of May–September for gions were counted as annuli (Thompson and Beckman 1995). stream fish species in this geographic region. This supports the Otolith edges were ranked using marginal increment analysis argument that otolith annulus formation is seasonal but not di- to assess the timing of annulus formation, plotting the percent- rectly related to spawning (Beckman and Wilson 1995); more age of individual fish exhibiting an opaque zone at the growing likely, formation is a result of temperature effects on otolith edge of the otolith throughout the year (Beckman et al. 1991). composition (Campana 1999) or reflects a seasonal increase in The outer growing edge of one otolith section from each fish fish growth or metabolism (Brothers 1979; Panella 1980). was classified as either opaque or translucent relative to the Our data, in conjunction with those of Thompson and Beck- adjacent, previously formed, region. Plots of the percentage of man (1995) for adults, indicate that White Suckers in this pop- otoliths with an opaque zone at the growing edge of the otolith ulation form a single opaque annulus each year in otoliths, be- by month were used to determine the timing of formation for ginning in summer months after the spring hatching period, the first two annuli in this population. and continuing each summer throughout life. Knowing the time period of birth and annulus formation allows for more precise and accurate age estimates than would be possible if age esti- RESULTS AND DISCUSSION mates were based simply on the count of annuli—for example, White Sucker spawning was documented in Bee Creek be- a fish sampled in fall with two otolith annuli would be aged ginning in early April and continuing through mid-May, 1994. at 1.5 years; a fish sampled in spring with two annuli would Downloaded by [Department Of Fisheries] at 19:15 20 July 2015 Larvae were first collected in early May. Otolith daily increment be aged at 2 years. This study documents the timing of initial counts (Calfee 1995) verified that larvae were produced from annulus formation only for the sampled population, and timing these early April through mid-May spawning events. The first of annulus formation could vary according to geographic re- annulus was formed in young-of-the-year fish primarily from gion (Beckman and Wilson 1995). Although a similar pattern is July to September, at ages 2–5 months (Figure 1). The second likely for White Suckers in other portions of its range, ideally annulus was formed the following June–July, at an approximate the timing of annulus formation should be documented for all age of 13–16 months. A total of 335 young-of-the-year fish age-groups in any fish aging study that uses hard parts (Beamish were collected, with monthly sample sizes ranging from 16 to and McFarlane 1983). 49 individuals. A total of 136 age-1 fish were collected, yielding an average monthly sample size of 13.6 individuals. Mean total length of individuals was 67 mm and 155 mm at the end of year REFERENCES 1 and year 2, respectively, assuming a hatch date of 1 April. Beamish, R. J., and G. A. McFarlane. 1983. The forgotten requirement for The seasonality of annulus formation documented in this age validation in fisheries biology. Transactions of the American Fisheries Society 112:735–743. study for the first two annuli was similar to that observed by Beckman, D. W. 2002. Comparison of aging methods and validation of age Thompson and Beckman (1995) for adult White Suckers, mostly estimates for the Rainbow Darter, Etheostoma caeruleum. Copeia 2002:830– older than 4 years. It is not known why annulus formation was 1– 835. MANAGEMENT BRIEF 1189

Beckman, D. W., and D. T. Howlett. 2013. Otolith annulus formation and Clifford, H. F. 1972. Downstream movements of White Sucker, Catostomus growth of two redhorse suckers (Moxostoma: Catostomidae). Copeia 2013: commersoni, fry in a brown-water stream of Alberta. Journal of the Fisheries 390–395. Research Board of Canada 29:1091–1093. Beckman, D. W., and C. A. Hutson. 2012. Validation of aging techniques and Curry, K. D., and A. Spacie. 1984. Differential use of stream habitat by spawning growth of the River Redhorse, Moxostoma carinatum,intheJamesRiver, catostomids. American Midland Naturalist 111:267–279. Missouri. Southwestern Naturalist 57:240–247. Panella, G. 1980. Growth patterns in fish sagittae. Pages 519–560 in D. C. Beckman, D. W., A. L. Stanley, J. H. Render, and C. A. Wilson. 1991. Age Rhoads and R. A. Lutz, editors. Skeletal growth of aquatic organisms. Plenum, and growth-rate estimation of Sheepshead Archosargus probatocephalus in New York. Louisiana waters using otoliths. U.S. National Marine Fisheries Service Fish- Pflieger, W. L. 1975. The fishes of Missouri. Missouri Department of Conser- ery Bulletin 89:1–8. vation, Jefferson City. Beckman, D. W., and C. A. Wilson. 1995. Seasonal timing of opaque zone Quist, M. C., Z. J. Jackson, M. R. Bower, and W. A. Hubert. 2007. Precision formation in fish otoliths. Pages 27–44 in D. H. Secor, J. M. Dean, and S. E. of hard structures used to estimate age of riverine catostomids and cyprinids Campana, editors. Recent developments in fish otolith research. University in the upper Colorado River basin. North American Journal of Fisheries of South Carolina Press, Columbia. Management 27:643–649. Bettinger, J. M., and S. Crane. 2011. Validation of annulus formation in otoliths Simmons, B. R., and D. W. Beckman. 2012. Age determination, growth, and of Notchlip Redhorse (Moxostoma collapsum) and Brassy Jumprock (Mox- population structure of the Striped Shiner and Duskystripe Shiner. Transac- ostoma sp.) in Broad River, South Carolina, with observation on their growth tions of the American Fisheries Society 141:846–854. and mortality. Southeastern Naturalist 10:443–458. Sylvester, R. M., and C. R. Berry. 2006. Comparison of White Sucker age Brothers, E. B. 1979. Age and growth studies on tropical fishes. Pages 119– estimates from scales, pectoral fin rays, and otoliths. North American Journal 136 in S. B. Saila and P. M. Roedel, editors. Proceedings of an interna- of Fisheries Management 26:24–31. tional workshop: stock assessment of tropical small-scale fisheries. Univer- Thompson, K. R., and D. W. Beckman. 1995. Validation of age estimates sity of Rhode Island, International Center for Marine Resource Development, from White Sucker otoliths. Transactions of the American Fisheries Soci- Kingston. ety 124:637–639. Calfee, J. W. 1995. Early life history of the White Sucker (Catostomus com- Trippel, E. A., and H. H. Harvey. 1987. Abundance, growth, and food supply mersoni) in Lake Taneycomo, Missouri. Master’s thesis. Southwest Missouri of White Suckers (Catostomus commersoni) in relation to lake morphometry State University, Springfield. and pH. Canadian Journal of Zoology 65:558–564. Campana, S. E. 1999. Chemistry and composition of fish otoliths: path- Wakefield, C., and D. W.Beckman. 2005. Life History attributes of White Sucker ways, mechanisms and applications. Marine Ecology Progress Series 188: (Catostomus commersoni) in Lake Taneycomo and associated tributaries in 263–297. southwest Missouri. Southwestern Naturalist 50:423–434. Downloaded by [Department Of Fisheries] at 19:15 20 July 2015 This article was downloaded by: [Department Of Fisheries] On: 20 July 2015, At: 19:21 Publisher: Taylor & Francis Informa Ltd Registered in England and Wales Registered Number: 1072954 Registered office: 5 Howick Place, London, SW1P 1WG

North American Journal of Fisheries Management Publication details, including instructions for authors and subscription information: http://www.tandfonline.com/loi/ujfm20 Timing of First Annulus Formation in White Sucker Otoliths Daniel W. Beckmana & John W. Calfeea a Department of Biology, Missouri State University, 901 South National Avenue, Springfield, Missouri 65897, USA Published online: 14 Nov 2014.

Click for updates

PLEASE SCROLL DOWN FOR ARTICLE

Taylor & Francis makes every effort to ensure the accuracy of all the information (the “Content”) contained in the publications on our platform. However, Taylor & Francis, our agents, and our licensors make no representations or warranties whatsoever as to the accuracy, completeness, or suitability for any purpose of the Content. Any opinions and views expressed in this publication are the opinions and views of the authors, and are not the views of or endorsed by Taylor & Francis. The accuracy of the Content should not be relied upon and should be independently verified with primary sources of information. Taylor and Francis shall not be liable for any losses, actions, claims, proceedings, demands, costs, expenses, damages, and other liabilities whatsoever or howsoever caused arising directly or indirectly in connection with, in relation to or arising out of the use of the Content. This article may be used for research, teaching, and private study purposes. Any substantial or systematic reproduction, redistribution, reselling, loan, sub-licensing, systematic supply, or distribution in any form to anyone is expressly forbidden. Terms & Conditions of access and use can be found at http:// www.tandfonline.com/page/terms-and-conditions North American Journal of Fisheries Management 34:1187–1189, 2014 C American Fisheries Society 2014 ISSN: 0275-5947 print / 1548-8675 online DOI: 10.1080/02755947.2014.951809

MANAGEMENT BRIEF

Timing of First Annulus Formation in White Sucker Otoliths

Daniel W. Beckman* and John W. Calfee Department of Biology, Missouri State University, 901 South National Avenue, Springﬁeld, Missouri 65897, USA

population, unrecognized aging errors can affect age-based Abstract conservation and management plans (Beamish and McFarlane The purpose of this study was to document the time of forma- 1983). Yearly aging studies of Catostomus species based on tion for the first two otolith annuli for White Sucker Catostomus otoliths have included mostly mature fish older than age 2 years, commersoni. Young-of-the-year and age-1 individuals were sampled monthly from a population in the White River drainage of the and age validation studies for White Suckers have been done Missouri Ozarks. The formation of a single annual opaque region with populations dominated by fish of age 5 years and older during the months of June–September of the first 2 years of life (Thompson and Beckman 1995). For reliable growth modeling was documented using edge analysis, whereby the progression of and population assessment, it is especially important to doc- otolith annulus formation is followed throughout the year in the ument whether the first annulus is formed in age-0 or age-1 population. The documentation that the first annulus was formed within the first 6 months following birth provides for more accu- fish. For example, White Suckers spawn primarily during spring rate and precise age estimations than previously available for this months (Wakefield and Beckman 2005), and otolith annulus species. form primarily in late summer to fall months (Thompson and Beckman 1995); therefore, the first annulus could form at an White Sucker Catostomus commersoni is a widespread approximate age of either 0.5 or 1.5 years. The current study catostomid found throughout much of Canada and the northern was designed to document the time of the formation for the first United States east of the Rocky Mountains; coolwater streams two otolith annuli by following Missouri Ozarks White Sucker in southern Missouri are near the southern edge of its range populations in the wild from hatch through age 2 years. (Pflieger 1975). In southwest Missouri, White Suckers moves from White River reservoirs into tributaries to spawn over gravel substrates in clear, shallow stream waters, being an important METHODS component of these aquatic ecosystems (Wakefield and Beck- This study was carried out on Bee Creek, a tributary of Lake man 2005). Growth of White Suckers is highly variable through- Taneycomo, a reservoir on the White River of southwest Mis- out their range (Trippel and Harvey 1987), and age and growth souri. Visual observations of White Suckers were made from Downloaded by [Department Of Fisheries] at 19:21 20 July 2015 data can thus provide a valuable measure of environmental con- boat or shore in Bee Creek upstream of its confluence with Lake dition. Obtaining reliable growth data requires valid age estima- Taneycomo beginning 3 March 1994 and approximately weekly tion methods. thereafter. Spawning was documented by observing behaviors Otolith annulus counts have been documented as a reliable associated with reproduction, including aggregation in shallows age estimator for several species of suckers (Catostomidae) in and clearing of nesting areas (Curry and Spacie 1984). Spawn- the genus Catostomus (Sylvester and Berry 2006; Quist et al. ing was verified by collecting eggs from nesting areas follow- 2007; Bettinger and Crane 2011). Aging methods were vali- ing spawning. Following initial spawning activities, sweepnet dated for White Suckers by Thompson and Beckman (1995), sampling was carried out three times per week in pelagic near- using edge analysis (Beckman and Wilson 1995). Although surface waters of shallow backwater areas, where White Sucker such validations have shown the reliability of otolith aging for larvae tend to congregate following swim-up (Clifford 1972). adult fish, immature fish have often been excluded from studies White Suckers were sampled monthly from May 1994 to because sample collection methods typically exclude smaller April 1995, except in December and January, in areas of fish. Unless validation is carried out for all age-groups in a Bee Creek where spawning was documented. Individuals were

*Corresponding author: [email protected] Received May 14, 2014; accepted July 23, 2014 1187 1188 BECKMAN AND CALFEE

An observer at the front of the boat netted all White Suckers Edge with Opaque Otolith Percent 10 0 observed to be affected by the current. May June July Aug Sep Oct Nov Dec Jan Feb Mar Apr The two lapilli otoliths were removed from each fish, one of 38 36 47 28 26 40 20 35 49 16 which was processed for aging by embedding in an epoxy resin. Month The embedded otolith was sectioned in the transverse plane by FIGURE 1. Percent of individuals with opaque zones at the growing edge of using a Struers Minitom low-speed saw with a diamond-tipped otoliths by month of capture for White Suckers in Bee Creek, White River system cutoff wheel (Thompson and Beckman 1995). Sections were of Missouri, as young-of-the-year (n = 335) and age 1 (n = 136). Numbers below sanded to a thickness of 0.50–0.75 mm and polished on a felt months indicate sample sizes for young-of-the-year individuals. µ pad using 0.3- m alumina suspension polishing compound. 2 months later in young-of-the-year fish than in age-1 fish (Fig- Daily growth increments were used to age larval fish up ure 1). These studies, along with other aging studies of Ozarks to 25 d posthatch (Calfee 1995). For juveniles, otolith trans- fishes (Beckman 2002; Beckman and Hutson 2012; Simmons verse sections were assessed using a compound microscope and Beckman 2012; Beckman and Howlett 2013), suggest that × with transmitted light at 40 magnification to identify concen- the predominant time of formation of the otolith opaque growth tric opaque (dark) and translucent (light) regions. Opaque re- zone (annulus) falls within the months of May–September for gions were counted as annuli (Thompson and Beckman 1995). stream fish species in this geographic region. This supports the Otolith edges were ranked using marginal increment analysis argument that otolith annulus formation is seasonal but not di- to assess the timing of annulus formation, plotting the percent- rectly related to spawning (Beckman and Wilson 1995); more age of individual fish exhibiting an opaque zone at the growing likely, formation is a result of temperature effects on otolith edge of the otolith throughout the year (Beckman et al. 1991). composition (Campana 1999) or reflects a seasonal increase in The outer growing edge of one otolith section from each fish fish growth or metabolism (Brothers 1979; Panella 1980). was classified as either opaque or translucent relative to the Our data, in conjunction with those of Thompson and Beck- adjacent, previously formed, region. Plots of the percentage of man (1995) for adults, indicate that White Suckers in this pop- otoliths with an opaque zone at the growing edge of the otolith ulation form a single opaque annulus each year in otoliths, be- by month were used to determine the timing of formation for ginning in summer months after the spring hatching period, the first two annuli in this population. and continuing each summer throughout life. Knowing the time period of birth and annulus formation allows for more precise and accurate age estimates than would be possible if age esti- RESULTS AND DISCUSSION mates were based simply on the count of annuli—for example, White Sucker spawning was documented in Bee Creek be- a fish sampled in fall with two otolith annuli would be aged ginning in early April and continuing through mid-May, 1994. at 1.5 years; a fish sampled in spring with two annuli would Downloaded by [Department Of Fisheries] at 19:21 20 July 2015 Larvae were first collected in early May. Otolith daily increment be aged at 2 years. This study documents the timing of initial counts (Calfee 1995) verified that larvae were produced from annulus formation only for the sampled population, and timing these early April through mid-May spawning events. The first of annulus formation could vary according to geographic re- annulus was formed in young-of-the-year fish primarily from gion (Beckman and Wilson 1995). Although a similar pattern is July to September, at ages 2–5 months (Figure 1). The second likely for White Suckers in other portions of its range, ideally annulus was formed the following June–July, at an approximate the timing of annulus formation should be documented for all age of 13–16 months. A total of 335 young-of-the-year fish age-groups in any fish aging study that uses hard parts (Beamish were collected, with monthly sample sizes ranging from 16 to and McFarlane 1983). 49 individuals. A total of 136 age-1 fish were collected, yielding an average monthly sample size of 13.6 individuals. Mean total length of individuals was 67 mm and 155 mm at the end of year REFERENCES 1 and year 2, respectively, assuming a hatch date of 1 April. Beamish, R. J., and G. A. McFarlane. 1983. The forgotten requirement for The seasonality of annulus formation documented in this age validation in fisheries biology. Transactions of the American Fisheries Society 112:735–743. study for the first two annuli was similar to that observed by Beckman, D. W. 2002. Comparison of aging methods and validation of age Thompson and Beckman (1995) for adult White Suckers, mostly estimates for the Rainbow Darter, Etheostoma caeruleum. Copeia 2002:830– older than 4 years. It is not known why annulus formation was 1– 835. MANAGEMENT BRIEF 1189

Beckman, D. W., and D. T. Howlett. 2013. Otolith annulus formation and Clifford, H. F. 1972. Downstream movements of White Sucker, Catostomus growth of two redhorse suckers (Moxostoma: Catostomidae). Copeia 2013: commersoni, fry in a brown-water stream of Alberta. Journal of the Fisheries 390–395. Research Board of Canada 29:1091–1093. Beckman, D. W., and C. A. Hutson. 2012. Validation of aging techniques and Curry, K. D., and A. Spacie. 1984. Differential use of stream habitat by spawning growth of the River Redhorse, Moxostoma carinatum,intheJamesRiver, catostomids. American Midland Naturalist 111:267–279. Missouri. Southwestern Naturalist 57:240–247. Panella, G. 1980. Growth patterns in fish sagittae. Pages 519–560 in D. C. Beckman, D. W., A. L. Stanley, J. H. Render, and C. A. Wilson. 1991. Age Rhoads and R. A. Lutz, editors. Skeletal growth of aquatic organisms. Plenum, and growth-rate estimation of Sheepshead Archosargus probatocephalus in New York. Louisiana waters using otoliths. U.S. National Marine Fisheries Service Fish- Pflieger, W. L. 1975. The fishes of Missouri. Missouri Department of Conser- ery Bulletin 89:1–8. vation, Jefferson City. Beckman, D. W., and C. A. Wilson. 1995. Seasonal timing of opaque zone Quist, M. C., Z. J. Jackson, M. R. Bower, and W. A. Hubert. 2007. Precision formation in fish otoliths. Pages 27–44 in D. H. Secor, J. M. Dean, and S. E. of hard structures used to estimate age of riverine catostomids and cyprinids Campana, editors. Recent developments in fish otolith research. University in the upper Colorado River basin. North American Journal of Fisheries of South Carolina Press, Columbia. Management 27:643–649. Bettinger, J. M., and S. Crane. 2011. Validation of annulus formation in otoliths Simmons, B. R., and D. W. Beckman. 2012. Age determination, growth, and of Notchlip Redhorse (Moxostoma collapsum) and Brassy Jumprock (Mox- population structure of the Striped Shiner and Duskystripe Shiner. Transac- ostoma sp.) in Broad River, South Carolina, with observation on their growth tions of the American Fisheries Society 141:846–854. and mortality. Southeastern Naturalist 10:443–458. Sylvester, R. M., and C. R. Berry. 2006. Comparison of White Sucker age Brothers, E. B. 1979. Age and growth studies on tropical fishes. Pages 119– estimates from scales, pectoral fin rays, and otoliths. North American Journal 136 in S. B. Saila and P. M. Roedel, editors. Proceedings of an interna- of Fisheries Management 26:24–31. tional workshop: stock assessment of tropical small-scale fisheries. Univer- Thompson, K. R., and D. W. Beckman. 1995. Validation of age estimates sity of Rhode Island, International Center for Marine Resource Development, from White Sucker otoliths. Transactions of the American Fisheries Soci- Kingston. ety 124:637–639. Calfee, J. W. 1995. Early life history of the White Sucker (Catostomus com- Trippel, E. A., and H. H. Harvey. 1987. Abundance, growth, and food supply mersoni) in Lake Taneycomo, Missouri. Master’s thesis. Southwest Missouri of White Suckers (Catostomus commersoni) in relation to lake morphometry State University, Springfield. and pH. Canadian Journal of Zoology 65:558–564. Campana, S. E. 1999. Chemistry and composition of fish otoliths: path- Wakefield, C., and D. W.Beckman. 2005. Life History attributes of White Sucker ways, mechanisms and applications. Marine Ecology Progress Series 188: (Catostomus commersoni) in Lake Taneycomo and associated tributaries in 263–297. southwest Missouri. Southwestern Naturalist 50:423–434. Downloaded by [Department Of Fisheries] at 19:21 20 July 2015 This article was downloaded by: [Department Of Fisheries] On: 20 July 2015, At: 19:15 Publisher: Taylor & Francis Informa Ltd Registered in England and Wales Registered Number: 1072954 Registered office: 5 Howick Place, London, SW1P 1WG

North American Journal of Fisheries Management Publication details, including instructions for authors and subscription information: http://www.tandfonline.com/loi/ujfm20 Effects of Appendaged Circle Hook Use on Catch Rates and Deep Hooking of Black Sea Bass in a Recreational Fishery Charles Bergmanna, William B. Driggers IIIa, Eric R. Hoffmayera, Matthew D. Campbella & Gilmore Pellegrina a National Marine Fisheries Service, Southeast Fisheries Science Center, Mississippi Laboratories, Post Office Drawer 1207, Pascagoula, Mississippi 39567, USA Published online: 19 Nov 2014.

Click for updates

To cite this article: Charles Bergmann, William B. Driggers III, Eric R. Hoffmayer, Matthew D. Campbell & Gilmore Pellegrin (2014) Effects of Appendaged Circle Hook Use on Catch Rates and Deep Hooking of Black Sea Bass in a Recreational Fishery, North American Journal of Fisheries Management, 34:6, 1199-1203, DOI: 10.1080/02755947.2014.956160 To link to this article: http://dx.doi.org/10.1080/02755947.2014.956160

PLEASE SCROLL DOWN FOR ARTICLE

Taylor & Francis makes every effort to ensure the accuracy of all the information (the “Content”) contained in the publications on our platform. However, Taylor & Francis, our agents, and our licensors make no representations or warranties whatsoever as to the accuracy, completeness, or suitability for any purpose of the Content. Any opinions and views expressed in this publication are the opinions and views of the authors, and are not the views of or endorsed by Taylor & Francis. The accuracy of the Content should not be relied upon and should be independently verified with primary sources of information. Taylor and Francis shall not be liable for any losses, actions, claims, proceedings, demands, costs, expenses, damages, and other liabilities whatsoever or howsoever caused arising directly or indirectly in connection with, in relation to or arising out of the use of the Content. This article may be used for research, teaching, and private study purposes. Any substantial or systematic reproduction, redistribution, reselling, loan, sub-licensing, systematic supply, or distribution in any form to anyone is expressly forbidden. Terms & Conditions of access and use can be found at http:// www.tandfonline.com/page/terms-and-conditions North American Journal of Fisheries Management 34:1199–1203, 2014 American Fisheries Society 2014 ISSN: 0275-5947 print / 1548-8675 online DOI: 10.1080/02755947.2014.956160

MANAGEMENT BRIEF

Effects of Appendaged Circle Hook Use on Catch Rates and Deep Hooking of Black Sea Bass in a Recreational Fishery

Charles Bergmann, William B. Driggers III,* Eric R. Hoffmayer, Matthew D. Campbell, and Gilmore Pellegrin National Marine Fisheries Service, Southeast Fisheries Science Center, Mississippi Laboratories, Post Ofﬁce Drawer 1207, Pascagoula, Mississippi 39567, USA

hooks across a number of fish taxa can reduce deep hooking— Abstract defined as internal hooking in a location other than the jaw Previous studies have demonstrated that circle hooks modified (e.g., eye, gill, esophagus, and stomach)—thus resulting in with an appendage can reduce the occurrence of deep hooking in some fishes. To determine whether this modification affects total fewer injuries to vital organs (e.g., Cooke and Suski 2004; catch and deep-hooking rate of Black Sea Bass Centropristis striata Sauls and Ayala 2012). in a recreational fishery off the coast of Florida, anglers were pro- A relatively recent innovation in design can reduce deep- vided with standardized gear that included one appendaged and hooking rates of circle hooks. The modified circle hook, here- one nonappendaged hook. Both hook types were fished equally, after referred to as an appendaged hook, incorporates a steel and a greater number of Black Sea Bass were caught on appen- n D n D wire extension (35 mm long, 1 mm wide) that extends at an daged hooks ( 301) than on nonappendaged hooks ( 221). There was no significant difference in the mean TL (P D 0.80) of approximately 45 downward angle from the eye (Figure 1). jaw- or deep-hooked fish during the study. Logistic regression was The extension creates a potential impediment to passage of the applied to the data using anatomical hooking location as the depen- hook from the mouth into the alimentary canal of a captured dent variable and angler, hook position, and hook type as categori- fish while negligibly affecting the hook’s size and weight. Wil- cal factors. Hook type was the only significant variable in the final model (P D 0.03) in which fewer fish were deep hooked on appen- lis and Millar (2001) reported that deep hooking of Squirefish daged hooks (0.96%) than on nonappendaged hooks (2.11%). Our Pagrus auratus (also known as Chrysophrys auratus) caught findings indicate appendaged hooks are capable of reducing deep on longline gear was significantly reduced when using appen- hooking of Black Sea Bass without reducing catch rates or altering daged hooks in lieu of standard circle hooks. However, Willis size composition of the catch. and Millar (2001) also found that catch rates of Squirefish were significantly lower with appendaged hooks. While lower- ing mortality rates resulting from deep hooking has obvious

Downloaded by [Department Of Fisheries] at 19:15 20 July 2015 Fishes within the snapper–grouper complex are among the conservation benefits, lower catch rates with a specific hook most sought after by recreational anglers off the southeastern type could result in recreational anglers being resistant to vol- coast of the United States (Coleman et al. 2000). Despite man- untarily using that hook (Cooke et al. 2012). agement measures designed to minimize the retention of Black Sea Bass Centropristis striata constitute a significant undersized fish, size limits alone could be inadequate to main- portion of the catch in the recreational fishery operating in the tain a stock at optimal levels due to high postrelease mortality western North Atlantic Ocean (NMFS 2014). Off the east rates (Rudershausen et al. 2007). While a number of factors, coast of the United States, this species ranges from Massachu- including capture depth, water temperature, handling time, setts to the Florida Keys, where until recently it has been con- fight duration, and bait type can affect postrelease survival, sidered overfished (SEDAR 2011). The recovery of Black Sea anatomical location of hooking is among the most important Bass in the region has been attributed to strict management of these factors (e.g., Bartholomew and Bohnsack 2005; measures, including catch quotas, bag limits, time and area Mapleston et al. 2008). When compared with J hooks, circle closures, size limits (currently 33 cm TL), and gear

*Corresponding author: [email protected] Received February 3, 2014; accepted August 14, 2014

1199 1200 BERGMANN ET AL.

FIGURE 1. Nonappendaged (10/0 Mustad EZ Baiter, model 39977) and appendaged (10/0 Mustad appendaged, modelAP39960D) circle hooks used to study catch rates and deep hooking of Black Sea Bass. Scale bar represents 20 mm.

restrictions (SEDAR 2011; NMFS 2014). For example, in 2010, a regulation was implemented requiring nonstainless steel circle hooks to be used when fishing with natural baits for fishes in the snapper–grouper complex. This regulation FIGURE 2. Sampling areas (black circles) for study of catch rates and deep specifically applied to federal waters of the western North hooking of Black Sea Bass off the eastern coast of Florida. Atlantic Ocean north of 28 latitude from Florida to North Carolina and did not prohibit the use of offset hooks (USOFR viable management alternative to reduce fishing mortality 2010). Data collected off the east coast of northern Florida by rates in the Black Sea Bass recreational fishery. observers on headboats (defined as vessels able to carry 10 or more paying passengers) indicate that anglers largely comply with this regulation and commonly use offset circle hooks (B. METHODS Sauls, Florida Fish and Wildlife Conservation Commission, In an effort to mimic conditions encountered within the personal communication). However, differences in circle hook Black Sea Bass recreational fishery operating in the southeast- designs, such as an offset point, can minimize the benefits of ern United States, sampling occurred on two headboats: the circle hook use (Cooke et al. 2012; Yokota et al. 2012). For Mayport Princess, of Mayport, Florida, and the Ocean Obses- example, Mapleston et al. (2008) reported that for eight spe- sion, of Cape Canaveral, Florida (Figure 2). Fifteen fishing cies of reef fishes, there was no significant difference in catch days were conducted on the vessels off the east coast of Flor- rates when offset and nonoffset circle hooks were compared; ida between June 13 and August 22, 2011. During each fishing

Downloaded by [Department Of Fisheries] at 19:15 20 July 2015 however, the incidence of deep hooking was greater when off- day, two experienced anglers, as identified by their familiarity set hooks were used for some species. to the crew, were selected to participate in the study. Each Numerous studies have demonstrated the relationships angler was provided a standardized rod (Penn SL2050C66T among hook design, anatomical hooking location, and postre- Slammer), reel (Penn 113HZ Senator), and tackle. The main- lease mortality of fishes (see Bartholomew and Bohnsack line consisted of 50-lb-test monofilament line (Momoi Hi- 2005 for a recent review). For example, Bugley and Shepherd Catch). Terminal tackle was constructed of one swivel (Rosco (1991) examined mortality rates of Black Sea Bass associated 6/0), monofilament leader material (89-lb-test Momoi Hi- with catch-and-release angling practices and determined that Catch), one weight (8–16-oz bank sinker), one 10/0 EZ Baiter hooking mortality could be as high as 7.8%. Those authors 10 offset circle hook (Mustad model 39977, hereafter referred also stated that all fish that died during their experiment were to as a nonappendaged hook), and one 10/0 nonoffset appen- deep hooked. The purpose of the current study was to compare daged circle hook (Mustad model number AP39960D) (Fig- anatomical hooking locations and catch rates of Black Sea ures 1, 3) (following USOFR 2010), we defined a circle hook Bass captured on appendaged hooks versus those of fish cap- as a “hook designed and manufactured so that the point is tured on commonly used hooks in the headboat fishery off the turned perpendicularly back to the shank to form a generally east coast of Florida. Furthermore, it was our intention to circular, or oval, shape”). The EZ Baiter hook, which meets determine whether the use of appendaged circle hooks is a the definition of a circle hook as defined above, is considered MANAGEMENT BRIEF 1201

Sea Bass that were captured and remained hooked were measured to the nearest millimeter TL by a trained observer. Fish that escaped a hook at the surface were not considered captured. Additionally, the anatomical location of hooking (e.g., jaw, eye, gill, esophagus, and stomach) and angler associated with each capture were noted. We used F-tests to determine whether the ratio of the standard deviations of mean TL from the two sample distributions was different from a value of 1 for both hook type and anatomical hooking location. Kolmogorov–Smirnov (K- S) tests were used to determine whether there was a significant difference in the sample distributions. If the assumptions of parametric statistics were met, a two-sided t-test was used to determine whether there was a significant difference in mean TL of fish caught on each hook type. If assumptions were violated, a Mann–Whitney–Wilcoxon W- test was performed to compare the median TL of the two samples. A chi-square test was used to determine whether there was a significant difference in the total catch of Black Sea Bass on appendaged and nonappendaged hooks. After categorizing the anatomical location of hooking as jaw or deep hooked (aggregate of eye-, gill-, esophagus-, or stomach-hooked fishes), a logistic regression with a backward selection procedure was used to examine relationships between the anatomical location of hooking and the explanatory variables: hook type, hook position (top or bottom), and angler. If a relationship between the dependent variable and any of the explanatory variables existed, chi-square tests of independence were utilized to further examine the relationship between the dependent and explanatory variables. Addition- ally, the odds ratio of each significant variable when compared to anatomical hooking location was calculated as [(n jaw hoo- kedA)/(n deep hookedA)] / [(n jaw hookedN)/(n deep hookedN)], where A and n represent appendaged and nonappendaged hook, respectively. All tests were conducted using Statgraphics Centurion XVI version 16.1.18 (StatPoint Tech- nologies) and were considered significant at a D 0.05. FIGURE 3. Schematic of terminal tackle used during the study of catch rates

Downloaded by [Department Of Fisheries] at 19:15 20 July 2015 and deep hooking of Black Sea Bass. RESULTS to make baiting easier and decrease difficulties associated with Over the course of the study, 522 Black Sea Bass were cap- dehooking captured fish, especially for inexperienced recrea- tured by the 10 participating anglers. Hook position had no tional anglers. As a result, the EZ Baiter hook is widely used effect on Black Sea Bass catch as an equal number were within the headboat fishery operating in the sampling area (G. caught on upper- (n D 261) and lower-positioned (n D 261) Strate, headboat operator, personal communication; E. Sander hooks. Length data for Black Sea Bass captured on both hook and B. Sauls, Florida Fish and Wildlife Conservation Commis- types were normally distributed and there was no significant sion, personal communication). The position (top or bottom) difference in the distributions (K-S statistic D 0.05, P D 0.84) of the nonappendaged and appendaged hooks was randomly or standard deviations (F D 0.89, P D 0.34) by hook type (Fig- assigned for each rig. Rigs were changed daily or whenever ure 4). The mean lengths of Black Sea Bass caught on each damage was incurred, and an equal number of both hooks hook type (nonappendaged hook D 268.8 mm, SD D 39.0; were fished during the study. All participating anglers used a appendaged hook D 269.7 mm, SD D 41.5; t D¡0.25, P D single bait type for each deployment. Bait was cut to an appro- 0.80) were not significantly different and both were less than priate size for a 10/0 hook and analyses were limited to those the current minimum size limit. There was a significant differ- Black Sea Bass captured with squid (Loligo sp.). All Black ence in the number of Black Sea Bass caught on each hook 1202 BERGMANN ET AL.

percentage of Black Sea Bass were hooked in the eye (25.00% [n D 4] versus 12.50% [n D 2]) when appendaged hooks were used (Table 1).

DISCUSSION The widespread acceptance of circle hook use in recreational fisheries can be attributed to the general positive attitude of anglers toward the utility of these hooks in assisting with conservation efforts (Cooke et al. 2012). However, Cooke and Suski (2004) stated that for circle hooks to be accepted by anglers the hooks must be equally or more efficient than J hooks in capturing targeted species. Clearly, the same logic would apply to the use of appendaged circle hooks when compared with other frequently used circle hooks. The results of the current study demonstrated that appendaged circle hooks were more efficient than the commonly used nonappendaged hook at capturing Black Sea Bass. In addition, since the length frequency distribution and mean size of Black Sea Bass caught during this study were not different between hook types, it is evident that appendaged and nonappendaged hooks sampled the same size composition from the fish population. It should be noted that our findings with regard to catch rate conflict with those of Willis and Millar (2001) and Swimmer et al. (2011), FIGURE 4. Length frequency distribution of Black Sea Bass (a) captured on the only studies we are aware of that have examined use of nonappendaged (white bars) and appendaged (dark bars) hooks and (b) jaw appendaged hooks. Both studies reported lower catch rates of (dark bars) and deep (white bars) hooked. Length data are displayed in 20-mm teleosts when using appendaged circle hooks. However, Willis size bins. and Millar (2001) reported that catch of small Squirefish was lower when using appendaged hooks, but the use of appendaged type (x2 D 12.26, P < 0.01) and catch was highest for Black hooks did not affect the catch rates of larger individuals. Sea Bass caught on appendaged hooks (n D 301 versus 221). The use of circle hooks in lieu of J hooks can reduce deep Among all Black Sea Bass captured, 2.11% and 0.96% were hooking and, ultimately, postrelease mortality of numerous deep hooked when using the nonappendaged hooks and appen- fishes (e.g., Cooke and Suski 2004; Bartholomew and Bohn- daged hooks, respectively. There was no significant difference sack 2005; Serafy et al. 2009; Godin et al. 2012). However, in the length distribution of Black Sea Bass that were jaw or while deep hooking has been reduced in many species through deep hooked (K-S statistic D 0.29, P D 0.12; Figure 4). Simi- the use of circle hooks, it has not been eliminated and continues larly, there was no significant difference in median TL of indi- to occur, albeit at a lower rate. We demonstrated that the use of viduals either jaw or deep hooked (W D 3651.00, P D 0.50).

Downloaded by [Department Of Fisheries] at 19:15 20 July 2015 Among the explanatory variables used in the logistic TABLE 1. Summary of number of captures, TL, and anatomical hooking regression, only hook type was retained after the backward location data for Black Sea Bass caught on nonappendaged and appendaged selection procedure was applied and was significantly related hooks. to anatomical hooking location (likelihood ratio test: x2 D 4.68, P D 0.03). The chi-square test for independence demon- Nonappendaged Appendaged strated that the hook type and anatomical hooking location Variable hook hook x2 D D D variables were not independent ( 4.71, df 1, P 0.03). Number of fish caught 221 301 The odds ratio of a Black Sea Bass being hooked in the jaw Size range (mm TL) 170–389 119–404 D was 3.10 (95% CI 1.06–9.06) times greater when the appen- Mean size (mm TL) (SD) 268.8 (39.0) 270.0 (41.5) daged hook was used than when the nonappendaged hook was Number jaw hooked 210 296 D used. Of the Black Sea Bass that were deep hooked (n 16), a Number deep hooked 11 5 higher percentage of fish caught on nonappendaged hooks Eye 2 4 compared with those caught on appendaged hooks were Gill 1 0 D D hooked in the gills (6.25% [n 1] versus 0.00% [n 0]), Esophagus 7 1 D D esophagus (43.75% [n 7] versus 6.25% [n 1]), and stom- Stomach 1 0 ach (6.25% [n D 1] versus 0% [n D 0]); however, a higher MANAGEMENT BRIEF 1203

appendaged circle hooks can reduce the incidence of deep Bacheler, N. M., and J. A. Buckel. 2004. Does hook type influence the catch hooking in Black Sea Bass, a finding consistent with Willis and rate, size, and injury of grouper in a North Carolina commercial fishery? Millar (2001) who indicated appendaged circle hooks Fisheries Research 69:303–311. Bartholomew, A., and J. A. Bohnsack. 2005. A review of catch-and-release decreased deep hooking of Squirefish from up to 30% of fish angling mortality with implications for no-take reserves. Reviews in Fish deep hooked using standard circle hooks to 2% of fish captured Biology and Fisheries 15:129–154. using an appendaged hook. While our results demonstrated a Bugley, K., and G. Shepherd. 1991. Effect of catch-and-release angling on the reduction in deep hooking when appendaged hooks were used, survival of Black Sea Bass. North American Journal of Fisheries Manage- the deep-hooking rate of Black Sea Bass on nonappendaged cir- ment 11:468–471. Coleman, F. C., C. C. Koenig, G. R. Huntsman, J. A. Musick, A. M. Eklund, cle hooks was relatively low (2.1%) and similar to rates (1.5– J.C.McGovern,R.W.Chapman,G.R.Sedberry,andC.B.Grimes. 9.5%) reported for other species caught within recreational fish- 2000. Long-lived reef fishes: the snapper–grouper complex. Fisheries 25 eries off the eastern coast of the United States (Bacheler and (3):14–21. Buckel 2004). Furthermore, several authors have demonstrated Cooke, S. J., V. M. Nguyen, K. J. Murchie, A. J. Danylchuk, and C. D. Suski. that deep-hooking rates can be species-specific (e.g., Cooke 2012. Scientific and stakeholder perspectives on the use of circle hooks in recreational fisheries. Bulletin of Marine Science 88:395–410. and Suski 2004) and can be positively related to fish length for Cooke, S. J., and C. D. Suski. 2004. Are circle hooks an effective tool for some species (e.g., Alos et al. 2009). Therefore, future studies conserving marine and freshwater recreational catch-and-release fisher- will need to ascertain whether a reduction in deep hooking asso- ies? Aquatic Conservation: Marine and Freshwater Ecosystems 14:299– ciated with appendaged hooks occurs for other fishes. 326. We compared an offset nonappendaged hook to a nonoffset Godin, A. C., J. K. Carlson, and V. Burgener. 2012. The effect of circle hooks on shark catchability and at-vessel mortality rates in longline fisheries. Bul- appendaged hook and cannot discount that the use of these letin of Marine Science 88:469–483. two hook types confounds deep-hooking effects of offset ver- Mapleston, A., D. Welch, G. A. Begg, M. McLennan, D. Mayer, and I. Brown. sus nonoffset hooks. Furthermore, the difference in shank 2008. Effect of changes in hook pattern and size on catch rate, hooking loca- length between the two hooks could have had an effect that we tion, injury and bleeding for a number of tropical reef fish species. Fisheries were not able to account for. We acknowledge considerable Research 91:203–211. NMFS (National Marine Fisheries Service). 2014. Recreational fisheries statis- further research is needed to examine the significance of these tics. Available: http://www.st.nmfs.noaa.gov/recreational-fisheries/access- potential interactions and, ideally, future investigations should data/run-a-data-query/queries/index. (January 2014). use identical hooks, which would isolate the effect of the Rudershausen, P. J., J. A. Buckel, and E. H. Williams. 2007. Discard composi- appendage. However, the findings of our study indicate that tion and release fate in the snapper and grouper commercial hook-and-line use of appendaged hooks could be a viable mechanism to fur- fishery in North Carolina, USA. Fisheries Management and Ecology 14:103–113. ther minimize mortality rates associated with deep hooking of Sauls, B., and O. Ayala. 2012. Circle hook requirements in the Gulf of Mexico: Black Sea Bass in recreational fisheries and suggest that these application in recreational fisheries and effectiveness for conservation of hooks can have similar positive attributes within other hook- reef fishes. Bulletin of Marine Science 88:667–679. based fisheries. Additionally, this result could be achieved SEDAR (Southeast Data Assessment Review). 2011. South Atlantic Black without reducing catch rates or altering size composition of Sea Bass. SEDAR 25, Stock Assessment Report, North Charleston, South Carolina. Available: www.sefsc.noaa.gov/sedar/download/ the Black Sea Bass catch. SEDAR25_BlackSeaBass_SAR.pdf?id=DOCUMENT. (December 2013). Serafy, J. E., D. W. Kerstetter, and P. H. Rice. 2009. Can circle hook use benefit billfishes? Fish and Fisheries 10:132–142. ACKNOWLEDGMENTS Swimmer, Y., J. Suter, R. Arauz, K. Bigelow, A. Lopez, I. Zanela, A. Bolanos,~ We thank the crews of the Mayport Princess and Ocean J. Ballestero, R. Suarez, J. Wang, and C. Boggs. 2011. Sustainable fishing gear: the case of modified circle hooks in a Costa Rican longline fishery. Downloaded by [Department Of Fisheries] at 19:15 20 July 2015 Obsession, including Becky Hogan, Captain George Strate, Marine Biology 158:757–767. Sandy Kelly, Captain Butch Kelly, and Captain Shawn Lind- USOFR (U.S. Office of the Federal Register). 2010. Fisheries of the say, as well as Kevin Rademacher and the volunteer anglers, Caribbean, Gulf of Mexico, and South Atlantic: snapper–grouper fishery including Leonard “Wild Bill” Reed, Greg Williams, Charlie off the southern Atlantic states; amendment 17A; emergency rule to Davis, and Lenny Zurick, for assisting in data collection. Men- delay effectiveness of the snapper-grouper area closure, final rule and temporary rule (50 CFR Part 622). Federal Register 75:236(9 December tion of specific vendors or products does not constitute 2010):76874–76890. endorsement by the National Marine Fisheries Service. Willis, T. J., and R. B. Millar. 2001. Modified hooks reduce incidental mortality of snapper (Pagrus auratus: Sparidae) in the New Zealand commercial longline fishery. ICES Journal of Marine Science 58:830–841. REFERENCES Yokota, K., T. Mituhasi, H. Minami, and M. Kiyota. 2012. Perspec- tives on the morphological elements of circle hooks and their per- Alos, J., G. Mateu-Vicens, M. Palmer, A. M. Grau, M. Cabanellas-Reboredo, formance in pelagic longline fisheries. Bulletin of Marine Science and A. Box. 2009. Performance of circle hooks in a mixed-species recrea- 88:623–629. tional fishery. Journal of Applied Ichthyology 25:565–570. This article was downloaded by: [Department Of Fisheries] On: 20 July 2015, At: 19:22 Publisher: Taylor & Francis Informa Ltd Registered in England and Wales Registered Number: 1072954 Registered office: 5 Howick Place, London, SW1P 1WG

North American Journal of Fisheries Management Publication details, including instructions for authors and subscription information: http://www.tandfonline.com/loi/ujfm20 Effects of Appendaged Circle Hook Use on Catch Rates and Deep Hooking of Black Sea Bass in a Recreational Fishery Charles Bergmanna, William B. Driggers IIIa, Eric R. Hoffmayera, Matthew D. Campbella & Gilmore Pellegrina a National Marine Fisheries Service, Southeast Fisheries Science Center, Mississippi Laboratories, Post Office Drawer 1207, Pascagoula, Mississippi 39567, USA Published online: 19 Nov 2014.

