Winter Feeding As an Overwintering Survival Strategy in Young-Of-The-Year Winter Flounder Richard J

This article was downloaded by: [Department Of Fisheries] On: 25 September 2012, At: 20:24 Publisher: Taylor & Francis Informa Ltd Registered in England and Wales Registered Number: 1072954 Registered office: Mortimer House, 37-41 Mortimer Street, London W1T 3JH, UK

Transactions of the American Fisheries Society Publication details, including instructions for authors and subscription information: http://www.tandfonline.com/loi/utaf20 Winter Feeding as an Overwintering Survival Strategy in Young-of-the-Year Winter Flounder Richard J. Bell a a Graduate School of Oceanography, University of Rhode Island, South Ferry Road, Narragansett, Rhode Island, 02882, USA Version of record first published: 13 Jun 2012.

To cite this article: Richard J. Bell (2012): Winter Feeding as an Overwintering Survival Strategy in Young-of-the-Year Winter Flounder, Transactions of the American Fisheries Society, 141:4, 855-871 To link to this article: http://dx.doi.org/10.1080/00028487.2012.675896

PLEASE SCROLL DOWN FOR ARTICLE

Full terms and conditions of use: http://www.tandfonline.com/page/terms-and-conditions 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. The publisher does not give any warranty express or implied or make any representation that the contents will be complete or accurate or up to date. The accuracy of any instructions, formulae, and drug doses should be independently verified with primary sources. The publisher shall not be liable for any loss, actions, claims, proceedings, demand, or costs or damages whatsoever or howsoever caused arising directly or indirectly in connection with or arising out of the use of this material. Transactions of the American Fisheries Society 141:855–871, 2012 C American Fisheries Society 2012 ISSN: 0002-8487 print / 1548-8659 online DOI: 10.1080/00028487.2012.675896

ARTICLE

Winter Feeding as an Overwintering Survival Strategy in Young-of-the-Year Winter Flounder

Richard J. Bell* Graduate School of Oceanography, University of Rhode Island, South Ferry Road, Narragansett, Rhode Island 02882, USA

Abstract For fish, the first winter of life can be a period of high mortality prior to recruitment. When resources are limiting or when fish are unable to feed due to low temperatures, starvation can often lead to size-dependent overwintering mortality. Larger individuals are typically better able to survive starvation because they have a higher percentage of energy reserves and a lower metabolic rate per unit of body mass than small individuals. Alternatively, fish that can feed during the winter are able to maintain their lipid stores and reduce their chance of starvation. The aim of this study was to examine the overwintering mortality and physiology of young-of-the-year (age-0) winter flounder Pseudopleuronectes americanus in relation to body size. Size-dependent mortality was investigated with 17 years of length-frequency data. In addition, I sampled the diet and whole-body crude lipid content of 309 age-0 winter flounder over the course of 1 year. Samples were taken in three estuaries in the northeastern USA during October– April. Whole-body crude lipid content ranged from 4.7% to 12.4% of dry weight. Larger age-0 winter flounder did not have higher lipid stores, and the age-0 fish did not exhibit size-dependent overwintering mortality. Age-0 winter flounder fed on amphipods and polychaetes throughout the winter, and their whole-body crude lipid content was maintained through the fall and winter. The physiology data lack temporal replication, but the spatial coherence in results (i.e., consistency among the three estuaries, representing two different stocks) suggests that the consumption and energy allocation patterns are real and that age-0 winter flounder follow an alternative overwintering survival strategy.

Juvenile ﬁsh in temperate latitudes face multiple sources Larger individuals typically have a higher proportion of of mortality during their ﬁrst year of life. During the summer, lipids and (based on allometric scaling relationships) a lower predation is typically a major source of mortality, whereas phys- metabolic rate per unit mass than small individuals (Bochdan- iological shock and starvation are often cited as major sources of sky and Leggett 2001). In numerous overwintering starvation

Downloaded by [Department Of Fisheries] at 20:24 25 September 2012 mortality during the winter (Conover and Present 1990; Sogard experiments, larger individuals within a year-class exhibited 1997; Schultz and Conover 1999; Hurst 2007). When fish are higher survival than smaller conspecifics, suggesting that size- faced with limited resources, energy allocation must be directed dependent overwintering mortality was an important source of toward the highest risk factors during different life stages. Dur- mortality prior to recruitment (Post and Evans 1989; Johnson ing the summer, energy is typically directed toward maximizing and Evans 1990; Pangle et al. 2004). These starvation experi- growth at the expense of lipid reserves to achieve large size and ments clearly demonstrated that when food was lacking or when to minimize predation. In the fall and early winter, juvenile fish fish were unable to feed, larger individuals had higher survival often switch from maximizing growth to maximizing energy than smaller fish. However, fish of some species feed during stores to prevent starvation during the food-limited winter (Post the winter, potentially reducing the risk of starvation and alter- and Parkinson 2001; Hurst and Conover 2003; Biro et al. 2005; ing the energy allocation dynamics that occur prior to the onset Heintz and Vollenweider 2005). of winter. With a reduced risk of starvation, age-0 fish would

*E-mail: [email protected] Received March 25, 2011; accepted January 4, 2012 Published online June 13, 2012 855 856 BELL

not need to switch their energy allocation from growth to lipid METHODS storage, and the overwintering survival advantage of larger individuals may not exist. Assuming that prey is available and Sample Collection that all size-classes are able to feed equally, the greater energy The length-frequency data for determining overwintering reserves in larger fish would not result in higher survival. Lipid mortality of age-0 winter flounder were collected during the levels would (1) remain relatively constant through the winter, RIDEM seine survey in October and the spring (April) trawl (2) vary with the availability of resources, or (3) both (Bystrom¨ survey from 1988 to 2005. A gear change in the spring trawl et al. 2006). survey prevented the inclusion of data from later years of the Winter flounder Pseudopleuronectes americanus are cold- survey. Age-0 winter flounder of the two coastal stocks remain adapted fish that are able to survive low temperatures by pro- in the nearshore areas and are susceptible to seining until the ducing antifreeze proteins (Fletcher 1981). Adults of the two fall (Saucerman and Deegan 1991); thereafter, they move into coastal stocks (Gulf of Maine and southern New England–Mid deeper water and are then susceptible to trawling (Dominion Atlantic Bight) move into estuaries and breed during the coldest Resources Services 2007). The seine survey initially sampled 15 part of the year at temperatures as low as −1◦C (Collette and stations throughout Narragansett Bay, increasing to 18 stations Klein-MacPhee 2002). However, the winter habitat of young-of- in subsequent years. Each station was sampled with a single the-year (age-0) winter flounder is not well known. It has been sweep of a seine, covering an area of approximately 500 m2 widely assumed that age-0 winter flounder move into deeper wa- from the shore to the shallow subtidal zone. The majority of ter within their native estuary and burrow into the mud during winter flounder caught were age 0; individuals greater than the coldest parts of the year (McCracken 1963). Most poten- 150 mm total length were excluded based on a visual inspection tial predators of winter flounder (predators larger than 30 mm) of the length-frequency distribution and previous studies (With- in the northeast leave the estuaries during the coldest months, erell and Burnett 1993). Samples from all stations were pooled thus limiting predation mortality during the winter (Collette and and divided into 5-mm size-classes. The abundance in each Klein-MacPhee 2002). Prey resources may be limiting, how- size-class was converted to individuals per 500 m2 (Delong et al. ever, which suggests that starvation could be important. It is 2001). unknown whether age-0 fish are feeding at these temperatures, The RIDEM spring trawl survey was a random stratified sur- living off energy reserves, or displaying some combination of vey that included 26 stations in Narragansett Bay. The RIDEM both strategies. Winter flounder primarily store energy as lipid vessel towed a bottom trawl at 4.63 km/h (2.5 knots) for 20 min. and utilize it during periods of starvation (Maddock and Bur- The net had a headrope length of 13.7 m and a footrope length ton 1994). If age-0 winter flounder are relying solely on stored of 18.3 m. The width was assumed to be 60% of the footrope energy, lipid levels would be expected to increase in the late (Laura Lee, RIDEM, personal communication), resulting in a fall and to become depleted through the winter. Larger individ- total area swept of 16,946 m2. The length-frequency data for uals would have a higher proportion of lipids and would most all winter flounder caught during the spring survey in each year likely exhibit higher survival. If winter flounder are meeting were separated into cohorts with NORMSEP software (Hassel- their daily requirements through feeding, lipid levels would not blad 1966; Gayanilo et al. 1995). NORMSEP assumes that the increase in the late fall but would remain relatively constant length-frequency distribution of a single cohort can be repre- through the winter; therefore, size-dependent starvation would sented by a normal distribution. I determined the age-1 length- not be considered an important component of overwintering frequency distribution in the spring for the 1998–2004 cohorts mortality. of winter flounder. The age-1 data for 1988–1998 were from The aim of this study was to determine the overwintering Delong (2003). The same method was used in both analyses