Click for updates

PLEASE SCROLL DOWN FOR ARTICLE

Taylor & Francis makes every effort to ensure the accuracy of all the information (the “Content”) contained in the publications on our platform. However, Taylor & Francis, our agents, and our licensors make no representations or warranties whatsoever as to the accuracy, completeness, or suitability for any purpose of the Content. Any opinions and views expressed in this publication are the opinions and views of the authors, and are not the views of or endorsed by Taylor & Francis. The accuracy of the Content should not be relied upon and should be independently verified with primary sources of information. Taylor and Francis shall not be liable for any losses, actions, claims, proceedings, demands, costs, expenses, damages, and other liabilities whatsoever or howsoever caused arising directly or indirectly in connection with, in relation to or arising out of the use of the Content. This article may be used for research, teaching, and private study purposes. Any substantial or systematic reproduction, redistribution, reselling, loan, sub-licensing, systematic supply, or distribution in any form to anyone is expressly forbidden. Terms & Conditions of access and use can be found at http:// www.tandfonline.com/page/terms-and-conditions North American Journal of Fisheries Management 34:1199–1203, 2014 American Fisheries Society 2014 ISSN: 0275-5947 print / 1548-8675 online DOI: 10.1080/02755947.2014.956160

MANAGEMENT BRIEF

Effects of Appendaged Circle Hook Use on Catch Rates and Deep Hooking of Black Sea Bass in a Recreational Fishery

Downloaded by [Department Of Fisheries] at 19:22 20 July 2015 Fishes within the snapper–grouper complex are among the conservation benefits, lower catch rates with a specific hook most sought after by recreational anglers off the southeastern type could result in recreational anglers being resistant to vol- coast of the United States (Coleman et al. 2000). Despite man- untarily using that hook (Cooke et al. 2012). agement measures designed to minimize the retention of Black Sea Bass Centropristis striata constitute a significant undersized fish, size limits alone could be inadequate to main- portion of the catch in the recreational fishery operating in the tain a stock at optimal levels due to high postrelease mortality western North Atlantic Ocean (NMFS 2014). Off the east rates (Rudershausen et al. 2007). While a number of factors, coast of the United States, this species ranges from Massachu- including capture depth, water temperature, handling time, setts to the Florida Keys, where until recently it has been con- fight duration, and bait type can affect postrelease survival, sidered overfished (SEDAR 2011). The recovery of Black Sea anatomical location of hooking is among the most important Bass in the region has been attributed to strict management of these factors (e.g., Bartholomew and Bohnsack 2005; measures, including catch quotas, bag limits, time and area Mapleston et al. 2008). When compared with J hooks, circle closures, size limits (currently 33 cm TL), and gear

*Corresponding author: [email protected] Received February 3, 2014; accepted August 14, 2014

1199 1200 BERGMANN ET AL.

Downloaded by [Department Of Fisheries] at 19:22 20 July 2015 however, the incidence of deep hooking was greater when off- day, two experienced anglers, as identified by their familiarity set hooks were used for some species. to the crew, were selected to participate in the study. Each Numerous studies have demonstrated the relationships angler was provided a standardized rod (Penn SL2050C66T among hook design, anatomical hooking location, and postre- Slammer), reel (Penn 113HZ Senator), and tackle. The main- lease mortality of fishes (see Bartholomew and Bohnsack line consisted of 50-lb-test monofilament line (Momoi Hi- 2005 for a recent review). For example, Bugley and Shepherd Catch). Terminal tackle was constructed of one swivel (Rosco (1991) examined mortality rates of Black Sea Bass associated 6/0), monofilament leader material (89-lb-test Momoi Hi- with catch-and-release angling practices and determined that Catch), one weight (8–16-oz bank sinker), one 10/0 EZ Baiter hooking mortality could be as high as 7.8%. Those authors 10 offset circle hook (Mustad model 39977, hereafter referred also stated that all fish that died during their experiment were to as a nonappendaged hook), and one 10/0 nonoffset appen- deep hooked. The purpose of the current study was to compare daged circle hook (Mustad model number AP39960D) (Fig- anatomical hooking locations and catch rates of Black Sea ures 1, 3) (following USOFR 2010), we defined a circle hook Bass captured on appendaged hooks versus those of fish cap- as a “hook designed and manufactured so that the point is tured on commonly used hooks in the headboat fishery off the turned perpendicularly back to the shank to form a generally east coast of Florida. Furthermore, it was our intention to circular, or oval, shape”). The EZ Baiter hook, which meets determine whether the use of appendaged circle hooks is a the definition of a circle hook as defined above, is considered MANAGEMENT BRIEF 1201

Downloaded by [Department Of Fisheries] at 19:22 20 July 2015 and deep hooking of Black Sea Bass. RESULTS to make baiting easier and decrease difficulties associated with Over the course of the study, 522 Black Sea Bass were cap- dehooking captured fish, especially for inexperienced recrea- tured by the 10 participating anglers. Hook position had no tional anglers. As a result, the EZ Baiter hook is widely used effect on Black Sea Bass catch as an equal number were within the headboat fishery operating in the sampling area (G. caught on upper- (n D 261) and lower-positioned (n D 261) Strate, headboat operator, personal communication; E. Sander hooks. Length data for Black Sea Bass captured on both hook and B. Sauls, Florida Fish and Wildlife Conservation Commis- types were normally distributed and there was no significant sion, personal communication). The position (top or bottom) difference in the distributions (K-S statistic D 0.05, P D 0.84) of the nonappendaged and appendaged hooks was randomly or standard deviations (F D 0.89, P D 0.34) by hook type (Fig- assigned for each rig. Rigs were changed daily or whenever ure 4). The mean lengths of Black Sea Bass caught on each damage was incurred, and an equal number of both hooks hook type (nonappendaged hook D 268.8 mm, SD D 39.0; were fished during the study. All participating anglers used a appendaged hook D 269.7 mm, SD D 41.5; t D¡0.25, P D single bait type for each deployment. Bait was cut to an appro- 0.80) were not significantly different and both were less than priate size for a 10/0 hook and analyses were limited to those the current minimum size limit. There was a significant differ- Black Sea Bass captured with squid (Loligo sp.). All Black ence in the number of Black Sea Bass caught on each hook 1202 BERGMANN ET AL.

percentage of Black Sea Bass were hooked in the eye (25.00% [n D 4] versus 12.50% [n D 2]) when appendaged hooks were used (Table 1).

Downloaded by [Department Of Fisheries] at 19:22 20 July 2015 Among the explanatory variables used in the logistic TABLE 1. Summary of number of captures, TL, and anatomical hooking regression, only hook type was retained after the backward location data for Black Sea Bass caught on nonappendaged and appendaged selection procedure was applied and was significantly related hooks. to anatomical hooking location (likelihood ratio test: x2 D 4.68, P D 0.03). The chi-square test for independence demon- Nonappendaged Appendaged strated that the hook type and anatomical hooking location Variable hook hook x2 D D D variables were not independent ( 4.71, df 1, P 0.03). Number of fish caught 221 301 The odds ratio of a Black Sea Bass being hooked in the jaw Size range (mm TL) 170–389 119–404 D was 3.10 (95% CI 1.06–9.06) times greater when the appen- Mean size (mm TL) (SD) 268.8 (39.0) 270.0 (41.5) daged hook was used than when the nonappendaged hook was Number jaw hooked 210 296 D used. Of the Black Sea Bass that were deep hooked (n 16), a Number deep hooked 11 5 higher percentage of fish caught on nonappendaged hooks Eye 2 4 compared with those caught on appendaged hooks were Gill 1 0 D D hooked in the gills (6.25% [n 1] versus 0.00% [n 0]), Esophagus 7 1 D D esophagus (43.75% [n 7] versus 6.25% [n 1]), and stom- Stomach 1 0 ach (6.25% [n D 1] versus 0% [n D 0]); however, a higher MANAGEMENT BRIEF 1203

appendaged circle hooks can reduce the incidence of deep Bacheler, N. M., and J. A. Buckel. 2004. Does hook type influence the catch hooking in Black Sea Bass, a finding consistent with Willis and rate, size, and injury of grouper in a North Carolina commercial fishery? Millar (2001) who indicated appendaged circle hooks Fisheries Research 69:303–311. Bartholomew, A., and J. A. Bohnsack. 2005. A review of catch-and-release decreased deep hooking of Squirefish from up to 30% of fish angling mortality with implications for no-take reserves. Reviews in Fish deep hooked using standard circle hooks to 2% of fish captured Biology and Fisheries 15:129–154. using an appendaged hook. While our results demonstrated a Bugley, K., and G. Shepherd. 1991. Effect of catch-and-release angling on the reduction in deep hooking when appendaged hooks were used, survival of Black Sea Bass. North American Journal of Fisheries Manage- the deep-hooking rate of Black Sea Bass on nonappendaged cir- ment 11:468–471. Coleman, F. C., C. C. Koenig, G. R. Huntsman, J. A. Musick, A. M. Eklund, cle hooks was relatively low (2.1%) and similar to rates (1.5– J.C.McGovern,R.W.Chapman,G.R.Sedberry,andC.B.Grimes. 9.5%) reported for other species caught within recreational fish- 2000. Long-lived reef fishes: the snapper–grouper complex. Fisheries 25 eries off the eastern coast of the United States (Bacheler and (3):14–21. Buckel 2004). Furthermore, several authors have demonstrated Cooke, S. J., V. M. Nguyen, K. J. Murchie, A. J. Danylchuk, and C. D. Suski. that deep-hooking rates can be species-specific (e.g., Cooke 2012. Scientific and stakeholder perspectives on the use of circle hooks in recreational fisheries. Bulletin of Marine Science 88:395–410. and Suski 2004) and can be positively related to fish length for Cooke, S. J., and C. D. Suski. 2004. Are circle hooks an effective tool for some species (e.g., Alos et al. 2009). Therefore, future studies conserving marine and freshwater recreational catch-and-release fisher- will need to ascertain whether a reduction in deep hooking asso- ies? Aquatic Conservation: Marine and Freshwater Ecosystems 14:299– ciated with appendaged hooks occurs for other fishes. 326. We compared an offset nonappendaged hook to a nonoffset Godin, A. C., J. K. Carlson, and V. Burgener. 2012. The effect of circle hooks on shark catchability and at-vessel mortality rates in longline fisheries. Bul- appendaged hook and cannot discount that the use of these letin of Marine Science 88:469–483. two hook types confounds deep-hooking effects of offset ver- Mapleston, A., D. Welch, G. A. Begg, M. McLennan, D. Mayer, and I. Brown. sus nonoffset hooks. Furthermore, the difference in shank 2008. Effect of changes in hook pattern and size on catch rate, hooking loca- length between the two hooks could have had an effect that we tion, injury and bleeding for a number of tropical reef fish species. Fisheries were not able to account for. We acknowledge considerable Research 91:203–211. NMFS (National Marine Fisheries Service). 2014. Recreational fisheries statis- further research is needed to examine the significance of these tics. Available: http://www.st.nmfs.noaa.gov/recreational-fisheries/access- potential interactions and, ideally, future investigations should data/run-a-data-query/queries/index. (January 2014). use identical hooks, which would isolate the effect of the Rudershausen, P. J., J. A. Buckel, and E. H. Williams. 2007. Discard composi- appendage. However, the findings of our study indicate that tion and release fate in the snapper and grouper commercial hook-and-line use of appendaged hooks could be a viable mechanism to fur- fishery in North Carolina, USA. Fisheries Management and Ecology 14:103–113. ther minimize mortality rates associated with deep hooking of Sauls, B., and O. Ayala. 2012. Circle hook requirements in the Gulf of Mexico: Black Sea Bass in recreational fisheries and suggest that these application in recreational fisheries and effectiveness for conservation of hooks can have similar positive attributes within other hook- reef fishes. Bulletin of Marine Science 88:667–679. based fisheries. Additionally, this result could be achieved SEDAR (Southeast Data Assessment Review). 2011. South Atlantic Black without reducing catch rates or altering size composition of Sea Bass. SEDAR 25, Stock Assessment Report, North Charleston, South Carolina. Available: www.sefsc.noaa.gov/sedar/download/ the Black Sea Bass catch. SEDAR25_BlackSeaBass_SAR.pdf?id=DOCUMENT. (December 2013). Serafy, J. E., D. W. Kerstetter, and P. H. Rice. 2009. Can circle hook use benefit billfishes? Fish and Fisheries 10:132–142. ACKNOWLEDGMENTS Swimmer, Y., J. Suter, R. Arauz, K. Bigelow, A. Lopez, I. Zanela, A. Bolanos,~ We thank the crews of the Mayport Princess and Ocean J. Ballestero, R. Suarez, J. Wang, and C. Boggs. 2011. Sustainable fishing gear: the case of modified circle hooks in a Costa Rican longline fishery. Downloaded by [Department Of Fisheries] at 19:22 20 July 2015 Obsession, including Becky Hogan, Captain George Strate, Marine Biology 158:757–767. Sandy Kelly, Captain Butch Kelly, and Captain Shawn Lind- USOFR (U.S. Office of the Federal Register). 2010. Fisheries of the say, as well as Kevin Rademacher and the volunteer anglers, Caribbean, Gulf of Mexico, and South Atlantic: snapper–grouper fishery including Leonard “Wild Bill” Reed, Greg Williams, Charlie off the southern Atlantic states; amendment 17A; emergency rule to Davis, and Lenny Zurick, for assisting in data collection. Men- delay effectiveness of the snapper-grouper area closure, final rule and temporary rule (50 CFR Part 622). Federal Register 75:236(9 December tion of specific vendors or products does not constitute 2010):76874–76890. endorsement by the National Marine Fisheries Service. Willis, T. J., and R. B. Millar. 2001. Modified hooks reduce incidental mortality of snapper (Pagrus auratus: Sparidae) in the New Zealand commercial longline fishery. ICES Journal of Marine Science 58:830–841. REFERENCES Yokota, K., T. Mituhasi, H. Minami, and M. Kiyota. 2012. Perspec- tives on the morphological elements of circle hooks and their per- Alos, J., G. Mateu-Vicens, M. Palmer, A. M. Grau, M. Cabanellas-Reboredo, formance in pelagic longline fisheries. Bulletin of Marine Science and A. Box. 2009. Performance of circle hooks in a mixed-species recrea- 88:623–629. tional fishery. Journal of Applied Ichthyology 25:565–570. This article was downloaded by: [Department Of Fisheries] On: 20 July 2015, At: 19:15 Publisher: Taylor & Francis Informa Ltd Registered in England and Wales Registered Number: 1072954 Registered office: 5 Howick Place, London, SW1P 1WG

North American Journal of Fisheries Management Publication details, including instructions for authors and subscription information: http://www.tandfonline.com/loi/ujfm20 Using the Right Tool for the Right Job Should Include Validation When Possible: Response to Comment Jan-Michael Hessenauera, Mary Tate Bremiganb & Kim T. Scribnerbc a Wildlife and Fisheries Conservation Center, Department of Natural Resources and the Environment, University of Connecticut, 1376 Storrs Road, Storrs, Connecticut 06269-4087, USA b Department of Fisheries and Wildlife, Michigan State University, 480 Wilson Road, Natural Resources Building, East Lansing, Michigan 48824, USA c Department of Zoology, Michigan State University, 288 Farm Lane, Natural Science Building, East Lansing, Michigan 48824, USA Click for updates Published online: 24 Nov 2014.

To cite this article: Jan-Michael Hessenauer, Mary Tate Bremigan & Kim T. Scribner (2014) Using the Right Tool for the Right Job Should Include Validation When Possible: Response to Comment, North American Journal of Fisheries Management, 34:6, 1207-1210, DOI: 10.1080/02755947.2014.959677 To link to this article: http://dx.doi.org/10.1080/02755947.2014.959677

PLEASE SCROLL DOWN FOR ARTICLE

Taylor & Francis makes every effort to ensure the accuracy of all the information (the “Content”) contained in the publications on our platform. However, Taylor & Francis, our agents, and our licensors make no representations or warranties whatsoever as to the accuracy, completeness, or suitability for any purpose of the Content. Any opinions and views expressed in this publication are the opinions and views of the authors, and are not the views of or endorsed by Taylor & Francis. The accuracy of the Content should not be relied upon and should be independently verified with primary sources of information. Taylor and Francis shall not be liable for any losses, actions, claims, proceedings, demands, costs, expenses, damages, and other liabilities whatsoever or howsoever caused arising directly or indirectly in connection with, in relation to or arising out of the use of the Content. This article may be used for research, teaching, and private study purposes. Any substantial or systematic reproduction, redistribution, reselling, loan, sub-licensing, systematic supply, or distribution in any form to anyone is expressly forbidden. Terms & Conditions of access and use can be found at http:// www.tandfonline.com/page/terms-and-conditions North American Journal of Fisheries Management 34:1207–1210, 2014 Ó American Fisheries Society 2014 ISSN: 0275-5947 print / 1548-8675 online DOI: 10.1080/02755947.2014.959677

COMMENT

Using the Right Tool for the Right Job Should Include Validation When Possible: Response to Comment

We hoped that the publication of Hessenauer et al. (2014) of our paper was to highlight the potential for error and bias would spark reflection and discussion among managers and and, more importantly, to offer methods for quantifying bias. scientists who use visual techniques to locate and track the Therefore, in so far as our paper contributes to the discussion success of fish nests. We are grateful for the opportunity to about this important aspect of fisheries research, the dialog continue the discussion about the need for validation in fisher- afforded by Stein et al. is productive. ies studies and to reply to the points made by Stein et al. We Secondly, the intent of our paper was not to provide a cri- begin by briefly discussing what the intent of our paper was tique of “the efficacy of snorkel surveys in detecting Large- and what it was not. The rest of our response pertains to spe- mouth Bass... nests” or to make disparaging remarks that cific remarks made in the Stein et al. comment. Ultimately, we “devalue, confound, or invalidate” the excellent body of work encourage both further evaluation of the factors underling the by the authors contributing to the Stein et al. comment or that variation in the extent of observational biases and broad dis- of other authors who utilize these methods. Snorkeling is a cussion of their implications under different research and man- valuable tool in fisheries research for some research questions. agement settings. Indeed, this topic strikes us as an ideal Our point was that validation methods such as those afforded opportunity for collaboration among research groups that span by genetic data can improve observational data and their use systems, species, and techniques. across a variety of situations and scales. The primary purpose of our paper was to comm.unicate that Scale in terms of the phenomena under study, sampling, methods are available to fisheries researchers which can vali- and analysis is important to consider in geography, spatial date studies of recruitment that rely (among other things) on statistics, and ecology (Dungan et al. 2002) as well as in observational methods. Specifically, genetic data and multi- population genetics (Anderson et al. 2010). Stein et al. stage sampling designs can be used to validate results obtained repeatedly refer to the whole-lake scale of our analysis. from visual observations (e.g., at the nest). Stein et al. suggest While we view it as a strength of our research that our anal- that expert opinion is invaluable in the design and carrying out yses of recruitment were conducted at the whole-lake scale, of field observations, including snorkeling. We agree. we stated that genetic data can be profitably used as a Unquestionably, Stein et al. have a vast set of experiences means of validation irrespective of the scale over which the with multiple species and multiple field settings, as demon- data are collected. We recommend validation of data strated by the large number of their research papers that they regardless of the spatial scale of analyses. cite, along with many others that are not cited. As readers of In their comment, Stein et al. express the opinion that the the North American Journal of Fisheries Management and prevalence of the two forms of bias we reported in Hessenauer

Downloaded by [Department Of Fisheries] at 19:15 20 July 2015 comparable journals, we agree that a certain level of credibil- et al. (2014) is unique to our choice of systems and techniques. ity is owed to authors based on their past experience and pro- We consider it common knowledge that the nature and magni- ductivity. The contributions of Stein et al. that are based on tude of biases and sampling error are highly likely to differ snorkeling observations and other methods are exemplary. among settings and methods. Here we respond to speciﬁc Stein et al. state that the biases we reported were associated criticisms made by Stein et al., emphasizing that comparisons with our survey design and should not be generalized to all of observational biases need to be made in a transparent and snorkeling surveys, especially when those surveys are con- data-driven fashion with an independent means of data valida- ducted by experienced personnel. Our references to examples tion, as is afforded by genetic methods. from the snorkeling literature (including publications involving Stein et al.) were not meant to suggest that there are substantive biases in all published surveys but rather to present ISSUE 1: OBSERVED NEST DETECTION BIAS examples from the literature in which observational methods Stein et al. believe that the occurrence of missed nests (those were used without independent validation. To our knowledge, not discovered by our observational methods) resulted from our none of the examples Stein et al. reference include the valida- ﬁeld methods and choice of study lake and is therefore fairly tion of visual methods. Thus, these papers prove neither the unique to our study. We realize that the rate at which we missed lack of bias nor the existence of it in this research. The purpose nests appears to be high and, as Stein et al. suggest, may in part

1207 1208 HESSENAUER ET AL.

be attributable to our study methods and system. However, we nest of origin. Sutter et al. (2012) studied six 0.25-ha ponds know of no published estimate of the efficiency of field methods stocked with 8 male and 7 female Largemouth Bass and traced for locating black bass Micropterus spp. nests that predates our all fall recruits back to a known male. In contrast, Warner study (but see the nearly concurrent Shaw and Allen 2014) and Lake is 24 ha in size with an unknown population size and sex occurs at the whole-lake scale; therefore, the uniqueness of the ratio. Barthel (2010) studied a much larger open riverine and rate at which we missed nests is rather difficult to assess. lacustrine system for Smallmouth Bass. Using a paternity anal- Because the missed nests in our study were discovered months ysis, Barthel (2010) assigned 84% of age-1 offspring to a can- later by genetic analysis, we were unable to observe the habitat didate male and suggested that the unassigned individuals features and guarding-male characteristics of the nests we either originated from unsampled males or were the result of missed. Therefore, it was not possible for us to evaluate how immigration through their system or genotyping errors. Simi- macrophyte cover, nest depth, and other factors might have larly, Gross and Kapuscinski (1997) studied a bay of Lake biased our field methods. A post hoc analysis of sampling effort Opeongo, Ontario, and assigned 62% of age-0 Smallmouth versus nest discovery appeared to reveal some bias. For exam- Bass to a nest of origin using nest-specific DNA fingerprints. ple, more nests were discovered on clear days than expected; These two studies occurred in open systems in which it was however, this analysis is dependent on nest availability during not possible to determine whether unassigned individuals orig- the sampling days. Bias was almost certainly present in the inated from outside the study area or from unobserved nests. methods we employed to locate nests. Stein et al. speculate that Both studies represent impressive examples of the power of this bias results in part from our choice of species. We acknowl- genetic analyses to inform the conservation of black bass spe- edge the differences between Largemouth Bass M. salmoides cies and may highlight differences in the efficiency of locating and Smallmouth Bass M. dolomieu nests that Stein et al. Smallmouth Bass nests and Largemouth Bass nests in the field. describe. We would hypothesize that the prevalence of missed We applaud the efforts of Shaw and Allen (2014) to evaluate nests is higher for Largemouth Bass than for Smallmouth Bass. the detection probability of Florida Bass M. floridanus broods Stein et al. also speculate that the missed nests in our study using duel snorkeler surveys, which are similar to the methods resulted from a reliance on boat-based observers. They certainly that we advocate in Hessenauer et al. (2014). Their study high- could be partially correct. However, the use of boat-based obser- lights the need for the training of field observers and suggests vation for Largemouth Bass nest detection is established in the that high rates of nest detection can be achieved by snorkeling. literature (e.g., Wagner et al. 2006; Reed and Pereira 2009; However, the detection probabilities calculated by such meth- Lawson et al. 2011). It is no accident that the ability of boat- ods apply only to broods available for detection by those meth- based “observers” (i.e., anglers) to spot nesting Largemouth ods. Other broods, e.g., broods located deeper than the Bass is what makes fishing during the nesting season controver- approximately 3.0-m depth to which the surveys of Shaw and sial and the impetus for much research. In addition to boat-based Allen (2014) extended (if extant), may have been missed. The observation, our method entailed towing a snorkeler behind the nest detection that we reported in Hessenauer et al. (2014) com- boat (propelled by a 40-lb thrust trolling motor) who actively pares two independent methods of nest detection: (1) an obser- searched for nests. We believed that this hybrid design (both vational boat–snorkel survey and (2) a genetic analysis of fall above- and underwater observation) would afford us the benefits age-0 populations. Of course, the boat–snorkel survey can only of each method while allowing us to sample at the whole-lake detect nests in the area surveyed at some unknown rate of scale (Warner Lake has approximately 2.5 km of shoreline). detection. The genetic analysis of fall age-0 populations pre- We fully agree with the conclusion of Stein et al. that our sumably reflected successful nests from the entire lake (and

Downloaded by [Department Of Fisheries] at 19:15 20 July 2015 sampling error rate only applies directly to the methods we detected them at some unknown rate of detection through age-0 used in Warner Lake during the 2010 Largemouth Bass sampling)—assuming that there was well-mixed dispersal from nesting season. We did not intend for the published text of the nest of origin, as reported by Hessenauer et al. (2012). We Hessenauer et al. (2014) to advocate the application of our feel that our comparison of independent methods is important speciﬁc sampling error rates to other systems or other stud- because some nests may not be available to snorkeling as a ies, nor do we believe that it does so. What we were hoping result of their location or habitat features. If one assumes that to impart to readers was simply that these error rates are snorkeling is 100% effective at detecting the nests available to largely unknown (or unreported) for other studies and may snorkelers, the total detection probability (lakewide) may still be larger than anticipated; this in no way rules out the possi- be low if a large number of nests are unavailable to snorkeling. bility that such errors are small. Studies are needed that compare the detection rates of vari- We feel that Stein et al.’s comparison of nest detection ous survey methods (e.g., boat observation, snorkeling, and a rates between our study and those of Parkos et al. (2011) and hybrid approach like the one that we used) in a variety of sys- Sutter et al. (2012) is misleading if one ignores the context of tems. Such work would aid researchers and managers in mak- those studies. Parkos et al. (2011) studied ﬁve 0.4-ha ponds ing a decision about what methods to employ based on the stocked with 10 male and 12 female Largemouth Bass and goals of the study, the size of the area to be sampled, and the reported that they were able to assign 86–91% of recruits to a conditions at the sample site. For example, if the goal of a COMMENT 1209

study were to detect the maximum possible number of nests in studies, nor to infer the rates at which they occurred. Rather, a large lake, should one employ methods that have a lower we only suggested that such error rates are unknown in previ- detection rate but that would allow the whole shoreline to be ous studies—and if such errors were present, they could sampled? Alternatively, should researchers employ methods impact the conclusions of a study, especially if the errors were that have a higher detection probability but that cannot feasi- nonrandom. Clearly, the error rates in our system that were bly sample an entire shoreline? detected by genetic methods would have been detrimental to an observational study attempting to make inferences about the features of successful Largemouth Bass nests if the error ISSUE 2: ERROR IN NEST SUCCESS/FAILURE rates were not known. Genetic assignment of age-0 individuals CLASSIFICATION to their nests of origin provided—for the first time—an inde- The second issue that we raised in Hessenauer et al. (2014) pendent check on visual observation methods in a wild popula- was that the observational method we employed (snorkelers tion supporting a real fishery, which in our opinion is examining known nest sites) frequently failed to detect swim- important. up fry (and thus successful nests as defined by Philipp et al. We appreciate the opportunity presented by Stein et al. to 1997) at nests to which age-0 individuals were later assigned continue the discussion of visual observation as a method for using genetic methods. Our observational methods scored black bass nest detection and tracking as well as to clarify nests as failed only if no offspring or guarding male were some of the conclusions of Hessenauer et al. (2014). In our observed over the course of three consecutive visits. We opinion, Hessenauer et al. (2014) does not invalidate any pre- attempted to evaluate the causes of this classification error vious study, nor was the intent to criticize particular studies. (e.g., nest depth and macrophyte cover) but had insufficient Rather, most of the studies that we cited were selected to point sample size to make robust conclusions. Stein et al. suggest out the variety of purposes for which visual observations of that our sampling interval (every 3–4 d) was not sufficiently black bass nests have been utilized. The extent to which the frequent to correctly associate swim-up fry schools with their bias or error we discovered in our study impacted any of the nests of origin. Although we agree that more frequent observa- studies we cited is unknown because the errors themselves are tion is always better, we disagree with their overall assess- unknown. Errors could have occurred if visual observation ment. Cooke et al. (2006) report that the average duration of limited the subset of nests available for manipulation, e.g., if a care for Largemouth Bass and Smallmouth Bass is 20 and 28 nest had a characteristic that made it particularly easy for d, respectively (Cooke et al. 2006: their Table 1). Although snorkelers to find, it would have been more likely to be Stein et al. state in their comment that an observational fre- included in the subset of nests located by snorkeling than nests quency of 1–2 d is required, the studies that they cite (which without that characteristic. Likewise, if the goal of a study informed our methods) typically used a 2–3-d interval (which were to determine the characteristics associated with nest suc- overlaps with ours). Cooke et al. (2006) also reported that cess, errors in determining such success may affect the conclu- Largemouth Bass and Smallmouth Bass continue parental care sions of the study. Detecting and describing the effects of error after offspring become free swimming and that the dispersion and bias is fundamental to scientific advancement; the out- of the swim-up fry school increases as offspring development comes of error, whether large or small, are of course depen- increases, which could increase the visibility of the school. As dent on the objectives of a given study. Hessenauer et al. noted by Stein et al., Cooke et al. (2006) documented that (2014) offered genetic pedigree reconstruction as a method of Largemouth Bass leave the nest site more quickly than do detecting observational errors associated with the surveying of

Downloaded by [Department Of Fisheries] at 19:15 20 July 2015 Smallmouth Bass, adding to the challenge of studying Large- black bass nests and suggested other methods to evaluate the mouth Bass. As we suggested in Hessenauer et al. (2014), a efficiency of observational nest surveys. We agree with Stein future study that evaluates the range of distances over which et al. that snorkel surveys and other visual observation meth- fry schools disperse from the nest site and the associated time ods are, and will remain, an important tool for the study of intervals would be very informative. Until then, we agree with black bass reproductive ecology and dynamics. Implementing Stein et al. that observations should occur at time intervals methods that can disclose the presence and magnitude of that are as short as possible. However, if the movement of off- errors associated with snorkeling and other observational spring from the original nest site is as rapid as Stein et al. sug- methods—and potentially correct for them—can only improve gest, the possibility of misclassifying successful nests as failed the utility of observational methods moving forward. seems plausible unless the snorkeler actually witnesses the moment when the larvae become free swimming. ACKNOWLEDGMENTS We would like to thank J. Vokoun and L. Nathan for com- CONCLUSIONS ments that improved the presentation of this manuscript. We It was not our intention to assert that the detection and clas- also thank Stein et al. for their comment and the subsequent sification errors that we highlighted have occurred in all other self-reflection that it incited. 1210 HESSENAUER ET AL.

JAN-MICHAEL HESSENAUER* Barthel, B. L. 2010. Influence of dispersal, natal environment, and variance in reproductive success on the genetic relationships within a popu- Wildlife and Fisheries Conservation Center, lation of freshwater fish. Doctoral dissertation. University of Illinois, Urbana. Department of Natural Resources and the Environment, Cooke, S. J., D. P. Philipp, D. H. Wahl, and P. J. Weatherhead. 2006. Energet- University of Connecticut, 1376 Storrs Road, ics of parental care in six syntopic centrarchid fishes. Oecologia 148:235– Storrs, Connecticut 06269-4087, USA 249. MARY TATE BREMIGAN Dungan, J. L., J. N. Perry, M. R. T. Dale, P. Legendre, S. Citron-Pousty, M.-J. Fortin, A. Jakomulska, M. Miriti, and M. S. Rosenberg. 2002. A balanced view of scale in spatial statistical analysis. Ecography 25:626–640. Department of Fisheries and Wildlife, Gross, M. L., and A. R. Kapuscinski. 1997. Reproductive success of Small- Michigan State University, mouth Bass estimated and evaluated from family-specific DNA fingerprints. 480 Wilson Road, Natural Resources Building, Ecology 78:1424–1430. East Lansing, Michigan 48824, USA Hessenauer, J. M., M. T. Bremigan, and K. T. Scribner. 2012. Genetic pedigree reconstruction facilitates estimates of age-0 Largemouth Bass dispersal. Transactions of the American Fisheries Society 141:1672–1681. KIM T. SCRIBNER Hessenauer, J. M., M. T. Bremigan, and K. T. Scribner. 2014. Genetic pedigree reconstruction detects bias in Largemouth Bass nest sampling procedures. Department of Fisheries and Wildlife, North American Journal of Fisheries Management 34:175–183. Michigan State University, Lawson, Z. J., J. W. Gaeta, and S. R. Carpenter. 2011. Coarse woody habitat, 480 Wilson Road, Natural Resources Building, lakeshore residential development, and Largemouth Bass nesting behavior. North American Journal of Fisheries Management 31:666–670. East Lansing, Michigan 48824, USA; and Parkos, J. J. III, D. H. Wahl, and D. P. Philipp. 2011. Influence of behavior and Department of Zoology, mating success on brood-specific contribution to fish recruitment in ponds. Michigan State University, Ecological Applications 21:2576–2586. 288 Farm Lane, Natural Science Building, Philipp, D. P., C. A. Toline, M. F. Kubacki, D. B. F. Philipp, and F. J. S. Phe- East Lansing, Michigan 48824, USA lan. 1997. The impact of catch-and-release angling on the reproductive success of Smallmouth Bass and Largemouth Bass. North American Journal of Fisheries Management 17:557–567. *Corresponding author: [email protected] Reed, J. R., and D. L. Pereira. 2009. Relationships between shoreline development and nest site selection by Black Crappie and Largemouth Bass. North American Journal of Fisheries Management 29:943–948. Shaw, S. L., and M. S. Allen. 2014. Localized spatial and temporal variation in reproductive effort of Florida Bass. Transactions of the American Fisheries Society 143:85–96. REFERENCES Sutter, D. A. H., C. D. Suski, D. P. Philipp, T. Klefoth, D. H. Wahl, P. Kersten, S. J. Cooke, and R. Arlinghaus. 2012. Recreational fishing selectively cap- Anderson, C., B. K. Epperson, M.-J. Fortin, R. Holderegger, P. James, M. tures individuals with the highest fitness potential. Proceedings of the Rosenberg, K. T. Scribner, and S. Spear. 2010. Consideration of scale in National Academy of Sciences of the USA 109:20960–20965. landscape genetics can improve predictive relationships between measures Wagner, T., A. K. Jubar, and M. T. Bremigan. 2006. Can habitat alteration and of spatial genetic structure and landscape pattern. Molecular Ecology spring angling explain Largemouth Bass nest success? Transactions of the 25:3565–3575. American Fisheries Society 135:843–852. Downloaded by [Department Of Fisheries] at 19:15 20 July 2015 This article was downloaded by: [Department Of Fisheries] On: 20 July 2015, At: 19:15 Publisher: Taylor & Francis Informa Ltd Registered in England and Wales Registered Number: 1072954 Registered office: 5 Howick Place, London, SW1P 1WG

North American Journal of Fisheries Management Publication details, including instructions for authors and subscription information: http://www.tandfonline.com/loi/ujfm20 Evaluation of Precision and Sample Sizes Using Standardized Sampling in Kansas Reservoirs Jeff D. Kocha, Ben C. Neelyb & Michael E. Colvinc a Kansas Department of Wildlife, Parks, and Tourism, 21514 South Yoder Road, Pretty Prairie, Kansas 67570, USA b Kansas Department of Wildlife, Parks, and Tourism, 5089 Road 2925, Independence, Kansas 67301, USA c Oregon Cooperative Fish and Wildlife Research Unit, Department of Fisheries and Wildlife, Oregon State University, 104 Nash Hall, Corvallis, Oregon 97331, USA Published online: 24 Nov 2014. Click for updates

To cite this article: Jeff D. Koch, Ben C. Neely & Michael E. Colvin (2014) Evaluation of Precision and Sample Sizes Using Standardized Sampling in Kansas Reservoirs, North American Journal of Fisheries Management, 34:6, 1211-1220, DOI: 10.1080/02755947.2014.959676 To link to this article: http://dx.doi.org/10.1080/02755947.2014.959676

PLEASE SCROLL DOWN FOR ARTICLE

Taylor & Francis makes every effort to ensure the accuracy of all the information (the “Content”) contained in the publications on our platform. However, Taylor & Francis, our agents, and our licensors make no representations or warranties whatsoever as to the accuracy, completeness, or suitability for any purpose of the Content. Any opinions and views expressed in this publication are the opinions and views of the authors, and are not the views of or endorsed by Taylor & Francis. The accuracy of the Content should not be relied upon and should be independently verified with primary sources of information. Taylor and Francis shall not be liable for any losses, actions, claims, proceedings, demands, costs, expenses, damages, and other liabilities whatsoever or howsoever caused arising directly or indirectly in connection with, in relation to or arising out of the use of the Content. This article may be used for research, teaching, and private study purposes. Any substantial or systematic reproduction, redistribution, reselling, loan, sub-licensing, systematic supply, or distribution in any form to anyone is expressly forbidden. Terms & Conditions of access and use can be found at http:// www.tandfonline.com/page/terms-and-conditions North American Journal of Fisheries Management 34:1211–1220, 2014 Ó American Fisheries Society 2014 ISSN: 0275-5947 print / 1548-8675 online DOI: 10.1080/02755947.2014.959676

ARTICLE

Evaluation of Precision and Sample Sizes Using Standardized Sampling in Kansas Reservoirs

Jeff D. Koch* Kansas Department of Wildlife, Parks, and Tourism, 21514 South Yoder Road, Pretty Prairie, Kansas 67570, USA Ben C. Neely Kansas Department of Wildlife, Parks, and Tourism, 5089 Road 2925, Independence, Kansas 67301, USA Michael E. Colvin Oregon Cooperative Fish and Wildlife Research Unit, Department of Fisheries and Wildlife, Oregon State University, 104 Nash Hall, Corvallis, Oregon 97331, USA

Abstract We evaluated the precision of samples and the number of stock-length fish collected by means of standard methods used for sampling North American freshwater fishes from 2010 to 2013 in Kansas. Additionally, we used resampling procedures to determine the number of gear deployments needed to achieve a relative standard error (RSE) of 25% for the CPUE and collect 100 stock-length individuals. Median RSE of electrofishing samples was generally less than 25% for Largemouth Bass Micropterus salmoides in all sizes of reservoirs and for Channel Catfish Ictalurus punctatus in medium (251–1,000 acres) and large reservoirs (greater than 1,000 acres). The RSE estimates were generally >25% for Bluegill Lepomis macrochirus and crappies Pomoxis spp. collected in trap nets and palmetto bass (female Striped Bass Morone saxatilis £ male White Bass M. chrysops) and Walleye Sander vitreus sampled in gill nets. With few exceptions, 100 stock-length individuals of all target species (e.g., Largemouth Bass, Bluegill, crappies, palmetto bass, Channel Catfish, Walleye) were not sampled at current levels of effort. Resampling procedures indicated that fewer than 20 deployments were usually needed to obtain an RSE 25% and 100 stock-length fish for Largemouth Bass in smaller impoundments (i.e., <250 acres); however, more than 20 deployments were needed in larger impoundments. The median effort needed to achieve an RSE 25% for Bluegills and crappies in trap nets varied and may exceed what some biologists find practical. Fewer gill-net deployments were needed to reach an RSE 25% for palmetto bass and Walleyes than to collect 100 stock-length fish. Our results indicate that more samples than are currently prescribed are generally needed to precisely sample Downloaded by [Department Of Fisheries] at 19:15 20 July 2015 sport fishes by means of standardized protocols in Kansas reservoirs. In some instances, obtaining precise samples may not be logistically feasible. In these situations, biologists should be aware of the potential shortcomings of sampling protocols and set objectives accordingly.

Fish sample data should be collected using protocols for spatial and temporal comparisons (Bonar et al. 2009a). designed to allow rigorous statistical comparisons (Willis and Two important characteristics of ﬁsh sample data are accuracy Murphy 1996). As such, biologists must consider whether and precision. Accuracy refers to the closeness of a metric to collected data will allow a proper evaluation of research or its true value, and precision refers to how well test results can management questions (Brown and Austen 1996). Addition- be reproduced. Because determining accuracy of ﬁsheries ally, data must be collected using standard methods that allow samples is generally not feasible, precision is most often used

*Corresponding author: [email protected] Received May 12, 2014; accepted August 23, 2014

1211 1212 KOCH ET AL.

to provide assurance that no unsuspected bias exists in the relative abundance metrics (i.e., CPUE) should also be measure (Brown and Austen 1996). If data are imprecise infer- considered. Several studies have recommended target preci- ences about metrics describing fish assemblages and popula- sion levels of a relative standard error (RSE) of at least tions may be limited (Brown and Austen 1996; Quist et al. 25% for a mean CPUE (hereafter RSE25) in fisheries man- 2009). Although the number of gear deployments is directly agement surveys (Robson and Regier 1964; Hardin and related to precision (Wilde 1995; Wilde and Fisher 1996; Conner 1992; Wilde and Fisher 1996; Dumont and Dumont and Schlechte 2004), few recommendations are avail- Schlechte 2004). Additional sampling objectives may able to guide the number of gear deployments for standardized include description of population size structure. Research- fish sampling. Quist et al. (2009) suggested that existing data ers have suggested a minimum sample size of 100 stock- be used to determine the number of gear deployments needed length individuals to adequately describe size structure to meet desired sampling objectives. This approach should be (Anderson and Neumann 1996), although specific sample considered when developing standardized sampling protocols sizes vary with survey objectives (Gustafson 1988; (Noble et al. 2007; Quist et al. 2009). Additionally, gear Miranda 1993, 2007; Vokoun et al. 2001; Quist et al. deployment estimators provide biologists with realistic expect- 2009). Regardless, variability of samples and number of ations of conclusions that can be drawn from fish sampling fish collected are important to consider when establishing data. or evaluating a sampling protocol. The Kansas Department of Wildlife,Parks,andTourism Evaluation of KDWPT fish sampledatacanprovide (KDWPT) adopted standard methods for sampling North insight into the quality of existing data and help guide American freshwater fishes in 2010 following recommen- future sampling efforts. Further, this evaluation provides dations from Bonar et al. (2009b). Kansas biologists stan- an examination of the influence of fish sampling standardi- dardized gears, seasons and times of sampling, and the zation as recommended by Bonar et al. (2009b). As such, manner in which sampling sites were chosen. As no sample our objectives were to (1) evaluate the precision of catch size criteria are provided in the methods proposed by estimates and the number of stock-length individuals col- Bonar et al. (2009b), the number of deployments per lected from standard sampling gears (i.e., electrofishing, impoundment were chosen based on logistical restraints trap-netting, gill netting) after implementation of standard associated with reservoir size (i.e., sample size increases methodsusedforsamplingNorthAmericanfreshwater with reservoir size; Table 1). Currently, the main objective fishes and (2) use resampling procedures to develop guide- of KDWPT sampling is to achieve these minimum sam- lines for the number of gear deployments required to pling efforts; however, obtaining precise data in regard to obtain an RSE25 and 100 stock-length fish.

TABLE 1. Prescribed minimum effort and median actual number of gear deployments (minimum and maximum actual effort in parentheses) expended by KDWPT using standard methods for sampling North American freshwater fishes from 2010 to 2013. Gill-net and trap-net effort was defined as the number of net-nights. Electrofishing effort was defined as the number of 10-min stations. Gear Reservoir size (acres) Prescribed minimum effort Actual effort

Experimental gill net <50 3 3 (1–5) 50–99 4 4 (2–8) Downloaded by [Department Of Fisheries] at 19:15 20 July 2015 100–299 6 6 (2–12) 300–999 8 8 (3–10) 1,000–2,499 10 10 (10–11) 2,500–4,999 12 12 (6–19) 5,000–8,999 15 16 (9–16) 9,000 20 20 (6–27) Trap net <100 2 2 (1–8) 100–299 4 4 (1–10) 300–999 8 8 (2–8) 1,000–1,499 10 10 (9–12) 1,500–1,999 12 8 (8–8) 2,000 16 16 (2–24) Electroﬁshing <250 6a 6 (1–15) 250 10 10 (1–65)

aOr a complete shoreline circuit if six stations cannot be completed. EVALUATION OF STANDARDIZED SAMPLING TECHNIQUES 1213

METHODS et al. 2009). The leads were 100 ft long and were constructed of 0.5-in bar mesh. Trap nets were deployed in randomly Fish Collection selected locations in water 3–16 ft deep in the afternoon and Data collected during 2010–2013 were analyzed from 184 retrieved the following morning, so sample periods encom- Kansas reservoirs varying in size from 1 to 16,020 acres. passed two crepuscular periods (Pope et al. 2009). Black Crap- Although physical and chemical characteristics of Kansas pie P. nigromaculatus and White Crappie P. annularis were reservoirs are diverse, most are relatively shallow, turbid, and combined because they are grouped for management activities do not stratify due to persistent winds (Dodds et al. 2006). in Kansas (e.g., length limits, creel limits). Trap-net effort var- The watersheds of most study reservoirs are dominated by ied by reservoir size from 2 to 16 net-nights (Table 1). either row crop or grassland. More detailed descriptions of Channel Catfish, palmetto bass, and Walleyes were sam- Kansas impoundments are given in Dodds et al. (2006). Tar- pled during autumn with gill nets that were generally fished get species for this project included Largemouth Bass Micro- concurrently with trap nets. Experimental sinking gill nets pterus salmoides, Bluegill Lepomis macrochirus,crappies were 80 ft long, 6 ft deep, and constructed of eight randomly Pomoxis spp., Channel Catfish Ictalurus punctatus,palmetto ordered 10-ft panels of 0.75–2.5-in bar measure monofilament £ bass (female Striped Bass Morone saxatilis male White mesh in 0.25-in increments (Pope et al. 2009). Similar to trap Bass M. chrysops), and Walleye Sander vitreus and were cho- nets, gill nets were deployed in the afternoon and retrieved the sen for their popularity with anglers, status of concern for next morning. Gill nets were set perpendicular to the shoreline managers, and widespread distribution in Kansas. For each in 6–16 ft of water with the orientation of nets randomly target species, data were included in the analysis when at least assigned for each deployment. Prescribed gill-net deployments one individual of that species was sampled from the respec- varied from 3 to 20 net-nights depending on reservoir size. tive impoundment during the study years. The number of res- Trap nets and gill nets that were fished for 12–28 h were ervoirs included for each target species varied from 157 to included in the analysis and corresponding effort was consid- 164 for Bluegill, Channel Catfish, crappies, and Largemouth ered as one net-night. Bass, while 55 reservoirs were included in the analysis for Because of variation in fish assemblages and reservoir char- both palmetto bass and Walleye. Discrepancies in these num- acteristics, study reservoirs were grouped by surface area. bers and total number of reservoirs included in the analysis Ponds were classified as those 0–20 surface acres. Most of were due to the absence of species or gears in some sampling these impoundments were small urban and municipal ponds. events. Reservoirs from 21 to 250 acres were classified as small, and Largemouth Bass were sampled by KDWPT in spring using encompassed state-owned Kansas State Fishing Reservoirs. daytime boat electrofishing. Electrofishing equipment (e.g., Medium impoundments varied from 251 to 1,000 acres and Smith Root, Coffelt, Midwest Lake Electrofishing Systems) generally represented larger municipal and county reservoirs. was variable among Kansas biologists; however, target power Lastly, reservoirs larger than 1,000 acres were classified as outputs of 2,750–3,250 W were used to achieve the desired large, which represented federal flood control reservoirs and fish response (Miranda 2009). Electrofishing surveys were power plant cooling supplies. conducted at randomly selected shoreline stations for 10 min. Reservoir maps with numbered-grid overlays were produced to assist in the selection of randomized sampling sites. Each Data Analysis grid was 330 £ 330 ft for impoundments smaller than 1,200 Observed data.—The precision of catch rates for stock- £ Downloaded by [Department Of Fisheries] at 19:15 20 July 2015 acres and 1,100 1,100 ft for impoundments larger than length fish (Gabelhouse 1984; Dumont and Neely 2011) was 1,200 acres. Prior to fish sampling each year, grids were ran- quantified by calculating the RSE for each combination of res- domly chosen for all gears from appropriate sampling loca- ervoir, target species, and sample year. As such, depending on tions (Bonar et al. 2009b). A minimum of 10 electrofishing how many years an impoundment was sampled, up to four stations were sampled in reservoirs greater than 250 acres, and RSE estimates were calculated for each reservoir and species. a minimum of six stations or one complete lap of the shoreline Similarly, the number of fish collected per sample was calcu- were sampled in reservoirs less than 250 acres (Table 1). lated as the sum of all stock-length individuals collected per Largemouth Bass were sampled during their spawning period year in each reservoir with each gear (e.g., number of stock- when water temperatures were 60–70 F. For this analysis, length Bluegills captured in trap nets at one reservoir in a only electrofishing samples collected between April 1 and given year). Stock-length fish were used in this analysis to mit- May 31 were used. igate potential effects of large and variable catches of age-0 Bluegills and crappies were collected with trap nets from individuals. October 1 to November 30 when water temperatures were Resampling analysis.—Data for resampling techniques 68 F or below. Trap nets consisted of two rectangular 3 £ 6-ft consisted of the number of stock-length fish of each target spe- frames, a 4-in opening in the frame, and four 2.5-ft-diameter cies captured during each gear deployment in each impound- hoops with a funnel attached to the first and third hoops (Pope ment. For electrofishing, one 10-min station was considered as 1214 KOCH ET AL.

one deployment, and for gill and trap nets, one net-night was 80 to over 100 surveys were needed to achieve this objective considered as one deployment. Catch data for each target spe- in 75% of surveys. cies and reservoir combination were pooled among years to avoid potential effects of small sample sizes, as some impoundments had few sample deployments (e.g., <5) annu- Trap-Netting ally. This resulted in each target species and reservoir combi- Catch rates of Bluegills and crappies in trap nets were nation having an array of numbers that represented catch per highly variable among years and reservoirs. Among species gear deployment during the entire study period. The resam- and reservoir sizes, the median RSE of stock CPUE was less pling approach (i.e., stochastic resampling procedure; Dumont than 25% for crappies only in large reservoirs. Among reser- and Schlechte 2004) began by randomly resampling (with voir sizes, the RSE25 for Bluegills was only reached in at least replacement) two numbers (i.e., two gear deployments) 25% of samples in ponds (Figure 1). In larger impoundments, 2,000 times and calculating the RSE for each of the 2,000 very few Bluegill CPUE estimates exhibited an RSE25. resamples. The proportion of 2,000 resamples achieving an Excluding crappies in large reservoirs, 100 stock-length indi- RSE 25 was then calculated. If the proportion was <0.8, viduals were not typically captured in trap-net samples then three deployments were evaluated. The threshold of 0.8 (Figure 2). was used in a similar analysis by Dumont and Schlechte Fewer trap nets were usually needed to precisely estimate (2004) as they indicated that 80% certainty in estimates was the CPUE for crappies than for Bluegills. In many instances, sufficient for these analyses. This process continued iteratively 20 or more trap nets were needed to reach an RSE25 for Blue- in increments of one deployment until the proportion of resam- gills regardless of reservoir size (Figure 3). Similarly, 20 trap pling events achieving precision (RSE25) objectives was nets were required to reach an RSE25 for crappies 50% of the 0.8. This process was repeated to evaluate gear deployments time in all sizes of reservoirs except ponds. Fewer gear deploy- needed to sample 100 stock-length fish. A minimum of two ments were generally required to capture 100 stock-length and a maximum of 100 deployments were set as the lower and Bluegills and crappies than to reach the RSE25 (Figure 4). upper boundaries to reflect the practical amounts of effort (Dumont and Schlechte 2004). Resampling analysis was conducted using the program R (R Core Development Team Gill Netting 2013). Relative standard errors of gill-net catch rates were generally lower in larger reservoirs where more effort was expended. The median RSE for CPUE of Channel Catfish RESULTS exceeded 25% in ponds and small reservoirs (Figure 1). Rarely were 100 stock-length Channel Catfish collected in any Electrofishing sampling event. The 90th percentile of Channel Catfish catch Observed precision of Largemouth Bass CPUE from 2010 was greater than 100 fish only in large reservoirs (Figure 2). to 2013 was relatively high, as RSE was often below 25% Palmetto bass and Walleyes were mostly absent from pond (Figure 1). Largemouth Bass catch precision appeared more samples and catch rates were highly variable. The median variable in larger waters, although median RSE was below RSE was below 25% for Walleyes only in large reservoirs; 25% for all reservoir sizes. With the exception of small reser- whereas, the median RSE for palmetto bass varied from 30% voirs, the target of 100 stock-length Largemouth Bass was to 50% (Figure 1). One hundred stock-length palmetto bass or

Downloaded by [Department Of Fisheries] at 19:15 20 July 2015 rarely achieved in annual samples (Figure 2). Walleyes were rarely collected in sampling events. Approxi- Resampling analyses indicated less than 10 deployments mately 25% of samples from large reservoirs contained 100 were needed to reach RSE25 for ponds and small impound- palmetto bass, and only around 10% of samples contained 100 ments (Figure 3). As reservoir size increased, more samples Walleyes (Figure 2). were needed to achieve an RSE25 for Largemouth Bass. In The number of gill nets needed to catch 100 stock-sized large reservoirs, the median number of samples needed to palmetto bass and Walleyes was generally high, often reach RSE25 was approximately 23; however, to achieve an approaching or exceeding 50 samples at the 50th percentile RSE25 in 75% of surveys, over 50 electrofishing stations (Figure 4). Generally, fewer than 20 gill nets were needed to would be needed. Fewer samples were needed to collect 100 reach an RSE25 for Channel Catfish in most samples in stock-length Largemouth Bass in ponds and small reservoirs medium and large reservoirs. Similarly, 10–15 gill nets were compared with larger reservoirs (Figure 4), where the number needed to obtain an RSE25 for half the time in ponds and of samples needed to collect 100 fish was highly variable. This small reservoirs. About 20–40 gill nets were needed to collect was evidenced by interquartile ranges for medium and large 100 stock-sized Channel Catfish, regardless of reservoir size, reservoirs of 88 and 64 samples, respectively. In medium and in at least 50% of samples (Figure 4). More than 20 samples large reservoirs, approximately 20–45 stations were needed to were usually needed to achieve an acceptable precision of collect 100 stock-length fish in 50% of surveys. Additionally, catch rates for palmetto bass in any size of reservoir at the EVALUATION OF STANDARDIZED SAMPLING TECHNIQUES 1215

100 100 Ponds Small

80 80

60 60

40 40

20 20

0 0 LMB BGL CRAPPIE CCF PBASS WAE LMB BGL CRAPPIE CCF PBASS WAE 100 100 Medium Large Relative standardRelative error 80 80

60 60

40 40

20 20

0 0 LMB BGL CRAPPIE CCF PBASS WAE LMB BGL CRAPPIE CCF PBASS WAE

FIGURE 1. Box plots of RSE of CPUE estimates for Largemouth Bass, Bluegills, crappies, Channel Catﬁsh, palmetto bass, and Walleyes from ponds (0–20 acres) and small (21–250 acres), medium (251–1,000 acres), and large reservoirs (>1,000 acres) in Kansas, 2010–2013. Each data point represents an RSE of the sample for a species in a given year. Lower and upper fences are 25th and 75th percentiles, and the median is in between. Bars represent 10th and 90th percentiles. Horizontal reference line represents RSE25.

50th percentile (Figure 3). Further, at the 75th percentile, over Our results are similar to those reported in a study examin- 40 gill-net deployments were needed to achieve an RSE25 for ing sampling precision and sample size requirements in Texas palmetto bass. The number of samples needed to precisely reservoirs (Dumont and Schlechte 2004). In that study, the sample Walleyes varied widely, but in large reservoirs, around only target species that exhibited a median RSE of 25% for 20 gill nets were needed to reach an RSE25 in 50% of samples. the CPUE of stock-length ﬁsh was Largemouth Bass.

Downloaded by [Department Of Fisheries] at 19:15 20 July 2015 Although median Largemouth Bass RSE was below 25% for all sizes of Kansas reservoirs, substantial variation in CPUE DISCUSSION was exhibited in medium and large reservoirs. Electrofishing With the exception of Channel Catfish in medium reser- data used for this study were collected during the day, and voirs and Largemouth Bass in all reservoir sizes, an RSE25 research has suggested that night electrofishing catch rates and was not achieved at the 50th percentile for target fishes with precision are higher than those collected during the day (Para- the current number of samples in ponds and small and large gamian 1989; Dumont and Dennis 1997). If biologists using Kansas reservoirs from 2010 to 2013. In large reservoirs, day electrofishing wish to improve data quality, night electro- which received more sampling effort, the median RSE of fishing may be used. Additionally, in situations where collect- CPUE was below 25% for Largemouth Bass, crappies, Chan- ing precise Largemouth Bass samples is important, decreasing nel Catfish, and Walleye. However, reaching an acceptable electrofishing segment length (e.g., 5 min) may increase the level of precision 50% of the time for these target species may number of gear deployments and improve sample precision, not be acceptable to fisheries managers. Additionally, mini- especially in instances where shoreline length may preclude mum recommended sample sizes for adequately describing large sample segments. Although Miranda et al. (1996) indi- size structure were rarely collected using current sampling cated that sample duration is inversely related to the variability protocols. of CPUE, given a fixed time allocation, more small samples 1216 KOCH ET AL.

300 300 Ponds Small

250 250

200 200

150 150

100 100

50 50

0 0 LMB BGL CRAPPIE CCF PBASS WAE LMB BGL CRAPPIE CCF PBASS WAE 800 800 Medium Large

700 700

Number of fish collected 600 600

500 500

400 400

300 300

200 200

100 100

0 0 LMB BGL CRAPPIE CCF PBASS WAE LMB BGL CRAPPIE CCF PBASS WAE

FIGURE 2. Box plots of the number of stock-length fish collected per sampling event for Largemouth Bass, Bluegills, crappies, Channel Catfish, palmetto bass, and Walleyes from ponds (0–20 acres) and small (21–250 acres), medium (251–1,000 acres), and large reservoirs (>1,000 acres) in Kansas, 2010–2013. Each data point represents stock-length catch of the sample for a species in a given year. Lower and upper fences are 25th and 75th percentiles, and the medianisin between. Bars represent 10th and 90th percentiles. Horizontal reference line represents 100 stock-length fish.

are better than a few large ones (Miranda and Boxrucker with multiple gear types. These corroborating findings suggest 2009). that refinement of sampling strategies is needed for Morone Similar to our study, Dumont and Schlechte (2004) indi- spp. Until then, fisheries managers should be aware of poten- cated the precision of catch rates for target species collected in tial shortcomings of temperate bass catch data and make deci- gill and trap nets generally did not meet precision objectives. sions accordingly.

Downloaded by [Department Of Fisheries] at 19:15 20 July 2015 Additionally, catches of target species in that study did not Channel Catfish catch rates were estimated with relatively generally meet thresholds for the minimum number of fish in high precision in our study regardless of reservoir size. How- Texas. Dumont and Schlechte (2004) reported that Texas also ever, our objective for catching 100 stock-length Channel Cat- used reservoir size to determine the number of gear deploy- fish was rarely met using standard protocols. Substantial ments. Currently in Kansas, 20 gill nets are prescribed as a sampling effort, beyond what is practical, would be required minimum effort for the largest reservoirs. Our results indicated to meet this objective. In situations where large numbers of that 20 nets will generally precisely sample Channel Catfish Channel Catfish are difficult to obtain with gill nets, biologists and Walleyes in large reservoirs; although, more are required might consider the use of baited, tandem hoop nets. Studies for palmetto bass. As many as 25 gill-net samples are needed indicate that catch rates of hoop nets in lentic environments to precisely sample palmetto bass in large reservoirs 50% of can be high (Michaletz and Sullivan 2002; Neely and Dumont the time; although, 80 may be needed in smaller impound- 2011; Stewart and Long 2012). Sullivan and Gale (1999) indi- ments. Several studies suggest difficulty in precisely sampling cated baited hoop nets caught more Channel Catfish per per- Morone spp. Wilde (1995) indicated that as many as 50 gill sonnel-hour than did gill nets. Although collecting large nets were needed to precisely sample temperate basses in numbers of Channel Catfish with hoop nets may achieve some Texas reservoirs. Additionally, Neumann et al. (1995) objectives, Michaletz and Sullivan (2002) indicated that reported difficulty obtaining precise samples of Morone spp. obtaining precise measures of CPUE using baited, tandem EVALUATION OF STANDARDIZED SAMPLING TECHNIQUES 1217

100 100 Ponds Small

80 80

60 60

40 40

20 20

0 0 LMB BGL CRAPPIE CCF PBASS WAE LMB BGL CRAPPIE CCF PBASS WAE 100 100 Medium Large

80 80

Number of Number samples needed to reach RSE25 60 60

40 40

20 20

0 0 LMB BGL CRAPPIE CCF PBASS WAE LMB BGL CRAPPIE CCF PBASS WAE

FIGURE 3. Box plots of sampling effort needed to achieve an RSE25 of CPUE effort for Largemouth Bass, Bluegills, crappies, Channel Catfish, palmetto bass, and Walleyes in ponds (0–20 acres) and small (21–250 acres), medium (251–1,000 acres), and large reservoirs (>1,000 acres) in Kansas. Each data point represents the number of deployments needed to achieve an RSE25. Units of effort are 10-min electrofishing stations for Largemouth Bass, one trap-net-night for Blue- gills and crappies, and one gill-net-night for Channel Catfish, palmetto bass, and Walleyes. Lower and upper fences are 25th and 75th percentiles, and the median is in between. Bars represent the 10th and 90th percentiles.

hoop nets was difﬁcult and that from 12 to 50 series deploy- examination of size structure. If collecting large numbers of ments were required to reach acceptable levels of precision. Walleyes is important to managers, other sampling methods Neely and Dumont (2011) reported that 5–18 series were (Isermann and Parsons 2011) should be considered. For required to reach an RSE25 with baited, tandem hoop nets. instance, surface gill nets resulted in higher CPUE of Walleyes

Downloaded by [Department Of Fisheries] at 19:15 20 July 2015 Buckmeier and Schlechte (2009) reported that hoop net series compared with benthic gill-net sets in Lake Erie (Isbell and provided accurate size structure data and consistent relative Rawson 1989). Additionally, researchers have reported suc- abundance estimates for adult Channel Catfish. These findings cessfully capturing Walleyes in trap nets, especially in shallow suggest that in instances where gill nets are not providing fish- water during spawning periods (Beard et al. 1997; Kocovsky eries scientists with sufficient data, a hoop net series may pro- and Carline 2001). Regardless of the sampling method, biolo- vide an alternative Channel Catfish sampling method. gists should be aware of the biases of each sampling method. Precision of Walleye CPUE was less than RSE25 more than In most cases, trap-netting protocols did not allow biolo- 50% of the time in large reservoirs, where most quality Wall- gists to collect precise catch data or collect enough fish for eye populations occur in Kansas. Low catch rates and high meaningful size structure analysis for Bluegills and crappies variability likely limit the value of information obtained in Kansas reservoirs without considerably more sampling regarding Walleyes in smaller reservoirs in Kansas. In most effort per impoundment. Due to logistic limitations for most instances, more samples were required to obtain an RSE25 Kansas biologists, this represents at least 3–4 d of sampling than to collect 100 stock-length Walleyes. These results sug- effort. Such a time investment is likely more than most biolo- gest that in cases where sampling time and resources are lim- gists would find acceptable for most ponds and small reser- ited, biologists should consider collecting precise catch data voirs. Additionally, relatively few samples collected at least rather than capturing large numbers of Walleyes for 100 stock-length Bluegills or crappies; although, fewer 1218 KOCH ET AL.

100 100 Ponds Small

80 80

60 60

40 40

20 20

0 0 LMB BGL CRAPPIE CCF PBASS WAE LMB BGL CRAPPIE CCF PBASS WAE

100 100 Medium Large

80 80

60 60 Number of samplesNumber needed to collect 100 stock-sized fish

40 40

20 20

0 0 LMB BGL CRAPPIE CCF PBASS WAE LMB BGL CRAPPIE CCF PBASS WAE

FIGURE 4. Box plots of sampling effort needed to collect 100 stock-length individuals of Largemouth Bass, Bluegills, crappies, Channel Catfish, palmetto bass, and Walleyes in ponds (0–20 acres) and small (21–250 acres), medium (251–1,000 acres), and large reservoirs (>1,000 acres) in Kansas. Each data point represents the number of deployments needed to collect 100 stock-length individuals. Units of effort are 10-min electrofishing stations for Largemouth Bass, one trap-net-night for Bluegills and crappies, and one gill-net-night for Channel Catfish, palmetto bass, and Walleyes. Lower and upper fences are 25th and 75th percentiles, and the median is in between. Bars represent the 10th and 90th percentiles.

samples were generally required to do so compared with those summer when hoop nets captured more White Crappies than required to reach acceptable levels of precision. Biologists did trap nets (Muoneke et al. 1993). Similarly, unbaited, tan- should determine which objectives (i.e., precision of CPUE dem hoop nets provided higher catch rates of Bluegills and

Downloaded by [Department Of Fisheries] at 19:15 20 July 2015 estimates or number of fish for size structure representation) crappies than standard trap nets in Iowa; however, catch rate are most important for a given impoundment. If high precision precision of the gears were similar (M. Flammang, Iowa of catch rates is important to biologists, the use of other meth- Department of Natural Resources, unpublished data). ods for sampling Bluegills may be appropriate; however, Har- Although standard protocols did not generally meet precision din and Conner (1992) indicated difficulty in obtaining precise and total catch goals for Bluegills and crappies in Kansas, fish- catches of Bluegills via electrofishing in Florida. Additionally, eries managers should set sampling objectives and consider Schultz and Haines (2005) suggested that fall trap-netting pro- using alternative sampling methods (e.g., multigear vided more accurate information about harvestable-length approaches) that meet specific objectives. Bluegills than did electrofishing in a large Kansas impound- Our results indicate that current sampling effort expended ment. For crappies, otter trawls can provide higher catch rates by KDWPT with standard methods for sampling North Ameri- and precision than trap nets in Florida lakes (Allen et al. can freshwater fishes is generally not providing precise esti- 1999). In Tennessee reservoirs, electrofishing collected larger mates of CPUE for stock-length fish. Additionally, the crappies than did trap-netting; however, fall trap nets were rec- minimum numbers of fish recommended to adequately ommended by those investigators for examination of year- describe size structure were usually not collected. Kansas class strength (Sammons et al. 2002). In an Oklahoma reser- Department of Wildlife, Parks, and Tourism is aware of the voir, hoop nets and trap nets had similar catch rates, except in shortcomings of some sampling procedures and is making EVALUATION OF STANDARDIZED SAMPLING TECHNIQUES 1219

efforts to improve the quality of future data. For example, communication, and reduces bias that is inherent in biological effort in smaller Kansas reservoirs should be increased if data collection (Bonar et al. 2009a). Benefits of standardiza- objectives of fisheries managers include obtaining precise tion are numerous (Bonar and Hubert 2002; Bonar et al. catch rates and adequate descriptions of size structure. Objec- 2009a), and our results by no means diminish the importance tive-based sampling should also be considered instead of the of standardization. We believe sample standardization should current protocol of determining minimum sampling efforts be continued in Kansas, and elsewhere, to promote develop- based on reservoir size. Along similar lines, KDWPT should ment of long-term databases and encourage regional compari- also consider basing minimum sampling effort on historical sons of fish populations. However, managers should recognize catch data (Quist et al. 2009) rather than on reservoir size. that data collected from standardized sampling regimes might Because our results indicate that data obtained from some not be sufficient to address particular questions. In these sampling protocols are often too variable to obtain precise esti- instances, managers might wish to supplement standardized mates of CPUE, biologists should focus on collecting numbers samples with alternative gears or different strategies for of fish needed to adequately describe size structure, as samples deployment of standard gears (e.g., time of year, time of day, that do not reach an RSE25 may only be useful for assessing fixed stations). large changes in relative abundance (Dumont and Schlechte 2004). If precise estimates of CPUE are desired, more effort should be expended on fewer reservoirs, which may represent a shift from the paradigm of sampling each reservoir annually. ACKNOWLEDGMENTS Managers may be better served if they collect a large number We thank K. Austin, C. Bever, J. Goeckler, S. Lynott, D. of statistically powerful samples from reservoirs on a 2- or 3- Nygren, R. Marteney, and S. Steffen for thoughtful discussions year cycle instead of obtaining marginal data every year. associated with this project. We also thank all KDWPT per- Finally, protocols for the use of supplemental, nonstandard sonnel that collected field data from 2010 to 2013. We thank gears are in place in Kansas (Marteney et al. 2010), and these S. Dumont, M. Quist, and five anonymous reviewers for alternative gears should be considered to supplement standard reviews that greatly improved the manuscript. Funding for this sampling procedures when objectives are not being met. study was provided by Kansas Department of Wildlife, Parks, Although standardized sampling often failed to achieve and Tourism. Fish sampling was supported by Federal Aid specified objectives, it should be noted that fisheries not spe- grant F-22-R. cifically managed for targeted species were included in analyses (i.e., fisheries of poor quality, low density, or of little interest to managers were included). Target species were chosen in part because of their widespread range in Kansas, but REFERENCES Allen, M. S., M. M. Hale, and W. E. Pine III. 1999. Comparison of trap nets small or remnant populations in some reservoirs might have and otter trawls for sampling Black Crappie in two Florida lakes. North influenced our findings. As such, biologists should recognize American Journal of Fisheries Management 19:977–983. that collection of high-quality data from marginal fisheries Anderson, R. O., and R. M. Neumann. 1996. Length, weight, and associated may not be feasible and that monitoring fisheries where sam- structural indices. Pages 447–482 in B. R. Murphy and D. W. Willis, edi- pling targets can be reached or viable populations exist should tors. Fisheries techniques, 2nd edition. American Fisheries Society, Bethesda, Maryland. be prioritized. These results highlight the inefficiency of Beard, T. D. Jr., S. W. Hewett, Q. Yang, R. M. King, and S. J. Gilbert. 1997. attempting to collect precise data from low-abundance popula- Prediction of angler catch rates based on Walleye population density. North

Downloaded by [Department Of Fisheries] at 19:15 20 July 2015 tions. One approach for limiting these inefficient samples American Journal of Fisheries Management 17:621–627. might be to classify reservoirs by the quality of important rec- Bonar, S. A., S. Contreras-Balderas, and A. C. Iles. 2009a. An introduction to reational fisheries. For example, a particular impoundment standardized sampling. Pages 1–12 in S. A. Bonar, W. A. Hubert, and D. W. Willis, editors. Standard methods for sampling North American freshwater might support a popular Walleye fishery but only maintains fishes. American Fisheries Society, Bethesda, Maryland. low-abundance Channel Catfish and palmetto bass populations Bonar, S. A., W. A. Hubert, and D. W. Willis, editors. 2009b. Standard meth- that garner little interest. A manager could classify this reser- ods for sampling North American freshwater fishes. American Fisheries voir as a Walleye fishery and sample until objectives were met Society, Bethesda, Maryland. for Walleyes but not necessarily for Channel Catfish or pal- Bonar, S. A., and W. A. Hubert. 2002. Standard sampling of inland fish: benefits, challenges and a call for action. Fisheries 27(3):10–16. metto bass. This might result in the manager ceasing gill-net Brown, M. L., and D. J. Austen. 1996. Data management and statistical techni- sampling effort when sample objectives for Walleyes were ques. Pages 17–62 in B. R. Murphy and D. W. Willis, editors. Fisheries met even though relatively poor samples were collected for techniques, 2nd edition. American Fisheries Society, Bethesda, Maryland. Channel Catfish and palmetto bass. Ultimately, this would Buckmeier, D. L., and J. W. Schlechte. 2009. Capture efficiency and size lead to more appropriate sampling effort allocation but still selectivity of Channel Catfish and Blue Catfish sampling gears. North American Journal of Fisheries Management 29:404–416. retain sample quality for target species. Dodds, W. K., E. Carney, and R. T. Angelo. 2006. Determining ecoregional Standardization of sampling methods allows easy spatial reference conditions for nutrients, Secchi depth, and chlorophyll a in Kansas and temporal comparison, encourages data sharing and lakes and reservoirs. Lake and Reservoir Management 22:151–159. 1220 KOCH ET AL.

Dumont, S. C., and J. A. Dennis 1997. Comparison of day and night electro- management of catfish: the second international symposium. American fishing in Texas reservoirs. North American Journal of Fisheries Manage- Fisheries Society, Symposium 77, Bethesda, Maryland. ment 17:939–946. Neumann, R. M., C. S. Guy, and D. W. Willis. 1995. Precision and size struc- Dumont, S. C., and B. C. Neely. 2011. A proposed change to palmetto bass ture of juvenile percichthyid samples collected with various gears from proportional size distribution length categories. North American Journal of Lake Texoma. North American Journal of Fisheries Management 15:956– Fisheries Management 31:722–725. 962. Dumont, S. C., and W. Schlechte. 2004. Use of resampling to evaluate a sim- Noble, R. L., D. J. Austen, and M. A. Pegg. 2007. Fisheries management study ple random sampling design for general monitoring of fishes in Texas reser- design considerations. Pages 31–50 in C. S. Guy and M. L. Brown, editors. voirs. North American Journal of Fisheries Management 24:408–416. Analysis and interpretation of freshwater fisheries data. American Fisheries Gabelhouse, D. W. Jr. 1984. A length-categorization system to assess fish Society, Bethesda, Maryland. stocks. North American Journal of Fisheries Management 4:273–285. Paragamian, V. L. 1989. A comparison of day and night electrofishing: size Gustafson, K. A. 1988. Approximating confidence intervals for indices of fish structure and catch per unit effort for Smallmouth Bass. North American population size structure. North American Journal of Fisheries Management Journal of Fisheries Management 9:500–503. 8:139–141. Pope, K. L., R. M. Neumann, and S. D. Bryan. 2009. Warmwater fish in small Hardin, S., and L. L. Conner. 1992. Variability of electrofishing crew effi- standing waters. Pages 13–25 in S. A. Bonar, W. A. Hubert, and D. W. Wil- ciency, and sampling requirements for estimating reliable catch rates. North lis, editors. Standard methods for sampling North American freshwater American Journal of Fisheries Management 12:612–617. fishes. American Fisheries Society, Bethesda, Maryland. Isbell, G. L., and M. R. Rawson. 1989. Relations of gill-net catches of Wal- Quist, M. C., K. I. Bonvecchio, and M. S. Allen. 2009. Statistical analysis and leyes and angler catch rates in Ohio waters of western Lake Erie. North data management. Pages 171–190 in S. A. Bonar, W. A. Hubert, and D. W. American Journal of Fisheries Management 9:41–46. Willis, editors. Standard methods for sampling North American freshwater Isermann, D. A., and B. G. Parsons. 2011. Harvest regulations and sam- fishes. American Fisheries Society, Bethesda, Maryland. pling. Pages 403–422 in B. A. Barton, editor. Biology, management, R Core Development Team. 2013. R: a language and environment for statisti- and culture of Walleye and Sauger. American Fisheries Society, cal computing. R Foundation for Statistical Computing, Vienna. Bethesda, Maryland. Robson, D. S., and H. A. Regier. 1964. Sample size in Peterson mark–recap- Kocovsky, P. M., and R. F. Carline. 2001. Dynamics of the unexploited Wall- ture experiments. Transactions of the American Fisheries Society 93:215– eye population of Pymatuning Sanctuary, Pennsylvania, 1997–1998. North 226. American Journal of Fisheries Management 21:178–187. Sammons, S. M., D. A. Isermann, and P. W. Bettoli. 2002. Variation in popula- Marteney, R., L. Aberson, C. Bever, S. Lynott, J. Reinke, and M. Shaw. 2010. tion characteristics and gear selection between Black and White crappies in Standard fish survey techniques for small lakes and reservoirs, 5th edition. Tennessee reservoirs: potential effects on management decisions. North Kansas Department of Wildlife, Parks, and Tourism, Emporia. American Journal of Fisheries Management 22:863–869. Michaletz, P. H., and K. P. Sullivan. 2002. Sampling Channel Catfish with tan- Schultz, R. D., and D. E. Haines. 2005. Comparison of seasonal Bluegill catch dem hoop nets in small impoundments. North American Journal of Fisheries rates and size distributions obtained with trap nets and electrofishing in a Management 22:870–878. large, heated impoundment. North American Journal of Fisheries Manage- Miranda, L. E. 1993. Sample sizes for estimating and comparing proportion- ment 25:220–224. based indices. North American Journal of Fisheries Management 13:383– Stewart, D. R., and J. M. Long. 2012. Precision of Channel Catfish catch esti- 386. mates using hoop nets in larger Oklahoma reservoirs. North American Jour- Miranda, L. E. 2007. Approximate sample sizes require to estimate length dis- nal of Fisheries Management 32:1108–1112. tributions. Transactions of the American Fisheries Society 136:409–415. Sullivan, K. P., and G. M. Gale. 1999. A comparison of Channel Catfish catch Miranda, L. E. 2009. Standardizing electrofishing power for boat electrofish- rates, size distributions, and mortalities using three different gears in a Mis- ing. Pages 223–230 in S. A. Bonar, W. A. Hubert, and D. W. Willis, editors. souri impoundment. Pages 293–300 in E. R. Irwin, W. A. Hubert, C. F. Standard methods for sampling North American freshwater fishes. Ameri- Rabeni, H. L. Schramm Jr., and T. Coon, editors. Catfish 2000: proceedings can Fisheries Society, Bethesda, Maryland. of the international symposium. American Fisheries Society, Symposium Miranda, L. E., and J. Boxrucker. 2009. Warmwater fish in large standing 24, Bethesda, Maryland. waters. Pages 29–42 in S. A. Bonar, W. A. Hubert, and D. W. Willis, edi- Vokoun, J. C., and C. F. Rabeni, and J. S. Stanovick. 2001. Sample-size tors. Standard methods for sampling North American freshwater fishes. requirements for evaluating population size structure. North American Jour- American Fisheries Society, Bethesda, Maryland. nal of Fisheries Management 21:660–665. Downloaded by [Department Of Fisheries] at 19:15 20 July 2015 Miranda, L. E., W. D. Hubbard, S. Sangare, and T. Holman. 1996. Optimizing Wilde, G. R. 1995. Gill net sample size requirements for temperate basses, electrofishing sample duration for estimating relative abundance of Large- shads, and catfishes. Proceedings of the Annual Conference of Southeastern mouth Bass in reservoirs. North American Journal of Fisheries Management Association of Fish and Wildlife Agencies 47(1993):588–595. 16:324–331. Wilde, G. R., and W. L. Fisher. 1996. Reservoir fisheries sampling and experi- Muoneke, M. I., O. E. Maughan, and C. C. Henry. 1993. Comparative capture mental design. Pages 397–409 in L. E. Miranda and D. R. DeVries, editors. efficiencies of frame and hoop nets for White Crappie (Pomoxis annularis Multidimensional approaches to reservoir fisheries management. American Rafinesque). Fisheries Research 18:231–240. Fisheries Society, Symposium 16, Bethesda, Maryland. Neely, B. C., and S. C. Dumont. 2011. Effect of soak duration on precision of Willis, D. W., and B. R. Murphy. 1996. Planning for sampling. Pages 1–15 in Channel Catfish catch with baited, tandem hoop nets. Pages 557–562 in P. B. R. Murphy and D. W. Willis, editors. Fisheries techniques, 2nd edition. H. Michaletz and V. H. Travnichek, editors. Conservation, ecology, and American Fisheries Society, Bethesda, Maryland. This article was downloaded by: [Department Of Fisheries] On: 20 July 2015, At: 19:15 Publisher: Taylor & Francis Informa Ltd Registered in England and Wales Registered Number: 1072954 Registered office: 5 Howick Place, London, SW1P 1WG

North American Journal of Fisheries Management Publication details, including instructions for authors and subscription information: http://www.tandfonline.com/loi/ujfm20 Absence of Handling-Induced Saprolegnia Infection in Juvenile Rainbow Trout with Implications for Catch- and-Release Angling Martin Schwabea, Thomas Meineltb, Thuy M. Phanb, Steven J. Cookecd & Robert Arlinghausae a Integrative Fisheries Management, Department for Crop and Animal Sciences, Faculty of Agriculture and Horticulture, Humboldt-Universität zu Berlin, 10115 Berlin, Germany b Department of Aquaculture and Ecophysiology, Leibniz-Institute of Freshwater Ecology and Inland Fisheries, 12587 Berlin, Germany c Institute of Environmental Science, Carleton University, 1125 Colonel By Drive, Ottawa, Ontario K1S 5B6, Canada Click for updates d Fish Ecology and Conservation Physiology Laboratory, Department of Biology, Carleton University, Ottawa, Ontario, Canada K1S 5B e Department of Biology and Ecology of Fishes, Leibniz-Institute of Freshwater Ecology and Inland Fisheries, 12587 Berlin, Germany Published online: 24 Nov 2014.