Downloaded by [Department Of Fisheries] at 20:24 25 September 2012 mortality and physiology in relation to body size of age-0 winter and yielded the same results. There were very few age-1 fish flounder. Mortality was calculated from 17 years of survey data in the 1993 spring survey (i.e., 1992 cohort), which took place collected in the fall and spring by the Rhode Island Department later in the year than usual; thus, curve fitting with NORM- of Environmental Management (RIDEM) in Narragansett Bay. SEP produced poor results. Due to the low number of fish and Age-0 winter flounder physiology was examined to determine the slightly longer growth period, all individuals smaller than whether they demonstrated an energy allocation pattern simi- 170 mm were considered age-1 fish for that survey. Both surveys lar to that observed in many other fish (lipid build-up through were conducted on multiple days during the month. The sample the late fall and early winter and subsequent lipid depletion date was calculated as the mean day of the month weighted by through the coldest parts of the year) or whether they instead abundance. The weighted mean spring sample date occurred in were able to maintain energy reserves through the winter. The 1- April for all years except two. The 1993 spring survey (1992 co- year physiology study measured whole-body crude lipid content hort) was conducted in June, and the 2001 spring survey (2000 and stomach contents of age-0 winter flounder during fall–spring cohort) occurred in May. and examined their relationship with a number of environmental Age-0 winter flounder in the physiology study were covariates in three estuaries (Narragansett Bay; Niantic River, collected as part of ongoing trawl surveys in three estuaries Connecticut; and Hampton Harbor, New Hampshire). along the northeastern USA: one estuary north of Cape Cod OVERWINTERING SURVIVAL STRATEGY 857

TABLE 1. Number of age-0 winter flounder collected on each sampling date laboratory’s vessel pulled an otter trawl (cod-end liner mesh = as part of ongoing trawl surveys in the three estuaries. Blanks represent dates 6 mm; headrope length = 9.1 m) over a distance of 0.69 km. All when trawls were not schedulated for that location. fish that were smaller than 150 mm were considered to be age-0 Date Narragansett Bay Niantic River Hampton Harbor winter flounder from the same cohort. Due to the low number of individuals in the trawl samples, length-frequency analysis 2007 samples was not possible for the majority of samples. The cut-off length Oct 22 5 was derived from the NORMSEP analysis conducted for Narra- Nov 19 4 gansett Bay as well as 20 years of analysis conducted by the Do- Nov 26 2 18 minion Nuclear Connecticut Millstone Environmental Labora- Nov 29 2 tory (Witherell and Burnett 1993; Dominion Resources Services Dec 6 13 2007). After fish were caught, they were stored at −20◦C until Dec 10 7 processed. Dec 11 27 Dec 18 2 Dec 21 4 Laboratory Procedures 2008 samples Diet analysis.—Each fish was measured to the nearest mil- Jan 2 11 limeter (total length) and weighed to the nearest milligram. Stan- Jan 6 3 dard length and total length were measured on approximately Jan 7 23 100 fish, and the relationship between the two measurement Jan 22 7 types was determined with a linear regression. The relationship Jan 24 5 was used to estimate the total lengths of winter flounder with Jan 29 2 damaged caudal fins. Total ingested prey weight was calculated Feb 4 23 as the weight of the full stomach minus the weight of the empty Feb 12 5 stomach. Shell weight was included as prey weight for all mol- Feb 19 4 lusks. All prey items were identified to the lowest taxon possible, Feb 26 8 enumerated, and weighed as a taxon group. All of the remaining Mar 11 16 8 organs, including the liver, were removed from the body cav- Mar 18 11 10 ity prior to lipid analysis. Winter flounder store the majority of Mar 24 12 their lipid reserves under the skin, particularly in a fatty deposit Apr 1 20 along the lateral line (Maddock and Burton 1994). A negligible Apr 5 22 quantity of lipid is stored in the liver; therefore, the removal Apr 9 16 of the liver and other organs would not affect the measure of Apr 21 19 whole-body crude lipids (Tyler and Dunn 1976; Maddock and Total 104 89 116 Burton 1994). Lipid analysis.—Whole-body crude lipid analysis was conducted according to the procedure described by Lee et al. (1996). Structural lipids (phospholipids and cholesterols) constitute a (Hampton Harbor) and two estuaries south of Cape Cod (Nar- portion of the whole-body crude lipid content but typically re- ragansett Bay and the Niantic River; Table 1). Winter flounder main relatively constant over the course of the winter and thus ◦ ◦ Downloaded by [Department Of Fisheries] at 20:24 25 September 2012 were collected monthly in Hampton Harbor (42 52 N, 70 46 W) were not separately considered in this study (Osako et al. 2003; at three different stations from October 2007 to April 2008 Heintz and Vollenweider 2005). After the heads and tails were by Normandeau Associates, Inc. The Normandeau Associates removed, fish were minced into a paste with a cleaver. I com- vessel pulled a fine-mesh otter trawl at 3.70 km/h (2 knots) for bined 3.0 g (wet weight) of fish paste in an Eberback blending 10 min. Winter flounder were collected weekly in Narragansett jar with 30 mL of a 1:1 chloroform : methanol solution and Bay at one station (Fox Island, 41◦34.5N, 71◦24.3W) from blended for 1 min. The homogenate was poured through coarse November 2007 to April 2008 by the University of Rhode Island filter paper (fast speed) into a 50-mL, glass-stoppered gradu- (URI) Graduate School of Oceanography (GSO). The URI ated cylinder. Twelve milliliters of 0.5% sodium chloride were vessel pulled an otter trawl (mesh size at the cod end = 5.1 cm; added to the filtered homogenate as a buffer, stoppered, mixed, headrope length = 11.9 m) at 3.70 km/h (2 knots) for 30 min. and allowed to stand for 30–60 min (until a clear separation Winter flounder were collected approximately monthly in the occurred). A 3-mL aliquot of the chloroform layer (lower layer) Niantic River (41◦32N, 72◦19W) at three different stations was transferred to a preweighed, 10-mL beaker and was allowed from November 2007 to April 2008 by the Dominion Nuclear to evaporate on a hotplate. The extracted lipid was the material Connecticut, Inc., Millstone Environmental Laboratory. The remaining in the beaker after the chloroform had evaporated. 858 BELL