To cite this article: Martin Schwabe, Thomas Meinelt, Thuy M. Phan, Steven J. Cooke & Robert Arlinghaus (2014) Absence of Handling-Induced Saprolegnia Infection in Juvenile Rainbow Trout with Implications for Catch-and-Release Angling, North American Journal of Fisheries Management, 34:6, 1221-1226, DOI: 10.1080/02755947.2014.959673 To link to this article: http://dx.doi.org/10.1080/02755947.2014.959673

PLEASE SCROLL DOWN FOR ARTICLE

Taylor & Francis makes every effort to ensure the accuracy of all the information (the “Content”) contained in the publications on our platform. However, Taylor & Francis, our agents, and our licensors make no representations or warranties whatsoever as to the accuracy, completeness, or suitability for any purpose of the Content. Any opinions and views expressed in this publication are the opinions and views of the authors, and are not the views of or endorsed by Taylor & Francis. The accuracy of the Content should not be relied upon and should be independently verified with primary sources of information. Taylor and Francis shall not be liable for any losses, actions, claims, proceedings, demands, costs, expenses, damages, and other liabilities whatsoever or howsoever caused arising directly or indirectly in connection with, in relation to or arising out of the use of the Content. This article may be used for research, teaching, and private study purposes. Any substantial or systematic reproduction, redistribution, reselling, loan, sub-licensing, systematic supply, or distribution in any form to anyone is expressly forbidden. Terms & Conditions of access and use can be found at http:// www.tandfonline.com/page/terms-and-conditions North American Journal of Fisheries Management 34:1221–1226, 2014 Ó American Fisheries Society 2014 ISSN: 0275-5947 print / 1548-8675 online DOI: 10.1080/02755947.2014.959673

MANAGEMENT BRIEF

Absence of Handling-Induced Saprolegnia Infection in Juvenile Rainbow Trout with Implications for Catch- and-Release Angling

Martin Schwabe* Integrative Fisheries Management, Department for Crop and Animal Sciences, Faculty of Agriculture and Horticulture, Humboldt-Universitat€ zu Berlin, 10115 Berlin, Germany Thomas Meinelt and Thuy M. Phan Department of Aquaculture and Ecophysiology, Leibniz-Institute of Freshwater Ecology and Inland Fisheries, 12587 Berlin, Germany Steven J. Cooke Institute of Environmental Science, Carleton University, 1125 Colonel By Drive, Ottawa, Ontario K1S 5B6, Canada; and Fish Ecology and Conservation Physiology Laboratory, Department of Biology, Carleton University, Ottawa, Ontario, Canada K1S 5B Robert Arlinghaus Integrative Fisheries Management, Department for Crop and Animal Sciences, Faculty of Agriculture and Horticulture, Humboldt-Universitat€ zu Berlin, 10115 Berlin, Germany; and Department of Biology and Ecology of Fishes, Leibniz-Institute of Freshwater Ecology and Inland Fisheries, 12587 Berlin, Germany

Abstract Globally, as many as two-thirds of angled fish are Catch-and-release angling is common in recreational fisheries. released (Cooke and Cowx 2004) because of legal require- During handling and dehooking, fish are subjected to stress and dermal injuries, which may result in infections by pathogens after ments or for voluntary reasons (Arlinghaus et al. 2007). the fish is released. The objective of this study was to evaluate the Any catch-and-release event involves handling fish during consequences of common handling practices used by anglers on the landing and dehooking processes. Handling may take Downloaded by [Department Of Fisheries] at 19:15 20 July 2015 the postrelease behavior and fate, particularly the susceptibility place using bare hands, towels, or other equipment used by Oncorhynchus mykiss to disease, of undersized Rainbow Trout . the angler to constrain the fish (Danylchuk et al. 2008). Behavior immediately following capture and subsequent release of fish was examined in a 40-L container, and long-term fate was Priortodehooking,fishmayalsobeexposedtoalanding studied for 2 weeks in tanks incubated with Saprolegnia parasitica net, which may remove some of the mucus from fish (Bar- zoospores. Trout were behaviorally impaired as indicated by the thel et al. 2003; De Lestang et al. 2008). Rough handling ease of being netted following the simulated fight associated with may affect survival postrelease due to the physical damage catch and release, but there were no further behavioral impacts to the epidermis resulting in infections, predation, or other due to subsequent handling. None of the Rainbow Trout developed fungal infections nor was any significant mortality observed forms of stress-induced mortality (Barthel et al. 2003; after 2 weeks; only 1 out of 137 fish died. Our data indicate that Colotelo and Cooke 2011). Relatively few studies have juvenile hatchery-reared Rainbow Trout have a high resilience to examined how fish behave after release (reviewed by Saprolegnia infection handling-induced stress. Donaldson et al. 2008), and aside from the duration of air exposure (Danylchuk et al. 2007; Arlinghaus et al. 2009)

*Corresponding author: [email protected] Received February 9, 2014; accepted August 25, 2014

1221 1222 SCHWABE ET AL.

no research known to the authors has evaluated the impact range for Rainbow Trout (water temperature D 18.7 § 0.5C of common handling practices by anglers on fish behavior [mean § SD], dissolved oxygen D 6.6 § 0.7 mg/L; Schmidt postrelease. It is pertinent to study how fish behavior and 1998). Fish were fed a maintenance diet of commercial trout probability of infection with pathogens vary with common pellets (i.e., 1.5% of body weight per day: Kaushik and Gomes handling techniques used by anglers during dehooking. 1988). One ubiquitous pathogen that may infect freshwater fish is During the experiments, fish were individually netted with Saprolegnia parasitica (Noga 1993; van West 2006). Sapro- a knotless landing net (Barthel et al. 2003) from the holding legnia parasitica is able to parasitize all species of freshwater tank (any excess fish captured were released immediately back fish, especially salmonids (Noga 1993). Infections are visible into the tank), identified using a PIT tag reader (TROVAN), as white to grey mold on the fish’s skin and often lead to death subsequently placed in a bucket filled with 40 L of water, and of the fish (Howe and Stehly 1998; van West 2006). In salmo- chased by tail pinching for 30 s to simulate a fight on the rod nids, predisposing factors leading to an infection include epi- as described in other simulated catch-and-release studies (e.g., dermal wounds (Harai and Hoshiai 1993) and physiological Kieffer 2000; Arlinghaus et al. 2009). Subsequently, fish were stress (e.g., Noga 1993). As handling by anglers after capture exposed to one of three randomly allocated handling treat- might interact with initial physical exhaustion during the fight, ments or one of two controls. Control groups included (1) net- it is possible that some released fish become infected with S. ting only (N D 28) and (2) netting and chasing (N D 27). parasitica on damaged epithelial layers. Treatments included netting, chasing, and handling with (3) Particularly under conditions of high predation threat post- dry hand (N D 28), (4) wet hand (N D 27), and (5) dry towel release (Cooke and Philipp 2004), rapid resumption of normal (N D 27). Each handling event was 5 s in duration (similar to behavior after catch and release is important to safeguard sur- experimental data in Danylchuk et al. 2007) that mimicked vival. Exercise alone already impairs the swimming of fish rapid dehooking of an undersized fish without actually hook- after release (Schreer et al. 2005; Arlinghaus et al. 2009), and ing the fish; this controlled for any confounding effects of any handling during dehooking may aggravate the situation. hooking injury to directly link the results to handling. Fish in Due to a lack of studies available it is currently unknown the unhandled groups were individually caught with a landing whether this actually is the case. net from the holding tank and either not chased at all or The objective of this study was to evaluate whether han- exposed to physical exhaustion by tail pinching only. dling practices used by anglers during the dehooking event To measure postrelease behavior, a behavioral impairment influence behavior and mortality of juvenile Rainbow Trout. score immediately after the simulated release event was We hypothesized that harsh handling practices would result in recorded following Gingerich et al. (2007). Impairment was Saprolegnia infections in juvenile Rainbow Trout. The impact defined as a state that would render the fish an easy prey to of handling on behavior postrelease was studied as well to bet- predators, as indicated by impairment of swimming or escape. ter understand how handling may affect fish behavior after All fish were individually released into a white container (40 their release. L; upper and lower diameters, 60 and 49 cm, respectively; height, 38 cm), where the behavior of each fish postrelease was observed for a total period of 300 s. At each of five inter- METHODS vals (0, 30, 60, 180, and 300 s), the fish’s behavioral state was Experimental fish were undersized juvenile Rainbow Trout categorized and assigned a value as follows: (mean TL D 19.4 § 1.49 cm [mean § SD], n D 137) obtained

Downloaded by [Department Of Fisheries] at 19:15 20 July 2015 from a commercial hatchery (Forellenzucht Werdermuhle,€ (0) fish appeared normal with normal swimming behavior, Niemegk, Germany). Juvenile fish were chosen because they (1) fish appeared normal with visible signs of fatigue, have to be released in most jurisdictions and may thus suffer (2) fish swam rapidly and randomly, and somewhat erratically, particularly from angling-induced handling. Fish were accli- (3) fish rested on the bottom of the tank, mated to experimental conditions over a period of 2 weeks in (4) fish displayed partial lateral equilibrium loss, an 800-L (1.1 £ 1.1 £ 1.0 m), aerated tank with a water inflow (5) fish displayed complete equilibrium loss (either floating on rate of »8 L/h. To ensure identification of each individual, fish its side or upside down). were tagged with 1.2 £ 0.2-cm PIT tags (TROVAN, model ID-100B; EURO I.D., Weilerswist, Germany), inserted into The five observed values were subsequently summed to yield a the dorsal musculature left of the dorsal fin 1 week prior to the release behavioral condition score after 300 s. Afterwards, as a experiment. Before experimentation, treatments were assigned second measure of behavioral impairment, the ease of netting at random to PIT tag numbers (see below). Behavior or mortal- the fish from the bucket was scored as either difficult (1) or ity of Rainbow Trout is not significantly influenced by PIT easy (0). Easy was scored when the fish was netted with the tags (Prentice and Park 1984). first attempt. The difficult category meant that the fish evaded Water quality during both acclimation and the subsequent the netting event actively, such that more than one netting experiment was measured daily and was within the optimal attempt was needed to capture the fish. This second behavioral MANAGEMENT BRIEF 1223

metric was used to simulate predator evasion ability postrelease. It was assumed that an easily netted fish would probably fall victim to a natural predator. To assess infections with S. parasitica and survival, after the 300-s observation period all fish from each treatments were evenly allocated into three replicate tanks so that in each tank all treatments were jointly represented to control for tank effects. Each tank had a capacity of 400 L (1.1 £ 1.1 £ 0.5 m; tap water inflow, »8 L/h; temperature, 18.7 § 0.5C; dissolved oxygen, 6.6 § 0.7 mg/L) and was stocked with trout at a density of 3.58 kg/m3 (tank 1, 46 individuals; tank 2, 46 individuals; tank 3, 45 individuals). To standardize the density of S. parasitica spores across tanks, each tank was inoculated with S. parasitica at a concentration of 4 zoospores/mL, which was slightly lower than the concentration targeted by Howe and Stehly (1998) in their study on Rainbow Trout. The strain FIGURE 1. Index of behavioral impairment summarized over 300 s postre- of S. parasitica originated from the American Type Culture lease in juvenile Rainbow Trout displayed as mean score (§SE). State of Collection (ATCC 22284) and was recultivated at least behavior was categorized as: (0) fish appeared normal with normal swimming behavior, (1) fish appeared normal with visible signs of fatigue, (2) fish swam 20 times. According to ATCC this strain of S. parasitica is rapidly and randomly in the tank, (3) fish rested on the bottom of the tank, (4) highly pathogenic, and Gieseker et al. (2006) were able to fish displayed partial lateral equilibrium loss, (5) fish displayed complete equi- infect Rainbow Trout using this strain. librium loss. Note that similar patterns existed at all time periods, but to save Seeds of hemp together with pieces from the pure culture of space results are only shown for the 300-s integrated measure. Significant dif- S. parasitica were put into glucose–yeast extract (GY) agar ferences in mean score are marked with an asterisk (*) (ANOVA and Tukey HSD post hoc test: F D 11.25, df D 4, P < 0.05). culture media (Howe and Stehly 1998; Meinelt et al. 2007). After 3 d, the seeds were moved into a liquid culture media homogeneous variances. The ease of netting was evaluated (50 ppm carbon), and after another 2 d, 1.2 L of a liquid cul- using a chi-square test using a standard contingency table ture media was produced and equally distributed to the three that compared expected and revealed frequencies. tanks. Before the beginning of the experiment, the number of spores was determined by streaking culture samples onto the GY agar and counting the number of spores. A 2-week obser- RESULTS vation period was chosen to capture delayed mortalities as a There were significant differences in postrelease behavioral result of handling and possibly infection by the fungus. After impairment scores among treatments (ANOVA and Tukey 2 weeks, each surviving fish was visually assessed for mortal- HSD post hoc test: F D 11.25, df D 4, P < 0.05; Figure 1). ity or infection by S. parasitica as either absent or present. Only the control group captured by landing net showed less Two weeks should be sufficient for S. parasitica to infect the behavioral impairment compared with all other groups (Fig- fish (Willoughby and Pickering 1977). In other experimental ure 1). A similar pattern was apparent based on the ease of net- infection studies, trout developed visible infections after only ting: the control group not chased was significantly more 2 to 5 d (Howe and Stehly 1998; Fregeneda Grandes et al. difficult to net than all other treatment groups (x2 D 32.92,

Downloaded by [Department Of Fisheries] at 19:15 20 July 2015 2001). Barnes et al. (2004) showed that for water hardness df D 4, P < 0.05; Figure 2). (calcium) of about 50–150 mg/L, 70–80 h were needed to During the 2-week observation period after release, we observe microscopic growth of S. diclina. Hardness of the observed only one mortality that occurred in the chased-only water used in the present study was similar: calcium, 94– control group. At the termination of the experiment, no fish 147 mg/L; magnesium, 8.4–18.8 mg/L (data retrieved from had visible signs of skin damage or S. parasitica infection. local drinking water industry). Therefore, an observation Therefore, the mortality rate as a function of handling treat- period of 2 weeks seemed valid. ment by dry hand, wet hand, or dry towel was 0%. Statistical comparisons of the behavioral impairment score upon release among treatments were conducted in SPSS (version 9.0.1) by a one factorial ANOVA at P < 0.05 DISCUSSION that assumed the ordinal behavioral scale has equidistant The results of this study suggest that handling does not qualities similar to a rating scale in survey research con- affect the behavior of juvenile hatchery Rainbow Trout fol- ducted with human subjects. Tests on homogeneity of varian- lowing a catch-and-release event; rather, exercise fatigue ces were conducted using Levene tests, and normality was appeared to be responsible for the statistical differences among tested using a Shapiro–Wilk test. Subsequently, a Tukey hon- treatments in terms of their postrelease behavior. These find- estly significantly different (HSD) post hoc test was used for ings agree with a number of other studies that show fish are 1224 SCHWABE ET AL.

of this study, such effects would likely be unrelated to common handling practices in Rainbow Trout during dehooking and rather be caused by physical exhaustion or air exposure. Mucus loss and damage of epithelia is to be expected during handling in any fish, e.g., handling with towels during dehooking or with dry hands. Indeed, Murchie et al. (2009) observed a maximum loss of slime in Bonefish Albula vulpes during handling by bare hands or gloved hands. Thus, the handling we exposed the juvenile Rainbow Trout to should also have caused some mucus abrasion. Abrasion promotes infection with S. parasitica in Rainbow Trout (Fregeneda Grandes et al. 2001). Because our handling did not result in Saproleg- nia infection within a 2-week observation period, it appears that handling-induced abrasion. common in recreational angling, may not be severe enough to cause adverse effects in Rainbow Trout under reasonably benign environmental conditions. FIGURE 2. The ease of netting juvenile Rainbow Trout after 300 s of chas- Our results can be explained by the fact that hatchery fish ing. The distributional differences were significant (x2 D 32.92, df D 4, P < 0.05). often show a reduced stress response (e.g., Weil et al. 2001) and overall are more resistant to handling (Rapp et al. 2012) than wild fish. Alternative explanations include the existence behaviorally impaired immediately after release when they of environmental conditions within an optimal range (Schmidt have been physically exhausted during the fight (e.g., Thomp- 1998) that prevented a further deterioration of the fish’s condi- son et al. 2008; Arlinghaus et al. 2009). This study adds to tion, use of generally well-conditioned fish (e.g., good nutri- this literature by reporting that rapid handling with limited air tional status from artificial feed), and short air exposure of exposure of 5 s does not aggravate the behavioral status of only 5 s during handling. It is also conceivable that our experi- Rainbow Trout postrelease relative to exhaustion alone. mental design failed to establish a large enough and long-last- The experiments were conducted at mild environmental ing enough zoospore load because we incubated the tanks just conditions within the thermal optimum range for Rainbow once and not repeatedly over the entire 2-week observation Trout (although at the upper part of the range), avoiding rapid period as was conducted, for example, by Howe and Stehly changes in water temperature or extended air exposure. Gin- (1998). Howe and Stehly (1998) also used a much harsher gerich et al. (2007) showed that air exposure and water tem- abrasion treatment involving daily dewatering of experimental perature may interact to cause behavioral changes and tanks and letting Rainbow Trout flop for 1 min over an abra- potentially large, delayed mortality in centrarchid fish. sive paper. Yet, despite this harsh treatment only 25% of those Increased handling time can also increase recovery times fish became infected with S. parasitica within 14 d at a contin- (White et al. 2008; Arlinghaus et al. 2009; Rapp et al. 2012) uous concentration of >5 zoospores/mL (Howe and Stehly and result in air exposure thresholds after which postrelease 1998). In their study, it was only when abrasion of the intensity swimming performance declines (Schreer et al. 2005). It just described was matched with temperature stress (simulated

Downloaded by [Department Of Fisheries] at 19:15 20 July 2015 appeared that our handling protocols did not exceed physio- by rapidly dropping water temperatures) that the mortalities logical thresholds for juvenile Rainbow Trout. Therefore, rose to 75% over 14 d. Similarly, Fregeneda Grandes et al. despite the lack of effects of rapid handling on behavior and (2001) exposed batches of 15 Rainbow Trout to shaking for mortality in our study, it is possible that negative effects of 2 min in a fan-shaped landing net made of a 6-mm knotless handling may be evident under actual angling conditions. For mesh at zoospore concentrations of 200–300 zoospores/ mL example, if handling coincides with longer air exposure than over 3 d with a subsequent 12-d observation period. Yet, tested in the present study or when release events occur during despite this harsh treatment, infection rates and mortalities unfavorable (e.g., too warm or too cold) temperature condi- varied between 0% and 100% depending on the Saprolegnia tions (Arlinghaus et al. 2007), an effect may be evident. More- morphotype and isolate. In light of these data, the lack of over, in systems where juvenile trout are released in the infections in response to a single, relatively mild, handling presence of cannibalistic larger trout or other predators, mor- event in our study does not appear unreasonable. Possibly, our tality postrelease might be more elevated than what was treatment was simply not harsh enough and led to only superﬁ- revealed in the present tank experiments because the physical cial mucus loss that was insufﬁcient to enable infection by impairment after exhaustion can limit escape abilities and Saprolegnia. Similarly, Pope et al. (2007) reported a very high result in predation postrelease (Danylchuk et al. 2007; survival rate of, on average, 96% in mouth-hooked Rainbow reviewed by Raby et al. 2014). However, in light of the result Trout under experimental laboratory conditions and found no MANAGEMENT BRIEF 1225

relationship between mortality and number of hooking events Cooke, S. J., and D. P. Philipp. 2004. Behavior and mortality of caught- (up to four). Hence, under similar environmental conditions as and-released bonefish (Albula spp.) in Bahamian waters with implica- those in the present study, juvenile Rainbow Trout of hatchery tions for a sustainable recreational fishery. Biological Conservation 118:599–607. origin are likely to tolerate moderate injury caused by handling Danylchuk, A. J., A. Adams, S. J. Cooke, and C. D. Suski. 2008. An evaluation and by hooking (Pope et al. 2007) without suffering infections of the injury and short-term survival of bonefish (Albula spp.) as influenced with S. parasitica and without experiencing significant by a mechanical lip-gripping device used by recreational anglers. Fisheries mortality. Research 93:248–252. The present study joins a number of recent examples that Danylchuk, S. E., A. J. Danylchuk, S. J. Cooke, T. L. Goldberg, J. Koppelman, and D. P. Philipp. 2007. Effects of recreational angling on the post-release have reported very low or even zero postrelease mortality in behavior and predation of bonefish (Albula vulpes): The role of equilibrium various freshwater fishes (e.g., Pope and Wilde 2004; Rapp status at the time of release. Journal of Experimental Marine Biology and et al. 2012), including Rainbow Trout (Pope et al. 2007). Neg- Ecology 346:127–133. ligible hooking mortality should generally be expected when De Lestang, P., R. Griffin, Q. Allsop, and B. S. Grace. 2008. Effects of two dif- environmental conditions are benign and the injury induced by ferent landing nets on injuries to the barramundi Lates calcarifer, an iconic Australian sport fish. North American Journal of Fisheries Management the capture event is mild (Arlinghaus et al. 2007; Pope et al. 28:1911–1915. 2007). However, based on the present work alone one should Donaldson, M. R., R. Arlinghaus, K. C. Hanson, and S. J. Cooke. 2008. not conclude that handling is generally not problematic for Enhancing catch-and-release science with biotelemetry. Fish and Fisheries fish. There is still merit for anglers to not artificially prolong 9:79–105. the fight to minimize behavioral impairment and to use wet FAO (Food and Agriculture Organization of the United Nations). 2012. Recre- ational fisheries. FAO, Technical Guidelines for Responsible Fisheries 13, hands or dehook fish entirely under water because this will Rome. Available: www.fao.org/fishery/publications/technical-guidelines/ minimize the stress response of the fish and maintain the fish en. (August 2013). in a better condition upon release (FAO 2012). If these actions Fregeneda Grandes, J. M., M. Fernandez Diez, and J. M. Aller Gancedo. 2001. are followed and deep hooking is avoided, the survival of Experimental pathogenicity in Rainbow Trout, Oncorhynchus mykiss (Wal- undersized Rainbow Trout of hatchery origin can approach baum), of two distinct morphotypes of long-spined Saprolegnia isolates obtained from wild brown trout, Salmo trutta L., and river water. Journal of 100%. Fish Diseases 24:351–359. Gieseker, C. M., S. G. Serfling, and R. Reimschuessel. 2006. Formalin treatment to reduce mortality associated with Saprolegnia parasitica in Rainbow ACKNOWLEDGMENTS Trout, Oncorhynchus mykiss. Aquaculture 253:120–129. This study was conducted under animal care protocol B09- Gingerich, A. J., S. J. Cooke, K. C. Hanson, M. R. Donaldson, C. T. Hasler, C. 11 under regulations of Carleton University, Ottawa, Ontario. D. Suski, and R. Arlinghaus. 2007. Evaluation of the interactive effects of air exposure duration and water temperature on the condition and survival The finalization of this study was financially supported by the of angled and released fish. Fisheries Research 86:169–178. Federal Ministry of Education and Research within the Besatz- Harai K., and G. Hoshiai. 1993. Characteristics of two Saprolegnia species iso- fisch Project (grant 01UU0907 to R.A.) and by the School of lated from Coho Salmon with saprolegniosis. Journal of Aquatic Animal Fishery and Aquatic Sciences, University of Florida, for sup- Health 5:115–118. porting a sabbatical of R.A. Feedback by one reviewer and the Howe, G. E., and G. Stehly. 1998. Experimental Infection of Rainbow Trout with Saprolegnia parasitica. Journal of Aquatic Animal Health 10:397–404. associate editor helped improve the paper. Kaushik, S. J., and E. F. Gomes. 1988. Effect of frequency of feeding on nitro- gen and energy balance in Rainbow Trout under maintenance conditions. Aquaculture 73:207–216. REFERENCES Kieffer, J. D. 2000. Limits to exhaustive exercise in fish. Comparative Bio- chemistry and Physiology Part A Molecular Integrative Physiology

Downloaded by [Department Of Fisheries] at 19:15 20 July 2015 Arlinghaus, R., S. J. Cooke, J. Lyman, D. Policansky, A. Schwab, C. D. Suski, S. G. Sutton, and E. B. Thorstad. 2007. Understanding the complexity of 126:161–179. catch-and-release in recreational fishing: an integrative synthesis of global Meinelt, T., A. Paul, T. M. Phan, E. Zwirnmann, A. Kruger,€ A. Wienke, and C. knowledge from historical, ethical, social, and biological perspectives. E. W. Steinberg. 2007. Reduction in vegetative growth of the water mold Reviews in Fisheries Science 15:75–167. Saprolegnia parasitica (Coker) by humic substance of different qualities. Arlinghaus, R., T. Klefoth, S. J. Cooke, A. J. Gingerich, and C. D. Suski. 2009. Aquatic Toxicology 83:93–103. Physiological and behavioural consequences of catch-and-release angling Murchie, K. J., S. E. Danylchuk, C. E. Pullen, E. Brooks, A. D. Shultz, C. D. on Northern Pike (Esox lucius L.). Fisheries Research 97:223–233. Suski, A. J. Danylchuk, and S. J. Cooke. 2009. Strategies for the capture Barnes, M. E., A. C. Gabel, D. J. Durben, T. R. Hightower, and T. J. Berger. and transport of bonefish, Albula vulpes, from tidal creeks to a marine 2004. Changes in water hardness influence colonization of Saprolegnia research laboratory for long-term holding. Aquaculture Research 40:1538– diclina. North American Journal of Aquaculture 66:222–227. 1550. Barthel, B. L., S. J. Cooke, C. D. Suski, and D. P. Philipp. 2003. Effects of Noga E. J., 1993. Water mold infections of freshwater fish. Annual Review of landing net mesh type on injury and mortality in a freshwater recreational Fish Diseases 3:291–304. fishery. Fisheries Research 63:275–282. Pope, K. L., and G. R. Wilde. 2004. Effect of catch-and-release angling on Colotelo, A. H., and S. J. Cooke. 2011. Evaluation of common angling- growth of largemouth bass, Micropterus salmoides. Fisheries Management induced sources of epithelial damage for popular freshwater sport fish using and Ecology 11:39–44. fluorescein. Fisheries Research 109:217–224. Pope K. L., G. R. Wilde, and D. W. Knabe. 2007. Effect of catch-and-release Cooke, S. J., and I. G. Cowx. 2004. The role of recreational fishing in global angling on growth and survival of Rainbow Trout, Oncorhynchus mykiss. fish crises. BioScience 54:857–859. Fisheries Management and Ecology 14:115–121. 1226 SCHWABE ET AL.

Prentice, E., and D. Park. 1984. A study to determine the biological feasibility Thompson, L. A., S. J. Cooke, M. R. Donaldson, K. C. Hanson, A. J. Ginger- of a new fish tagging system. Annual Report to the Bonneville Power ich, T. Klefoth, and R. Arlinghaus. 2008. Physiology, behavior, and survival Administration, Project 83-19, Portland, Oregon. of angled and air-exposed Largemouth Bass. North American Journal of Raby, G. D., J. R. Packer, A. J. Danylchuk, and S. J. Cooke. 2014. The under- Fisheries Management 28:1059–1068. studied and underappreciated role of predation in the mortality of fish van West, P. 2006. Saprolegnia parasitica, an oomycete pathogen with a fishy released from fishing gears. Fish and Fisheries 15:489–505. appetite: new challenges for an old problem. Mycologist 20:99–104. Rapp,T.,J.Hallermann,S.J.Cooke,S.K.Hetz,S.Wuertz,andR. Weil, L., T. Barry, and J. Malison. 2001. Fast growth in Rainbow Trout is cor- Arlinghaus. 2012. Physiological and behavioural consequences of cap- related with a rapid decrease in post-stress cortisol concentrations. Aquacul- tureandretentionincarpsacksonCommonCarp(Cyprinus carpio ture 193:373–380. L.), with implications for catch-and-release recreational fishing. Fisher- White, A. J., J. F. Schreer, and S. J. Cooke. 2008. Behavioral and physiological ies Research 125–126:57–68. responses of the congeneric Largemouth (Micropterus salmoides) and Schmidt G. W. 1998. Bewirtschaftung der forelle. [Cultivation of trout.] Pages Smallmouth bass (M. dolomieu) to various exercise and air exposure dura- 265–467 in W. Sch€aperclaus and M. von Lukowicz, editors. Lehrbuch der tions. Fisheries Research 89:9–16. teichwirtschaft. [Textbook of aquaculture]. Ullmer, Berlin. Willoughby L., and A. Pickering. 1977. Viable Saprolegniaceae spores on the Schreer, J. F., D. M. Resch, M. L. Gately, and S. J. Cooke. 2005. Swimming epidermis of the salmonid fish Salmo trutta and Salvelinus alpinus. Transac- performance of Brook Trout after simulated catch-and-release angling: tions of the British Mycological Society 68:91–95. looking for air exposure thresholds. North American Journal of Fisheries Management 25:1513–1517. Downloaded by [Department Of Fisheries] at 19:15 20 July 2015 This article was downloaded by: [Department Of Fisheries] On: 20 July 2015, At: 19:15 Publisher: Taylor & Francis Informa Ltd Registered in England and Wales Registered Number: 1072954 Registered office: 5 Howick Place, London, SW1P 1WG

North American Journal of Fisheries Management Publication details, including instructions for authors and subscription information: http://www.tandfonline.com/loi/ujfm20 A Simulation-Based Evaluation of In-Season Management Tactics for Anadromous Fisheries: Accounting for Risk in the Yukon River Fall Chum Salmon Fishery Matthew J. Catalanoa & Michael L. Jonesa a Quantitative Fisheries Center, Department of Fisheries and Wildlife, Michigan State University, 153 Giltner Hall, East Lansing, Michigan 48824, USA Published online: 01 Dec 2014.

Click for updates

To cite this article: Matthew J. Catalano & Michael L. Jones (2014) A Simulation-Based Evaluation of In-Season Management Tactics for Anadromous Fisheries: Accounting for Risk in the Yukon River Fall Chum Salmon Fishery, North American Journal of Fisheries Management, 34:6, 1227-1241, DOI: 10.1080/02755947.2014.951801 To link to this article: http://dx.doi.org/10.1080/02755947.2014.951801

PLEASE SCROLL DOWN FOR ARTICLE

Taylor & Francis makes every effort to ensure the accuracy of all the information (the “Content”) contained in the publications on our platform. However, Taylor & Francis, our agents, and our licensors make no representations or warranties whatsoever as to the accuracy, completeness, or suitability for any purpose of the Content. Any opinions and views expressed in this publication are the opinions and views of the authors, and are not the views of or endorsed by Taylor & Francis. The accuracy of the Content should not be relied upon and should be independently verified with primary sources of information. Taylor and Francis shall not be liable for any losses, actions, claims, proceedings, demands, costs, expenses, damages, and other liabilities whatsoever or howsoever caused arising directly or indirectly in connection with, in relation to or arising out of the use of the Content. This article may be used for research, teaching, and private study purposes. Any substantial or systematic reproduction, redistribution, reselling, loan, sub-licensing, systematic supply, or distribution in any form to anyone is expressly forbidden. Terms & Conditions of access and use can be found at http:// www.tandfonline.com/page/terms-and-conditions North American Journal of Fisheries Management 34:1227–1241, 2014 C American Fisheries Society 2014 ISSN: 0275-5947 print / 1548-8675 online DOI: 10.1080/02755947.2014.951801

ARTICLE

A Simulation-Based Evaluation of In-Season Management Tactics for Anadromous Fisheries: Accounting for Risk in the Yukon River Fall Chum Salmon Fishery

Matthew J. Catalano*1 and Michael L. Jones Quantitative Fisheries Center, Department of Fisheries and Wildlife, Michigan State University, 153 Giltner Hall, East Lansing, Michigan 48824, USA

Abstract Salmon fisheries are managed under uncertainties in abundance, population dynamics, and fishery implementation. These uncertainties create risk, which can be accounted for using probabilistic harvest control rules. Probabilistic control rules ensure management actions are consistent with the level of acceptable risk. Within a fishing season, managers seek to meet escapement objectives by using preseason forecasts and in-season data to set fishery openings and catch targets. We used a stochastic simulation model of the Yukon River fall Chum Salmon Oncorhynchus keta fishery to evaluate the effects on fishery performance of (1) the degree of in-season risk tolerance and (2) two methods for estimating daily abundance projections. We defined risk in terms of the probability of failing to meet escapement objectives, which is consistent with the emphasis on meeting escapement objectives in Alaska’s Sustainable Salmon Fisheries Policy. The analysis revealed expected trade-offs between harvest and escapement objectives. However, risk tolerance did not strongly influence fishery performance mainly because outcomes were highly uncertain due to process errors. Average subsistence harvest decreased from 97,000 to 88,000 Chum Salmon and the probability of failing to meet escapement goals in four of the five most recent years decreased from 0.04 to 0.01 when a risk-averse approach was implemented. However, when preseason forecasts were relatively low and close to thresholds where no fishing would be allowed, then a risk-averse approach increased the probability of meeting escapement objectives. A Bayesian approach to estimating the projected run abundance performed similarly to a simpler approach that relied entirely on the preseason forecast until the average first quarter point of the run. Our results suggest that assessment approaches and the degree of in-season risk tolerance are likely less important in determining fishery performance for Yukon River Fall Chum Salmon than are the escapement goals used to manage the fishery. Downloaded by [Department Of Fisheries] at 19:15 20 July 2015 In-season salmon management decisions must be made on such as escapement goals (Bue et al. 2009; Hilsinger et al. the basis of uncertain run size projections and with imperfect 2009). For example, in-season escapement data cannot be used management control over outcomes. Run size projections are in managing harvest on the Yukon River because harvesting has uncertain because daily cumulative indices of abundance are concluded long before salmon reach spawning tributaries where poor predictors of the total run size due to interannual varia- escapement is measured (Bue et al. 2009; Hilsinger et al. 2009). tion in run timing and size and observation errors (Hyun et al. In such cases, managers must rely on imprecise preseason fore- 2005). Management control over outcomes is imperfect, espe- casts and noisy in-season data (Bue et al. 2009; Hilsinger et al. cially in large systems such as the Yukon and Kuskokwim rivers 2009). in Alaska, where time lags exist between when harvest man- The uncertainties of in-season salmon management create agement decisions are made and when the outcomes of those the risk of negative outcomes resulting from failure to meet decisions can be measured relative to management objectives objectives under a particular set of management decisions. Thus,

*Corresponding author: [email protected] 1Present address: School of Fisheries, Aquaculture, and Aquatic Sciences, Auburn University, Auburn, Alabama 36849, USA. Received November 18, 2013; accepted July 28, 2014 1227 1228 CATALANO AND JONES

managers and policy makers should explicitly consider uncer- minimum threshold than would a risk-tolerant manager before tainty and risk in evaluating the performance of potential man- opening the fishery. This definition of risk-aversion employs the agement strategies (Peterman and Anderson 1999; Peterman narrow interpretation of the term “risk” to refer to the possibil- 2004). In general, risk can be defined as the possibility of suf- ity of an unfavorable biological outcome, which could occur if fering harm or loss and thus is a function of both the probability escapement goals are not met. We use this definition of risk here- that an unfavorable outcome occurs and the degree of unfavor- after because it is consistent with current Alaska salmon policy, ability if it were to occur. Several studies have incorporated which prioritizes meeting escapement objectives. For fisheries uncertainty into evaluations of in-season salmon management of Yukon River fall Chum Salmon Oncorhynchus keta (YFC), strategies using stochastic simulation models (Link and Peter- managers have recently become more risk-averse in prosecuting man 1998; Robb and Peterman 1998; Su and Adkison 2002). In the commercial fishery following extremely low returns from their most sophisticated form, these simulation approaches are 1997 to 2002. During that time, escapement goals were not met often referred to as a Management Strategy Evaluation (MSE). in 4 of 6 years (Bue et al. 2009). Assessing whether a risk-averse, An MSE uses closed loop simulations of a fish population along in-season management approach for YFC is expected to better with the entire management process to evaluate the long-term meet multiple objectives (escapement, subsistence, commercial) performance of management strategies (i.e., data collection pro- than is a more-risk tolerant approach will be informative for fu- cedures, assessment models, harvest control rules) under the un- ture management of this stock. More generally, investigating the certain states of nature, which are often represented by Bayesian influence of the degree of risk tolerance associated with a proba- posterior distributions (Butterworth et al. 1997; Cooke 1999; bilistic harvest control rule would be relevant to the management Smith et al. 1999). Thus, an MSE is a useful tool for objec- of other fisheries, as few such studies exist. tively evaluating trade-offs associated with different manage- Salmon managers should carefully consider how probabil- ment strategies in a multi-objective context and in the face of ity distributions for projected abundance are estimated, given uncertainty (Walters and Martell 2004; Butterworth 2007). the importance of these distributions when using probabilis- In addition to employing MSE approaches in policy eval- tic control rules. In-season salmon abundance projections typ- uation, uncertainty and risk can be incorporated directly into ically take into account a preseason forecast and cumulative harvest control rules using a probability-based approach to set- data collected during the run using formal or informal methods ting catch levels (Prager et al. 2003; Shertzer et al. 2008). The (Walters and Buckingham 1975; Fried and Hilborn 1988; Hyun probability-based approach uses harvest control rules that en- et al. 2005). As an example of a formal method, Bayes’ theo- sure target catch levels are consistent with the level of risk rem could be employed where the preseason forecast is treated acceptable to managers (Shertzer et al. 2008). For instance, if as the prior on the projected abundance and the cumulative in- managers are interested in maintaining the stock above some season information is treated as the data (Fried and Hilborn threshold abundance level Xlim, the harvest rate could be set to 1988). However, most managers use informal procedures that satisfy rely on “rules of thumb” and intuition derived from experience. For example, they may choose to rely on a preseason forecast until some arbitrary estimated proportion of the run has entered P(Xˆ t ≤ Xlim) ≤ P*, (1) the river. Regardless of the method, reliance on in-season data ˆ and certainty in the run size generally increases throughout the where Xt is the projected stock abundance at time t under a run as in-season data become more informative relative to the proposed harvest rate and P* represents the level of risk ac- preseason forecast. Understanding the influence on fishery per- ceptable to managers. Hereafter we refer to the acceptable risk Downloaded by [Department Of Fisheries] at 19:15 20 July 2015 formance of in-season risk tolerance and methods for estimating level (P*) as the degree of “risk tolerance.” This probabilistic the projected run size would aid salmon mangers in better meet- approach differs from traditional rules that set target catches ing fishery objectives. In this study, we used an MSE model of (or harvest rates) based on point estimates of projected abun- the in-season dynamics of the YFC run and fishery to evaluate dance. One advantage of the probabilistic approach is that man- the relative performance of (1) different levels of risk tolerance agers and stakeholders must decide upon and declare a value for in prosecuting the fishery and (2) two different methods for P*, which increases transparency in the management process weighting preseason forecasts versus in-season data in the esti- (Shertzer et al. 2008). Probabilistic control rules are becom- mation of the probability distribution of the projected run during ing more common in marine stock assessments due to increased the season. adoption of the precautionary approach to fisheries management (Shertzer et al. 2008). Despite the growing use of probabilistic control rules, we Yukon River Fall Chum Salmon are unaware of their use for in-season salmon management. Fall Chum Salmon are one of two distinct groups of Chum Most salmon populations are managed for spawning escape- Salmon stocks that spawn in the Yukon River; the other be- ment goals. In this context, a risk-averse manager would require ing the summer Chum Salmon stock. Yukon River fall Chum a smaller probability (P*) that escapement is less than some Salmon differ from the summer stock in spawning locations, MANAGEMENT TACTICS FOR ANADROMOUS FISHERIES 1229

during the season using two test ﬁsheries (Big Eddy–Middle Mouth, rkm 40; Mountain Village, rkm 139) and sonar at Pilot Station (rkm 197).

METHODS We constructed a model that simulated a self-sustaining Chum Salmon population that return annually to spawn at ages 3–5 and are exposed to a downstream commercial and an upstream subsistence fishery in the Yukon River. A single iteration of the model comprised a 50-year time series of salmon runs and associated outcomes for harvest and escapement, with each year consisting of a 90-d simulation of in-season run dynamics and harvesting. Thus, by linking the escapement and future returns over multiple years, the model accounted for the long-term effects of management strategies on future returns. The model was iterated 500 times to generate a distribution of outcomes for a particular management strategy given FIGURE 1. The Yukon River study area indicating fishing districts (Y1–Y6) and major spawning areas of the Yukon River fall Chum Salmon stock (adapted four sources of uncertainty (Williams 1997): (1) structural: un- from Bue et al. 2009). certainty in the values of YFC life history parameters, (2) observation: the collection of imperfect data, (3) implementation: differences between desired and actual harvests, and (4) process: body morphology, and timing of the run. The YFC enter the natural variation in YFC population dynamics. river from mid-July to mid-September and spawn in spring-fed The model.—For each year of a simulation, a run size (Ry) reaches mostly located in the upper Yukon River basin including was generated as the sum of the brood year recruitments re- the Tanana River and in tributaries of the upper main stem in turning at ages 3–5, which was determined from the number of Canada (Figure 1; Bue et al. 2009), whereas the summer stock spawners (Sy) in past generations via a Ricker stock–recruitment enters the river during June and July and spawns in tributaries function: of the lower half of the drainage (Bue et al. 2009). The flesh of YFC has a high fat content because they need to cope with 5 −βSy−a greater migration distances than do the summer stocks, which Ry = αy−a Sy−ae py−a,a εy−a, (2) makes YFC valuable for subsistence and commercial fisheries. a=3 The Yukon River is divided into six management districts (Figure 1). Historically, most of the commercial harvest has where αy is the brood-year-specific productivity, β is the density- occurred in the three lower districts (Y1–Y3), which extend dependent parameter, py,a are brood year proportions returning from the mouth of the river to river kilometer (rkm) 482. Some at age, and εy are annual recruitment deviations. Three sources of commercial harvest for roe also occurs in the lower Tanana process uncertainty were incorporated in the stock–recruitment River and in district Y5 of the main stem near the Tanana function: (1) decadal-scale variation in productivity associated

Downloaded by [Department Of Fisheries] at 19:15 20 July 2015 River confluence, as well as in the Canadian main stem. Subsis- with oceanic conditions by modeling the αy as a random walk 2 tence harvest occurs throughout the main stem and tributaries with variance σα (Peterman et al. 2003), (2) additional interan- but is concentrated in upstream districts, where fish are har- nual recruitment variation (i.e., noise) by modeling εy as lognor- σ2 vested for many uses with an emphasis on food for sled team mally distributed with a mean value of 1 and a variance of R, dogs. and (3) brood year variation in maturation schedules by assum- The fishery is managed by the Alaska Department of Fish ing the py,a are Dirichlet distributed with parameter vector (γ3, and Game (ADFG) and federal agencies for a hierarchical set of γ4, γ5). Structural uncertainty was incorporated by assuming objectives with escapement goals as the top priority, followed by parameter values were drawn from the posterior distributions of subsistence fishing opportunities, and lastly commercial harvest. an age-structured, stock–recruitment analysis for Yukon River A basin-wide escapement goal range of 300,000–600,000 Chum fall Chum Salmon (Jones et al. 2011). The stock–recruitment Salmon exists (based on stock–recruitment analysis) as well as modeling approach followed closely a Bayesian age-structured, several goals for individual tributaries and a negotiated goal state-space model by Fleischman et al. (2013). for escapement into Canada. Escapement is measured on the The fate of Chum Salmon as they migrated upstream was sim- primary spawning tributaries, but as previously mentioned, these ulated using a simple box-car representation (Starr and Hilborn data cannot be used for in-season management due to time lags 1988) that divided the river into 30 reaches (r), each represent- between harvest and spawning. Instead, run strength is assessed ing one day (d) of upstream travel. The daily number of salmon 1230 CATALANO AND JONES

large enough that the current projection of the total run (Nˆ d,y) exceeded a given management threshold run size. Therefore, these quantities represented the link between the degree of risk tolerance and the prosecution of the harvest in our simulation. We used time-invariant threshold run sizes as defined in the current ADFG management plan for YFC (Bue et al. 2009). Under this plan the run size thresholds were no harvest if Nˆ d,y < 300,000 fish, restricted subsistence if Nˆ d,y ≥ 300,000, unrestricted subsistence if Nˆ d,y ≥ 500,000, and restricted commercial harvest if Nˆ d,y ≥ 600,000 fish. A given fishery was opened if the projected run abundance exceeded the corresponding management threshold with critical probability, P*f, y. Therefore, the control rules were

⎧ ∗ ⎪ 0ifP(Nˆ , < 300, 000) > P ⎪ d y sub,y ⎨ . ˆ < , ≤ ∗ 0 5ifP(Nd,y 300 000) Psub,y and Ksubd,y ∗ (5) ⎪ P(Nˆ d,y < 500, 000) > P , ⎩⎪ sub y ˆ < , > ∗ 1if P(Nd,y 500 000) Psub,y

and FIGURE 2. The daily proportions of the run entering the river (i.e., within- year run timing) from 28 observed Yukon River fall Chum Salmon run timings from 1980 to 2007 (thin lines; B. Borba, ADFG, unpublished data). The average ˆ < , > ∗ cumulative proportions of the run returning on each date (t¯ ) are indicated by 0ifP(Nd,y 600 000) Pcom,y d Kcomd,y ∗ , (6) the thick solid line. Thick dashed lines represent ± 1SD(σ ) around the mean 1ifP(Nˆ , < 600, 000) ≤ P , t¯d d y com y cumulative run proportions.

(Nr,d,y ) entering the first reach of the lower river (r = 1) was where P*f, y represents the degree of escapement risk tolerance assumed by managers with respect to prosecution of fishery f N1,d,y = Rytd,y, (3) in year y. Values closer to zero represent a risk-averse approach to management and values closer to 1.0 are more risk toler- where td,y is the proportion of the total run entering the river on ant. For example, under an assumed P*f,y of 0.5, harvest for a = day d of the run in year y (i.e., run timing; d td,y 1). Run particular fishery would be allowed only if the median of the timings were generated by resampling from 28 observed YFC run projection on a given day exceeded the corresponding har- run timings from 1980 to 2007 (B. Borba, ADFG, unpublished vest threshold for that fishery. The daily process of computing data; Figure 2). The observed run timings were a composite av- P(Nˆ d,y< threshold) and comparing it with P*f,y sought to mimic erage of the two lower-river test fisheries and Pilot Station sonar. a salmon manager’s daily decision process regarding whether

Downloaded by [Department Of Fisheries] at 19:15 20 July 2015 Run timings from the different projects were standardized to the to open a particular fishery given available data. We evaluated timing at the mouth by lagging the Mountain Village timing by the relative performance of different levels of managerial risk 2 d and the Pilot Station timing by 3 d. All individuals entering tolerance by simulating different approaches to setting values the river on the same day were assumed to move upstream at the of P*f,y (see later section Management strategies). same rate and experience commercial and subsistence harvest. Instantaneous fishing rates were updated daily as a function Daily reach-specific abundances were: of the most recent run projection (Nˆ d,y), the harvest policy, and expected subsistence utilization. We assumed commercial −Kcomd,y Fcomr,d,y −Ksubd,y Fsubr,d,y Nr+1,d+1,y = Nr,d,ye , (4) harvest occurred within the first 10 reaches of the river (i.e., first 10 d in river; 483 km; fishing districts Y1, Y2, and Y3) where Fcomr,d ,y and Fsubr,d ,y are instantaneous commercial and and subsistence harvest in reaches 11–30. Thus, Fcomr,d ,y = subsistence fishing rates (given as per day). A probabilistic har- 0forr > 10 and Fsubr,d ,y = 0forr ≤ 10. This amount of vest control rule was used to set fishery openings and closures spatial segregation between fisheries does not actually occur via the Kcomd,y and Ksubd,y parameters, which determined the in the Yukon River, but our simplification reflects the general degree to which a particular fishery was open or closed on a pattern in which commercial harvest is concentrated in the lower given day of the run. A fishery was opened (Kcomd,y or Ksubd,y river and subsistence fisheries are more prevalent upstream. Fish > 0) on a given day of the run only if there was a probability surviving to reach 31 were counted as escapement, estimated as MANAGEMENT TACTICS FOR ANADROMOUS FISHERIES 1231

follows: Similar to commercial harvest, subsistence harvest rates were assumed equal among reaches 11–30 on a given day of the Sy = N31,d,y. (7) run and the fishing effort was assumed to be evenly distributed d over the run. Implementation uncertainty in subsistence harvesting was incorporated via the εd,y, which were lognormally Escapement estimates were then used to generate future distributed errors with a mean of 1.0 and an SD of σsub. recruits (equation 2). Implementation error variance for the subsistence fishery was Commercial harvest rates were computed such that the har- simply the variance around the recent mean subsistence catch vest summed over the entire run would match a target catch if because subsistence catch was not strongly related to run size. harvest was allowed on all possible future days. The annual tar- For estimation of both commercial and subsistence implemen- get catch for the commercial fishery (Hd,y) was calculated as the tation uncertainty, we used only data from recent years (1992– point estimate of the projected run in excess of (i.e., constrained 2009) because of changes in markets and processing capacity to nonnegative) the commercial threshold and was updated each and in subsistence utilization since the mid-1970s. day of the run as follows: Estimates of the in-season run projection (Nˆ d,y ) were updated daily by combining a preseason forecast (Oˆ y ) with a run estimate Hd,y = max(0, Nˆ d,y − 600,000). (8) based on cumulative in-season data up through day d (Iˆd,y). Note that each of these estimates pertains to the total expected run Thus, the probabilistic control rule was concerned with for the year, not numbers of salmon entering the river on a whether the fishery was open on a particular day and not the particular day of the season. The point estimate of the preseason harvest rate applied; if the fishery was open, target harvest was forecast (Oˆ y; i.e., median forecasted run size) was generated based on the point estimate of the run projection. Target catch as a bias-corrected lognormal random deviate of the true run was set to zero if the threshold exceeded the run projection. The size: annual target catch was converted each day to a reach-specific instantaneous harvest rate by dividing the annual instantaneous σ2 Oˆ ∼ lognormal ln(R ) − O , σ2 , (11) harvest rate by the number of reaches (i.e., 10 reaches) in which y y 2 O each migrating cohort of fish was exposed to the commercial σ2 fishery according to the equation, where the forecast variance ( O ) was estimated from observed historical deviations between preseason forecasts and recon- ln 1 − Hd,y structed run sizes from 1990 to 2009 (ADFG, unpublished data; Nˆ d,y Fcom , , =− ε , , (9) Table 1). r d y 10 d y Daily run estimates from cumulative in-season data (Iˆd,y) were simulated to approximate estimates that would be ob- where implementation uncertainty was incorporated as lognor- tained from downriver test fisheries and Pilot Station sonar. mally distributed errors (εd,y) with a mean of 1.0 and an SD of The point estimate of the Iˆd,y was generated as an estimated σcom. Thus, we assumed commercial fishing rates equal for the cumulative passage to date past the downstream reach (cˆ , ) first 10 reaches of river on a given day and effort was evenly dis- d y divided by the average cumulative proportion of the run re- tributed across the run because we sought to mimic a manager’s turning to date (t¯ ) based on the 27 years of historical run- attempt to spread the harvest over the run. Implementation er- d timing data (Walters and Buckingham 1975; Hyun et al. 2005) as ror variance for the commercial fishery was obtained by fitting Downloaded by [Department Of Fisheries] at 19:15 20 July 2015 follows: linear relationships between reconstructed annual run size estimates and observed commercial harvests (Collie et al. 2012). cˆd,y The implementation error variance was the estimated error vari- Iˆd,y = . (12) t¯ ance of the linear models. d The target catch for the subsistence fishery was assumed to Cumulative passage estimates (cˆd,y) were generated as log- be the average amount necessary for subsistence (ANS; i.e., normal deviates of the true cumulative passage (i.e., d N1,d,y) 129,300; Bue et al. 2009) as determined by ADFG, which rep- with an observation error variance taken as the estimated vari- resents the average subsistence utilization in recent years. Sub- ance of deviations between annual passage estimates at Pilot sistence harvest rates were then computed as Station sonar and reconstructed run sizes from a basin-wide assessment model (Fleischman and Borba 2009; Table 1). ln 1 − ANS Variance of the daily in-season run estimates (σ2 )was Nˆ d,y −Hy Iˆd,y Fsub , , =− ε , , (10) r d y 20 d y a function of the cumulative passage estimate on day d (cˆd,y), the average cumulative proportion of the run return- where the harvest rate takes into account the expected down- ing on that day (t¯d ), and variance in the historical cumula- stream cumulative removals by the commercial fishery (H ). tive proportions returning (σ2 ; Walters and Buckingham 1975), y t¯d 1232 CATALANO AND JONES

TABLE 1. Parameter values used in the in-season simulation model. Upper and lower limits for the 95% credible intervals are shown for the posterior distribution of parameters obtained from a basin-wide, stock–recruitment analysis for Yukon River fall Chum Salmon (Jones et al. 2011). Structural uncertainty was incorporated into simulations by drawing parameter sets (one set for each model iteration) from the posterior distribution of the stock–recruitment parameters.

Lower limit of Upper limit of Model parameter Symbol Estimate 95% interval 95% interval Process error Alphaa α 2.63 1.06 4.87 Betaa β 9.10 × 10−07 3.10 × 10−07 1.64 × 10−06 a 2 Recruitment variance σ R 0.13 0.01 0.38 a 2 Alpha random walk variance σ α 0.12 0.01 0.40 a Proportion maturing at age 3 γ3 0.04 0.02 0.05 a Proportion maturing at age 4 γ4 0.70 0.66 0.73 a Proportion maturing at age 5 γ5 0.27 0.23 0.31 b Daily cumulative run proportions td,y seeFigure2 Observation error c 2 Preseason forecast variance σ H 0.38 d 2 Cumulative passage observation error variance σ c 0.04 b Average of daily cumulative run proportions t¯d seeFigure2 Variance of daily cumulative run proportionsb σ2 seeFigure2 t¯d Implementation error 2 Commercial implementation error variance σ com 0.51 2 Subsistence implementation error variance σ sub 0.13

a Jones et al. (2011). b From observed historical daily run proportions. c From observed deviations between historical forcasts and estimated run sizes. d Fleischman and Borba (2009).

determined as used a time-invariant, risk-neutral threshold for the subsistence fishery (P*sub,y = 0.5) and a conservative one for the commercial 2 σ2 σ2 = cˆ , ¯ ¯ fishery (P*com,y 0.1). State law gives subsistence fisheries pri- σ2 = d y td 1 + 2 td . (13) Iˆ , 4 ority over commercial fisheries; therefore, we wished to assess d y t¯ t¯d d whether management performance could be improved by adopting a more conservative threshold for the commercial fishery. ˆ Total run size projections (Nd,y) were updated daily by com- The third scenario used a risk-neutral threshold for both fisheries ˆ puting the weighted average of the preseason forecast (Od,y) (P* = 0.5), but switched to a conservative one (P* = 0.1) for ˆ f,y f,y and the in-season run estimates (Id,y): the commercial fishery when a “stock of concern” designation was declared by failing to meet the minimum escapement goal

Downloaded by [Department Of Fisheries] at 19:15 20 July 2015 Nˆ d,y = wd,y Oˆ y + (1 − wd,y)Iˆd,y, (14) in 4 of the last 5 years. This scenario was important to consider because managers may become more conservative in managing where wd,y is the weighting of the preseason forecast. We tested the commercial fishery if recent management performance (or two approaches for setting the wd,y (see Management strategies run size) has been poor. below). Fishery performance should be more sensitive to the degree Management strategies.—We evaluated the performance of of managerial risk tolerance when the run size is closest to the three risk tolerance scenarios, each representing a different ap- management thresholds. Therefore, we conducted an additional proach to setting risk tolerance (P*f,y) values. The first approach set of analyses to determine the influence of the degree of risk involved using a constant risk tolerance value between fisheries tolerance when run sizes were between 300,000 and 800,000. and over time. We evaluated a range of constant P*f,y values from We obtained the results of the individual year simulations from 0.5 to 0.1, with 0.5 representing a risk-neutral approach and 0.1 a each of the 500 model iterations (500 iterations × 50 years = risk-averse approach. The maximum P*f,y valueusedinallsce- per iteration 25,000 individual years). We then focused on narios was 0.5. Larger, more risk-tolerant values were unrealistic the subset of those years in which the preseason forecast was because they would allow harvesting when the point estimate between 300,000 and 800,000 salmon. Basing the subset on the (posterior median) of the run projection was less than a man- value of the imperfect forecast and not the true run size made agement threshold (Shertzer et al. 2008). The second scenario the analysis more relevant to managers, who do not know the MANAGEMENT TACTICS FOR ANADROMOUS FISHERIES 1233

true run size and must make decisions based on noisy presea- salmon and subsistence catch of 107,000 salmon (Figure 3a, b). son forecasts and in-season data. We divided the forecast range A constant risk-averse P*f,y of 0.1 for both fisheries resulted in (300,000–800,000) into five 100,000-fish intervals and assessed an average commercial catch of 179,000 salmon and subsistence differences in the performance indicators under different P*f,y catch of 103,000 salmon (Figure 3a, b). The probability of com- values and among intervals. mercial closure increased from 0.30 with a risk-neutral approach We evaluated two different methods for weighting the to 0.42 with a risk-averse approach (Figure 3c). The probability preseason forecast and cumulative in-season data in computing of meeting subsistence needs decreased as risk tolerance de- the total run projections (Nˆ d,y). The first method used Bayes’ creased (Figure 3d). Escapement increased from 429,000 with theorem to obtain the posterior distribution of the run projection a risk-neutral approach to 459,000 salmon with a risk-averse by weighting the preseason forecast and cumulative in-season approach (Figure 3e). The median probability of a stock of data inversely proportional to their variances (Fried and Hilborn concern designation was 0.0 for all scenarios and the upper 1988). The second method used the forecast exclusively (i.e., quartile decreased from 0.02 to 0.0 with increased risk-aversion wd,y = 1.0) until the day on which the first 25% of the run (Figure 3f). has historically entered the river (i.e., t¯d = 0.25), after which When risk tolerance differed between fisheries, a risk-neutral the in-season data were used exclusively (i.e., wd,y = 0). We stance toward the subsistence fishery coupled with a risk-averse refer to this method hereafter as the quartile method. Yukon stance toward the commercial fishery resulted in not only the River fall Chum Salmon managers within ADFG currently use largest average subsistence catch (113,000 salmon; Figure 3b) a hybrid of these two approaches. The quartile method is used but also the highest probability of commercial closure (0.44; periodically by ADFG depending on the magnitude and uncer- Figure 3c). The time-varying risk tolerance scenario with in- tainty in the preseason forecast and depending on any available creased risk aversion under a stock of concern designation re- auxiliary data. The Bayesian method is not used explicitly sulted in minimal changes in performance indicators relative to (i.e., the computations are not formally carried out), but ADFG the baseline risk-neutral approach (i.e., constant P*f,y for both managers often subjectively attempt to balance the forecast fisheries; Figure 3). This finding was not unexpected given the and the in-season data depending on their confidence in either overall low probability of a stock of concern designation. estimate. Pairwise scatter plots of performance indicators revealed the Performance indicators.—The performance indicators were shape of the trade-off responses (Figure 4). No clear trade-off the mean subsistence catch, mean commercial catch, the prob- existed between probability of meeting subsistence needs and ability of commercial closure for an entire season, the proba- probability of commercial closure (Figure 4a). The relationship bility of providing minimum amounts necessary for subsistence between each of the fishery risk indicators and average escape- (lower bound of the ANS range: 89,500; Bue et al. 2009), the ment was linear (probability of meeting minimum subsistence probability of a stock of concern designation (failure to meet harvest versus average escapement; Figure 4b) or moderately the minimum escapement goal for 4 of the most recent 5 years), nonlinear (probability of commercial closure versus average es- and the probability of meeting the minimum escapement goal capement; Figure 4c), and thus “win–win” scenarios clearly did of 300,000 Chum Salmon. Mean catch performance indicators not exist among this set of indicators. The fishery-specific har- were calculated as the mean catch over each 50-year model sim- vest control rule provided increased probability of meeting sub- ulation. Probability performance indicators were calculated as sistence needs without sacrificing escapement (Figure 4b) but the proportion of the 50 years of simulated salmon runs satis- simultaneously increased the probability of commercial closure fying a particular criterion (e.g., commercial closure, minimum (Figure 4c).

Downloaded by [Department Of Fisheries] at 19:15 20 July 2015 ANS, minimum escapement). Although performance indicators were only modestly influenced by risk tolerance levels, differences were evident in individual year simulations, particularly when the run size was RESULTS near the management thresholds. As an example, the daily time The values of performance indicators varied substantially series of run size estimates, commercial harvest, and subsis- among the 500 model iterations regardless of the risk tolerance tence harvest for a single year simulation is shown in Figure 5. level or data-weighting approach. For example, the interquartile The true run size for this example was 620,000 salmon and range of the average commercial catch for a constant risk-averse the preseason forecast indicated a run of 825,000 ± 330,000 management approach ranged from 108,000 to 269,000 Chum salmon. Under the risk-neutral approach, the subsistence catch Salmon (Figure 3a). This variability was attributable to variation was 124,000 salmon and the commercial catch was 68,000 in salmon returns, which resulted mainly from stationary process salmon. Not unexpectedly, the risk-averse approach resulted in a error (interannual recruitment variation) and nonstationarity in complete commercial closure. Interestingly, the risk-averse sce- stock productivity. nario resulted in a greater subsistence catch (129,000 salmon) The degree of risk tolerance modestly affected fishery per- than the risk-neutral approach because more fish were avail- formance indicators (Figure 3). Adopting a constant risk-neutral able in the upstream subsistence zone due to the commercial P*f,y of 0.5 resulted in an average commercial catch of 178,000 closure. 1234 CATALANO AND JONES Downloaded by [Department Of Fisheries] at 19:15 20 July 2015

FIGURE 3. Boxplots showing the distribution of performance indicators for Yukon River fall Chum Salmon from 500 model iterations for increasingly risk-averse P* values ranging from 0.5 to 0.1, and for variable P* between ﬁsheries (“ﬁsh”) and variable P* over time as a function of stock of concern designation (“time”).

In years when the preseason forecast was near the manage- the median commercial catch decreased from 42,000 to 8,000 ment thresholds of 300,000, 500,000, and 600,000 salmon, the salmon with increasing risk aversion (Figure 6a), but these effects of changes in risk tolerance were more pronounced for losses were not accompanied by substantial increases in other some of the performance indicators (Figure 6). For example, performance indicators such as subsistence catch (Figure 6b). when the forecast was between 600,000 and 700,000 salmon, In another example, when the forecast was between 500,000 MANAGEMENT TACTICS FOR ANADROMOUS FISHERIES 1235

0.80 (a)

fish 0.75 0.5

0.3 0.4

0.70 0.2

0.1 0.65

P(Minimum Subsistence) 0.60 0.45 0.40 0.35 0.30 P(Commercial Closure)

0.80 (b)

fish 0.75 0.5 0.40.3

0.70 0.2 0.1 P(Minimum Subsistence) P(Minimum 0.60 0.65 420 430 440 450 460 470 480 Average Escapement (thousands)

0.4 FIGURE 5. Example time series of a single year in-season simulation for 0.3 Yukon River fall Chum Salmon showing the (a) daily run size estimates ( ± 90% 0.35 0.30 credible interval [CI]), (b) subsistence catch, and (c) commercial catch. Solid lines in panels (b) and (c) represent the outcomes of using a risk-neutral P*of0.5, 0.2 whereas dashed lines represent a risk-averse P* of 0.1. Horizontal dashed lines 0.40 0.1 in panel (a) show the management threshold run sizes for escapement (300,000

Downloaded by [Department Of Fisheries] at 19:15 20 July 2015 fish salmon), full subsistence harvest (500,000 salmon), and full commercial harvest (600,000 salmon). An identical random seed was used for both P* values so the

0.45 results of the two simulations are comparable. P(Commercial Closure) P(Commercial 420 430 440 450 460 470 480 and 600,000 salmon, increasing risk aversion resulted in a sub- Average Escapement (thousands) stantial reduction in the probability of meeting minimum ANS (Figure 6d) and an increase in the probability of commercial clo- FIGURE 4. Pairwise trade-off plots showing median performance indica- sure (Figure 6c), but with minimal improvement in escapement tor values across six in-season management strategies for Yukon River fall Chum Salmon representing a range of P* values from 0.1 to 0.5, and for indicators (Figure 6e, f). Only at forecasts between 300,000 and variable P* between fisheries (“fish”). The performance indicator pairs are 400,000 was a risk-averse approach advantageous. In this case, as follows: (a) probability of meeting minimum amounts necessary for sub- taking a risk-averse approach resulted in a nearly 50% reduction sistence [P(minimum subsistence)] versus probability of commercial closure in the probability of failing to meet the minimum escapement of [P(commercial closure)]; (b) P(minimum subsistence) versus average escape- 300,000 salmon when compared with the risk-neutral approach ment; (c) P(commercial closure) versus average escapement. Variable P* over time (“time”) is not shown because median values were identical to the P* = (Figure 6f), and this improvement was not accompanied by less 0.5 scenario. P(commercial closure) was plotted in reverse order so that all in- desirable values of the other performance indicators. Thus, only dicators improved with increasing values, which enhanced visualization of the when forecasts were very low did performance benefit from a Pareto frontier. risk-averse, in-season management approach. 1236 CATALANO AND JONES Downloaded by [Department Of Fisheries] at 19:15 20 July 2015

FIGURE 6. Average values of performance indicators for Yukon River fall Chum Salmon as a function of the probability threshold (P*) for a range of values of the preseason forecast (different line types). For example, the solid line represents the values of performance indicators when the preseason forecast was between 300,000 and 399,999 salmon. The averages were calculated across each of the individual year simulations and across the 500 model iterations (500 iterations × 50 years per iteration = 25,000 individual years). MANAGEMENT TACTICS FOR ANADROMOUS FISHERIES 1237

The Bayesian and quartile methods for obtaining daily esti- gested that this range was too low to produce MSY based on mates of the expected run size performed similarly (Figure 7). a reanalysis of the stock–recruit data using a Bayesian state- The average subsistence catch (median of the average 50-year space model. However, Fleischman and Borba (2009) did not catch over 500 model iterations) was 107,000 salmon for both recommend increasing the goal due to the large amount of uncer- methods (Figure 7b). Probability of commercial closure was tainty in the MSY escapement level and potential disruptions of 0.26 for the quartile method and 0.30 for the Bayesian method local fisheries. Uncertainty and risk related to YFC harvest man- (Figure 7c). The probability of meeting minimum amounts nec- agement is probably better addressed by MSE analyses of es- essary for subsistence of 89,500 salmon was 0.7 for both meth- capement goals rather than by testing of probabilistic in-season ods (Figure 7d). All other performance indicators were similar control rules. Collie et al. (2012) developed a risk assessment between the two methods (Figure 7). framework to evaluate the performance of a range of potential Although the two methods performed similarly overall, their escapement goal policies for Chum Salmon in a multiobjec- relative performance varied among individual year simulations tive setting with the explicit consideration of uncertainty. Collie (Figure 8). For example, run estimates obtained with the two et al.’s (2012) analysis revealed clear trade-offs among escape- methods agreed strongly in some years, resulting in similar ment, subsistence, and commercial fishery objectives. Such an subsistence and commercial catches (Figure 8a–c). In other analysis could be used by managers and stakeholders to select years, performance of the two methods differed substantially an escapement goal policy that provides adequate harvesting op- when in-season data from early in the run diverged from the portunities while ensuring sustainable yields. Harvest policies forecast. In such cases, the estimated run size changed sub- determined from this process might differ from MSY escape- stantially when we shifted from the forecast to the in-season ment goal policies. data (Figure 8d–f). Nevertheless, the distribution of outcomes Not unexpectedly, the degree of risk tolerance had a greater over 500 model iterations suggested that the two methods per- influence on fishery performance when the run size was near formed similarly over the 50-year time horizon and across the management threshold abundance. However, close exami- simulations. nation of the trends in performance indicators at low run sizes suggests that a risk-averse approach that results in reduced harvest is not necessarily accompanied by substantial gains toward DISCUSSION escapement objectives. Because these are relatively small run Poor YFC runs from 1998 to 2002 caused significant hardship sizes, the amount of additional commercial harvest allowed un- for subsistence users and commercial fishers along the Yukon der a risk-tolerant approach is also likely to be small during River (Bue et al. 2009). Processing capacity that was lost during most years, and commensurate increases in escapement were these small runs has yet to fully return (Bue et al. 2009). As equally small. Thus, there may be little to gain by a risk-averse well, managers have become quite averse to the risk of failing to approach, even in years with poor runs. Only when the forecast meet escapement and subsistence objectives. Risk aversion has was between 300,000 and 400,000 salmon did the probability led managers to delay commercial openings until in-season run of meeting the escapement goal increase substantially under a estimates become more certain, which has undoubtedly resulted more risk-averse approach. in foregone harvesting opportunities. Our analysis suggests that, Management performance was similar between the Bayesian integrated over the possible future states of the stock, the degree and quartile run-size projections. Generally, the in-season data of in-season risk tolerance as defined by a probabilistic control become more informative around the first quarter point of the rule is not a strong driver of performance for this fishery. Specif- run. Thus, the Bayesian method was mostly weighted toward

Downloaded by [Department Of Fisheries] at 19:15 20 July 2015 ically, we found that increased risk aversion resulted in modest the in-season data after the quarter point because precision of gains in escapement and subsistence objectives while modestly in-season estimates began to exceed the preseason forecast at increasing the probability of commercial closures. An impor- that time. Therefore it was not surprising that the two methods tant implication of this finding is that probabilistic in-season performed similarly. Two conclusions could be drawn from this control rules that allow for more (or less) risk aversion are un- finding. The first is that adopting the Bayesian approach results likely to perform substantially better for this stock under these in no loss of performance. The other is that there is nothing to management thresholds than do simpler control rules that rely gain by switching to the more complicated Bayesian approach. on point estimates of run projections relative to management Although either of these conclusions is valid given our findings, thresholds. there could be additional conceptual advantages of the Bayesian The component of the overall management strategy most approach not accounted for in our analysis. likely driving YFC fishery performance is the escapement A Bayesian method provides a quantitative, defensible, and goal policy from which the in-season management thresholds transparent method for estimating the projected run size from a were obtained. The current escapement goal (300,000–600,000 preseason forecast and in-season data (Fried and Hilborn 1988) salmon) was determined from a traditional stock–recruitment when compared with the quartile method or some other more analysis with the intention of attaining maximum sustainable subjective weighting scheme. Fried and Hilborn (1988) reported yield (MSY; Eggers 2001). Fleischman and Borba (2009) sug- that a Bayesian approach to integrating several in-season run 1238 CATALANO AND JONES Downloaded by [Department Of Fisheries] at 19:15 20 July 2015

FIGURE 7. Boxplots showing the distribution of performance indicators from 500 model iterations for two methods (Bayesian and quartile methods) usedto obtain daily in-season estimates of the expected run size for Yukon River fall Chum Salmon. The conﬁdence threshold (P*) was 0.5 for these simulations. MANAGEMENT TACTICS FOR ANADROMOUS FISHERIES 1239

Downloaded by [Department Of Fisheries] at 19:15 20 July 2015 FIGURE 8. Two example in-season time series of the (a, d) daily run size estimates ( ± 90% credible interval), (b, e) daily subsistence catches, and (c, f) daily commercial catches for Yukon River fall Chum Salmon under the P* = 0.5 policy. Solid lines represent time series in which the Bayesian method was used to estimate the run size, whereas dashed lines represent the quartile method. Horizontal dashed lines in panels (a) and (d) represent the management threshold run sizes for escapement (300,000 salmon), full subsistence harvest (500,000 salmon), and full commercial harvest (600,000 salmon). For each time series, an identical random seed was used for the two methods so that the results are comparable between methods.

projections and a preseason forecast for Bristol Bay Sockeye continuity of decision making following management staff Salmon O. nerka was always more accurate than the least changes. accurate in-season projection and sometimes more accurate Other analyses of in-season management performance have than the most accurate projection. The Bayesian approach not attempted to model the daily Bayesian decision-making also has intuitive appeal because, for example, few reasonable process for setting ﬁshery openings. Su and Adkison (2002) as- managers would base decisions on a highly uncertain forecast sumed managers relied on a preseason forecast through the third when a relatively precise in-season estimate is available. week of the run of Pink Salmon O. gorbuscha in southeastern Furthermore, a Bayesian approach lends itself to increased Alaska, then relied on in-season data thereafter. This approach and more structured thinking about uncertainty, which could was similar to our quartile method because it did not attempt foster improved communication with stakeholders and improve to weigh the relative merits of the two sources of information 1240 CATALANO AND JONES

from a Bayesian perspective. Link and Peterman (1998) used escapement can also be risky because it would likely involve a simulation model to assess the value of additional in-season foregone harvest, which has economic consequences, and may monitoring projects in the Nass River Sockeye Salmon fishery, reduce stock productivity if overcompensatory recruitment dy- British Columbia, but they assumed in-season management de- namics prevail. Our formulation of the probabilistic control rules cisions relied only on several sources of in-season data rather to emphasize biological risks is broadly consistent with precau- than simulating a preseason forecast. Two studies have incorpo- tionary management practices in the field of stock assessment rated the Bayesian approach into in-season abundance estima- and decision making (Prager et al. 2003). However, consideration (Walters and Buckingham 1975; Fried and Hilborn 1988), tion of other risks such as economic or cultural risks is necessary but these studies did not evaluate the pairing of the Bayesian for a comprehensive vetting of potential harvest policies. The approach and probabilistic control rules. advantage of casting the control rules in the MSE framework is In an effort to focus on the effects of risk tolerance, we that it can account for many types of risks via a diverse set of chose to simplify other aspects of our in-season model. For ex- performance indicators, despite the narrow focus of the control ample, we used relatively simple rules for distributing harvest rule. We encourage future studies to further explore risk from among reaches and over time when abundance was sufficient multiple perspectives while concurrently engaging stakeholders to allow fishing. Closer investigation of different approaches to in the exploration of management trade-offs. distributing fishing effort would be a fruitful avenue for research as managers in these systems must rely on intuition rather than tested strategies. Several studies have used historical relation- ACKNOWLEDGMENTS ships between abundance and fishing effort to predict daily har- We thank members of the Arctic–Yukon–Kuskokwim Sus- vest rates during salmon runs (Walters and Buckingham 1975; tainable Salmon Initiative (AYKSSI) Escapement Goals Expert Link and Peterman 1998; Su and Adkison 2002). Future studies Panel: M. Adkison, B. Bue, S. Fleischman, A. Munro, and E. of the in-season dynamics of YFC could develop and employ Volk. Helpful discussion and feedback on this work were pro- these relationships to more realistically depict fishing effort dy- vided by R. Peterman and staff of the Alaska Department of namics. As another simplification, we allowed risk tolerance to Fish and Game: B. Borba, B. Clark, J. Conitz, J. Estensen, M. affect the fishery only through the setting of openings via the Evanson, and J. Linderman. A special thanks is extended to J. probabilistic control rule but not in the calculation of catch tar- Spaeder, K. Gillis, and K. Williams of the Bering Sea Fisher- gets. More sophisticated probabilistic control rules to compute men’s Association. This work was funded by a grant to M. Jones catch targets could be tested (Shertzer et al. 2010). These con- from AYKSSI and the Bering Sea Fishermen’s Association. trol rules could be further modified to account for uncertainty not only in abundance but also in the management thresholds themselves, if the thresholds were a function of estimates of the REFERENCES Bue, F. J., B. M. Borba, R. Cannon, and C. C. Krueger. 2009. Yukon River biological productivity of the stock (Shertzer et al. 2010). fall Chum Salmon fisheries: management, harvest, and stock abundance. Our findings are relevant to other large systems with spatial Pages 703–742 in C. C. Krueger and C. E. Zimmerman, editors. Pacific and temporal separation between harvesting and escapement. salmon: ecology and management of western Alaska’s populations. American We would expect the influence of in-season risk tolerance to be Fisheries Society, Bethesda, Maryland. lower in systems such as Bristol Bay, Alaska, where certainty Butterworth, D. S. 2007. Why a management procedure approach? Some posi- tives and negatives. ICES Journal of Marine Science 64:613–617. in projections of run abundance and management control over Butterworth, D. S., K. L. Cochrane, and J. A. A. D. Oliveria. 1997. Management harvest are higher. In these fisheries, increased certainty in run procedures: a better way to manage fisheries? The South African experience.

Downloaded by [Department Of Fisheries] at 19:15 20 July 2015 projections would reduce the frequency of exceeding risk toler- Pages 83–90 in E. K. Pikitch, D. D. Huppert, and M. P. Sissenwine, editors. ance thresholds (i.e., P*) when the run is actually substantially Global trends: fisheries management. American Fisheries Society, Sympo- larger than abundance thresholds. In addition, more conserva- sium 20, Bethesda, Maryland. Collie, J. S., R. M. Peterman, and B. M. Zuehlke. 2012. A fisheries risk as- tive escapement goals would also increase the influence of risk sessment framework to evaluate trade-offs among management options in tolerance because run sizes would more frequently be near man- the presence of time-varying productivity. Canadian Journal of Fisheries and agement reference points. The large amount of temporal varia- Aquatic Sciences 69:209–223. tion in stock productivity was influential in our study, and this Cooke, J. G. 1999. Improvement of fishery management advice through sim- dynamic is likely shared by many other salmon stocks. ulation testing of harvest algorithms. ICES Journal of Marine Science 56: 797–810. The concept of risk is important to define when evaluating the Eggers, D. E. 2001. Biological escapement goals for Yukon River fall Chum performance of harvest policies. The probabilistic control rules Salmon. Alaska Department of Fish and Game, Division of Commercial we evaluated characterized risk in terms of failure to meet es- Fisheries, Regional Information Report 3A01–10, Anchorage. capement goals. This approach assumes that risks are primarily Fleischman, S. J., and B. M. Borba. 2009. Escapement estimation, spawner- biological and only occur when escapement goals are not met, recruit analysis, and escapement goal recommendation for fall Chum Salmon in the Yukon River drainage. when in fact there are risks on either side of the escapement Fleischman, S. J., M. J. Catalano, R. A. Clark, and D. R. Bernard. 2013. An age- goal. Clearly there are biological risks associated with failing structured state-space stock–recruit model for Pacific salmon (Oncorhynchus to provide adequate spawning escapement. Allowing too much spp.). Canadian Journal of Fisheries and Aquatic Sciences 70:401–414. MANAGEMENT TACTICS FOR ANADROMOUS FISHERIES 1241

Fried, S. M., and R. Hilborn. 1988. Inseason forecasting of Bristsl Bay, Alaska, Robb, C. A., and R. M. Peterman. 1998. Application of Bayesian decision Sockeye Salmon (Oncorhynchus nerka) abundance using Bayesian probabil- analysis to management of a Sockeye Salmon (Oncorhynchus nerka)fishery. ity theory. Canadian Journal of Fisheries and Aquatic Sciences 45:850–855. Canadian Journal of Fisheries and Aquatic Sciences 55:86–98. Hilsinger, J. R., E. Volk,G. Sandone, and R. Cannon. 2009. Salmon management Shertzer, K. W., M. H. Prager, and E. H. Williams. 2008. A probability-based ap- in the Arctic-Yukon-Kuskokwim region of Alaska: past, present, and future. proach to setting annual catch levels. U.S. National Marine Fisheries Service Pages 495–519 in C. C. Krueger and C. E. Zimmerman, editors. Pacific Fishery Bulletin 106:225–232. salmon: ecology and management of western Alaska’s populations. American Shertzer, K. W., M. H. Prager, and E. H. Williams. 2010. Probabilistic ap- Fisheries Society, Bethesda, Maryland. proaches to setting acceptable biological catch and annual catch targets for Hyun, S.-Y., R. Hilborn, J. J. Anderson, and B. Ernst. 2005. A statistical model multiple years: reconciling methodology with national standards guidelines. for in-season forecasts of Sockeye Salmon (Oncorhynchus nerka) returns to Marine and Coastal Fisheries: Dynamics, Management, and Ecosystem Sci- the Bristol Bay districts of Alaska. Canadian Journal of Fisheries and Aquatic ence [online serial] 2:451–458. Sciences 62:1665–1680. Smith, A. D. M., K. J. Sainsbury, and R. A. Stevens. 1999. Implementing Jones, M. L., E. Volk, M. Adkison, B. Bue, S. Fleischman, A. Munro, and M. J. effective fisheries-management systems – management strategy evaluation Catalano. 2011. Management strategies for AYK salmon stocks: accounting and the Australian partnership approach. ICES Journal of Marine Science for uncertainty. Final Report to the Arctic-Yukon-Kuskokwim Sustainable 56:967–979. Salmon Initiative, Anchorage, Alaska. Starr, P., and R. Hilborn. 1988. Reconstruction of harvest rates and stock con- Link, M. R., and R. M. Peterman. 1998. Estimating the value of in-season tribution in gauntlet salmon fisheries: application to British Columbia and estimates of abundance of Sockeye Salmon (Oncorhynchus nerka). Canadian Washington sockeye (Oncorhynchus nerka). Canadian Journal of Fisheries Journal of Fisheries and Aquatic Sciences 55:1408–1418. and Aquatic Sciences 45:2216–2229. Peterman, R. M. 2004. Possible solutions to some challenges facing fisheries Su, Z., and M. D. Adkison. 2002. Optimal in-season management of Pink scientists and managers. ICES Journal of Marine Science 61:1331–1343. Salmon (Oncorhynchus gorbuscha) given uncertain run sizes and seasonal Peterman, R. M., and J. L. Anderson. 1999. Decision analysis: a method for changes in economic value. Canadian Journal of Fisheries and Aquatic Sci- taking uncertainties into account in risk-based decision making. Human and ences 59:1648–1659. Ecological Risk Assessment 5:231–244. Walters, C. J., and S. Buckingham. 1975. A control system for intraseason Peterman, R. M., B. J. Pyper, and B. W. MacGregor. 2003. Use of the Kalman salmon management. Pages 105–137 in IIASA (International Institute for filter to reconstruct historical trends in productivity of Bristol Bay Sockeye Applied Systems Analysis), editor. Proceedings of a workshop on salmon Salmon (Oncorhynchus nerka). Canadian Journal of Fisheries and Aquatic management. IIASA, CP-75-2, Laxenburg, Austria. Sciences 60:809–824. Walters, C. J., and S. J. D. Martell. 2004. Fisheries ecology and management. Prager, M. H., C. E. Porch, K. W. Shertzer, and J. F. Caddy. 2003. Targets Princeton University Press, Princeton, New Jersey. and limits for management of fisheries: a simple probability-based approach. Williams, B. K. 1997. Approaches to the management of waterfowl under North American Journal of Fisheries Management 23:349–361. uncertainty. Wildlife Society Bulletin 25:714–720. Downloaded by [Department Of Fisheries] at 19:15 20 July 2015 This article was downloaded by: [Department Of Fisheries] On: 20 July 2015, At: 19:15 Publisher: Taylor & Francis Informa Ltd Registered in England and Wales Registered Number: 1072954 Registered office: 5 Howick Place, London, SW1P 1WG

North American Journal of Fisheries Management Publication details, including instructions for authors and subscription information: http://www.tandfonline.com/loi/ujfm20 Importance of Ultrasonic Field Direction for Guiding Juvenile Blueback Herring Past Hydroelectric Turbines Christopher W. D. Gurshina, Matthew P. Balgea, Michael M. Taylora & Benjamin E. Lenzb a Normandeau Associates, Inc., 30 International Drive, Suite 6, Portsmouth, New Hampshire 03801, USA b New York Power Authority, 123 Main Street, White Plains, New York 10601, USA Published online: 01 Dec 2014.