Whole-body crude lipid was calculated as and

Na loge N lipid extracted (g) M = o . (3) lipid (%) = −t fish paste (g wet weight) chloroform added (15 mL) × × 100. (1) To account for the two different gears, the October abundance 3mL (No) and April abundance (Na) were both scaled to the number of fish per 500 m2.Timet was in months. The estimate of M In addition, 65 fish from Hampton Harbor were dried to a (units per month) was a relative index of mortality, however, constant weight at 65◦C to determine moisture content, energy because potential differences in catchability were not included stores per gram of dry weight, and total energy content of in the model. Mortality was regressed against the mean total l ε each fish. Total energy content for each fish was calculated length in October ( )pluserror( ) to determine whether body as the energy content of lipids plus the energy content of lean size was related to overwintering mortality: tissue. For these calculations, I used a mean energy content of M = β + β · l + ε, 36.43 kJ/g dry weight for lipids and 20.1 kJ/g dry weight for 0 1 (4) lean tissue mass (Brett 1995). where β0 is the intercept and β1 is the coefficient for length. Winter flounder have small mouths and typically consume Environmental Data epibenthic prey or the exposed part of a prey item (e.g., worm The relationship between environmental factors and lipids heads or clam siphons). Because the stomach contents often con- was examined with the data available from each estuary. The stituted pieces of animals, it was challenging to enumerate them environmental factors were included in a multiple regression effectively. However, due to the winter flounder’s small mouths, as described below (Statistical Analysis section). Photope- the numerically abundant prey in this study also accounted for riod was calculated from data available through the National the majority of the prey mass as measured by weight. Prey Estuarine Research Reserve System (http://nerrs.noaa.gov/; weight alone was used to compare the prey items. The diet http://cdmo.baruch.sc.edu/). Photoperiod has been linked to the proportion of a given taxon consumed by an individual winter buildup of antifreeze proteins in winter flounder (Fletcher 1981) flounder was calculated by dividing the weight of that taxon in and to changes in the behavior of juveniles (Casterlin and the stomach by the summed weight of all taxa in the stomach. Reynolds 1982). The data from Great Bay Estuary, New Hamp- The average diet in each estuary over the entire study was cal- shire (43◦355.09N, 70◦4958.32W), were used for Hampton culated as the mean diet proportion of each taxon for all winter Bay. The data for Narragansett Bay came from the National Es- flounder in the estuary. The seasonal diet was calculated as the tuarine Research Reserve station within the bay; the same data mean of the summed proportions of all polychaete taxa and the were used for Narragansett Bay and the Niantic River. Hours summed proportions of all amphipod taxa in the stomach con- of daylight for each day were calculated as the total number of tents of all winter flounder captured per sample date within each hours (during the period 0400–2130 hours) for which the photon estuary. irradiance was greater than 16.7 mol·m−2·s−1. The photoperiod A crude estimate of stomach fullness was calculated by di- on each sampling date was calculated as the mean of the pho- viding the ingested prey weight by the winter flounder’s total toperiod for the7dpriortothesamplingdate.Temperature body weight and averaging over all fish caught on each sampling and salinity were measured with a conductivity–temperature– date. A large number of factors affect the quantity of food in a

Downloaded by [Department Of Fisheries] at 20:24 25 September 2012 depth instrument at each Narragansett Bay and Niantic River fish’s stomach, such as water temperature, metabolic demand, station. Dissolved oxygen was also measured in Narragansett and the time of day. Taking these caveats into consideration, the Bay. Chlorophyll was measured weekly at the Fox Island sta- ratio of ingested prey weight to fish body weight was believed tion in Narragansett Bay; these data are available through the to contain some relevant information on stomach fullness. URI-GSO (www.gso.uri.edu/phytoplankton/). Two permanent The relationship between winter flounder size and lipid con- stations in Hampton Harbor collected temperature, salinity, and tent was examined by regressing percent lipid content against to- dissolved oxygen data. The environmental collection stations tal length, the sample date (as a categorical covariate, cat[date]), were within a few kilometers of the three trawling stations. and the total length × sample date interaction:

lipids = β0 + β1 · length + β2 · cat(date) Statistical Analysis + β · × + ε. Overwintering mortality (M) was calculated with a simple 3 length cat(date) (5) exponential equation (Quinn and Deriso 1999), The variation in whole-body crude lipid content through the winter was examined with backward-elimination multiple re- −Mt Na = Noe (2) gression by testing environmental and physical variables against OVERWINTERING SURVIVAL STRATEGY 859

lipid content: part of a time series:

lipids = β0 + β1 · x1 + β2 · x2 +···+βk · xk + ε. (6) lipids = β0 + β1 · date + εt , (7)

ε = φε + v I tested the following variables: fish total length (x1), salinity where t t−1 t (x2), hours of daylight (x3), temperature (x4), date (x5), depth of the sampling stations (x6), estuary (x7), fish weight (x8), dis- 2 vt ∼ NID(0,σ ) and |φ| < 1. (8) solved oxygen (x9), and chlorophyll (x10). Correlated independent variables (R2 > 0.80) were removed from the model. The trend in lipid content over time was examined with a gener- The residual error (εt) is equal to the residual error in the pre- alized least-squares linear model fit. To account for potential vious time step (εt−1) plus an independent normally distributed autocorrelation in the residuals, an autoregressive lag-1 corre- error term (vt). As long as the absolute value of the autocorrela- lation structure was added to the model because samples were tion parameter (φ) is less than 1.0, the autocorrelation decays.

1995

Oct Apr Abundance/500 m^2 Abundance/500 Downloaded by [Department Of Fisheries] at 20:24 25 September 2012 0.00 0.05 0.10 0.15 0.20

0 5 10 15 20 Length (cm)

FIGURE 1. Length frequency of age-0 winter ﬂounder abundance per 500 m2 in October 1995 (solid line) and the following spring (April 1996). The spring abundance (dashed line) is substantially lower than October abundance. 860 BELL

1988 1989 1990 0.8 0.020 Oct Apr 0.04 0.6 0.8 0.4 0.6 0.4 Oct N Oct N Oct N Apr N Apr N Apr N 0.02 0.2 0.0 0.5 1.0 1.5 0.0 0.2 0.0 0.00 0.000 0.010 0.000 0.010 0.020 0 5 10 15 20 0 5 10 15 20 0 5 10 15 20 1991 1992 1993 4 0.20 23 Oct N Oct N Oct N Apr N Apr N Apr N 0.10 1 0 0.00 0.00 0.10 0.20 0.000 0.002 0.004 0.000 0.004 0.008 0.012 0.000 0.004 0 5 10 15 20 0 5 10 15 20 0 5 10 15 20 1994 1995 1996 0.5 0.30 0.20 0.4 0.20 Oct N Oct N Oct N Apr N Apr N Apr N 0.2 0.3 0.0 0.1 0.00 0.10 0.00 0.10 0.000 0.010 0.020 0.000 0.002 0.004 0.000 0.004 0.008 0.012 0 5 10 15 20 0 5 10 15 20 0 5 10 15 20

FIGURE 2. Length frequency of age-0 winter ﬂounder abundance (N) per 500 m2 in October (solid line) and the following spring (dashed line), 1988–1996. The x-axis describes total length (cm). Spring abundance is presented on the secondary y-axis to facilitate comparison of the length-frequency distributions.

RESULTS threshold below which fish would not survive the winter. Nar- ragansett Bay winter flounder did not exhibit size-dependent Overwintering Mortality overwintering mortality. The age-0 winter flounder in Narragansett Bay decreased in In the 1-year physiology study, 309 fish were sampled for abundance between October and spring, and in many years the diet and 296 were analyzed for lipids in the three estuaries mean of the length-frequency distribution increased substan- (Table 1). Fish ranged in size from 40 to 150 mm total length, tially (Figures 1–3). The object of this study was not to examine

Downloaded by [Department Of Fisheries] at 20:24 25 September 2012 and due to differences in sampling gear the length-frequency overwintering growth, but it was clear that simple mortality distributions were not the same among estuaries (Figure 5). in the smaller size-classes would not have accounted for the The sampling gear in Narragansett Bay had a larger mesh size changes in the length-frequency distributions between sampling and caught fish ranging from 80 to 160 mm (mean ± SD = periods. Differences in gear would also not have accounted for 121 ± 15 mm). The Niantic River survey used a trawl with the changes in distribution because the trawl gear contained a fine-mesh cod-end liner that retained small winter flounder and < other species of flatfish ( 5 cm). The changes in distribution TABLE 2. Results for regression of overwintering mortality (M) against mean indicated that growth occurred during a number of years. Over- total length in October (l) for age-0 winter flounder (M = β0 + β1 · l + ε; β = β = l ε = wintering mortality varied between 0.36 and 0.79 per month 0 intercept, 1 regression coefficient for ; error term). (mean M ± SD = 0.63 ± 0.13 per month). The regression of Parameter Estimate SE tP mortality on mean total length in October was not significant = (overall model P 0.12; Table 2; Figure 4). The distributions β0 0.119 0.309 0.386 0.705 of mean length in October, mean length in the spring, and mor- β1 0.068 0.041 1.665 0.117 tality did not exhibit any clear breaks that would indicate a size OVERWINTERING SURVIVAL STRATEGY 861

1997 1998 1999 0.5 0.4

Oct 0.008 0.4 Apr 0.4 0.6 0.2 0.3 Oct N Oct N Oct N Apr N Apr N Apr N 0.2 0.3 0.2 0.1 0.0 0.0 0.0 0.1 0.000 0.002 0.004 0.000 0.004 0.0000 0.0015 0.0030 0 5 10 15 20 0 5 10 15 20 0 5 10 15 20 2000 2001 2002 0.20 0.30 Oct N Oct N Oct N Apr N Apr N Apr N 0.0 0.4 0.8 1.2 0.0 0.2 0.4 0.00 0.10 0.000 0.004 0.000 0.001 0.002 0.003 0.000 0.002 0.004 0 5 10 15 20 0 5 10 15 20 0 5 10 15 20 2003 2004 0.8 0.4 0.6 0.4 0.6 Oct N Oct N Apr N Apr N 0.2 0.2 0.0 0.0 0.000 0.010 0.020 0.000 0.010 0.020 0 5 10 15 20 0 5 10 15 20

FIGURE 3. Length frequency of age-0 winter ﬂounder abundance (N) per 500 m2 in October (solid line) and the following spring (April; dashed line), 1997–2004. The x-axis describes total length (cm). Spring abundance is presented on the secondary y-axis to facilitate comparison of the length-frequency distributions.

a very small mesh size and typically caught fish ranging from counted for 68% of the diet by weight; unidentifiable food items 40 to 100 mm (mean ± SD = 72 ± 19 mm). The Hampton constituted 31% of the diet by weight (Table 3). Polychaetes and Harbor survey caught age-0 fish covering the full range of amphipods each made up 34% of the total prey weight. In Hamp- lengths (mean ± SD = 105 ± 25 mm). ton Harbor, polychaetes and amphipods accounted for 83% of the diet by weight (Table 3). Polychaetes made up 56% and am- Diet phipods made up 27% of the total prey weight. Unidentifiable The food habits data showed that 289 of the 309 fish sam- food items constituted 12.5% of the diet by weight. The propor-

Downloaded by [Department Of Fisheries] at 20:24 25 September 2012 pled during the study had food in their stomachs; only 20 of the tion of amphipods and polychaetes in the stomach contents of stomachs were empty. Amphipods and polychaetes dominated winter flounder varied over time and among the three estuaries the diet, constituting greater than 90% of the total stomach con- (Figure 6). tent weight in 71% of the fish. Other prey items were relatively The crude measure of stomach fullness (ingested prey rare: two Atlantic mud crabs Panopeus herbstii and two her- weight divided by total body weight) was within the same range mit crabs Pagurus sp. were observed in Narragansett Bay diet for all three estuaries and varied over the course of the winter samples; nine bivalve shells were present in Narragansett Bay (Figure 7). In Narragansett Bay and Hampton Harbor, stomach and Hampton Harbor samples; and four gastropod shells and fullness was higher in the fall–early winter and late winter–early the substrate itself were also observed. Substrate included sand spring than in early January and mid-February. The decline in and mud and was often part of an animal tube. Narragansett Bay appears to be large because only two fish were In Narragansett Bay, amphipods (particularly Unciola irro- caught on January 29 and both had empty stomachs; however, rata) and polychaetes (particularly bamboo worms, family Mal- during the entire study this was the only trawl sample in which danidae) dominated the winter flounder diets. These two prey all of the fish had empty stomachs. In the Niantic River, stomach groups constituted 92% of the total stomach content weight fullness varied over the winter but did not show a midwinter de- (Table 3). In the Niantic River, polychaetes and amphipods ac- cline. The measure of stomach fullness appeared to be relatively 862 BELL

FIGURE 4. Overwintering instantaneous mortality (per month) of age-0 winter ﬂounder versus the mean total length of the cohort in October. There was no signiﬁcant linear relationship between the two variables (P = 0.12).

constant through the fall–early winter and then increased for the found that the total energy content of adult winter ﬂounder

Downloaded by [Department Of Fisheries] at 20:24 25 September 2012 remainder of the winter–early spring. The water temperature varied between 14 and 24 kJ/g dry weight. never decreased below 4◦C in the Niantic River (Figure 8). The relationship between lipids and fish size varied among the estuaries. In Narragansett Bay, there was a negative relation- Lipid Content ship between total length and lipid content (β1 =−0.012711% Whole-body crude lipid content ranged from 0.89% to 4.85% lipid/mm; adjusted R2 = 0.79, P < 0.0001). The interaction term of total wet weight across the three estuaries (mean ± SD = β3 was not significant, indicating that the slope β1 was not sig- 1.94 ± 0.72% of wet weight). The whole-body crude lipid nificantly different between sampling periods. The interaction content expressed as a percentage of dry weight (measured in term was not included in the selected regression model. The cat- 65 fish from Hampton Harbor) ranged from 4.7% to 12.4% of egorical date term β2 determined whether the intercept varied dry weight (mean ± SD = 7.66 ± 1.68% of dry weight). Total between sampling dates. The categorical date term was not sig- energy content (lipids + lean tissue) was calculated from the nificantly different for the majority of the winter but increased dry weight percentages and ranged from 20.09 to 22.13 kJ/g dry in March and April (i.e., the last three sampling periods), when weight; these values are similar to those found in other studies. it was significantly higher than in the initial sampling period Steimle and Terranova (1985) estimated total winter flounder (Table 4). There was a significant negative relationship between energy content at 16.4 kJ/g dry weight, and Plante et al. (2005) lipid content and fish size on the initial sample date in the OVERWINTERING SURVIVAL STRATEGY 863

TABLE 3. Stomach contents of age-0 winter ﬂounder sampled in three es- Narragansett Bay, RI tuaries. For each estuary, the diet is reported as the proportional contribution of each diet item to the total stomach content weight (Prop.) averaged over the sampling dates and the total number of each prey item summed over the entire study (Abund.). Unreported abundance values represent prey items that constituted only part of an organism (e.g., worm heads).

Narragansett Niantic Hampton 468 12 Bay River Harbor 2 0 Diet item Prop. Abund. Prop. Abund. Prop. Abund. 40 60 80 100 120 140 160 Niantic River, CT Worms (total) 0.47 0.34 0.56 Polychaeta 0.155 0.172 0.406 Lumbrineri- 0.001 0.000 0.068 dae Maldanidae 0.319 0.164 0.075 Annelida 0.000 0.000 0.010 Frequency Amphipods 0.45 0.34 0.27 0246810 (total) 40 60 80 100 120 140 160 Amphipoda 0.029 57 0.191 174 0.204 1,048 Hampton Harbor, NH Unciola 0.404 731 0.036 27 0.062 136 irrorata Ampelisci- 0.015 74 0.117 119 0.002 6 dae Listriella 0.003 1 0.000 0 0.000 0 spp. Valvifera 0.000 0 0.000 0 0.013 36 Cumacea 0.001 2 0.011 11 0.022 94 0 5 10 15 40 60 80 100 120 140 160 Crustacea 0.002 4 0.001 1 0.002 1 Length (mm) Mollusca 0.004 11 0.000 1 0.002 7 Mud 0.010 0.001 0.006 FIGURE 5. Length frequency of age-0 winter ﬂounder in three estuaries over Vegetation 0.000 0.000 0.002 the winter (Narragansett Bay, Rhode Island; Niantic River, Connecticut; and Hampton Harbor, New Hampshire). The x-axis describes total length (mm). Unknown 0.046 0.306 0.125