Click for updates

To cite this article: Christopher W. D. Gurshin, Matthew P. Balge, Michael M. Taylor & Benjamin E. Lenz (2014) Importance of Ultrasonic Field Direction for Guiding Juvenile Blueback Herring Past Hydroelectric Turbines, North American Journal of Fisheries Management, 34:6, 1242-1258, DOI: 10.1080/02755947.2014.963749 To link to this article: http://dx.doi.org/10.1080/02755947.2014.963749

PLEASE SCROLL DOWN FOR ARTICLE

Taylor & Francis makes every effort to ensure the accuracy of all the information (the “Content”) contained in the publications on our platform. However, Taylor & Francis, our agents, and our licensors make no representations or warranties whatsoever as to the accuracy, completeness, or suitability for any purpose of the Content. Any opinions and views expressed in this publication are the opinions and views of the authors, and are not the views of or endorsed by Taylor & Francis. The accuracy of the Content should not be relied upon and should be independently verified with primary sources of information. Taylor and Francis shall not be liable for any losses, actions, claims, proceedings, demands, costs, expenses, damages, and other liabilities whatsoever or howsoever caused arising directly or indirectly in connection with, in relation to or arising out of the use of the Content. This article may be used for research, teaching, and private study purposes. Any substantial or systematic reproduction, redistribution, reselling, loan, sub-licensing, systematic supply, or distribution in any form to anyone is expressly forbidden. Terms & Conditions of access and use can be found at http:// www.tandfonline.com/page/terms-and-conditions North American Journal of Fisheries Management 34:1242–1258, 2014 Ó American Fisheries Society 2014 ISSN: 0275-5947 print / 1548-8675 online DOI: 10.1080/02755947.2014.963749

ARTICLE

Importance of Ultrasonic Field Direction for Guiding Juvenile Blueback Herring Past Hydroelectric Turbines

Christopher W. D. Gurshin,* Matthew P. Balge, and Michael M. Taylor1 Normandeau Associates, Inc., 30 International Drive, Suite 6, Portsmouth, New Hampshire 03801, USA Benjamin E. Lenz New York Power Authority, 123 Main Street, White Plains, New York 10601, USA

Abstract Populations of Blueback Herring Alosa aestivalis, an important anadromous forage fish in the northeastern USA, have declined from historic levels. Measures to reduce mortality from many sources, including entrainment by hydroelectric turbines, are cited by fishery managers to be important to restoring populations. At the Crescent Hydroelectric Project (Crescent) on the Mohawk River, New York, pulsed ultrasound (122–128 kHz) was used to deter adult and juvenile Blueback Herring migrating out to sea from entering the intake channel to the headrace and turbines, where mortality may occur. To increase the deterrence rate, the sound field was extended further upriver to expose juvenile Blueback Herring to an increasing sound gradient as they migrate downriver and allow them more time to avoid the intake channel. When juvenile Blueback Herring were present upstream of Crescent and exposed to an ultrasound field, mean catch per unit effort (CPUE) by pelagic trawling in the main channel downriver of the ultrasound was 94% of the mean CPUE in the upriver trawl region and 250% of the mean CPUE in the intake channel. Repeated mobile echo sounder surveys revealed that the abundance of juvenile Blueback Herring averaged 35 times higher in the downriver main channel region than in the intake channel region. During the peak migration period of September 20 through October 14, 2012, continuous fixed-location horizontal echo sounding showed the percentage of net downstream passage of juvenile Blueback Herring that bypassed the intake channel significantly increased from 31.1% to 76.5% after redirecting the ultrasonic field upriver. These results demonstrate that pulsed ultrasound, when properly directed, can be used effectively as a fish deterrent system to provide safe downstream passage of juvenile Blueback Herring at hydropower dams. At Crescent, this could increase annual survival by tens of thousands of fish or more. Downloaded by [Department Of Fisheries] at 19:15 20 July 2015 Blueback Herring Alosa aestivalis is an important anadro- pollution, channelization and dredging, barriers to migration, mous forage species of fish in the northeastern USA and is and mortality from fish entrainment by turbines at hydroelec- currently managed together with Alewife A. pseudoharengus tric facilities and intakes at power plants. (the two species collectively referred to as river herring) The FMP identifies technologies for fish guidance as miti- under the interstate fishery management plan (FMP) prepared gation measures to deter fish from entering intakes at water by the Atlantic States Marine Fisheries Commission (ASMFC withdrawal facilities. These technologies use sensory stimuli, 2009). The most recent stock assessment report declared such as light, sound, flow, turbulence, and electric fields, to stocks from many river systems, including the Hudson and change local fish distributions at hydropower dams to improve Mohawk rivers, as being substantially depleted from historic fish passage (Schilt 2007). Ultrasound (frequencies greater levels (ASMFC 2012). The FMP identifies multiple factors than 20 kHz) can be detected by alosines but not by other clu- contributing to their population decline, such as overfishing, peids (Mann et al. 1997; Popper et al. 2004), with some of

*Corresponding author: [email protected] 1Present address: RPS Evans-Hamilton, 4608 Union Bay Place Northeast, Seattle, Washington 98105-4027, USA. Received October 4, 2013; accepted August 26, 2014

1242 ULTRASONIC FIELD DIRECTION 1243

these frequencies eliciting avoidance responses in Blueback Horizontally aimed ultrasonic projectors were used Herring (Nestler et al. 1992; Dunning and Ross 2010). When between 2008 and 2012, but the orientation with respect to captive Blueback Herring were exposed to loud sounds rang- flow was modified in 2010. In 2008, two sets of four projectors ing from 0.10 to 420 kHz, Nestler et al. (1992) found the were pointed in opposite directions and perpendicular to flow strongest avoidance response between 120 and 130 kHz. Field to form an acoustic barrier across the entrance of the intake evaluations later revealed that transducers transmitting 124.6 channel (Figure 1C). Dunning and Gurshin (2012) suggested and 130.9 kHz at a source level of 187 or 200 decibels refer- that the deterrence rate might be increased if the projectors enced to 1 mPa at 1 m (dB re 1 mPa at 1 m) partially repelled were redirected to extend the effective sound field further Blueback Herring over a distance of 60 m for periods of up to upriver so that fish encounter an increasing sound gradient. 1 hour. Dunning and Ross (2010) found band-limited sound This would provide more time for juvenile Blueback Herring between 122 and 128 kHz at sound pressure levels (SPLs) to respond to the ultrasound and avoid entering the intake ranging from 145 to 163 dB re 1 mPa to elicit avoidance reac- channel regardless of water velocity. As a result, beginning in tions from adult Blueback Herring held in a tank. 2010, four western projectors were redirected upriver by 45 The use of ultrasound for guiding Blueback Herring and toward the main channel (Figure 1C). other alosines safely downstream is supported by experimental In this study, fish passage at Crescent was monitored by an evidence and field studies that show that ultrasound reduces intensive sampling effort from September 8 through October entrainment and impingement. On the Savannah River in 26, 2012, to (1) determine whether the majority of juvenile Georgia, Ploskey et al. (1995) showed that ultrasound signifi- Blueback Herring migrated down the main channel in the pres- cantly reduced the mean hourly rates of Blueback Herring pas- ence of ultrasound and (2) evaluate the effectiveness of the sage through the turbines at the Richard B. Russell reconfigured ultrasonic projectors for guiding out-migrating hydroelectric dam by 56%. At Annapolis Tidal Hydroelectric juvenile Blueback Herring away from the turbines to the main Generating Station located in Nova Scotia, Gibson and Myers channel for downstream passage. (2002) demonstrated that a 122–128 kHz ultrasonic system reduced fish passage of Blueback Herring through the turbines by 49%. Dunning et al. (1992) found that Alewife schooled METHODS and strongly avoided broadband sound between 117 and Multiple acoustic and net sampling methods were used in a 133 kHz with an SPL of 157 dB re 1 mPa at 1 m or greater. complementary fashion to measure Blueback Herring abun- When ultrasound was used in Lake Ontario at the cooling dance at different scales in time and space. Stratified-random water intake of the James A. FitzPatrick Nuclear Power Plant, pelagic trawl surveys and systematic mobile echo sounder sur- the density of Alewife near the cooling water intake decreased veys were repeatedly conducted to compare the density and by as much as 96% and impingement of Alewife was reduced abundance of juvenile Blueback Herring over time and among by as much as 87% (Ross et al. 1993, 1996). regions in the Mohawk River upstream of Crescent. Fish pas- Adult Blueback Herring are known to pass through several sage estimates from continuous stationary horizontal echo locks of the Erie Canal during their spring upriver spawning sounding upriver and downriver of the ultrasonic projectors migration up the Mohawk River in New York (Figure 1A, B; were used to compare the number of fish migrating downriver Schmidt et al. 2003). Adults migrating downriver following in the main channel to the number expected under two null spawning and juveniles out-migrating to sea several months hypotheses. Acoustic Doppler current profiler (ADCP) tech- later may encounter dams at six hydropower facilities. The nology was used to continuously measure water velocity at

Downloaded by [Department Of Fisheries] at 19:15 20 July 2015 Crescent Hydroelectric Project (Crescent), operated by the each site and measure daily river discharge. Computations and New York Power Authority, is a facility where out-migrating statistics of discharge, trawl, and processed acoustic data were juvenile Blueback Herring are known to encounter hydroelec- performed using SAS software version 9.3 (SAS Institute, tric turbines (Dunning and Gurshin 2012). An ultrasonic pro- Cary, North Carolina) and Matlab software version R2011b jector array was installed in 2008 and since then has operated (Mathworks, Natick, Massachusetts). at Crescent from May through mid-November to deter Blue- back Herring adults during their postspawn downstream migration (May–July) and juveniles during their downstream Study Area migration (August–November) from entering the channel that Crescent is located on the Mohawk River approximately leads to the Crescent headrace and turbines. In a hydroacous- 8 km upstream of the confluence to the Hudson River, New tic study of fish passage at Crescent during the fall of 2008, York (Figure 1). Dams A and B, separated by an island, the percentage of juvenile Blueback Herring that migrated impound the Mohawk River. The river channel that flows downriver past the headrace and turbines (31.3%) was signifi- down the west side of the island conveys water through four cantly higher than expected (11.5%) if entrainment (and den- hydroelectric turbines during power generation (hereafter, the sity) is proportional to river flow (Dunning and Gurshin “intake channel”). The primary navigation channel (hereafter, 2012). the “main channel”) continues to the east of the island toward 1244 GURSHIN ET AL. Downloaded by [Department Of Fisheries] at 19:15 20 July 2015

FIGURE 1. Location of the study area (A) in New York and (B) in the Mohawk River, New York. A model was created (C) predicting the ultrasonic ﬁeld produced by eight integrated projector assemblies placed at the entrance of the intake channel at the Crescent Hydroelectric Project, after redirecting the four western projectors by a 45 angle to point upriver in the main channel of the Mohawk River; the previous orientation of the ultrasonic ﬁeld is shown as the dashed outline. The maps of the study area in the main channel upriver and downriver of the ultrasonic projectors show (D) pelagic trawl regions (polygons), locations of split- beam transducers aimed horizontally (open circle) and vertically (crossed circles), locations of horizontally aimed acoustic Doppler current meters (triangles), and (E) mobile echo sounder survey regions (polygons) sampled along transects (dashed lines).

Dam A and the entrance to Waterford Flight Lock 6 of the Erie 24.4-m-wide and 0.3-m-high opening in the flashboards Canal System. Out-migrating juvenile Blueback Herring cur- installed on top of Dam A when the water level is above the rently have three turbine bypass routes for downstream pas- fixed crest elevation of the dam (56.1 m), (2) in the water spill- sage at Crescent: (1) down the main channel and through a ing over the flashboards of both Dams A and B during ULTRASONIC FIELD DIRECTION 1245

high-flow events, and (3) down the main channel and through in the downstream direction only (i.e., setting negative Vm to the Waterford Flight Lock 6. A pond elevation of 56.4 m or 0). A Wilcoxon signed-rank test for paired data was used to higher was maintained to produce a minimum river discharge test whether the mean hourly difference in Vm or downstream (7.1 m3/s) through the flashboard opening at Dam A through- flow between the upriver and downriver site was significantly out the study. The observed maximum water depths of the different than 0. Average bottom water temperature was sampled sections of the river during the study period were described from measurements taken every 15 min by two 4.9 m in the intake channel, 10.0 m upriver of the island in the water-temperature data loggers (HOBO Pro v2; Onset Com- main channel, and 12.8 m east of the island in the main chan- puter, Bourne, Massachusetts) attached to the center trans- nel. Dunning and Gurshin (2012) provide additional details ducer mount at the upriver site. about Crescent and the surrounding water bodies. The ultrasonic projectors (Ultra Electronics Ocean Sys- tems, Braintree, Massachusetts) at Crescent consisted of eight Trawl Sampling of Fish Populations integrated projector assemblies (hereafter, “projectors”) con- The size and abundance of juvenile Blueback Herring figured in two 2 £ 2 block arrays (0.3 m apart from the center acoustically observed were verified by sampling the fish popu- of each transducer face). Each projector contained all the elec- lations with a pelagic trawl. A 3-m-long cone-shaped net was tronics necessary for signal generation, amplification, and attached to an aluminum frame (1.74 m wide and 1.14 m transmission. Each transducer of a projector had a nominal high). The stretched mesh size was 13 mm for the body and source level of 196 dB re 1 mPa at 1 m and a 13 £ 30 beam 4 mm for the cod end. A 2.25-kg steel depressor was fastened width within 3 dB of the major response axis. The projectors with a 1-m chain to each bottom corner of the net frame. A simultaneously transmitted 0.5-s pulses at 1‑s intervals and 36‑cm-diameter polyurethane float-and-line from each top cor- 125-kHz center frequency (122–128 kHz bandwidth contain- ner of the net frame held the top of the net 1 m below the sur- ing »99% of the acoustic energy). Figure 1C shows the spatial face while it was towed. A 7.5-m vessel was used to deploy, extent of the SPLs (root-mean-square [rms] pressure) between tow, and retrieve the net. When fully deployed, the net was 122 and 128 kHz within the ultrasonic field. Sound level approximately 30 m behind the vessel and towed upstream measurements made using a hydrophone at several locations along a 200-m towpath at 1.2 m/s. and depths on June 28, 2011, showed that the effective SPLrms Four trawl-sampling regions were defined in the river based known for an avoidance response by Blueback Herring ranged on the depths and potential navigation hazards (Figure 1D). from 177 to 180 dB re 1 mPa within a few meters north of the The “upriver” trawl region was located upriver of the fixed- reconfigured projectors and decreased to 166 dB re 1 mPa at location hydroacoustic monitoring site where the SPL was less 140 m and 156–160 dB re 1 mPa at 350 m, depending on than 145 dB re 1 mPa. The “ultrasound gradient” trawl region depth. was defined as the region of the main channel between the upriver acoustic monitoring site and the ultrasonic projectors at the entrance to the intake channel. As Blueback Herring River Discharge and Temperature Monitoring migrate downstream, they encounter increasing SPLs within To estimate the expected downstream passage based on this region, particularly in the western main channel. The flow, a 500-kHz ADCP (Argonaut SL; SonTek/YSI, San “downriver” trawl region was defined as the area in the main Diego, California) was horizontally aimed to continuously channel between the downriver acoustic monitoring site and measure and average the along-channel water velocity along the buoy perimeter for the Dam A forebay. The “intake

Downloaded by [Department Of Fisheries] at 19:15 20 July 2015 the river axis over a 34 or 38-m range (Vx) and river height (H) channel” trawl region was defined as the area in the intake every 2.5 min at two locations: in the main channel upriver channel downriver of the ultrasonic projectors where water (hereafter, the “upriver site”) and in the main channel down- depths permit trawling. river (hereafter, the “downriver site”) of the ultrasonic projec- Each trawl region was sampled by three randomly selected tor array (Figure 1D, E). River discharge (Q) through the tows along planned towpaths. The HYPACK navigation map- entire channel of the upriver and downriver sites was esti- ping software (HYPACK, Middletown, Connecticut) and the mated as the product of the total cross-sectional area of the Global Positioning System (Trimble DSM-232 receiver with channel in the vertical plane (A) and the mean water velocity submeter accuracy) were used for real-time positioning of the of the entire channel (Vm). vessel and trawl operation along the towpaths. Between the An estimate of Vm across the entire channel was made daily start and end of each tow, acoustic data were concurrently col- along a cross-river transect at each site by a vessel-mounted lected by a downward-looking split-beam echo sounder 1‑MHz ADCP. The concurrent Vm, Vx, and H measurements (Model 241; Hydroacoustic Technology, Seattle) that operated were used to construct a velocity‑index regression model for at a nominal center frequency of 420 kHz. Trawl surveys were predicting a complete time series of Vm and Q from the contin- made on 11 nights between September 9 and October 25. All uous fixed-location ADCP measurements (Morlock et al. fish were identified to species and enumerated, and the total 2002; Dunning and Gurshin 2012). Hourly Q was based on Vm length (TL) of up to 50 individuals of the same species per 1246 GURSHIN ET AL.

tow was measured to the nearest millimeter. All fish were Simmonds and Fryer 1996). A split-beam transducer with a released back to the river. The target strength (TS) distribution 15 circular half-power beam width was mounted to a pole of the Blueback Herring catch was estimated from the TS–TL attached midship and 36 cm below the water surface. A 1.25- relation for Alewife (Warner et al. 2002), which is also repre- ms frequency-modulated chirp signal, sweeping from 415 to sentative of juvenile Blueback Herring (Gurshin 2012). 425 kHz, was transmitted every 0.1 s at an effective pulse Catch per unit effort (CPUE) from pelagic trawling, defined duration of 0.18 ms and pulse length of 0.27 m after filter as the number of fish per 200-m tow, provided a relative index compression to a narrow echo pulse to provide good spatial of abundance. Young-of-the-year Blueback Herring were resolution and improve signal-to-noise ratio (Ehrenberg and identified by their length. The CPUE of young-of-the-year Torkelson 2000). The echo sounder frequency was selected to Blueback Herring was compared among trawl sampling minimize an avoidance response since the hearing sensitivity regions during the period before the majority migrated out of of Blueback Herring is much less at 420 kHz than at lower the study area. The CPUE data were log10(x C 1) transformed echo sounder frequencies (Nestler et al. 1992; Mann et al. to meet the assumption of normality for analysis of variance 1997). (ANOVA) based on the Shapiro–Wilk test statistic. The variance of the transformed CPUE values was homogeneous among the trawl sampling regions based on the Levene’s test Continuous Echo Sounding of Fish Passage for homogeneous variance. A Tukey’s studentized range test Fixed-location echo sounding has been previously used to for multiple pairwise comparisons was used to detect which monitor schooling clupeids in shallow channels (Guillard regions were significantly different in CPUE of Blueback 1998; Pedersen and Trevorrow 1999; Dunning and Gurshin Herring. 2012). Similar to other stationary acoustic surveys in shallow water (Kubecka and Wittingerova 1998; Enzenhofer and Cronkite 2000; Krumme and Saint-Paul 2003), horizontally Mobile Echo Sounder Surveys of Fish Abundance aimed transducers (i.e., horizontal beaming) were used in this The abundance of Blueback Herring near Crescent was esti- study because the relatively simple bottom topography, calm mated from seven mobile echo sounder surveys during the water surface conditions, and few storm events allowed fish to day. Blueback Herring were easier to classify in the echograms be detected at ranges greater than the depth of water and, thus, during the day when they were distributed as dense groups sampled a larger volume and cross-sectional area of the river than during the night when they were distributed as single fish channel than would vertically aimed transducers. As a result, echoes throughout the water column (based on echograms ver- echo integration of the volume backscatter collected by fixed- ified by nighttime pelagic trawling; Luecke and Wurtsbaugh location horizontal beaming was used to derive estimates of 1993; Freon et al. 1996; Petitgas and Levenez 1996). Addi- juvenile Blueback Herring passage (Pedersen and Trevorrow tionally, Blueback Herring were the dominant schooling spe- 1999; Dunning and Gurshin 2012). cies during the study period in the Mohawk River (Dunning A linear array of three horizontally aimed split-beam trans- and Gurshin 2012). For each survey, the vessel equipped with ducers (hereafter, “horizontal transducers”) and one vertically the same split-beam echo sounder used during trawling sur- aimed split-beam transducer (hereafter, “vertical transducer”) veyed at 2.0–2.3 m/s along transects within four river regions: was installed to monitor fish passage at each of two sites; in (1) the intake channel, (2) upriver and (3) downriver of the the main channel upriver (hereafter, the “upriver site”) and in ultrasonic projectors in the main channel, and (4) a northern the main channel downriver (hereafter, the “downriver site”)

Downloaded by [Department Of Fisheries] at 19:15 20 July 2015 extension of the upriver main channel region (Figure 1E; of the ultrasonic projectors (Figure 1D, E). The transducers Table 1). were attached to mounts placed on the riverbed and multi- Transects were perpendicular to the river channel and about plexed to a digital echo sounder (HTI Model 243; Hydroacous- 50 m apart. A systematic survey design often leads to a more tic Technology, Seattle). The nominal center frequency of precise abundance estimate of patchily distributed ﬁsh and these transducers was 420 kHz. Nominal half-power beam offers higher sampling efﬁciency (Simmonds et al. 1992; widths were 6 for horizontal transducers and 15 for vertical

TABLE 1. Mobile echo sounder survey effort in 2012 for estimating the abundance of juvenile Blueback Herring in the intake channel and upriver and downriver of the ultrasonic projectors in the main channel of the Mohawk River near the Crescent Hydroelectric Project. See Figure 1E. Echo sounder survey region Area (m2) Number of transects Survey dates

Intake channel 69,261 9 September 9, 13, 19; October 3, 9, 19, 26 Downriver main channel 122,756 11 September 9, 13, 19; October 3, 9, 19, 26 Upriver main channel 240,392 22 September 9, 13, 19; October 3, 9, 19, 26 Upriver extension 288,846 22 September 9, 19; October 19, 26 ULTRASONIC FIELD DIRECTION 1247

of all targets within the sampled volume as defined by MacLennan et al. (2002). A range‑dependent minimum Sv threshold equivalent to a minimum echo strength threshold of ¡61dB was applied to the echograms to reduce the contribution of background noise and smaller scatterers to the density estimate of juvenile Blueback Herring (Rudstam et al. 2009). 2 Target strength (TS [dB re 1 m ] D 10 log10sbs) of single echo detections (SEDs) was determined from the echogram when fish density was low enough to avoid bias due to overlapping fish echoes, which was determined by echogram strata with an Sawada index less than 0.1 (Sawada et al. 1993; Rud- stam et al. 2009). Mobile echo sounder surveys.—The echogram for each sur- FIGURE 2. Schematic diagram from the side view looking upstream show- vey transect was stratified into 1-m depth layers and 20-m ing the cross-sectional area effectively sampled by the transducers (dark gray transect segments called elementary distance sampling units triangles) deployed from the eastern shore (on right). The light gray lines strat- (EDSUs). The surface exclusion zone (0.9 m depth) and bot- ify the river channel into three zones for allocating representative sampling of fish passage by the horizontally aimed transducers. tom exclusion zone (approximately 0.75 m above the bottom detection) in the echograms were manually reviewed and modified as necessary to exclude interference or bias from bubble transducers. While the maximum depth and mean cross-sec- reverberation or bottom backscatter. Each echogram of vol- tional area at the upriver (9.1 m and 1,497 m2) and downriver ume backscatter was manually scrutinized, and clusters of sites (10.7 m and 1,277 m2) were similar, there were some dif- overlapping echoes were visually classified as “schools.” Juve- ferences in their channel geometry (Figure 2). The horizontal nile Blueback Herring was the species most likely represented transducers sampled approximately 13–16% of the channel by these schools because (1) it was the most abundant school- cross section and monitored fish passage that was assumed ing pelagic species previously found here at this time of year representative of the entire channel. The vertical transducer (Dunning and Gurshin 2012, (2) it was the most abundant spe- was used primarily to aid interpretation of the echoes in the cies in pelagic trawl catches (see results), and (3) similar echo echograms obtained by the horizontal transducers and verify characteristics were observed during echo sounding of the acoustic estimates of fish movement. trawl zone. These regions of classified school echoes were All transducers were calibrated using a standard transducer echo integrated and exported as an area backscattering coeffi- of known sensitivity at the manufacturer prior to deployment cient (sa), defined by MacLennan et al. (2002) as the sv inte- (Johannesson and Mitson 1983; ANSI/ASA 2012). A 0.18-ms grated over depth. The remaining volume backscatter of frequency-modulated chirp pulse was transmitted every 0.2 s unclassified fish echoes above the threshold was apportioned for horizontal transducers and every 0.1 s for vertical trans- to juvenile Blueback Herring by the proportion of SEDs within ducers. The systematic sampling scheme consisted of the a dorsal-aspect TS range expected for juvenile Blueback Her- transducers sequentially collecting raw acoustic backscatter ring (¡50 to ¡42 dB; Gurshin 2012). for consecutive 2.5-min intervals, starting with the most east- The areal density (D; fish/m2) of juvenile Blueback Herring ern transducer at the beginning of each hour. This sampling for the full water column of each 20-m EDSU (j) within each

Downloaded by [Department Of Fisheries] at 19:15 20 July 2015 scheme resulted in each transducer collecting six 2.5-min survey region (i) was then estimated as intervals per hour and spaced 10 min apart from September 8 ÀÁÀÁ through October 26, 2012. D D s =〈s 〉 C s £ P =〈s 〉 ; i;j ai;j;school bs i;JBH ai;j;nonschool i;JBH bs i;all (1) Acoustic Data Processing

All raw acoustic data were imported into Echoview signal where sai,j,school is the sa attributable to echoes classified as processing software (version 5.0; Myriax Software, Hobart, schools of juvenile Blueback Herring along EDSU j of survey Australia) using preferred parameterization and settings fol- region i, sa, i, j, nonschool is the remaining sa (i.e., total sa,–sa, lowing standard practices (De Robertis and Higginbottom school) along EDSU j in survey region i, hsbsi i, JBH is the in 2007; Parker-Stetter et al. 2009; Rudstam et al. 2009). The situ mean TS for SEDs in survey region i that are within the ¡1 volume backscattering coefficient (sv;m ) and its decibel juvenile Blueback Herring TS range, and hsbs ii, all is the in ¡1 equivalent, volume backscattering strength (Sv;dBrem ), situ mean TS for all SEDs in survey region i. The total abun- from echo integration was assumed to be proportional to fish dance of juvenile Blueback Herring for each region and survey density (Foote 1983) and quantified the acoustic energy was estimated by multiplying the mean areal density by the 2 reflected back from the backscattering cross sections (sbs;m) area of the region. 1248 GURSHIN ET AL.

A two-stage geostatistical approach was taken to map the with fewer than 10 tracked echoes classified as juvenile Blue- distribution of juvenile Blueback Herring from spatially auto- back Herring. correlated data collected by the systematic echo sounder After the most effective algorithm parameters were deter- surveys (Petitgas 1993; Paramo and Roa 2003; Mello and mined, split-beam tracking was used to estimate swimming Rose 2005; Gurshin et al. 2013). During the first stage, the direction, downstream rate of movement, and TS of individual spatial structure of the data (i.e., the spatial covariance) was fish echoes within the TS range of juvenile Blueback Herring. explored through semivariogram modeling (also known as During the first few weeks of monitoring, nighttime vertical “variograms”; see Cressie and Hawkins 1980; Petitgas 1993; migrations of midge insect larvae Chaoborus sp. (Teraguchi Rivoirard et al. 2000), which describes the amount of spatial and Northcote 1966), which were verified by net sampling, autocorrelation as a function of distance between any two almost completely overwhelmed the echograms with volume locations. The semivariogram model parameters were then backscatter from their gas sacs (Malinen et al. 2005; Knudsen used to spatially predict fish density at unsampled locations et al. 2006). Therefore, the mean echogram sv was not a good within each survey region. A first-order polynomial on the index of abundance for juvenile Blueback Herring without tak- coordinates was used to first detrend the data before semivar- ing steps to remove the acoustic contributions of several iogram modeling. The omnidirectional semivariograms dynamic sources. assumed no geometric anisotropy after examining directional The sv attributable to Chaoborus sp. and variable back- semivariograms. Universal kriging with a first-order trend on ground noise levels attributable to scattering from other inver- coordinates and a 40-point moving local neighborhood was tebrates, particles, wind, rain, or flow turbulence (sv, noise) that used to spatially interpolate Blueback Herring densities at exceeded the threshold was estimated for each diel period and each node of the 100 £ 100 point grid within each survey date in the study and then was subtracted from the mean sv for region. Semivariogram modeling and kriging were performed each echogram (sv, total) to remove these sources from the using the package “geoR” in R statistical computing software acoustic estimate of fish density. Daytime was defined as the (Ribeiro and Diggle 2001; R Development Core Team 2012). period between 1 h after sunrise and 1 h before sunset. Night- The vertical distribution of juvenile Blueback Herring in the time was defined as the period between 1 h after sunset and water column was described by the proportion of the total sa in 1 h before sunrise. Dawn and dusk were defined as the periods each 1-m depth layer throughout each region. within 1 h or less of sunrise or sunset, respectively. The daily Fish passage estimation.—The total net passage of juvenile diel-specific mean sv, noise was estimated from a randomly Blueback Herring (Nnet) through the upriver and downriver selected subset of five echograms from each daytime and sites based on continuous monitoring by fixed-location hori- nighttime period of every other date, defining a 100‑ping zontal transducers was estimated by summing the Nnet, i esti- sequence in each selected echogram, and then estimating the mates from three river channel zones (i), which were mean sv for that sequence, excluding fish echoes or other echo estimated as traces (e.g., ascending bubbles). The mean sv, noise for days and nights not examined were linearly interpolated. Mean sv, Nnet;i D Nd;i ¡ Nu;i; (2) noise for dawn and dusk was linearly interpolated between adjacent day and night periods.

where Nd and Nu are the number of fish moving downriver and Removal of acoustic contributions from gas bubbles rising upriver, respectively, through each channel zone. These at various rates from the sediment at the two monitoring sites parameters were estimated for each hour by the following was necessary to accurately estimate fish density. A randomly Downloaded by [Department Of Fisheries] at 19:15 20 July 2015 equations: selected subset of echograms from the vertical transducer were examined to manually identify the distinctive (left-to- ÀÁ N D T £ A £ s = h s i £ R £ P ; right) diagonal echo pattern of rising bubbles and to determine d v;JBH bs tx ¡ diel d d (3) ÀÁ their mean sv. Given the dynamic nature of bubble formation, Nu D T £ A £ sv; = h s i £ Ru £ ðÞ¡ Pd ; JBH bs tx ¡ diel 1 (4) the proportion of mean echogram sv attributable to bubbles (Pbubbles) was estimated for each selected echogram. First, the where T is the duration of the period in seconds, hourly mean echo strength from tracked SEDs was converted to sv sv,JBHis the sv attributable to juvenile Blueback Herring, (Rudstam et al. 2009). The converted sv values of all tracked hsbs i is derived from the TS mode specific to each transducer SEDs (total) and tracked SEDs classified as bubbles were (tx) and diel period pooled across the time series, Rd and Ru summed for each echogram and divided by the total number are the net rates of downstream and upstream movement in the of valid digital samples in the echograms to derive an equiva- horizontal plane of tracked fish echoes classified as juvenile lent proxy for the mean bubble and total sv, and then the bub- Blueback Herring averaged for each daily diel period, and Pd ble-to-total sv ratio was used as an estimate for Pbubbles. is the proportion of the fish echoes classified as juvenile Blue- Because the Pbubbles estimates by manual classification versus back Herring that are moving downriver for each hour. For automated tracking were shown to be similar for the subset estimating Pd, data from adjacent hours were pooled for hours from the vertical transducer data, automated tracking was used ULTRASONIC FIELD DIRECTION 1249

to classify the bubbles for each echogram collected by the hor- The chi-square test was used to test for significant differen- izontal transducers and then Pbubbles was estimated as previ- ces between the observed proportion in the cumulative net pas- ously described. The mean Pbubbles was 0.17–0.41 at the sage of juvenile Blueback Herring (in thousands) at the downriver site and 0.08–0.12 at the upriver site. downriver site and the expected proportion based on either the The acoustic contribution of residential fish species to the proportion of 0.313 observed in 2008 by Dunning and Gurshin acoustic estimate of fish passage of juvenile Blueback Herring (2012) for the original ultrasonic field or the downriver-to- was minimized by subtracting a constant sv representing their upriver ratio of measured river discharge. The expected net contribution (sv, residents). A time-series average of the mini- passage (Nnet, e) of juvenile Blueback Herring that migrated mum mean sv within each daily diel period, after noise and downriver in the main channel was estimated from the bubble removal, was used to estimate sv, residents. In summary, observed net passage upriver (Nnet, upriver)as the allocation of sv by these steps led to the extraction of peaks

in sv from the time series of mean echogram sv and the inter- Nnet;e D Nnet;upriver £ 0:313 (6) pretation of the peaks as migration episodes of juvenile Blue- back Herring. The s was expressed as v,JBH under null hypothesis 1 and ÀÁ ÀÁ sv;JBH D sv;total ¡ sv;noise £ ðÞ1 ¡ Pbubbles ¡ sv;residents: (5) Nnet;e D Nnet;upriver £ Qdownriver=Qupriver (7)

The number of tracked fish echoes classified as juvenile under null hypothesis 2. Blueback Herring that moved in a downstream direction These comparisons were made for the full time series (Sep- divided by the number moving in either a downstream or tember 8 through October 26) and the peak migration period. upstream direction was used to estimate P of all juvenile d The peak migration period was chosen to exclude the early Blueback Herring, assuming individual movements were rep- period of the study when observed migration events (i.e., resentative of individuals not tracked (i.e., in schools). Down- peaks in N ) were successively absent at the upriver site and stream and upstream swimming direction was estimated as the net fish were staging prior to out-migration. The migration period net direction between the first and last SEDs in the track § also excluded the time when juvenile Blueback Herring were restricted to being within 45 of the horizontal plane and not present or scarce, indicating the out-migration from within §45 of being perpendicular to the major response axis upstream was complete. of the beam. The distance in the horizontal plane between the The densities of juvenile Blueback Herring estimated by first and last SEDs of tracked echoes classified as juvenile trawl and mobile echo sounder surveys were also compared Blueback Herring moving downstream or upstream divided by among channel regions to provide further evidence of safe the elapsed time in the beam was used for R and R . d u downstream passage. All available information from ADCPs, trawls, mobile echo sounder surveys, and fixed-location hydro- Evaluation of Ultrasound Effectiveness on Downriver acoustics was used in a complementary fashion to aid interpre- Passage tation of the net fish passage estimates of juvenile Blueback Herring. The effectiveness of the ultrasound in guiding out-migrating juvenile Blueback Herring past the intake channel was determined by testing two null hypotheses. First, if ultrasound RESULTS Downloaded by [Department Of Fisheries] at 19:15 20 July 2015 has no effect on downriver passage, then the proportion of fish migrating downriver past the intake channel after redirecting River Discharge and Temperature the projectors should be equal to the proportion observed in The velocity-index models were Vm D 0.024 – Vx(70.049 – 2 the presence of the original ultrasonic field. In comparison to 7.635H)(F2, 46 D 43.24, P < 0.001;R D 0.65) at the upriver Dunning and Gurshin (2012), the second null hypothesis site and Vm D 0.024 – Vx(70.049 – 7.635H)(F2, 46 D 56.48, P assumes that passage and distribution of juvenile Blueback < 0.001;R2 D 0.71) at the downriver site. There were three Herring through the main and intake channels in the absence high-flow events of varying magnitude during the study: the of ultrasound were proportional to Q. While the extent to first peaking at 76 m3/s on September 19, the second peaking which juvenile Blueback Herring behave functionally as at 101 m3/s on October 16, and the third peaking at 234 m3/s plankton has not been empirically tested, this proportional-to- on October 21(Figure 3A). The change in river stage increased flow assumption was based on the following: emigration pat- from 0.1 to 0.5 m during these high-flow events (Figure 3B). terns of river herring are related to water flow (Pardue 1983; River discharge only substantially increased at the downriver Kosa and Mather 2001), alosines are attracted to higher current site during the highest-flow event (October 19–23) when dis- velocities (Barry and Kynard 1986; Moser et al. 2000), and charge through the turbines at Crescent was coincidentally water flow is a predictor of entrainment of migratory pelagic minimal. The mean hourly averaged Vm and Q at the upriver species (ASMFC 2009; Grimaldo et al. 2009). site (0.037 m/s and 57 m3/s, respectively) was significantly 1250 GURSHIN ET AL.

higher than that at the downriver site (0.015 m/s and 20 m3/s; Shiner Notropis hudsonius, 53 Brook Silverside Labidesthes P < 0.001). The near-bottom water temperature in the middle sicculus, 6 Gizzard Shad Dorosoma cepedianum, 2 Black of the channel at the upriver site decreased from 24.0Cto Crappie Pomoxis nigromaculatus, 1 Bluegill Lepomis macro- 12.2C during the study (Figure 3C). chirus, 1 Channel Catfish Ictalurus punctatus, and 1 Common Shiner Luxilus cornutus. The total length of Blueback Herring averaged 70 mm and ranged from 52 to 111 mm. The mode of Trawl Estimates of Abundance the TS distribution predicted from the length distribution of Juvenile Blueback Herring was the most abundant species the Blueback Herring catch closely matched the observed TS caught before out-migration was complete. Of the 10,717 fish modes from echo sounding (Figure 4). and 9 species caught in nighttime trawl surveys, 98.8% were The abundance of juvenile Blueback Herring differed Blueback Herring. The other species caught were 62 Spottail among regions and over time. Mean CPUE of Blueback Her- ring increased during the first 2 weeks in all regions, decreased over the next two surveys on September 21 and 28, and then slightly increased on October 2 before continuing to decrease (Figure 5A). Between September 9 and October 10 when Blueback Herring catches were substantial, the mean log10 (x C 1) transformed CPUE was significantly different among some regions (ANOVA: F3, 80 D 5.74; P D 0.001). Trans- formed CPUE of Blueback Herring in the main channel upriver and downriver of the ultrasonic projectors was significantly higher than in the intake channel region, but not different than in the ultrasound gradient region (Figure 5B). The arithmetic mean CPUE in the downriver region (161 fish/tow) was approximately 94% of the CPUE in the upriver region (172 fish/tow) and was 2.5 times higher than the CPUE in the intake channel (64 fish/tow). On October 15, few Blueback Herring were caught by the pelagic trawl and by October 25 only two individuals were caught in 12 tows (Figure 5C).

Mobile Echo Sounder Survey Estimates of Abundance and Distribution Echograms from the mobile echo sounder surveys contained predominantly clusters of overlapping echoes classiﬁed as juvenile Blueback Herring that were patchily distributed in the mid to upper water column. Geostatistical analysis revealed that an average of 79% of the variance in the density Downloaded by [Department Of Fisheries] at 19:15 20 July 2015

FIGURE 3. Characteristics of the Mohawk River, including (A) the hourly mean river discharge estimates based on a velocity-index equation using continuous 2.5-min observations by a 500-kHz horizontally aimed acoustic Dopp- ler current proﬁler located near the eastern shore of the main channel upriver and downriver of the ultrasonic projectors, and hourly discharge through the turbines calculated by the power generation at the Crescent Hydroelectric Proj- FIGURE 4. Frequency distribution of the predicted target strength (TS) ect, (B) the change in hourly mean river stage, and (C) the mean water temper- based on the length of Blueback Herring caught by pelagic trawl (Trawl ature measured every 15 min by two HOBO Pro v2 data loggers attached to a Catch), mean TS of echo traces from continuous ﬁxed-location echo sounding transducer mount within 1 m of the river bottom and located at the center of (Echo Traces), and in situ TS of all single echo detections from all daytime the main channel upriver of the ultrasonic projectors. mobile echo sounder surveys combined (Mobile SEDs). ULTRASONIC FIELD DIRECTION 1251

FIGURE 5. Catch per unit effort (CPUE) of juvenile Blueback Herring caught during the night by pelagic trawl showing (A) the survey-mean CPUE from Sep- tember 9 through October 10, 2012, and the mean CPUE § 95% confidence intervals (CI) of all trawls during (B) September 9 through October 10, 2012, (locations that share a letter are not significantly different) and (C) October 15 through October 25, 2012. The trawls were done in four regions of the Mohawk River: upriver of the 122–128 kHz ultrasonic projectors and 420-kHz fixed-location echo-sounding site (upriver), in the area of increasing downriver ultrasound gradient (ultrasound gradient), in the intake channel downriver of the ultrasonic projectors (intake channel), and in the main channel downriver of the ultrasonic projectors (downriver main channel).

of juvenile Blueback Herring was explained by spatial auto- were not signiﬁcantly different based on paired t-tests of the correlation that ranged to an average distance of 145 m. Juve- inter-regional differences among the seven echo sounder surveys nile Blueback Herring were mostly found within 3 m of the (Table 2). However, the mean density and total abundance of surface during the day, where they accounted for about 80– juvenile Blueback Herring in the upriver and downriver main 90% of the mean sa averaged among the seven echo sounder channel regions were both signiﬁcantly greater than the density surveys. and abundance in the intake channel. The downriver main Juvenile Blueback Herring progressed from being present channel averaged 35 times more juvenile Blueback Herring than throughout each survey region on September 9 to being nearly the intake channel and averaged 91% of the combined abun-

Downloaded by [Department Of Fisheries] at 19:15 20 July 2015 absent in the river on October 26 (Figures 6, 7). Juvenile Blue- dance in the intake and downriver main channel regions. back Herring were distributed at similar densities throughout all regions surveyed at the start of the study, although the abundance in the intake channel was more than 50% less than Continuous Monitoring of Downstream Fish Passage in the other regions. Thereafter, densities were relatively low The mean Rd for juvenile Blueback Herring measured by in the intake channel and higher in the downriver main chan- horizontal split‑beam tracking ranged from 0.15 to 0.27 m/s nel. By September 3, clusters of juvenile Blueback Herring (2.1–3.9 body lengths/s) when Rd was averaged by day, diel were found far upriver, in front of the ultrasonic projectors, period, and site. The mean Pd ranged from 0.45 in the east and downriver in the main channel, but only in a few clusters main channel at the downriver site during dawn hours (0500– of low density in the intake channel. The highest concentra- 0700 hours) to 0.68 in the center main channel at the upriver tions of juvenile Blueback Herring were seen in the downriver site during night hours (2000–0400 hours). The Pd was higher main channel region on October 3 but largely dissipated on at the upriver site than at the downriver site during the night subsequent surveys. During the same surveys, juvenile Blue- and dawn, while the sites were less different during the day back Herring were scarce in the intake channel. and at dusk. When Pd was less than 0.5, net downstream pas- 28w?>Mean density and total abundance of juvenile Blue- sage was negative and was interpreted as more ﬁsh moving back Herring in the upriver and downriver main channel regions through the acoustic beam in the upstream direction. 1252 GURSHIN ET AL. Downloaded by [Department Of Fisheries] at 19:15 20 July 2015 FIGURE 6. Kriged maps of the juvenile Blueback Herring density estimated from data collected by a 420-kHz split-beam echo sounder used on seven systematic echo sounder surveys at the Crescent Hydroelectric Project during the day between September 9 and October 26, 2012. The red triangle marks the location of the ultrasonic projectors.

A diel pattern was generally present in the echograms asso- migration, and postmigration periods (Figure 9). Net down- ciated with a passage event of juvenile Blueback Herring stream passage was consistently low before September 20 and (Figure 8). During the night, individual echoes were observed then peaked multiple times at the upriver site. Net downstream while clusters of multiple echoes (i.e., “schools”) were largely passage at the downriver site was relatively low between Sep- absent. At dawn, echoes became clustered and school echoes tember 20 and October 6 until two large pulses of juvenile formed. Schools of juvenile Blueback Herring appeared to Blueback Herring passed through on October 6 and 8. Peak move through the beams in upriver and downriver directions net downstream passage occurred at both sites during October at both sites. When peaks in sv,JBHwere observed, they typi- 5–6 and then declined in the following days. This difference cally occurred between 0600 and 1200 hours. resulted in different inﬂection points in the cumulative curves The time series of net downstream passage show three pat- for net downstream passage at the two sites (Figure 10). While terns of variability interpreted as premigration, active out- net downstream passage after October 14 was variably low, ULTRASONIC FIELD DIRECTION 1253

higher than the percentage observed by Dunning and Gurshin (2012) (31.3%; x2 D 7,905.8, P < 0.001) and the percentage expected based on river flow (49.1%; x2 D 2,499.8, P < 0.001). Within the peak migration period, the coefficient of variation (standard error/mean) of the mean daily net downstream passage was about 45% at the upriver site and 38% at the downriver site. The coefficient of variation of the mean hourly net downstream passage for each day within the migration period averaged 5% at the upriver site and 11% at the downriver site. Inclusion of the net passage during the premigration and postmigration periods, which could be in either direction at times, may not accurately represent passage of actively out-migrating juvenile Blueback Herring. However, even with the premigration and postmigration periods included, the percentage of the net downstream passage of juvenile Blueback Herring at the upriver site that moved downriver in the main channel was about 43.7%, which was significantly higher than 31.3% for the first ultrasonic projector configuration observed by Dunning and Gurshin (2012) (x2 D 825.4, P < 0.001) and the percentage expected based on river flow (40.9%; x2 D 36.8, P < 0.001).

FIGURE 7. (A) Mean density and (B) total abundance of juvenile Blueback Herring sampled along cross-river transects that were surveyed by a vessel- mounted 420-kHz split-beam echo sounder within four regions of the Mohawk DISCUSSION River near the Crescent Hydroelectric Project: (1) the northern upriver exten- These results demonstrate that the majority of out-migrat- sion (sampled on four occasions; n/s indicates the dates not sampled), the main ing juvenile Blueback Herring avoided the intake channel after channel (2) upriver and (3) downriver of ultrasonic (122–128 kHz) projectors the ultrasonic field was redirected. After directing the ultra- aimed to deter fish from entering the intake channel, and (4) the intake sonic projector upriver, continuous acoustic monitoring channel. showed a significant increase in the proportion (76.5%) of juvenile Blueback Herring migrating downriver in the main migration during this period was considered complete based channel during the peak migration period. Repeated echo on juvenile Blueback Herring being scarcely observed in the sounder surveys show that the 35-fold difference in abundance trawl and echo sounder surveys. The peak migration period of between the downriver main channel and the intake channel September 20 through October 14 represented 72% of the was an order of magnitude higher than the average fivefold cumulative net downstream passage for the full time series. difference observed prior to redirecting the projectors During the peak migration period, 76.5% of the out-migrat- (Dunning and Gurshin 2012). The net downstream passage of ing juvenile Blueback Herring passage at the upriver site juvenile Blueback Herring through the main channel was sig- moved downriver in the main channel, which was significantly nificantly higher than expected, assuming passage was Downloaded by [Department Of Fisheries] at 19:15 20 July 2015

TABLE 2. Arithmetic means and paired t-tests for differences in the density (ﬁsh/m2) and total abundance (thousands) of juvenile Blueback Herring between regions of the main channel upriver and downriver of the ultrasonic projectors and in the intake channel among seven mobile echo sounder surveys. The null hypothesis is the mean difference equals 0. Region Density Abundance

Means Upriver main channel (U) 0.91 218.28 Downriver main channel (D) 1.43 175.48 Intake channel (I) 0.33 22.51 Paired t-tests (t, P) U–D ¡0.52 (¡1.68, 0.143) 42.81 (0.95, 0.379) U–I 0.58 (4.24, 0.005) 195.77 (4.24, 0.005) D–I 1.10 (2.78, 0.032) 152.96 (3.10, 0.021) 1254 GURSHIN ET AL.

FIGURE 9. Hourly (gray) and daily (black) estimates of net downstream passage of juvenile Blueback Herring in the main channel (A) upriver and (B) downriver of ultrasonic projectors at the Crescent Hydroelectric Project on the Mohawk River from September 8 through October 26, 2012. The vertical dashed lines delineate the out-migration period from September 20 through October 14. Note that negative net downstream passage indicates upstream movement.

used in regulatory practice (ASMFC 2009; USEPA 2014) that entrainment and impingement is directly proportional to the ﬂow of water withdrawn by a power generation facility. Previ- ous studies evaluated the effectiveness of ultrasound as a ﬁsh guidance technology by on–off or before-after–control-impact paired experimental designs (Ploskey et al. 1995; Ross et al. 1996; Skalski et al. 1996). In this study, the ultrasonic projectors were operating throughout the study for the same reasons given by Dunning and Gurshin (2012): the unknown patchy temporal distribution of the downstream migration of juvenile Blueback Herring limits the assurance of an adequately

FIGURE 8. Volume backscattering strength apportioned to juvenile Blue-

back Herring (JBH Sv) for the centrally located horizontal transducers upriver

Downloaded by [Department Of Fisheries] at 19:15 20 July 2015 (upper group of panels) and downriver (lower group of panels) of the ultrasonic projectors for the full survey duration (top panels within each upriver and downriver panel group), for a 24-hour period on October 6, 2012, during the peak migration (middle panels, and also the grey bar in the top panels), and for echograms sampled at A–D (labeled on the middle panels) within that 24-hour period showing the diel variability (bottom panels).

proportional to river discharge, which further supports the idea that the increased passage of juvenile Blueback Herring through the main channel was a result of redirecting the ultrasonic ﬁeld and not a change in river discharge between the two studies. FIGURE 10. Cumulative net downstream passage of juvenile Blueback Her- The evaluation of the effectiveness of ultrasound as a ﬁsh ring in the main channel upriver and downriver of ultrasonic projectors at the Crescent Hydroelectric Project from September 8 through October 26, 2012. deterrent at Crescent compared the observed cumulative net The vertical dashed lines delineate the out-migration period from September downstream passage estimates of juvenile Blueback Herring 20 through October 14. Note that negative net downstream passage indicates to the expected number based on the assumption commonly upstream movement. ULTRASONIC FIELD DIRECTION 1255

designed on–off experiment and the regulatory preference that The direction and magnitude of net passage of juvenile most of the juvenile Blueback Herring migrate down the main Blueback Herring determined by continuous horizontal echo channel rather than through the turbines. sounding in a shallow noisy riverine environment is subject to Dunning and Gurshin (2012) interpreted their results as several dynamically changing sources of variability and uncer- either (1) the flow null hypothesis is a valid assumption and tainty. Measurement error stemming from system calibration, the ultrasonic projectors were partially effective at diverting TS, species classification, behavior, or environmental condi- juvenile Blueback Herring from entering the intake channel or tions can behave randomly over many observations and is (2) the observed proportion (31.3%) of juvenile Blueback Her- often considered negligible to the sampling error (Demer ring moving downriver was evidence that the flow assumption 2004) or is incorporated in variance estimators (Skalski et al. is invalid and could be explained by other abiotic and biotic 1996). In contrast to the imprecision caused by random error, factors. If (1) was the correct assumption, by reconfiguring the potential sources of uncertainty associated with TS or behavior ultrasound to allow more exposure time to increasing SPLs for can contribute to systematic deviation from the true value juvenile Blueback Herring as they migrate downriver, then (bias) up to 50% (Simmonds and MacLennan 2005). However, their proportion moving downriver should increase if the ultra- biased estimates can still provide useful relative indices of sound is an effective deterrent. If there was no relative abundance (Rose et al. 2000). increase of juvenile Blueback Herring moving downriver in In contrast to other fish passage studies that either did not the main channel, then perhaps an alternative mechanism quantify uncertainty or did not account for variability in noise explains the observed downstream passage through the main or TS (Mulligan and Kieser 1996; Ransom et al. 1996), a channel. multistep processing approach was used in this study to The results from intensive spatial and temporal sampling derive an index of abundance and net downstream passage by three sampling methods demonstrated that the preponder- of juvenile Blueback Herring based on parameters that ance of juvenile Blueback Herring migrated downriver in the accounted for periodic changes in fish swimming direction main channel during the fall of 2012 and avoided the intake and speed, TS, and acoustic contributions from other scatter- channel leading to the hydroelectric turbines at Crescent, thus ers (i.e., Chaoborus sp., bubbles, and other fish). The removal reducing the potential for turbine passage mortality. The techniques of acoustic contributions of background noise, acoustic size and visual patterns of the echoes in the echo- Chaoborus sp., bubbles, and resident fish used in this study grams, coupled with the species identification and CPUE by were appropriate and similar to other hydroacoustic studies trawls, confirmed that juvenile Blueback Herring were the in noisy environments (Malinen et al. 2005; De Robertis and principal pelagic species observed. The general migratory Higginbottom 2007). patterns observed were that juvenile Blueback Herring were Acoustic estimates of fish passage used the modal TS spe- moving back and forth between sites or “milling” around in cific to each diel period and transducer to account for potential the Dam A pond before out-migration, net downstream pas- TS differences known to arise from diel variation in behavior sage peaked at the upriver site and CPUE steadily declined and different sound incidence angles among transducers throughout all regions during the migration period, and juve- (Foote 1980; Luecke and Wurtsbaugh 1993; Hjellvik et al. nile Blueback Herring were nearly absent following a high- 2004; Henderson et al. 2007). The direction of the horizontal flow event on October 15. transducers across the river channel was assumed to generally The time series of net downstream passage and the cumula- produce a side-aspect TS distribution when fish swim through tive net downstream passage show that the majority of the fish the beam during their downstream migration. The TS mode

Downloaded by [Department Of Fisheries] at 19:15 20 July 2015 passed the downriver site on October 6 after several peaks in observed by horizontal transducers ranged from ¡50 to net downstream passage were observed upriver. Net down- ¡44 dB, which was similar in magnitude to Gurshin (2012) stream passage showed periods of upstream movement, while and Brooking and Rudstam (2009). The estimates used for the the mobile echo sounder survey revealed over time that juve- net rate of downstream or upstream movement (averaged 2.4– nile Blueback Herring congregated in the downriver survey 2.6 body lengths/s) were consistent with the range of swim- region at high densities. The back-and-forth movements at the ming speeds reported for Alewives from previous split-beam downriver site and the accumulation of juvenile Blueback Her- tracking by Arrhenius et al. (2000). ring downriver in the main channel provided some evidence of While unquantified individual sources of error make the residency duration until the fish either exited through Water- accuracy of the net downstream passage estimates difficult ford Flight Lock 6 or a high-flow event like the one on October to assess as absolute abundance estimates, these estimates 15 facilitated their passage through the opening in the flash- as suitable relative measures for testing the null hypotheses boards or over the dam. This result is consistent with the in net downstream passage were corroborated by the behavior of juvenile Blueback Herring moving downstream in observed trends and verification provided by trawl and response to environmental cues, such as high flow (Kosa and mobile echo sounder surveys. Diagnosing and quantifying Mather 2001) or decreasing temperature (O’Leary and Kynard the many potential sources of uncertainty in the down- 1986; Iafrate and Oliveira 2008). stream fish passage estimates was beyond the scope of this 1256 GURSHIN ET AL.

study. Further investigation into estimating uncertainty in Arrhenius, F., B. J. A. M. Benneheij, L. G. Rudstam, and D. Boisclair. 2000. multiparameter estimates of fish passage using simulation Can stationary bottom split-beam hydroacoustics be used to measure fish or resampling techniques (Rose et al. 2000) or Bayesian swimming speed in situ? Fisheries Research 45:31–41. ASMFC (Atlantic States Marine Fisheries Commission). 2009. Amendment 2 statistics (Punt and Hilborn 1997; Anderson et al. 2007; to the interstate fishery management plan for shad and river herring. € Fassler et al. 2009) is warranted for providing reliable ASMFC, Washington, D.C. absolute estimates of abundance of migratory anadromous ASMFC (Atlantic States Marine Fisheries Commission). 2012. River herring fishes. benchmark stock assessment. ASMFC, Stock Assessment Report 12-02, After directing the ultrasonic field upriver, the percentage Washington, D.C. Barry, T., and B. Kynard. 1986. Attraction of adult American Shad to fish lifts of out-migrating juvenile Blueback Herring that bypassed the at Holyoke Dam, Connecticut River. North American Journal of Fisheries turbine channel increased from 31.3% to 76.5%. At Crescent, Management 6:233–241. this improvement in downstream passage through the main Brooking, T. E., and L. G. Rudstam. 2009. Hydroacoustic target strength dis- channel represents the potential survival of an additional tributions of alewives in a net-cage compared with field surveys: decipher- 76,840 juvenile Blueback Herring, based on an out-migration ing target strength distributions and effect on density estimates. Transactions of the American Fisheries Society 138:471–486. of 1 million Blueback Herring (a similar magnitude to what Cada, G. F. 2001. The development of advanced hydroelectric turbines to was observed in this study), a short-term turbine passage sur- improve fish passage survival. Fisheries 26(9):14–23. vival of 96% for the two Kaplan-type turbines (Mathur et al. Cressie, N., and D. M. Hawkins. 1980. Robust estimation of the variogram: I. 1996) and 70% for the two Francis-type turbines (Cada 2001), Journal of the International Association for Mathematical Geology 12:115– equal turbine use, and near 100% survival through the main 125. De Robertis, A., and I. Higginbottom. 2007. A post-processing technique to channel. This study has demonstrated how pulsed ultrasound estimate the signal-to-noise ratio and remove echosounder background can improve downstream passage of juvenile Blueback Her- noise. ICES Journal of Marine Science 64:1282–1291. ring at hydropower dams when the ultrasonic field direction Demer, D. A. 2004. An estimate of error for the CCAMLR 2000 survey esti- optimizes exposure to allow sufficient time for approaching mate of krill biomass. Deep-Sea Research Part II Topical Studies in Ocean- fish to respond and avoid turbine intakes. The improved sur- ography 51:1237–1251. Dunning, D. J., and C. W. D. Gurshin. 2012. Downriver passage of juvenile vival from ultrasonic fish deterrent systems can be considered Blueback Herring near an ultrasonic field in the Mohawk River. North beneficial to conservation efforts for restoring population lev- American Journal of Fisheries Management 32:365–380. els of Blueback Herring, and possibly other alosines, espe- Dunning, D. J., and Q. E. Ross. 2010. Effect of radio-tagging on escape reac- cially if ultrasound is used at multiple dams on the same river tions of adult Blueback Herring to ultrasound. North American Journal of or at sites with high turbine passage mortality. Fisheries Management 30:26–32. Dunning D. J., Q. E. Ross, P. Geoghegan, J. J. Reichle, J. K. Menezes, and J. K. Watson. 1992. Alewives avoid high-frequency sound. North American Journal of Fisheries Management 12:407–416. ACKNOWLEDGMENTS Ehrenberg, J. E., and T. C. Torkelson. 2000. FM slide (chirp) signals: a technique for significantly improving the signal-to-noise performance in hydroa- This study was funded by the New York Power Authority coustic assessment systems. Fisheries Research 47:193–199. (Contract Number 4500190871). Fish were collected under Enzenhofer, H. J., and G. Cronkite. 2000. Fixed location hydroacoustic estima- Scientific Permit 1250 issued by the New York State Depart- tion of fish migration in the riverine environment: an operational manual. ment of Environmental Conservation. Fieldwork was per- Canadian Technical Report of Fisheries and Aquatic Sciences 2312:46. € formed under Canal Permit C2W100058 issued by the New Fassler, S. M. M., A. S. Brierley, and P. G. Fernandes. 2009. A Bayesian approach to estimating target strength. ICES Journal of Marine Science York State Canal Corporation. Assistance with instrumenta- 66:1197–1204. tion from K. Kumagai, B. McFadden, and P. Nealson (Hydroa- Foote, K. G. 1980. Effect of fish behaviour on echo energy: the need for meas- Downloaded by [Department Of Fisheries] at 19:15 20 July 2015 coustic Technology) was greatly appreciated. Assistance with urements of orientation distributions. Journal du Conseil International pour analysis and preparation of this manuscript was received from l’Exploration de la Mer 39:193–201. D. Ettere, P. Geoghegan, J. Howarth, and B. O’Mara. Data col- Foote, K. G. 1983. Linearity of fisheries acoustics, with addition theorems. Journal of the Acoustical Society of America 73:1932–1940. lection would not have been possible without the field assis- Freon, P., F. Gerlotto, and M. Soria. 1996. Diel variability of school structure tance of C. Baker, E. Fel’Dotto, A. Frizzell, S. Maxwell, S. with special reference to transition periods. ICES Journal of Marine Science Schanke, E. Sobo, C. Thompson, and J. Wollis. 53:459–464. Gibson, A. J., and R. A. Myers. 2002. Effectiveness of a high-frequency sound fish diversion system at the Annapolis Tidal Hydroelectric Generating Sta- tion, Nova Scotia. North American Journal of Fisheries Management REFERENCES 22:770–784. Anderson, C. I. H., J. K. Horne, and J. Boyle. 2007. Classifying multi-fre- Grimaldo, L. F., T. Sommer, N. V. Ark, G. Jones, E. Holland, P. B. Moyle, B. quency fisheries acoustic data using a robust probablistic classification tech- Herbold, and P. Smith. 2009. Factors affecting fish entrainment into massive nique. Journal of the Acoustical Society of America 121:230–237. water diversions in a tidal freshwater estuary: can fish losses be managed? ANSI/ASA (American National Standard Institute/Acoustical Society of North American Journal of Fisheries Management 29:1253–1270. America). 2012. ANSI/ASA S1.20-2012 procedures for calibration of Guillard, J. 1998. Daily migration cycles of fish populations in a tropical estu- underwater electroacoustic transducers. ASA, S1 Acoustics, Melville, New ary (Sine-Salloum, Senegal) using a horizontal-directed split-beam trans- York. ducer and multibeam sonar. Fisheries Research 35:23–31. ULTRASONIC FIELD DIRECTION 1257

Gurshin, C. W. D. 2012. Target strength measurements of juvenile Blueback Connecticut River. Transactions of the American Fisheries Society Herring from the Mohawk River, New York. North American Journal of 115:529–536. Fisheries Management 32:381–386. Paramo, J., and R. Roa. 2003. Acoustic-geostatistical assessment and habitat- Gurshin, C. W. D., W. H. Howell, and J. M. Jech. 2013. Synoptic acoustic and abundance relations of small pelagic fish from the Colombian Caribbean. trawl surveys of spring-spawning Atlantic Cod in the Gulf of Maine cod Fisheries Research 60:309–319. spawning area. Fisheries Research 141:44–61. Pardue, G. B. 1983. Habitat suitability index models: Alewife and Blueback Henderson, M. J., J. K. Horne, and R. H. Towler. 2007. The influence of beam Herring. U.S. Fish and Wildlife Service FWS/OBS-82/10.58. position and swimming direction on fish target strength. ICES Journal of Parker-Stetter, S. L., L. G. Rudstam, P. J. Sullivan, and D. M. Warner. 2009. Marine Science 65:236–237. Standard operating procedures for fisheries acoustic surveys in the Great Hjellvik, V., O. R. Godø, and D. Tjøstheim. 2004. Diurnal variation in acoustic Lakes. Great Lakes Fishery Commission, Special Publication 09-01, Ann densities: why do we see less in the dark? Canadian Journal of Fisheries and Arbor, Michigan. Aquatic Sciences 61:2237–2254. Pedersen, B., and M. V. Trevorrow. 1999. Continuous monitoring of fish in a Iafrate, J., and K. Oliveira. 2008. Factors affecting migration patterns of river shallow channel using a fixed horizontal sonar. Journal of the Acoustical herring in a coastal Massachusetts stream. Environmental Biology of Fishes Society of America 105:3126–3135. 81:101–110. Petitgas, P. 1993. Geostatistics for fish stock assessments: a review and an Johannesson, K. A., and R. B. Mitson. 1983. Fisheries acoustics: a practical acoustic application. ICES Journal of Marine Science 50:285–298. manual for aquatic biomass estimation. FAO Fisheries Technical Paper 240. Petitgas, P., and J. J. Levenez. 1996. Spatial organization of pelagic fish: echo- Knudsen, F. R., P. Larsson, and P. J. Jakobsen. 2006. Acoustic scattering from gram structure, spatio-temporal condition, and biomass in Senegalese a larval insect (Chaoborus flavicans) at six echosounder frequencies: impli- waters. ICES Journal of Marine Science 53:147–153. cation for acoustic estimates of fish abundance 79:84–89. Ploskey, G., J. Nestler, G. Weeks, and C. Schilt. 1995. Evaluation of an inte- Kosa, J. T., and M. E. Mather. 2001. Processes contributing to variability in grated fish-protection system. Pages 162–171 in J. J. Cassidy, editor. Water- regional patterns of juvenile Blueback Herring abundance across small power ‘95 proceedings of the international conference on waterpower. coastal systems. Transactions of American Fisheries Society 130:600–619. American Society of Civil Engineers, New York. Krumme, U., and U. Saint-Paul. 2003. Observations of fish migration in a mac- Popper, A. N., D. T. T. Plachta, D. A. Mann, and D. Higgs. 2004. Response of rotidal mangrove channel in northern Brazil using a 200-kHz spli-beam clupeid fish to ultrasound: a review. ICES Journal of Marine Science sonar. Aquatic Living Resources 16:175–184. 61:1057–1061. Kubecka, J., and M. Wittingerova. 1998. Horizontal beaming as a crucial com- Punt, A. E., and R. Hilborn. 1997. Fisheries stock assessment and decision ponent of acoustic fish stock assessment in freshwater reservoirs. Fisheries analysis: the Bayesian approach. Reviews in Fish Biology and Fisheries Research 35:99–106. 7:35–63. Luecke, C., and W. A. Wurtsbaugh. 1993. Effects of moonlight and daylight R Development Core Team. 2012. R: a language and environment for statisti- on hydroacoustic estimates of pelagic fish abundance. Transactions of cal computing, version 2.13.2. R Foundation for Statistical Computing, American Fisheries Society 122:112–120. Vienna. MacLennan, D. N., P. G. Fernandes, and J. Dalen. 2002. A consistent approach Ransom, B. H., T. W. Steig, and P. A. Nelson. 1996. Comparison of hydroa- to definitions and symbols in fisheries acoustics. ICES Journal of Marine coustic and net catch estimates of Pacific salmon smolt (Oncorhynchus sp.) Science 59:365–369. passage at hydroelectric dams in the Columbia River basin, USA. ICES Malinen, T., A. Tuomaala, and H. Peltonen. 2005. Hydroacoustic fish stock Journal of Marine Science 53:477–481. assessement in the presence of dense aggregations of Chaoborus larvae. Ribeiro, P. J. Jr., and P. J. Diggle. 2001. geoR: a package for geostatistical Canadian Journal of Fisheries and Aquatic Sciences 62:245–249. analysis. R-News (June):15–18. Mann, D. A., Z. Lu, and A. N. Popper. 1997. A clupeid fish can detect ultra- Rivoirard, J., J. Simmonds, K. G. Foote, P. Fernandes, and N. Bez. 2000. Geo- sound. Nature 389:341. statistics for estimating fish abundance. Blackwell Scientific Publications, Mathur, D., P. G. Heisey, K. J. McGrath, and T. R. Tatham. 1996. Juvenile Malden, Massachusetts. Blueback Herring (Alosa aestivalis) survival via turbine and spillway. Water Rose, G., S. Gauthier, and G. Lawson. 2000. Acoustic surveys in Resources Bulletin 32:155–161. the full monte: simulating uncertainty. Aquatic Living Resources 13:367– Mello, L. G. S., and G. A. Rose. 2005. Using geostatistics to quantify seasonal 372. distribution and aggregation patterns of fishes: an example of Atlantic Cod Ross, Q. E., D. J. Dunning, J. K. Menezes, M. J. Kenna Jr., and G. Tiller. 1996. (Gadus morhua). Canadian Journal of Fisheries and Aquatic Sciences Reducing impingement of Alewives with high-frequency sound at a power Downloaded by [Department Of Fisheries] at 19:15 20 July 2015 62:659–670. plant intake on Lake Ontario. North American Journal of Fisheries Manage- Morlock, S. E., H. T. Nguyen, and J. H. Ross. 2002. Feasibility of acous- ment 16:548–559. tic doppler velocity meters for the production of discharge records from Ross, Q. E., D. J. Dunning, R. Thorne, J. K. Menezes, G. Tiller, and J. K. Wat- U.S. Geological Survey streamflow-gaging stations. U.S. Geological son. 1993. Response of Alewives to high-frequency sound at a power plant Survey, Water-Resources Investigations Report 01-4157, Indianapolis, intake on Lake Ontario. North American Journal of Fisheries Management Indiana. 13:291–303. Moser, M. L., M. Darazsdi, and J. R. Hall. 2000. Improving passage efficiency Rudstam, L. G., S. L. Parker-Stetter, P. J. Sullivan, and D. M. Warner. 2009. of adult American Shad at low-elevation dams with navigation locks. North Towards a standard operating procedure for fishery acoustic surveys in the American Journal of Fisheries Management 20:376–385. Laurentian Great Lakes, North America. ICES Journal of Marine Science Mulligan, T. J., and R. Kieser. 1996. A split-beam echo-counting model for 66:1391–1397. riverine use. ICES Journal of Marine Science 53:403–406. Sawada, K., M. Furusawa, and N. J. Williamson. 1993. Conditions for the pre- Nestler, J. M., G. R. Ploskey, and J. Pickens. 1992. Response of Blueback Her- cise measurement of fish target strength in situ. Journal of the Marine ring to high-frequency sound and implications for reducing entrainment at Acoustical Society of Japan 20:73–79. hydropower dams. North American Journal of Fisheries Management Schilt, C. R. 2007. Developing fish passage and protection at hydropower 12:667–683. dams. Applied Animal Behaviour Science 104:295–325. O’Leary, J. A., and B. Kynard. 1986. Behavior, length, and sex ratio of Schmidt, R. E., B. M. Jessop, and J. E. Hightower. 2003. Status of river herring seaward-migrating juvenile American Shad and Blueback Herring in the stocks in large rivers. Pages 171–182 in K. E. Limburg and J. R. Waldman, 1258 GURSHIN ET AL.

editors. Biodiversity, status, and conservation of the world’s shads. Ameri- the Columbia River, Washington. Canadian Journal of Fisheries and can Fisheries Society, Symposium 35, Bethesda, Maryland. Aquatic Sciences 53:2188–2198. Simmonds, E. J., and R. J. Fryer. 1996. Which are better, random or systematic Teraguchi, M., and T. G. Northcote. 1966. Vertical distribution and migration acoustic surveys? A simulation using North Sea herring as an example. of Chaborus flaviscans larvae in Corbett Lake, British Columbia. Limnol- ICES Journal of Marine Science 53:39–50. ogy and Oceanography 11:164–176. Simmonds, E. J., N. J. Williamson, F. Gerlotto, and A. Aglen. 1992. Acoustic USEPA (U.S. Environmental Protection Agency). 2014. National pollutant survey design and analysis procedure: a comprehensive review of current discharge elimination system—final regulations to establish requirements practice. ICES Cooperative Research Report 187:127. for cooling water intake structures at existing facilities and amend require- Simmonds, J., and D. MacLennan. 2005. Fisheries acoustics: theory and ments a phase I facilities. Federal Register 79:158(15 August 2014):48300– practice, second edition. Blackwell Scientific Publications, Oxford, 48439. UK. Warner, D. M., L. G. Rudstam, and R. A. Klumb. 2002. In situ target strength Skalski, J. R., G. E. Johnson, C. M. Sullivan, E. Kudera, and M. W. Erho. of Alewives in freshwater. Transactions of the American Fisheries Society 1996. Statistical evaluation of turbine bypass efficiency at Wells Dam on 131:212–223. Downloaded by [Department Of Fisheries] at 19:15 20 July 2015 This article was downloaded by: [Department Of Fisheries] On: 20 July 2015, At: 19:15 Publisher: Taylor & Francis Informa Ltd Registered in England and Wales Registered Number: 1072954 Registered office: 5 Howick Place, London, SW1P 1WG

North American Journal of Fisheries Management Publication details, including instructions for authors and subscription information: http://www.tandfonline.com/loi/ujfm20 Movement and Spatial Distribution of Common Carp in a South Dakota Glacial Lake System: Implications for Management and Removal Matthew J. Hennena & Michael L. Browna a Northern Plains Biostress Laboratory, Department of Natural Resource Management, South Dakota State University, Box 2140B, Brookings, South Dakota 57007, USA Published online: 10 Dec 2014.

Click for updates

To cite this article: Matthew J. Hennen & Michael L. Brown (2014) Movement and Spatial Distribution of Common Carp in a South Dakota Glacial Lake System: Implications for Management and Removal, North American Journal of Fisheries Management, 34:6, 1270-1281, DOI: 10.1080/02755947.2014.959674 To link to this article: http://dx.doi.org/10.1080/02755947.2014.959674

PLEASE SCROLL DOWN FOR ARTICLE

Taylor & Francis makes every effort to ensure the accuracy of all the information (the “Content”) contained in the publications on our platform. However, Taylor & Francis, our agents, and our licensors make no representations or warranties whatsoever as to the accuracy, completeness, or suitability for any purpose of the Content. Any opinions and views expressed in this publication are the opinions and views of the authors, and are not the views of or endorsed by Taylor & Francis. The accuracy of the Content should not be relied upon and should be independently verified with primary sources of information. Taylor and Francis shall not be liable for any losses, actions, claims, proceedings, demands, costs, expenses, damages, and other liabilities whatsoever or howsoever caused arising directly or indirectly in connection with, in relation to or arising out of the use of the Content. This article may be used for research, teaching, and private study purposes. Any substantial or systematic reproduction, redistribution, reselling, loan, sub-licensing, systematic supply, or distribution in any form to anyone is expressly forbidden. Terms & Conditions of access and use can be found at http:// www.tandfonline.com/page/terms-and-conditions North American Journal of Fisheries Management 34:1270–1281, 2014 Ó American Fisheries Society 2014 ISSN: 0275-5947 print / 1548-8675 online DOI: 10.1080/02755947.2014.959674

ARTICLE

Movement and Spatial Distribution of Common Carp in a South Dakota Glacial Lake System: Implications for Management and Removal

Matthew J. Hennen* and Michael L. Brown Northern Plains Biostress Laboratory, Department of Natural Resource Management, South Dakota State University, Box 2140B, Brookings, South Dakota 57007, USA

Abstract Common Carp Cyprinus carpio form dense populations that often negatively impact water quality and clarity, native fish communities, and aquatic plant growth in aquatic ecosystems outside their native range, including the U.S. Midwest. In an attempt to guide future management and control activities, we used acoustic telemetry to quantify monthly spatial distributions and movement patterns for adult Common Carp in Brant and Round lakes, South Dakota, during open-water periods for 2 years. During a span of 12 months of manual tracking in the spring, summer, and fall seasons, we obtained 530 locations from 19 acoustically tagged fish. In general, fish displayed a clear tendency to aggregate in shallow depths close to shore during the spawning period in May and June and also postspawn during August. In contrast, fish dispersed offshore into deeper water during July. Mean hourly movement rates during 24-h tracking sessions ranged from 100.3 m/h to 165.9 m/h. Although nonsignificant, the highest movement rates occurred in June and October and during the dusk period. Our results, along with similar studies in the region, indicate predictable and repeatable seasonal aggregations and distributions of Common Carp that will likely aid in the development of sampling protocols and control programs for temperate lakes. These results can be used to develop integrated control techniques that would target adult fish during periods of seasonal aggregations and dispersions, which is critical for effective population control.

Common Carp Cyprinus carpio, native to drainages of the shallow aquatic habitats of temperate regions, alter water qual- Black, Caspian, and Aral seas (i.e., the Ponto-Caspian region; ity and clarity, impact native fish communities, and disturb Balon 1995), is one of the most widely distributed and intro- aquatic plant growth through their spawning and benthic feed- Downloaded by [Department Of Fisheries] at 19:15 20 July 2015 duced fish species on the planet (McCrimmon 1968; Panek ing behaviors (Chumchal et al. 2005; Bajer et al. 2009; Weber 1987). These large, benthivorous cyprinids were intentionally and Brown 2009). Additionally, changes in macroinvertebrate introduced into all of the major watersheds of the USA as a and zooplankton communities (Zambrano and Hinojosa 1999; potential source of food desired by European immigrants in Parkos et al. 2003; Miller and Crowl 2006) and increased the late 1800s (McCrimmon 1968). Since their introduction, phosphate and ammonia levels (Lougheed et al. 1998) have they have become naturalized in all 48 contiguous states and been observed. are one of the most abundant aquatic invasive species in North The movement of Common Carp in freshwater systems is America (Panek 1987). Common Carp are considered a nui- driven by climatic and seasonal factors in combination with sance in many aquatic systems because they can develop dense autecology, system morphometry, and habitat characteristics populations that inflict extensive ecological and economic (Stuart and Jones 2006; Penne and Pierce 2008; Bajer et al. damage (Bonneau 1999; Koehn 2004; Weber and Brown 2011). In general, fish in northern temperate systems form 2009). High density populations, primarily found in the loose, prespawn aggregations in early spring followed by

*Corresponding author: [email protected] Received March 26, 2014; accepted August 23, 2014

1270 MOVEMENT AND SPATIAL DISTRIBUTION OF COMMON CARP 1271

dense spawning aggregations in areas of aquatic or inundated spp. in the secluded western basin, covers less than 1% of the terrestrial vegetation in late spring and early summer (Swee lake area. Bottom substrate composition is homogenous with and McCrimmon 1966; Penne and Pierce 2008; Bajer and about 70% being classified as silt and 24% as muck (Stukel Sorensen 2009). Postspawn fish are typically broadly distrib- 2003). uted and move between shallow and deep water areas through- Round Lake, third in the chain of lakes, has a surface area out the summer and early fall prior to forming overwintering of 76 ha (43560N, 96590W; Figure 1) and shares an unim- aggregations in deeper offshore basins in the late fall prior to peded connection with the northwestern portion of Brant Lake ice formation (McCrimmon 1968; Penne and Pierce 2008; via Silver Creek. Fish movement between Brant and Round Bajer et al. 2011). Assessments of population dynamics, along lakes is unrestricted and fish move between the two basins dur- with developing effective management and removal strategies ing the open-water period. Round Lake is shallow with a maxi- to reduce abundance, require knowledge of seasonal move- mum depth of 1.7 m and mean depth of 1.1 m in the main ment patterns and spatial distributions in a range of aquatic basin. The bottom substrate is primarily silt and muck and systems (Clark et al. 1991; Pope and Willis 1996; Bajer and large cattail marshes are located on the southeastern and north- Sorensen 2012). Common Carp behavioral studies assist man- western areas of the lake. Common Carp density estimates in agers in selecting appropriate sampling gears and methodolo- Brant and Round lakes are over 100 fish/ha (M. J. Weber and gies to obtain reliable estimates of CPUE, size and age coauthors, South Dakota State University, unpublished data). structure, growth, and condition (Pope and Willis 1996; Bajer Fish collection and transmitter implantation.—We sam- and Sorensen 2012). pled adult Common Carp (n D 20; 586–807 mm TL; mean D Currently, a growing fraction of Common Carp research in 697 mm) from Brant Lake using shoreline boat electrofishing the Upper Midwest region of the USA is focused towards in fall 2007 and surgically implanted them with acoustic trans- understanding movement patterns and advancing removal pro- mitters (CHP-87-L; Sonotronics). The implanted fish weighed grams to ultimately reduce abundance and reverse associated between 2,800 and 7,900 g (mean D 4,623 g). Eleven female negative ecological impacts (Bajer et al. 2011; Weber et al. fish and nine male fish were implanted with transmitters that 2011; Colvin et al. 2012). These fish have exhibited predict- were 90 mm in length, 18 mm in diameter, and weighed able and repeatable seasonal aggregations and distributions, 11.5 g in water. indicating a potential for large-scale removal programs (John- We placed fish in an aerated holding tank prior to being sen and Hasler 1977; Penne and Pierce 2008; Bajer et al. anesthetized in a separate tank containing a 100-mg/L solution 2011). Movement patterns vary among latitudes and bodies of of tricaine methanesulfonate (MS-222; Argent Chemical Lab- water (i.e., natural lakes, rivers, and culture ponds) (Bauer and oratories). Following sedation (loss of equilibrium) of the fish, Schlott 2004; Stuart and Jones 2006; Penne and Pierce 2008); we recorded total length (mm) and weight (g) and the fish therefore, our study was designed to contribute to the knowl- were placed ventral side up into a U-shaped surgery trough. edge of Common Carp spatial distributions in temperate lakes. Surgery to implant the acoustic transmitters into the abdominal Our goal was to determine movement and distribution patterns cavity followed published methods (Summerfelt and Smith in two interconnected lakes to aid in formulating management 1990; Winter 1996). We sterilized all surgical equipment with and removal strategies for dense populations of these fish pres- 70% isopropyl alcohol. We continuously flushed the gills of ent in lakes throughout the Midwest. anesthetized fish with fresh lake water during the surgical procedure. We removed three to five scales in a row and an incision (3–5 cm) was made with a scalpel, posterior to the pelvic

Downloaded by [Department Of Fisheries] at 19:15 20 July 2015 METHODS girdle and lateral to the midventral line. After the incision was Study area.—Brant Lake is a 421-ha eutrophic glacial lake open, gender and maturity were determined by visual inspec- located in Lake County, South Dakota (43550N, 96570W; tion of the gonads. Prior to implanting the transmitters, we Figure 1). It is fourth in a chain of four interconnected natural sterilized them with sodium hypochlorite followed by isopro- lakes and forms the headwaters of Skunk Creek, which flows pyl alcohol and then they were rinsed with deionized water. into the Big Sioux River and eventually the Missouri River. We inserted the cylindrical-shaped tags into the abdominal The lake is subcircular with one main basin and one relatively cavity and closed the wound using a simple, interrupted suture small (»15-ha), shallow basin (West Brant) in the western pattern. Sterile nonabsorbable surgical sutures (FSLX, Prolene portion. It is a relatively shallow, polymictic lake with a mean 3-0 Blue monofilament and FS-1, Ethilon 3-0 Black monofila- depth of 2.9 m and a maximum depth of 4.7 m. Brant Lake ment nylon; Ethicon) were used due to cool water tempera- does not thermally stratify due to shallow depths and wind and tures (range D 6.3–17.5C) that may have inhibited wound wave action. It possesses a simple bathymetry with a shoreline healing. Following surgery, we held each fish for a 30-min that gradually slopes into the nearly flat main basin. Partial recovery period in fresh, aerated lake water and then released winterkills occur periodically on Brant Lake. Aquatic vegeta- them near the surgery location. tion, in the form of scattered beds of sago pondweed Stuckenia Battery life expectancy for the tags was 18 months, and the pectinata, bulrushes Scirpus spp., and stands of cattails Typha transmitters were less than 2% of the individual weight of 1272 HENNEN AND BROWN

each fish, as recommended by Winter (1996). Acoustic trans- and dusk, the dusk period was 2 h before and after sunset, and mitters were required due to the high water conductivity mea- the nocturnal period occurred from the end of dusk to dawn. sured in Brant Lake (1,787 mS/cm, Stukel 2003), which When movements overlapped two diel periods, we designated attenuates radio signals transmitted through water and into the them to the period in which the majority of the movement air (Winter 1996). likely occurred. Telemetry.—We conducted manual telemetry from a boat Data analysis.—We examined Common Carp spatial distri- during the open-water season from May to November 2008 butions and movement patterns at a monthly scale. We treated and from April to August 2009. We began manual tracking individual fish as the experimental unit, and analyses of depth sessions at either the southeastern boat ramp of Brant Lake or and distance to shore were based on individual monthly the Silver Creek channel between Brant and Round lakes (Fig- means. We created a perimeter map and shape file of Brant ure 1). If the starting location was the Brant Lake boat ramp, a and Round lakes in ArcMap 9.3 (ESRI, Redlands, California) direction was randomly selected (clockwise or counterclock- by digitizing a 2008 aerial photo. We plotted locations on a wise) to start tracking. If the tracking session began in Silver perimeter map of Brant and Round lakes and the Near Analy- Creek, tracking began in the channel and continued into Round sis Tool was used in ArcMAP 9.3 to calculate distance from Lake. We used three independent 5-h tracking periods during each location to shore. the open-water season (morning, midday, and evening) to To quantify aggregations and identify patterns within obtain seasonal movement and distribution. The morning point location data, we conducted average Nearest Neigh- tracking period began around sunrise, generally occurring bor Index (NNI) and multidistance spatial cluster analyses between 0600 and 1100 hours Central Standard Time. Midday (Ripley’s K-function) using the Spatial Statistics Tools in tracking occurred from approximately 1100 to 1600 hours to ArcGIS 9.3 (Clark and Evans 1954; Ripley 1976; Wiegand span an equal time interval between the morning and evening and Moloney 2004). The NNI, expressed as a ratio, divides tracking periods. The evening tracking period generally the mean distance from each point location to its nearest occurred from 1600 to 2100 hours. We repeated each tracking neighbor point location by the expected mean distance for period at least two times per month. We defined tracking ses- a hypothetical random distribution given the size of the sions by a complete search of Brant Lake, Round Lake, or area and the number of points (a D 0.05) (Clark and Evans both lakes and selected based primarily on weather conditions, 1954; Krebs 1999). The null hypothesis states that the such as wind and precipitation. locations are randomly distributed. If the index was less We located individual fish from a boat using a hand- than 1, the distribution of locations exhibited clustering. held directional hydrophone (DH-2; Sonotronics) and The pattern is considered to be highly clustered as values receiver (USR-96; Sonotronics). Following the initial approach 0. If the index is greater than 1, the locations are detection, we maneuvered the boat until a strong, omnidi- considered dispersed. In contrast to the NNI, Ripley’s K- rectional signal was obtained. We then considered the boat function statistic considers the complete distribution of to be positioned above the fish (likely within 10 m) and a locations and summarizes spatial dependence (i.e., cluster- Garmin GPSMAP 76 was used to record GPS coordinates ing or dispersion) over a range of distances (Ripley 1976; (Universal Transverse Mercator, North American Datum of Wiegand and Moloney 2004; DiMaggio et al. 2008). 1983). We recorded transmitter frequency and the associ- Ripley’s K-function analysis in ArcGIS counts the number ated identification number at each location along with date, of neighboring locations within a given distance of each time, water depth, and surface temperature. Tagged fish location and evaluates the spatial distribution in relation to