TABLE 4. Lipids model for age-0 winter flounder sampled in Narragansett 2 = = = β + β · + β · Niantic River (adjusted R 0.29, P 0.0007). However, the Bay: lipids 0 1 length 2 cat(date). The number following cat(date) β indicates the sampling day as measured from the start of the collection period, interaction term 3 was significant, indicating that the relation- October 22. See Methods for definition of model terms. The interaction term ship between size and lipids varied over the winter (β1 ranged β · × 3 length cat(date) was not significant and was removed from the model. from −0.04223 to 0.00882% lipid/mm). The relationship was negative on four of the Niantic River sampling dates and pos- Parameter Estimate SE tP itive on two sampling dates (Table 5). For Hampton Harbor, Downloaded by [Department Of Fisheries] at 20:24 25 September 2012 β0 1.0429 0.1983 5.2599 <0.0001 there was also a negative relationship between fish length and 2 Length −0.0055 0.0015 −3.6033 0.0006 lipid content (β1 =−0.003707% lipid/mm; adjusted R = 0.30, Cat(date) 35 0.0418 0.1684 0.2485 0.8044 P < 0.0001). The interaction term β3 was not significant, but Cat(date) 49 0.2543 0.1221 2.0829 0.0405 the categorical date term β2 was significant, thus indicating that Cat(date) 60 −0.0730 0.1376 −0.5304 0.5973 the mean lipid content varied between sampling dates (Table 6). Cat(date) 72 0.0435 0.1193 0.3648 0.7163 There was a weak negative relationship between length and Cat(date) 76 0.1180 0.1483 0.7954 0.4288 lipid content for fish from all three estuaries on the majority of Cat(date) 94 −0.0425 0.1317 −0.3226 0.7478 sampling dates, indicating that larger fish did not have a higher Cat(date) 99 0.1496 0.2177 0.6870 0.4941 percentage of whole-body crude lipids than small fish. The lin- Cat(date) 113 0.2823 0.1304 2.1646 0.0335 ear models fit the data reasonably well. The residuals exhibited Cat(date) 120 0.0827 0.1374 0.6020 0.5489 no patterns and were normally distributed (Figure 9). Cat(date) 127 0.0709 0.1205 0.5883 0.5580 < Cat(date) 141 0.4920 0.1123 4.3798 0.0001 Mean Lipid Content over Time < Cat(date) 148 0.6528 0.1162 5.6203 0.0001 The whole-body crude lipid content of age-0 winter flounder < Cat(date) 166 0.9367 0.1078 8.6893 0.0001 varied by individual, over the course of the winter, and by 864 BELL

Narragansett Bay 0.0 0.2 0.4 0.6 0.8 1.0 Oct 22 Dec 22 Jan 22 Feb 22 Mar 22 Apr 22

Niantic

Poly Amph Unkn Proportion 0.0 0.2 0.4 0.6 0.8 1.0 Oct 22 Dec 22 Jan 22 Feb 22 Mar 22 Apr 22

Hampton Harbor 0.0 0.2 0.4 0.6 0.8 1.0 Oct 22 Dec 22 Jan 22 Feb 22 Mar 22 Apr 22 Downloaded by [Department Of Fisheries] at 20:24 25 September 2012 FIGURE 6. Proportion (by weight) of polychaetes (Poly), amphipods (Amph), and unidentifiable (unknown) items (Unkn) in the stomach contents of age-0 winter flounder sampled in three estuaries during the winter. The tick marks on the x-axis represent the 22nd day of each month from October to April. Polychaetes, amphipods, and unidentifiable items constituted greater than 95% of the total diet (by weight) in all three estuaries.

estuary (Figure 10). In Hampton Harbor, boat problems pre- trend, and then increased rapidly to the highest observed values vented sampling from the end of January through February. The in March and April. There was an increase in temperature and a mean whole-body crude lipid content of fish sampled in the fall sharp increase in salinity during March and April (Figure 8). A and early winter was generally higher than that of fish sampled simple linear model could not be fitted to the Narragansett Bay in March and April. A linear regression with an autoregressive data because the mean lipid content increased sharply and was lag-1 error structure to account for autocorrelation indicated not linear. An initial fit produced strong patterns in the residuals, a small, significant negative trend in lipid content over the and these patterns could not be corrected with transformations. −3 sampling period (β1 =−3.85 × 10 % lipid/d, φ = 0.129, P The model was instead fitted to the data from October–February, < 0.0001; Table 7). Lipid levels in fish from Narragansett Bay with the sharp increase in March–April removed. The slope of were variable through the fall and winter, with no discernible the line was not significantly different from zero (β1 =−5.0 × OVERWINTERING SURVIVAL STRATEGY 865

FIGURE 7. A crude measure of stomach fullness (ingested prey weight divided by total wet body weight) for age-0 winter ﬂounder in three estuaries (values are medians calculated for each sampling date). [Figure available online in color.] Downloaded by [Department Of Fisheries] at 20:24 25 September 2012

FIGURE 8. Environmental variables—temperature (◦C), photoperiod (daylight hours), salinity (‰ [ppt]), and dissolved oxygen (mg/L)—in each estuary over the course of the winter (NB = Narragansett Bay, Rhode Island; CT = Niantic River, Connecticut; NH = Hampton Harbor, New Hampshire). [Figure available online in color.] 866 BELL Downloaded by [Department Of Fisheries] at 20:24 25 September 2012

FIGURE 9. Residuals and Q–Q plots for the regression of age-0 winter ﬂounder lipids against ﬁsh total length with date in the three estuaries (stand = standard). In the Q–Q plots, standard residuals are approximately normal if they are on the one to one line when plotted against theoretical values from a normal distribution. OVERWINTERING SURVIVAL STRATEGY 867

TABLE 5. Lipids model for age-0 winter ﬂounder sampled in the Niantic River: lipids = β0 + β1·length + β2·cat(date) + β3·length × cat(date). The number following cat(date) indicates the sampling day as measured from the start of the collection period, October 22. See Methods for deﬁnition of model terms.

Parameter Estimate SE tP

β0 7.0899 2.2482 3.1536 0.0025 Length −0.0422 0.0182 −2.3221 0.0237 Cat(date) 50 −5.4847 2.2795 −2.4061 0.0193 Cat(date) 92 −5.3014 4.1288 −1.2840 0.2042 Cat(date) 105 −3.8537 2.2864 −1.6855 0.0972 Cat(date) 148 −5.4870 2.3390 −2.3459 0.0224 Cat(date) 162 −5.7096 2.3795 −2.3995 0.0196 Length × cat(date) 50 0.0399 0.0188 2.1252 0.0378 Length × cat(date) 92 0.0422 0.0772 0.5471 0.5863 Length × cat(date) 105 0.0251 0.0190 1.3239 0.1907 Length × cat(date) 148 0.0448 0.0202 2.2173 0.0305 Length × cat(date) 162 0.0510 0.0215 2.3758 0.0208 FIGURE 10. Mean ( ± SE) whole-body crude lipid content of age-0 winter ﬂounder by date in the three estuaries. [Figure available online in color.]

−5 φ = = 2 10 % lipid/d, 0.018, P 0.9767). Mean lipid content in β0 + β1·length + β4·temperature (R = 0.27, P < 0.0001). A the Niantic River varied over the course of the sampling period single regression model combining all three estuaries resulted and exhibited a small but significant increase (β1 =−4.41 × in the following significant model, with estuary as a categori- −3 φ = = 10 % lipid/d, 0.195, P 0.0025). cal variable: lipids = β0 + β3·photoperiod + β4·temperature 2 The best-fitting multiple regression model differed among + β6·depth + β7·cat(estuary) (R = 0.20, P < 0.0001). The the three estuaries (Table 8). With backward elimination, the multiple regression analysis did not provide a consistent lin- full model for Narragansett Bay reduced to lipids = β0 + ear relationship between the environmental variables and lipid 2 β1·length + β2·salinity + β3·photoperiod (R = 0.72, P < content in the three estuaries. 0.0001). The high correlation was largely the result of the increase in lipids and photoperiod during March and April. For the Niantic River, none of the environmental variables were DISCUSSION significant. With backward elimination, the model reduced to Size-selective mortality is predicated on the idea that smaller 2 the date of sampling: lipids = β0 + β5·date (R = 0.13, P = individuals within a cohort are more susceptible to starvation 0.001). For Hampton Harbor, the full model reduced to lipids = and predation (Sogard 1997; Post and Parkinson 2001; Hurst 2007). The overwintering mortality and lipid levels indicated that smaller winter flounder were not more susceptible than large TABLE 6. Lipids model for age-0 winter flounder sampled in Hampton Har- fish; these findings provided a direct mechanism for the lack of = β + β · + β · bor: lipids 0 1 length 2 cat(date). The number following cat(date) size-selective mortality in Narragansett Bay. Lipid stores did not indicates the sampling day as measured from the start of the collection period, Downloaded by [Department Of Fisheries] at 20:24 25 September 2012 increase with size in age-0 winter flounder. Under food-limited October 22. See Methods for definition of model terms. The interaction term β3·length × cat(date) was not significant and was removed from the model. conditions, larger individuals would not have additional energy reserves and would exhibit the same overwintering survival as Parameter Estimate SE tPsmaller individuals. This study did not examine predation, and estimated mortality was not partitioned into starvation and pre- β 2.4675 0.3002 8.2189 <0.0001 0 dation components. However, the lack of a size signal suggests Length −0.0037 0.0017 −2.1632 0.0328 that the impacts of both starvation and predation were evenly Cat(date) 35 −0.0344 0.2087 −0.1650 0.8693 distributed across all size-classes. Cat(date) 45 0.0435 0.2222 0.1957 0.8452 Few studies have documented lipid content in overwintering Cat(date) 57 −0.2371 0.3298 −0.7190 0.4738 juvenile fish; those studies that are available show elevated lipid Cat(date) 77 −0.3757 0.2125 −1.7686 0.0799 levels in the fall and early winter and depleted lipid levels in the Cat(date) 141 −0.7013 0.2344 −2.9919 0.0035 spring (Miranda and Hubbard 1994; Biro et al. 2004). Striped Cat(date) 154 −0.6907 0.2181 −3.1674 0.0020 bass Morone saxatilis in the northeast began building reserves Cat(date) 170 −0.5668 0.2130 −2.6607 0.0090 around November, with maximum lipid levels occurring in Cat(date) 182 −0.4936 0.2024 −2.4389 0.0164 late December–January (Hurst et al. 2000); lipid levels then 868 BELL

TABLE 7. Results for regression of mean lipids in age-0 winter ﬂounder versus time in each estuary (lipids = β0 + β1 · date + εt ). See Methods for deﬁnition of model terms.

Estuary Parameter Estimate SE tP

Narragansett Bay β0 1.67353 0.15535 10.77268 <0.0001 Date −0.00005 0.00175 −0.02932 0.97673 φ 0.01763 Niantic River β0 1.30848 0.16496 7.93204 <0.0001 Date 0.00441 0.00141 3.13981 0.00249 φ 0.19493 Hampton Harbor β0 2.12822 0.08422 25.26997 <0.0001 Date −0.00385 0.00069 −5.57217 <0.0001 φ 0.12923

declined as the winter progressed. Walleye pollock Theragra the result of energetic demand outpacing caloric intake, while chalcogramma in Alaska exhibited maximum lipid content in any increase in lipid requires a fish to ingest prey at a level that December followed by minimum levels in March (Heintz and satisfies its daily requirements plus a surplus that can be put Vollenweider 2005). Sampling for the present study began in into storage. Because winter flounder consumed prey during the October and only occurred during 1 year, but in two of the three entire sampling period, the majority of the variation in lipid con- study estuaries, sampling did not capture any accumulation tent during the winter can most likely be attributed to variation period or subsequent decline. Lipid levels in the third estuary, in the quantity and quality of available prey. Hampton Harbor, were slightly lower during late winter Winter flounder are benthivores that are known to consume compared with the fall–early winter, but due to boat problems epibenthic organisms (Klein-MacPhee 1978; Stehlik and Meise samples could not be collected during late January through 2000; Link et al. 2002). As in previous studies, amphipods February. Lipid levels may have simply declined during this and polychaetes were the most important prey taxa found interval, but given the fact that lipid levels of fish in the other in the winter flounder stomachs. Winter flounder are visual estuaries varied above and below the levels observed during feeders, typically consuming the abundant available prey at the the fall, it would be difficult to speculate on the lipid levels of substrate–water interface (MacDonald and Green 1986; Franz Hampton Harbor winter flounder during the same time period. and Tanacredi 1992; Steimle et al. 1994; Carlson et al. 1997; Changes in physical factors (temperature, salinity, and dis- Wanat 2002). Although benthic data were not collected, the solved oxygen) can have important effects on fish osmoregu- varying diets of winter flounder within the three estuaries over lation and metabolism and, consequently, energy use. Seasonal time probably reflected the abundance or activity of epibenthic changes in the environment of the three estuaries (temperature species in those estuaries. The winter–spring amphipod U. and photoperiod) were significantly correlated with variation irrorata has a much higher energetic value than many other am- in the lipid content of winter flounder. These seasonal changes phipod species (Steimle and Terranova 1985) and increases in also affect the abundance and composition of other species in abundance in January–February, peaking around May (Steimle the estuary. Any decrease in whole-body crude lipid content is et al. 1994). Unciola irrorata was an extremely important prey

Downloaded by [Department Of Fisheries] at 20:24 25 September 2012 TABLE 8. Parameter estimates for the multiple linear regression of age-0 winter flounder lipid content with physical and environmental covariates in each estuary and in all estuaries combined. See Methods for definition of model terms; only significant terms were retained and are displayed here. Estuary was included as a categorical variable.

Parameter Narragansett Bay Niantic River Hampton Harbor All estuaries

β0 −8.0492 1.343507 1.7671 −0.1844 Length (mm) −0.0115 −0.0052 Salinity (‰) 0.1670 Photoperiod (h) 0.6239 0.0837 Temperature (◦C) 0.1049 0.0754 Date 0.0041 Depth 0.0427 Estuary (Narragansett Bay) 0.7852 Estuary (Hampton Harbor) 0.1127 OVERWINTERING SURVIVAL STRATEGY 869