Downloaded by [Department Of Fisheries] at 19:15 20 July 2015 were considered to be either dead or to have expelled their complete spatial randomness. The distribution is considered transmitters if we located them multiple times (three or to be clustered if the number of locations found within a more) in a small area (< 50-m radius) and no subsequent given distance is greater than that for a random distribu- locations were taken outside this area. Data collected from tion. A square-root transformation of the K-statistic, i.e., tagged fish prior to the determined transmitter loss or fish an L function transformation [L(d)], was used to create a mortalitywasusedintheanalyses. linear function, stabilize the variance, and simplify inter- In addition to regular tracking sessions, we performed one pretation (Wiegand and Moloney 2004). This allows deter- 24-h tracking session in June, July, August, and October 2008 mination of clustering at various distances by comparing to assess diel movement rates. We continuously located four the observed L(d) values to the expected L(d)values.Loca- randomly selected tagged fish every 1–2 h during each 24-h tions were determined to be clustered when observed L(d) session. Due to safety concerns of navigating shallow water- values were greater than the expected L(d)valuesand ways in Round Lake, we limited assessments of 24-h move- greater differences indicate more clustering (DiMaggio ment rates to individuals located in Brant Lake. We et al. 2008). These analyses were done to compare the dis- categorized the data into four diel periods: dawn, diurnal, tribution of fish locations in 2008 and 2009. dusk, and nocturnal. The dawn period occurred 2 h before and We used a repeated-measures ANOVA using the MIXED after sunrise, the diurnal period was the time between dawn procedure in SAS version 9.2 (SAS Institute 2008) to test for MOVEMENT AND SPATIAL DISTRIBUTION OF COMMON CARP 1273

FIGURE 1. Bathymetric map of Round and Brant lakes, South Dakota, where open-water Common Carp movement and spatial distribution were examined from May to November 2008 and from April to August 2009. The inset at the lower left displays the location of the two lakes within South Dakota.

monthly and diel period differences in movement rates, dis- Spatial Distribution tance to shore, and depth at fish location. Movement rates The NNI analysis indicated that the locations of fish in were loge transformed to normalize residuals allowing the use May, June, July, and August 2008 and June 2009 (Figure 2) of parametric analysis. We applied Bonferroni’s correction to were more clustered than a random distribution (P 0.03) statistically adjust P-values to allow for conservative multiple (Table 1). The greatest amount of clustering was observed in comparisons. June 2009 (NNI ratio D 0.62) when the NNI was closest to 0. The least amount of clustering occurred in July 2008 when the NNI ratio was 0.87. Conversely, fish were randomly distrib- RESULTS uted during October 2008 and July 2009 (P 0.13). A total of 380 locations (234 female, 146 male) from 19 The multidistance spatial cluster analyses indicated a clus- acoustically tagged adult Common Carp were obtained from tered distribution of locations up to 350 m in the months of Downloaded by [Department Of Fisheries] at 19:15 20 July 2015 May 14 to November 4, 2008, and 150 locations were obtained May, June, July, and August 2008 when compared to a random from 11 fish from April 22 to August 18, 2009. The average distribution (Table 2). May 2008 locations were considered number of locations collected per detected fish was 20 (mini- clustered to the maximum distance analyzed (600 m). Locations mum D 2; maximum D 32) in 2008 and 14 in 2009 (minimum became more dispersed than a random distribution at distances D 3; maximum D 20). The total locations collected for an indi- of 600 m, 400 m, and 450 m for June, July, and August 2008, vidual fish ranged from 2 to 52 (mean D 28). One female fish respectively. Locations in May, June, and August were more was never found during manual telemetry, potentially due to clustered at all distances than locations in July, as indicated by emigration or transmitter failure. In addition, 6 of 19 (32%) the higher L(d) values. A peak in the amount of clustering tagged fish were determined to have either died or expelled within a given month occurred at a distance of 150 m in the their transmitter and 3 fish found during manual telemetry in months May [L(d) D 280.20], June [L(d) D 296.98], and 2008 were not detected in 2009. Locations collected prior to August [L(d) D 239.61] when the difference (DiffK) between the deaths or transmitter expulsions were used for data analy- the observed and the expected L values was greatest. The great- ses because movements were similar to other tagged fish est amount of clustering in July occurred at a distance of 100 m movement patterns. Ten fish had active transmitters at the con- [L(d) D 156.44]. Overall, fish were highly clustered at 150 m clusion of the study. in June and most dispersed at a distance of 600 m in July. 1274 HENNEN AND BROWN

May 2008 N = 17 (61)

June 2009 June 2008 N = 11 (38) N = 17 (81)

July 2008 July 2009 N = 17 (118) N = 9 (71)

August 2008 N = 16 (56) 1 km

FIGURE 2. Monthly distributions for adult Common Carp in Brant and Round lakes, South Dakota, from May through August 2008 and June and July 2009. The N indicates the number of Common Carp located at least once, with the total number of locations in parentheses. Downloaded by [Department Of Fisheries] at 19:15 20 July 2015

TABLE 1. Monthly sample size (N) and nearest neighbor index for adult Common Carp locations obtained via acoustic telemetry in Brant and Round lakes, South Dakota, during May through October 2008 and June and July 2009. Month Year N Nearest neighbor ratio Z score P Spatial pattern

May 2008 61 0.854 ¡2.18 0.03 Clustered June 2008 81 0.746 ¡4.38 <0.01 Clustered July 2008 118 0.870 ¡2.70 <0.01 Clustered August 2008 56 0.712 ¡4.13 <0.01 Clustered October 2008 32 0.873 ¡1.37 0.17 Random June 2009 38 0.623 ¡4.44 <0.01 Clustered July 2009 71 1.094 1.51 0.13 Random MOVEMENT AND SPATIAL DISTRIBUTION OF COMMON CARP 1275

TABLE 2. Multidistance spatial cluster analyses (Ripley’s K-function), using an L transformation of the K values, of adult Common Carp locations obtained via acoustic telemetry in Brant and Round lakes, South Dakota, during May through August 2008. Locations are considered more clustered then a random distribution when observed L(d) values at a given distance are greater than the expected. Larger differences (DiffK) indicate more clustering and negative values indicate dispersal. May 2008 June 2008 July 2008 August 2008 Expected L(d) Observed Observed Observed Observed value L(d) DiffK L(d) DiffK L(d) DiffK L(d) DiffK

50 121.1 71.1 154.5 104.5 103.7 53.7 115.4 65.4 100 223.7 123.7 239.8 139.8 156.4 56.4 178.3 78.3 150 280.2 130.2 297.0 147.0 200.1 50.1 239.6 89.6 200 320.4 120.4 337.0 137.0 247.1 47.1 281.0 81.0 250 362.1 112.1 374.0 124.0 288.9 38.9 318.6 68.6 300 416.4 116.4 407.6 107.6 317.9 17.9 350.7 50.7 350 457.9 107.9 440.9 90.9 352.1 2.1 374.8 24.8 400 494.1 94.1 473.4 73.4 384.4 ¡15.6 405.0 5.0 450 528.7 78.7 500.0 50.0 414.2 ¡35.8 446.0 ¡4.0 500 569.6 69.6 523.3 23.3 450.9 ¡49.1 486.6 ¡13.4 550 607.0 57.0 551.0 1.0 479.9 ¡70.1 516.3 ¡33.7 600 627.2 27.2 577.7 ¡22.3 505.7 ¡94.3 540.5 ¡59.5

Monthly Depth and Distance to Shore Fish were found in deeper water (>2.4 m) from September Mean monthly depth at location ranged from 1.3 m in June through November in 2008. 2008 to 2.9 m in November 2008 (Figure 3). Fish were found Common Carp monthly mean distance to shore varied at the shallowest depths (<1.8 m) in May and June of both greatly from 63.4 m in June 2008 to 190.2 m in November years during peak spawning activity. A similar postspawn pat- 2008 (Figure 4). Similar trends occurred both years; fish were tern was observed both years when fish were located in deeper found less than 100 m from shore in May and June (Figure 4). water in July (>2.3 m) and then back in shallower depths in Fish then moved further offshore in July (>160 m) and dis- August (<1.9 m). Fish were located in significantly shallower played inshore movements during August of both years. Fish water during May, June, and August of 2008 and June and were located significantly closer to shore during May and June August of 2009 when compared to July of 2008 (P < 0.04). of 2008 and June of 2009 when compared to July of 2008 (P Downloaded by [Department Of Fisheries] at 19:15 20 July 2015

FIGURE 4. Mean monthly minimum distance (m) from shore for adult Com- FIGURE 3. Mean monthly depth (m) at location of adult Common Carp mon Carp (solid line) and mean monthly water temperatures ( C; dashed line) (solid line) and mean monthly water temperatures ( C; dashed line) in Brant in Brant and Round lakes, South Dakota, during open-water periods in 2008 and Round lakes, South Dakota, during open-water periods in 2008 (May– (May–November) and 2009 (April–August). Error bars represent § 1 standard November) and 2009 (April–August). Error bars represent § 1 standard error. Error bars that share a letter are not signiﬁcantly different (P 0.05). error. Error bars that share a letter are not signiﬁcantly different (P 0.05). 1276 HENNEN AND BROWN

TABLE 3. Mean monthly minimum distance from shore (m) across diel periods for adult Common Carp in Brant and Round lakes, South Dakota, during May through November 2008 and April through August 2009. The months of May through August were combined because multiple years of data were collected. Stan- dard errors are given in parentheses. Within a column, monthly means followed by the same letter are not signiﬁcantly different (P 0.05). Month Year Morning Midday Evening

May 2008, 2009 89.8 (27.5) z 92.6 (25.8) zyx 52.9 (16.4) y June 2008, 2009 92.9 (18.5) z 44.0 (8.4) x 58.7 (12.7) y July 2008, 2009 148.9 (16.5) z 164.2 (22.4) z 155.7 (23.5) z August 2008, 2009 113.0 (23.3) z 36.0 (8.6) yx 104.8 (25.0) zy September 2008 245.3 (92.9) z 143.7 (55.2) zyx 77.0 (36.3) zy October 2008 164.6 (22.3) zy 203.7 (52.6) z November 2008 155.5 (66.1) z 236.5 (54.9) z April 2009 145.4 (107.3) z 12.2 (5.4) zy

< 0.03). Mean distance to shore was greatest in September, and June 2008 (mean D 1.2 m; SE D 0.2) and June 2009 October, and November of 2008, but only fish in October were (mean D 0.6 m; SE D 0.1) than in July (mean D 2.9 m; SE D statistically further from shore than those in June 2008 (P < 0.2) and October 2008 (mean D 3.0 m; SE D 0.4) (P < 0.01). 0.01). Male and female fish were located at similar depths and A statistical difference (P < 0.05) in depth at location was distances from shore in all months except June 2008 when detected when comparing the midday period of July 2008 to female fish were found at significantly shallower depths July 2009. (female mean D 1.1 m; male mean D 1.6 m) and closer to shore (female mean D 28.5 m; male mean D 101.1 m) (P < 0.05). Movement Rates (24-h Tracking) Mean hourly movement rates during 24-h tracking sessions ranged from 100.3 m/h (SE D 25.1) in August to 165.9 m/h Diel Period Distances to Shore and Depth (SE D 28.1) in October (Figure 5). Hourly movement rates No differences in distance to shore were observed among varied considerably for individual fish, ranging from 40.8 m/h diel periods (morning, midday, and evening) within the to 241.9 m/h (mean D 133.9 m/h). Mean hourly movement months of May, June, July, and August 2008 to May, June, rates decreased from June to August and peaked in October. July, and August 2009. Therefore, data for these months were However, mean hourly movement rates did not statistically combined to compare monthly diel period distance to shore. differ between June, July, August, and October indicating that During the morning period, fish were located closest to shore in May (mean D 89.8 m; SE D 27.5) and farthest from shore during September (mean D 245.3 m; SE D 92.9; Table 3). TABLE 4. Mean monthly depth (m) at location across diel periods for adult No statistical differences in distance to shore were detected Common Carp in Brant and Round lakes, South Dakota, during May through in the morning period among months. Fish were found fur- November 2008 and April through August 2009. Standard errors are provided in parentheses. Within a column, monthly means followed by the same letter thest offshore during the midday period in November (mean are not significantly different (P 0.05). Downloaded by [Department Of Fisheries] at 19:15 20 July 2015 D 236.5 m; SE D 54.9) and closest to shore in August (mean D 36.0 m; SE D 8.6). Fish were found significantly closer to Month Year Morning Midday Evening shore during the midday period in June and August when May 2008 2.3 (0.3) zy 2.0 (0.4) zyx 0.9 (0.1) y compared with July and November (P < 0.05). In the even- June 2008 1.6 (0.2) y 1.1 (0.1) x 1.2 (0.2) y ing period, fish were found statistically closer to shore (P < July 2008 2.3 (0.2) zy 2.8 (0.2) z 2.9 (0.2) z 0.05) in May (mean D 52.9 m; SE D 16.4) and June (mean August 2008 2.2 (0.3) zy 0.7 (0.2) x 1.8 (0.4) zy D 58.7 m; SE D 12.7) than in July (mean D 155.7 m; SE D September 2008 2.8 (0.6) zy 2.7 (0.5) zyx 0.8 (0.4) zy 23.5) and October (mean D 203.7 m; SE D 52.6). October 2008 2.5 (0.3) zy 3.0 (0.4) z During the morning period, fish were found at statistically November 2008 2.5 (0.6) zy 3.4 (0.4) zy shallower depths in June 2008 (mean D 1.6 m; SE D 0.2) April 2009 2.9 (0.6) zy 0.7 (0.2) zy and August 2009 (mean D 0.6 m; SE D 0.2) than in July May 2009 3.3a zy 1.5 (0.5) zyx 2.1 (0.8) zy 2009 (mean D 3.0 m; SE D 0.2) (P < 0.03; Table 4). Fish June 2009 2.0 (0.3) zy 1.8 (0.4) zyx 0.6 (0.1) y were found in shallower depths during the midday period in July 2009 3.0 (0.2) z 1.5 (0.2) yx 1.9 (0.3) zy June and August 2008 than in July, October, and November August 2009 0.6 (0.2) y 1.5 (0.4) zyx 1.7 (0.6) zy of 2008 (P < 0.01). During the evening period, fish were found at shallower depths in May (mean D 0.9 m; SE D 0.1) aStandard error was not available. MOVEMENT AND SPATIAL DISTRIBUTION OF COMMON CARP 1277

200 knowledge of Common Carp movement and seasonal distribution patterns assists in developing standardized sampling 180 methods and selecting the appropriate gear(s) to produce reli-

160 able estimates of population characteristics (Clark et al. 1991; Pope and Willis 1996; Bajer and Sorensen 2012). Stan- 140 dard sampling techniques that managers use for routine monitoring and population assessments in temperate lakes may 120 involve the use of fyke nets or gill nets, which produce relative abundance estimates (CPUE) (Miranda and Boxrucker Movement Rate (m/h) 100 2009). However, adult ﬁsh tend to avoid passive capture tech-

80 niques (Clark et al. 1991; Bajer and Sorensen 2012). Thus, gill nets require a large effort to produce reliable estimates,

60 and fyke nets are ineffective for this species (Clark et al. June July August September October 1991). Boat electrofishing is an effective gear for estimating Month Common Carp densities (Bajer and Sorensen 2012), and our results suggest that boat electrofishing would be most effec- FIGURE 5. Mean minimum displacement per hour for adult Common Carp tagged with ultrasonic transmitters in Brant Lake, South Dakota, from June to tive for sampling high numbers of fish during May and June October 2008. Error bars represent § 1 standard error. in north temperate lakes. However, this sampling may be biased towards larger individuals, and other gears, such as specifically designed traps (Schwartz 1986; Stuart et al. fish moved consistently at the same rate during this period 2006; Thwaites et al. 2010) and bag seines (<5-cm bar (F D 1.28, P D 0.33). Movement rates did not differ between mesh), may be needed to sample the range of fish sizes pres- diel periods across months or diel periods within months. Fish ent (Weber et al. 2011). On the other hand, sampling at ran- were most active during dusk periods (mean D 159.6 m/h; domly chosen sites in July, when fish are more randomly SE D 25.7) and least active during dawn periods (mean D distributed, may be the best option for standard monitoring 113.5 m/h; SE D 10.6; Figure 6). programs to reduce biases for indices, such as abundance and length structure, which are often used to make comparisons among waters and over time (Miranda and Boxrucker 2009). Additionally, the observed movement patterns suggest a DISCUSSION scenario for large-scale removals using an integrated approach The tagged population displayed annual open-water move- to reduce Common Carp abundances in systems similar to ment patterns and spatial distributions, which will aid fisher- Brant and Round lakes. Potential integrated control techniques ies managers in developing management strategies and that may be useful include removal via commercial seining selecting appropriate control techniques in glacial lakes. The (Rose and Moen 1952; Fritz 1987; Colvin et al. 2012), traps (Schwartz 1986; Thwaites et al. 2010), electrofishing (Bajer and Sorensen 2012), and localized chemical treatments (Mark- ing 1992; Bonneau 1999; Wydoski and Wiley 1999) and using barriers to prevent spawning migrations (Verrill and Berry

Downloaded by [Department Of Fisheries] at 19:15 20 July 2015 1995; Lougheed and Chow-Fraser 2001). Combining all of the information suggests that active gears (e.g., electrofishing) would be effective for collecting fish during the morning and midday periods in August when fish were aggregated in shallow nearshore habitats and exhibit the lowest movement rates. Additionally, commercial seining is a promising strategy for controlling invasive Common Carp populations and is likely the most effective gear when fish are dispersed in offshore habitats (Rose and Moen 1952; Weber et al. 2011; Colvin et al. 2012). However, the desirable results from seining removals are often short-lived and commercial seines are generally size selective for larger individuals (Bajer et al. 2009; FIGURE 6. Mean minimum displacement per hour across diel periods for Weber et al. 2011; Colvin et al. 2012). Thus, it is important adult Common Carp tagged with ultrasonic transmitters in Brant Lake, South that other gears designed to target smaller-bodied individuals Dakota, from June, July, August, and October 2008. Error bars represent § 1 are used in conjunction with seining to most effectively con- standard error. trol fish populations (Weber et al. 2011). 1278 HENNEN AND BROWN

A small number of fish bearing transmitters (i.e., Judas fish) Pierce (2008) described the largest monthly mean core activity may be used to target removal strategies specifically during area of adult fish during July, indicating dispersal during this annual aggregations (Johnsen and Hasler 1977; Penne and period. These observed behaviors may be driven by prey avail- Pierce 2008; Bajer et al. 2011). However, previous studies ability and both interspecific and intraspecific competition for have indicated expulsion or death rates of 0–100% for adult resources. High-density Common Carp populations, in addi- Common Carp surgically implanted with transmitters, which tion to other invertebrate-consuming fishes present in littoral is likely due to factors such as water temperature, surgical pro- habitats, may reduce benthic prey resources (Zambrano and cedures, and suture material (Okland et al. 2003; Bauer and Hinojosa 1999; Michaletz et al. 2005; Miller and Crowl Schlott 2004; Daniel et al. 2009). The 32% transmitter loss 2006). Thus, Common Carp may move offshore after spawn- rate that we observed was slightly lower than other long-term ing to exploit areas of greater prey abundance or to increase tracking studies (Stuart and Jones 2002; Penne 2007), but this individual foraging efficiency by reducing the amount of clus- potential loss must be accounted for prior to undertaking the tering and competition for prey resources (Fernandez-Delgado Judas fish technique. 1990; Bajer et al. 2010). Competition among Goldfish Caras- Common Carp were located in shallow peripheral habitats sius auratus, a close relative of Common Carp, in increasing close to shore in higher proportions during the spawning sea- cluster sizes has been shown to reduce individual foraging effi- son (i.e., May and June), as has been previously observed ciency, thus fish may avoid aggregating during certain times of (Otis and Weber 1982; Penne and Pierce 2008; Bajer and Sor- the year to avoid resource competition (Stenberg and Persson ensen 2009). These spawning aggregations have been sug- 2005). Additionally, Stenberg and Persson (2005) indicated gested to be driven by chemical cues or pheromones that Goldfish assess patches in their environment for food (Schwartz 1987; Kobayashi et al. 2002; Sisler and Sorensen resources and persistently forage in areas where food is 2008) and typically form in areas of aquatic vegetation and located. The ability to utilize a range of habitats to exploit submersed terrestrial vegetation, which provide suitable peak prey abundance is advantageous; laboratory and field broadcast spawning substrate and nursery habitats that are studies have shown that many teleost fishes are able to use important for recruitment (McCrimmon 1968; Penne and visual and olfactory cues to locate food and remember its loca- Pierce 2008; Bajer and Sorensen 2009). Adult fish can likely tion (Braithwaite and de Perera 2006; Bajer et al. 2010). be targeted in large numbers using various control techniques Following the postspawn offshore movement observed dur- when spawning aggregations are formed in nearshore habitats ing July, adult fish displayed an inshore movement in August (Penne and Pierce 2008). In systems having morphology simi- and their distribution became more clustered. The same trend lar to the study lakes, electrofishing, traps, and piscicide treat- was shown for Common Carp in Clear Lake, Iowa (Penne and ments would likely be most effective in these peripheral Pierce 2008), and in populations of White Bass Morone chrys- habitats, where commercial seining may not be feasible. Addi- ops and Black Crappie Pomoxis nigromaculatus in South tionally, engineering barriers, with trapping and removal capa- Dakota glacial lakes (Guy et al. 1992; Beck and Willis 2000). bilities, between overwintering (i.e., deeper main basins) and This behavior is likely not indicative of protracted or fractional spawning habitats (i.e., vegetated littoral areas) in intercon- spawning (Swee and McCrimmon 1966; Phelps et al. 2008) or nected glacial lake systems may be important for limiting fish related to mean water temperatures as they were similar migrations into suitable spawning habitats, thus potentially between July and August. limiting reproduction and reducing recruitment (Verrill and Although we present limited data from September through Berry 1995; Penne and Pierce 2008; Bajer and Sorensen November, the offshore movement and distribution of fish

Downloaded by [Department Of Fisheries] at 19:15 20 July 2015 2009). Another possible strategy for interconnected systems in locations during fall is associated with cooling water tempera- north temperate lakes would be to allow fish to access shallow tures during this period and the onset of overwinter aggrega- vegetated areas for spawning and then erect barriers prevent- tions that occur regularly in temperate Common Carp ing movement back into the deeper overwintering basins (Ver- populations (Johnsen and Hasler 1977; Penne and Pierce rill and Berry 1995; Bajer and Sorensen 2009). Fish would 2008; Bajer et al. 2011). The aggregation and dispersal pat- then be trapped in shallow areas making them vulnerable to terns observed may be contingent on resource availability and removal efforts or winter hypoxia. quality, with increased aggregations likely occurring in areas Postspawn adult fish movement and spatial distribution fol- of highest quality (Parrish and Edelstein-Keshet 1999; Sten- lowed similar annual trends, which are likely driven by a com- berg and Persson 2005; Penne 2007). Additional research to bination of physiological, environmental (e.g., prey resources determine the mechanisms driving postspawn fish behaviors and avian predation risk), and climatic factors (e.g., tempera- may be warranted to further refine removal techniques. ture, wind and wave action). Similar midsummer offshore Information on fine-scale, hourly movements of Common movement and dispersal patterns were also observed in an Carp from telemetry studies is limited and no information Iowa glacial lake, where adult Common Carp mean distance regarding fish hourly movement rates based on hourly detec- to shore and depth at location were greater in July than in May tions was found in the literature. However, individual move- and June (Penne and Pierce 2008). Additionally, Penne and ments have been shown to be highly variable in open water MOVEMENT AND SPATIAL DISTRIBUTION OF COMMON CARP 1279

(Otis and Weber 1982; Crook 2004; Stuart and Jones 2006). Spengler, and A. Wiering, and the South Dakota Department Hourly estimated movement rates of Common Carp in the of Game, Fish, and Parks personnel, including B. Blackwell, Murray River, Australia, ranged from 150 to 900 m/h, but D. Lucchesi, and T. Kaufman, for their assistance with data these estimates may be biased because they were based on collection and field work. We thank B. Graeb, M. Weber, D. long-range movements (>10 km) over several hours (>16 h) Willis, R. Eades, and three anonymous reviewers for providing or days (<21 d) (Stuart and Jones 2002). Observed hourly insight that improved this manuscript. Special thanks also go movement rates of fish did not statistically differ by month or to D. White and M. Gregor of Sonotronics for assistance with time of day despite changes in weather, season, and spawning various aspects of acoustic telemetry. Partial funding for this activity. The trend in decreasing movement rates from June to project was provided through the Federal Aid in Sport Fish August may be driven by the change from spawning to post- Restoration Act Study 1513 (Project F-15-R-42) administered spawn behavior. The peak in movement rates observed in through South Dakota Department of Game, Fish, and Parks October may be attributed to decreasing water temperatures and the South Dakota Agricultural Experiment Station. and increased foraging activity prior to the formation of ice cover. Cooke and Schreer (2003) indicated that Common Carp are more active during times of fluctuating temperatures and least active during stable periods. Observed activity varied REFERENCES Bajer, P. G., C. J. Chizinski, and P. W. Sorensen. 2011. Using the Judas tech- greatly among individuals with one fish moving over 5 km in nique to locate and remove wintertime aggregations of invasive Common a 24-h period while another fish only moved 950 m. The wide Carp. Fisheries Management and Ecology 18:497–505. range in movement rates can be partially explained by individ- Bajer, P. G., H. Lim, M. J. Travaline, B. D. Miller, and P. W. Sorensen. 2010. ual behaviors of less mobile fish using restricted home ranges Cognitive aspects of food searching behavior in free-ranging wild Common or nomadic fish using larger areas of the system, as is often Carp. Environmental Biology of Fishes 88:295–300. Bajer, P. G., and P. W. Sorensen. 2009. Recruitment and abundance of an inva- seen in Common Carp tracking studies (Otis and Weber 1982; sive fish, the Common Carp, is driven by its propensity to invade and repro- Stuart and Jones 2002; Crook 2004). Extensive and erratic duce in basins that experience winter-time hypoxia in interconnected lakes. movements of Common Carp within or between lakes were Biological Invasions 12:1101–1112. also observed in a study done in eastern Wisconsin (Otis and Bajer, P. G., and P. W. Sorensen. 2012. Using boat electrofishing to estimate Weber 1982). the abundance of invasive Common Carp in small Midwestern lakes. North American Journal of Fisheries Management 32:817–822. Overall, adult Common Carp exhibited predictable and Bajer, P. G., G. Sullivan, and P. W. Sorensen. 2009. Effects of a rapidly repeatable yearly movement patterns and spatial distributions increasing population of Common Carp on vegetative cover and waterfowl during open-water periods, with a clear tendency to aggregate in a recently restored Midwestern shallow lake. Hydrobiologia 632:235–245. in shallow depths that were close to shore during their spawn- Balon, E. K. 1995. Origin and domestication of the wild carp, Cyprinus carpio: ing period. Adult fish then dispersed offshore into deeper from Roman gourmets to the swimming flowers. Aquaculture 129:3–48. Bauer, C., and G. Schlott. 2004. Overwintering of farmed Common Carp (Cyp- waters for a period of time postspawn before returning to shal- rinus carpio L.) in the ponds of a central European aquaculture facility– low depths near shore during the late summer. In combination measurement of activity by radio telemetry. Aquaculture 241:301–317. with other studies in the region (Penne and Pierce 2008; Bajer Beck, H. D., and D. W. Willis. 2000. Biotelemetry of White Bass in a South et al. 2011; Weber et al. 2011), the information provided here Dakota glacial lake. Journal of Freshwater Ecology 15:229–236. will aid aquatic resource managers in developing sampling Bonneau, J. L. 1999. Ecology of a fish biomanipulation in a Great Plains reservoir. Doctoral dissertation. University of Idaho, Moscow. and control techniques for temperate lakes. These results can Braithwaite, V. A., and T. B. de Perera. 2006. Short-range orientation in fish: be used to target integrated control techniques during periods how fish map space. Marine and Freshwater Behaviour and Physiology

Downloaded by [Department Of Fisheries] at 19:15 20 July 2015 of seasonal aggregations and dispersions, which is essential 39:37–47. for effectively controlling the population (Weber et al. 2011; Chumchal, M. M., W. H. Nowlin, and R. W. Drenner. 2005. Biomass-depen- Colvin et al. 2012). However, further research is warranted dent effects of Common Carp on water quality in shallow ponds. Hydrobio- logia 545:271–277. into the techniques, application, and efficacy of an integrated Clark, P. J., and F. C. Evans. 1954. Distance to nearest neighbor as a measure removal approach to reduce abundance and effectively control of spatial relationships in populations. Ecology 35:445–453. populations over a prolonged period of time. Clark, S. W., D. W. Willis, and C. R. Berry. 1991. Indexing of Common Carp populations in large palustrine wetlands of the northern plains. Wetlands 11:163–172. ACKNOWLEDGMENTS Colvin, M. E., C. L. Pierce, T. W. Stewart, and S. E. Grummer. 2012. Strate- gies to control a Common Carp population by pulsed commercial harvest. The authors thank the Department of Natural Resource North American Journal of Fisheries Management 32:1251–1264. Management, South Dakota State University, Brookings, Cooke, S. J., and J. F. Schreer. 2003. Environmental monitoring using physio- South Dakota, for financial support and the use of equipment logical telemetry–a case study examining Common Carp responses to ther- and facilities. We also thank numerous technicians and gradu- mal pollution in a coal-fired generating station effluent. Water, Air, and Soil Pollution 142:113–136. ate students, including A. Ahrens, C. Dekker, G. Douglas, C. Crook, D. A. 2004. Movements associated with home-range establishment by Funk, B. Johnson, S. Lundgren, L. Pierce, K. Rounds, T. two species of lowland river fish. Canadian Journal of Fisheries and Aquatic Schaeffer, C. Schoenebeck, W. Schreck, N. Siepker, D. Sciences 61:2183–2193. 1280 HENNEN AND BROWN

Daniel, A. J., B. J. Hicks, N. Ling, and B. O. David. 2009. Acoustic and radio- Penne, C. R. 2007. Radio telemetry study of Common Carp in Clear Lake, transmitter retention in Common Carp (Cyprinus carpio) in New Zealand. Iowa, to guide future management. Master’s thesis. Iowa State University, Marine and Freshwater Research 60:328–333. Ames. DiMaggio, C., A. Bucciarelli, K. J. Tardiff, D. Vlahov, and S. Galea. 2008. Penne, C. R., and C. L. Pierce. 2008. Seasonal distribution, aggregation, and Spatial analytic approaches to explaining the trends and patterns of drug habitat selection of Common Carp in Clear Lake, Iowa. Transactions of the overdose deaths. Pages 447–464 in Y. F. Thomas, D. Richardson, and I. American Fisheries Society 137:1050–1062. Cheung, editors. Geography and drug addiction. Springer, New York. Phelps, Q. E., B. D. S. Graeb, and D. W. Willis. 2008. First year growth and Fernandez-Delgado, C. 1990. Life history patterns of the Common Carp, Cyp- survival of Common Carp in two glacial lakes. Fisheries Management and rinus carpio, in the estuary of the Guadalquivir River in south-west Spain. Ecology 15:85–91. Hydrobiologia 206:19–28. Pope, K. L., and D. W. Willis. 1996. Seasonal influences on freshwater fisher- Fritz, A. W. 1987. Commercial fishing for carp. Pages 17–30 in E. L. Cooper, ies sampling data. Reviews in Fisheries Science 4:57–73. editor. Carp in North America. American Fisheries Society, Bethesda, Ripley, B. D. 1976. The second-order analysis of stationary point processes. Maryland. Journal of Applied Probability 13:255–266. Guy, C. S., R. M. Neumann, and D. W. Willis. 1992. Movement patterns of Rose, E. T., and T. Moen. 1952. The increase in game-fish populations in east adult Black Crappie, Pomoxis nigromaculatus, in Brant Lake, South Dakota. Okoboji Lake, Iowa, following intensive removal of rough fish. Transac- Journal of Freshwater Ecology 7:137–147. tions of the American Fisheries Society 82:104–114. Johnsen, P. B., and A. D. Hasler. 1977. Winter aggregations of carp (Cyprinus SAS Institute. 2008. SAS procedures guide for personal computers, version carpio) as revealed by ultrasonic tracking. Transactions of the American 9.2. SAS institute, Cary, North Carolina. Fisheries Society 106:556–559. Schwartz, F. J. 1986. A leadless stackable trap for harvesting Common Carp. Kobayashi, M., P. W. Sorensen, and N. E. Stacey. 2002. Hormonal and phero- North American Journal of Fisheries Management 6:596–598 monal control of spawning behavior in the Goldfish. Fish Physiology and Schwartz, F. J. 1987. Homing behavior of tagged and displaced carp, Cyprinus Biochemistry 26:71–84. carpio, in Pymatuning Lake, Pennsylvania/Ohio. Ohio Journal of Science Koehn, J. D. 2004. Carp (Cyprinus carpio) as a powerful invader in Australian 87:15–22. waterways. Freshwater Biology 49:882–894. Sisler, S. P., and P. W. Sorensen. 2008. Common Carp and Goldfish discern Krebs, C. J. 1999. Ecological methodology, 2nd edition. Addison Wesley conspecific identity using chemical cues. Behaviour 145:1409–1425. Longman, Menlo Park, California. Stenberg, M., and A. Persson. 2005. The effects of spatial food distribution and Lougheed, V. L., B. Crosbie, and P. Chow-Fraser. 1998. Predictions on group size on foraging behavior of benthic fish. Behavioural Processes the effect of Common Carp (Cyprinus carpio) exclusion on water qual- 70:41–50. ity, zooplankton, and submergent macrophytes in a Great Lakes wet- Stuart, I. G., and M. J. Jones. 2002. Ecology and management of Common land. Canadian Journal of Fisheries and Aquatic Sciences 55:1189– Carp in the Barmah-Millewa Forest. Arthur Rylah Institute, Final Report, 1197. Heidelberg, Victoria, Australia. Lougheed, V. L., and P. Chow-Fraser. 2001. Spatial variability in the response Stuart, I. G., and M. J. Jones. 2006. Movement of Common Carp, Cyprinus of lower trophic levels after carp exclusion from a freshwater marsh. Journal carpio, in a regulated lowland Australian river: implications for manage- of Aquatic Ecosystem Stress and Recovery 9:21–34. ment. Fisheries Management and Ecology 13:213–219. Marking, L. L. 1992. Evaluation of toxicants for the control of carp and other Stuart, I. G., A. Williams, J. McKenzie, and T. Holt. 2006. Managing a migra- nuisance fishes. Fisheries 17(6):6–13. tory pest species: a selective trap for Common Carp. North American Jour- McCrimmon, H. R. 1968. Carp in Canada. Fisheries Research Board of Can- nal of Fisheries Management 26:888–893. ada Bulletin 165:1–93. Stukel, S. M. 2003. Assessing the sustainability of fish communities in Michaletz, P. H., K. E. Doisy, and C. F. Rabeni. 2005. Influence of productiv- glacial lakes: habitat inventories and relationships between lake attrib- ity, vegetation, and fish on macroinvertebrate abundance and size in Mid- utes and fish communities. Master’s thesis. South Dakota State Univer- western USA impoundments. Hydrobiologia 543:147–158. sity, Brookings. Miller, S. A., and T. A. Crowl. 2006. Effects of Common Carp (Cyprinus car- Summerfelt, R. C., and L. S. Smith. 1990. Anesthesia, surgery, and pio) on macrophytes and invertebrate communities of a shallow lake. Fresh- related techniques. Pages 213–272 in C. B. Schreck and P. B. Moyle, water Biology 51:85–94. editors. Methods for fish biology. American Fisheries Society, Miranda, L. E., and J. Boxrucker. 2009. Warmwater fish in large standing Bethesda, Maryland. waters. Pages 29–42 in S. A. Bonar, W. A. Hubert, and D. W. Willis, edi- Swee, U. B., and H. R. McCrimmon. 1966. Reproductive biology of the carp, Downloaded by [Department Of Fisheries] at 19:15 20 July 2015 tors. Standard methods for sampling North American freshwater fishes. Cyprinus carpio L., in Lake St. Lawrence, Ontario. Transactions of the American Fisheries Society, Bethesda, Maryland. American Fisheries Society 95:372–380. Okland, F., C. J. Hay, T. F. Naesje, N. Nickandor, and E. B. Thorstad. 2003. Thwaites, L. A., B. B. Smith, M. Decelis, D. Fleer, and A. Conallin. 2010. A Learning from unsuccessful radio tagging of Common Carp in a Namibian novel push trap element to manage carp (Cyprinus carpio L.): a laboratory reservoir. Journal of Fish Biology 62:735–739. trial. Marine and Freshwater Research 61:42–48. Otis, K. J., and J. J. Weber. 1982. Movement of carp in the Lake Winnebago Verrill, D. D., and C. R. Berry. 1995. Effectiveness of an electrical bar- system determined by radio telemetry. Wisconsin Department of Natural rier and lake drawdown for reducing Common Carp and Bigmouth Buf- Resources Technical Bulletin 134. falo abundances. North American Journal of Fisheries Management Panek, F. M. 1987. Biology and ecology of carp. Pages 1–15 in E. L. Cooper, 15:137–141. editor. Carp in North America. American Fisheries Society, Bethesda, Weber, M. J., and M. L. Brown. 2009. Effects of Common Carp on aquatic Maryland. ecosystems 80 years after “carp as a dominant”: ecological insights for fish- Parkos, J. J. III, V. J. Santucci Jr., and D. H. Wahl. 2003. Effects of adult Com- eries management. Reviews in Fisheries Science 17:524–537. mon Carp (Cyprinus carpio) on multiple trophic levels in shallow meso- Weber, M. J., M. J. Hennen, and M. L. Brown. 2011. Simulated population cosms. Canadian Journal of Fisheries and Aquatic Sciences 60:182–192. responses of Common Carp to commercial exploitation. North American Parrish, J. K., and L. Edelstein-Keshet. 1999. Complexity, pattern, and evolu- Journal of Fisheries Management 31:269–279. tionary trade-offs in animal aggregation. Science (Washington, D.C.) Wiegand, T., and K. A. Moloney. 2004. Rings, circles, and null-models for 284:99–101. point pattern analysis in ecology. Oikos 104:209–229. MOVEMENT AND SPATIAL DISTRIBUTION OF COMMON CARP 1281

Winter, J. 1996. Advances in underwater biotelemetry. Pages 555–590 in B. R. ﬁsheries management in North America. American Fisheries Society, Murphy and D. W. Willis, editors. Fisheries techniques, 2nd edition. Ameri- Bethesda, Maryland. can Fisheries Society, Bethesda, Maryland. Zambrano, L., and D. Hinojosa. 1999. Direct and indirect effects of carp (Cyp- Wydoski, R. S., and R. W. Wiley. 1999. Management of undesirable ﬁsh spe- rinus carpio L.) on macrophyte and benthic communities in experimental cies. Pages 403–430 in C. C. Kohler and W. A. Hubert, editors. Inland shallow ponds in central Mexico. Hydrobiologia 408/409:131–138. Downloaded by [Department Of Fisheries] at 19:15 20 July 2015 This article was downloaded by: [Department Of Fisheries] On: 20 July 2015, At: 19:15 Publisher: Taylor & Francis Informa Ltd Registered in England and Wales Registered Number: 1072954 Registered office: 5 Howick Place, London, SW1P 1WG

Click for updates

To cite this article: (2014) Erratum, North American Journal of Fisheries Management, 34:6, 1282-1282, DOI: 10.1080/02755947.2014.987629 To link to this article: http://dx.doi.org/10.1080/02755947.2014.987629

PLEASE SCROLL DOWN FOR ARTICLE

Taylor & Francis makes every effort to ensure the accuracy of all the information (the “Content”) contained in the publications on our platform. However, Taylor & Francis, our agents, and our licensors make no representations or warranties whatsoever as to the accuracy, completeness, or suitability for any purpose of the Content. Any opinions and views expressed in this publication are the opinions and views of the authors, and are not the views of or endorsed by Taylor & Francis. The accuracy of the Content should not be relied upon and should be independently verified with primary sources of information. Taylor and Francis shall not be liable for any losses, actions, claims, proceedings, demands, costs, expenses, damages, and other liabilities whatsoever or howsoever caused arising directly or indirectly in connection with, in relation to or arising out of the use of the Content. This article may be used for research, teaching, and private study purposes. Any substantial or systematic reproduction, redistribution, reselling, loan, sub-licensing, systematic supply, or distribution in any form to anyone is expressly forbidden. Terms & Conditions of access and use can be found at http:// www.tandfonline.com/page/terms-and-conditions North American Journal of Fisheries Management 34:1282, 2014 Ó American Fisheries Society 2014 ISSN: 0275-5947 print / 1548-8675 online DOI: 10.1080/02755947.2014.987629

ERRATUM

Please make the following correction in a recent issue of this journal:

Volume 34(5), October 2014: “Annual Variation of Spawning Cutthroat Trout in a Small Western USA Stream: A Case Study with Implications for the Conservation of Potamodromous Trout Life History Diversity,” by Stephen Bennett, Robert Al-Chokhachy, Brett B. Roper, and Phaedra Budy, pages 1033–1046.

Pages 1041–1042. The captions to Figures 4, 5, and 6 appear out of order. They should read as follows:

FIGURE 4. Relationship between the mean maximum daily water temperature (8C) during weeks 1–13 (prior to spawning) and on the day of the year of onset (5th percentile), the median day (50th percentile), and the ending day (95th percentile) of Cutthroat Trout spawning in Spawn Creek, 2006–2012. [Color ﬁgure available online.]

FIGURE 5. Proportion of all Cutthroat Trout redds observed by year and by 500-m segment of stream starting at the mouth (0–500 m) and moving upstream. Stream segments are lumped together for display, but the spatial distribution of redds was assessed for spatial pattern (uniform, random, and clumped) using 250-m segments.

FIGURE 6. Cumulative proportion of Cutthroat Trout redds observed by week in Spawn Creek relative to stream distance from the mouth, 2006–2012. Downloaded by [Department Of Fisheries] at 19:15 20 July 2015

1282