item for winter flounder in Narragansett Bay but was negligible cally migrate into cold water (4–6◦C) during the day to reduce in the diets of fish sampled in the other estuaries. The large predation and decrease energetic demand in food-limited lakes increase in whole-body crude lipid content (March–April) in (Brett 1995). Although the cold temperatures provide a poten- Narragansett Bay winter flounder may be linked to an increase tial thermal refuge from predation, age-0 winter flounder may in U. irrorata abundance and thus consumption of this taxon. also be taking advantage of a lower metabolic rate and reduced Food resources, metabolism, and daily energy requirements competition for resources relative to fish that occupy warmer typically decrease during the winter as the temperature declines. offshore wintering areas. Age-0 winter flounder continued to eat throughout the winter; The strategy of feeding through the winter and not allocating however, the crude measurement of stomach fullness suggested energy for storage may also allow winter flounder to continue that feeding decreased in two of the estuaries during the coldest to grow during the coldest months of the year. Growth was part of the year. Unfortunately, it was not possible to determine not explicitly addressed in this study, but based on the change whether the fish were responding to the temperature or to the in length distributions between fall and spring, it is clear that availability of prey. Stomach fullness did not decline in the growth occurred during a number of years. Fish that switch Niantic River, where the temperatures remained slightly warmer, between summer growth and fall energy storage typically stop but again there were no data on prey availability. growing during the winter (Hurst and Conover 1998). Other age- Winter feeding is not unique to winter flounder, yet this 0 estuarine fish (black sea bass Centropristis striata, tautog Tau- species appears to be the only example of a marine fish that toga onitis, cunner Tautogolabrus adspersus, and smallmouth does not build up energy reserves prior to the onset of winter flounder Etropus microstomus) had negligible growth during a but instead feeds at cold temperatures and maintains its lipid 135-d overwintering study conducted in the laboratory (Hales stores. A few studies of freshwater fish have reported a similar and Able 2001). The life history of age-0 winter flounder sug- alternative overwintering survival strategy. In a pre-alpine lake, gests an alternative overwinter survival strategy compared with age-0 ruffe Gymnocephalus cernuus and age-0 Eurasian perch the strategy used by many other age-0 fish. Perca fluviatilis did not exhibit size-dependent mortality because Starvation has been hypothesized as a major source of over- they fed through the winter. The ruffe, which fed on a benthic wintering mortality for age-0 individuals of temperate-latitude diet, actually increased lipid content during the winter (Eck- species because many fish are unable to feed at low tempera- mann 2004). Age-0 walleyes Sander vitreus did not experience tures and prey resources are often limited. Based on the risk of size-specific mortality during overwintering mesocosm studies starvation, many age-0 fish exhibit a survival strategy in which even when predators were present, and the walleyes showed they direct energy away from somatic growth and toward lipid no distinct relationship between size and weight-specific lipid reserves prior to the onset of winter. Larger individuals often stores (Pratt and Fox 2002). Walleyes fed during the winter, have an advantage under these starvation conditions because which reduced the risk of starvation and reduced the number lipid reserves increase with size and metabolic rate per unit of fish in poor condition (i.e., starving fish) that would be easy mass decreases with size (Schultz and Conover 1999; Post and prey for predators. Fish that feed at low temperatures can sur- Parkinson 2001; Biro et al. 2004, 2005; Pangle et al. 2004; Hurst vive if the daily food requirements that are necessary to support 2007). Alternatively, fish that are able to feed at low tempera- temperature-reduced metabolism do not exceed the available tures would have a reduced risk of starvation. Such fish would resources. When resources are limited but not absent, smaller not need to switch from summer growth to fall energy storage, fish may even have an advantage because they have a lower crit- and lipid reserves would remain relatively constant through the ical resource demand, requiring less total food (Pratt and Fox winter. Although there was only 1 year of physiology sampling, 2002; Bystrom¨ et al. 2006). Pratt and Fox (2002) suggested that larger age-0 winter flounder did not have higher lipid stores

Downloaded by [Department Of Fisheries] at 20:24 25 September 2012 the smaller walleyes may have come closer to consuming their than smaller individuals, and the age-0 fish did not exhibit size- full daily ration than did larger individuals, thus eliminating the dependent overwintering mortality. Age-0 winter flounder fed expected advantage of larger fish. throughout the course of the winter and maintained their whole- One caveat of this type of work is that only the survivors body crude lipid content during fall and winter. The data lack are sampled. It is unclear whether the fish that experienced temporal replication, but the spatial coherence in results (i.e., mortality had different lipid levels or a different food intake. consistency among the three estuaries, which represent two dif- The goal of this study was to examine mortality and physiology ferent stocks) suggests that the consumption and energy alloca- in relation to body size. Although the sampled fish only represent tion patterns are real and that age-0 winter flounder follow an the survivors, they covered the full length distribution of age-0 alternative overwintering survival strategy. winter flounder, indicating that the range of body sizes were feeding and maintaining their lipid levels. Unlike many marine fish, age-0 winter flounder of the two ACKNOWLEDGMENTS coastal stocks do not move offshore, which could serve as a I gratefully acknowledge the hard work of the individuals physiological adaption to reduce overall energy expenditures who collected fish during the four trawl surveys, including Erin during the winter. Sockeye salmon Oncorhynchus nerka verti- Bohaboy (URI-GSO weekly fish trawl); RIDEM; Don Danila 870 BELL

(Dominion Nuclear Connecticut, Inc., Millstone Environmental Hales, L. S., and K. W. Able. 2001. Winter mortality, growth and behavior of Laboratory); and Normandeau Associates, Inc. I thank Chong young-of-the-year of four coastal fishes in New Jersey (USA) waters. Marine Lee (Food Sciences, URI) for providing laboratory space and Biology 139:45–54. Hasselblad, V. 1966. Estimation of parameters for a mixture of normal distribu- techniques for processing fish lipids; I am also grateful to tions. Technometrics 8:431–444. Chris Calabretta, Sheldon Pratt, and Jason Graff for assisting Heintz, R. A., and J. J. Vollenweider. 2005. Seasonal variation in energy alloca- with benthic invertebrate identification. Manuscript comments tion strategies of walleye pollock. Alaska Fisheries Science Center Quarterly and suggestions by Jeremy Collie are greatly appreciated. The Report (October-November-December):1–7. manuscript was also greatly improved by three reviewers. The Hurst, T. P. 2007. Causes and consequences of winter mortality in fishes. Journal of Fish Biology 71:315–345. Nature Conservancy provided funding for this project. Hurst, T. P., and D. O. Conover. 1998. Winter mortality of young-of-the-year Hudson River striped bass (Morone saxatilis): size-dependent patterns and effects on recruitment. Canadian Journal of Fisheries and Aquatic Sciences REFERENCES 55:1122–1130. Biro, P. A., A. E. Morton, J. R. Post, and E. A. Parkinson. 2004. Over- Hurst, T. P., and D. O. Conover. 2003. Seasonal and interannual variation in winter lipid depletion and mortality of age-0 rainbow trout (Oncorhynchus the allometry of energy allocation in juvenile striped bass. Ecology 84:3360– mykiss). Canadian Journal of Fisheries and Aquatic Sciences 61:1513– 3369. 1519. Hurst, T. P., E. T. Schultz, and D. O. Conover. 2000. Seasonal energy dynamics of Biro, P. A., J. R. Post, and M. V. Abrahams. 2005. Ontogeny of energy alloca- young-of-the-year Hudson River striped bass. Transactions of the American tion reveals selective pressure promoting risk-taking behaviour in young fish Fisheries Society 129:145–157. cohorts. Proceedings of the Royal Society B 272:1443–1448. Johnson, T. B., and D. O. Evans. 1990. Size-dependent winter mortality of Bochdansky, A., and W. Leggett. 2001. Winberg revisited: convergence of rou- young-of-the-year white perch: climate warming and invasion of the Lauren- tine metabolism in larval and juvenile fish. Canadian Journal of Fisheries and tian Great Lakes. Transactions of the American Fisheries Society 119:301– Aquatic Sciences 58:220–230. 313. Brett, J. R. 1995. Energetics. Pages 1–63 in C. Groot, L. Margolis, and W. C. Klein-MacPhee, G. 1978. Synopsis of biological data for the winter flounder Clarke, editors. Physiological ecology of Pacific salmon. University of British Pseudopleuronectes americanus (Walbaum). NOAA Technical Report NMFS Columbia Press, Vancouver. Circular 414. Bystrom,¨ P., J. Andersson, A. Kiessling, and L. O. Eriksson. 2006. Size and Lee, C. M., B. Trevino, and M. Chaiyawat. 1996. A simple and rapid sol- temperature dependent foraging capacities and metabolism: consequences vent extraction method for determining total lipids in fish tissue. Journal of for winter starvation mortality in fish. Oikos 115:43–52. AOAC (Association of Official Analytical Chemists) International 79:487– Carlson, J. K., D. J. Randall, and M. E. Mroczka. 1997. Feeding habits of 492. winter flounder (Pleuronectes americanus) in a habitat exposed to anthro- Link, J., K. Bolles, and C. G. Milliken. 2002. The feeding ecology of flatfish pogenic disturbance. Journal of Northwest Atlantic Fishery Science 21:65– in the northwest Atlantic. Journal of Northwest Atlantic Fishery Science 73. 30:1–17. Casterlin, M. E., and W. W. Reynolds. 1982. Thermoregulatory behavior and MacDonald, J. S., and R. H. Green. 1986. Food resource utilization by five diel activity of yearling winter flounder, Pseudopleuronectes americanus species of benthic feeding fish in Passamaquoddy Bay, New Brunswick. (Walbaum). Environmental Biology of Fishes 7:177–180. Canadian Journal of Fisheries and Aquatic Sciences 43:1534–1546. Collette, B. B., and G. Klein-MacPhee, editors. 2002. Bigelow and Schroeder’s Maddock, D. M., and M. P. M. Burton. 1994. Some effects of starvation on the fishes of the Gulf of Maine. Smithsonian Books, Washington, D.C. lipid and skeletal muscle layers of the winter flounder, Pseudopleuronectes Conover, D. O., and T. M. C. Present. 1990. Countergradient variation in growth americanus. Canadian Journal of Fisheries and Aquatic Sciences 72:1672– rate: compensation for length of the growing season among Atlantic silver- 1679. sides from different latitudes. Oecologia 83:316–324. McCracken, F. D. 1963. Seasonal movements of the winter flounder, Pseu- Delong, A. K. 2003. A key factor analysis of Narragansett Bay winter flounder. dopleuronectes americanus on the Atlantic coast. Journal of the Fisheries University of Rhode Island, Kingston. Research Board of Canada 20:551–586. Delong, A. K., J. S. Collie, C. J. Meise, and J. C. Powell. 2001. Estimating Miranda, L., and W. Hubbard. 1994. Length-dependent winter survival and lipid growth and mortality of juvenile winter flounder, Pseudopleuronectes ameri- composition of age-0 largemouth bass in Bay Springs Reservoir, Mississippi. canus, with a length-based model. Canadian Journal of Fisheries and Aquatic Transactions of the American Fisheries Society 123:80–87. Downloaded by [Department Of Fisheries] at 20:24 25 September 2012 Sciences 58:2233–2246. Osako, K., K. Kuwahara, H. Saito, M. A. Hossain, and Y. Nozaki. 2003. Dominion Resources Services. 2007. Monitoring the marine environment of Effect of starvation on lipid metabolism and stability of DHA content of Long Island Sound at Millstone Power Station: 2006 annual report. Millstone lipids in horse mackerel (Trachurus japonicus) tissues. Lipids 38:1263– Power Station, Millstone Environmental Laboratory, Waterford, Connecticut. 1267. Eckmann, R. 2004. Overwinter changes in mass and lipid content of Perca Pangle, K. L., T. M. Sutton, R. E. Kinnunen, and M. H. Hoff. 2004. Overwinter fluviatilis and Gymnocephalus cernuus. Journal of Fish Biology 65:1498– survival of juvenile lake herring in relation to body size, physiological con- 1511. dition, energy stores and food ratio. Transactions of the American Fisheries Fletcher, G. L. 1981. Effects of temperature and photoperiod on the plasma Society 133:1235–1246. freezing point depression, Cl− concentration, and protein “antifreeze” in Plante, S., C. Audet, Y. Lambert, and J. de la Noue.¨ 2005. Alternative methods winter flounder. Canadian Journal of Zoology 59:193–201. for measuring energy content in winter flounder. North American Journal of Franz, D. R., and J. T. Tanacredi. 1992. Secondary production of the amphipod Fisheries Management 25:1–6. Ampelisca abdita Mills and its importance in the diet of juvenile winter Post, J. R., and D. O. Evans. 1989. Size-dependent overwinter mortality flounder (Pleuronectes americanus) in Jamaica Bay, New York. Estuaries of young-of-the-year yellow perch (Perca flavescens): laboratory, in situ, 15:193–203. and field experiments. Canadian Journal of Fisheries and Aquatic Sciences Gayanilo, F. C., P. Sparre, and D. Pauly. 1995. The FAO-ICLARM stock as- 46:1958–1968. sessment tools (FISAT) user’s guide, volume 8. Food and Agriculture Orga- Post, J. R., and E. A. Parkinson. 2001. Energy allocation strategy in young fish: nization of the United Nations, Rome. allometry and survival. Ecology 82:1040–1053. OVERWINTERING SURVIVAL STRATEGY 871

Pratt, T. C., and M. G. Fox. 2002. Influence of predation risk on the overwinter americanus, in the New York Bight apex. U.S. National Marine Fisheries mortality and energetic relationships of young-of-year walleyes. Transactions Service Fishery Bulletin 92:608–619. of the American Fisheries Society 131:885–898. Steimle, F. W., and R. J. Terranova. 1985. Energy equivalents of marine organ- Quinn, T. J., and R. B. Deriso. 1999. Quantitative fish dynamics. Oxford Uni- isms from the continental shelf of the temperate northwest Atlantic. Journal versity Press, New York. of Northwest Atlantic Fishery Science 6:117–124. Saucerman, S. E., and L. A. Deegan. 1991. Lateral and cross-channel movement Tyler, A. V., and R. S. Dunn. 1976. Ration, growth, and measures of so- of young-of-the-year winter flounder (Pseudopleuronectes americanus)in matic and organ condition in relation to meal frequency in winter floun- Waquoit Bay, Massachusetts. Estuaries 14:440–446. der, Pseudopleuronectes americanus, with hypotheses regarding popula- Schultz, E. T., and D. O. Conover. 1999. The allometry of energy reserve tion homeostasis. Journal of the Fisheries Research Board of Canada 33: depletion: test of a mechanism for size-dependent winter mortality. Oecologia 63–75. 119:474–483. Wanat, J. M. 2002. Using habitat suitability models to identify essen- Sogard, S. M. 1997. Size-selective mortality in the juvenile stage of teleost tial fish habitat for the winter flounder (Pseudopleuronectes ameri- fishes: a review. Bulletin of Marine Science 60:1129–1157. canus) in Great Bay estuary, N.H. Master’s thesis. University of New Stehlik, L. L., and C. J. Meise. 2000. Diet of winter flounder in a New Jersey Hampshire, Durham. estuary: ontogenetic change and spatial variation. Estuaries 23:381–391. Witherell, D. B., and J. Burnett. 1993. Growth and maturation of winter flounder, Steimle, F. W., D. Jeffress, S. A. Fromm, R. N. Reid, J. J. Vitaliano, and A. Pleuronectes americanus, in Massachusetts. U.S. National Marine Fisheries Frame. 1994. Predator-prey relationships of winter flounder, Pleuronectes Service Fishery Bulletin 91:816–820. Downloaded by [Department Of Fisheries] at 20:24 25 September 2012 This article was downloaded by: [Department Of Fisheries] On: 25 September 2012, At: 20:25 Publisher: Taylor & Francis Informa Ltd Registered in England and Wales Registered Number: 1072954 Registered office: Mortimer House, 37-41 Mortimer Street, London W1T 3JH, UK

Transactions of the American Fisheries Society Publication details, including instructions for authors and subscription information: http://www.tandfonline.com/loi/utaf20 Evidence of Local Adaptation in Westslope Cutthroat Trout Daniel P. Drinan a e , Alexander V. Zale b , Molly A. H. Webb c , Mark L. Taper d , Bradley B. Shepard a f & Steven T. Kalinowski d a Montana Cooperative Fishery Research Unit, Department of Ecology, Montana State University, Bozeman, Montana, 59717, USA b U.S. Geological Survey, Montana Cooperative Fishery Research Unit and Department of Ecology, Montana State University, Bozeman, Montana, 59717, USA c U.S. Fish and Wildlife Service, Bozeman Fish Technology Center, Bozeman, Montana, 59715, USA d Department of Ecology, Montana State University, Bozeman, Montana, 59717, USA e School of Aquatic and Fishery Sciences, University of Washington, Box 355020, Seattle, Washington, 98195, USA f Wildlife Conservation Society, 301 North Willson Avenue, Bozeman, Montana, 59715, USA Version of record first published: 11 Jun 2012.

To cite this article: Daniel P. Drinan, Alexander V. Zale, Molly A. H. Webb, Mark L. Taper, Bradley B. Shepard & Steven T. Kalinowski (2012): Evidence of Local Adaptation in Westslope Cutthroat Trout, Transactions of the American Fisheries Society, 141:4, 872-880 To link to this article: http://dx.doi.org/10.1080/00028487.2012.675907

PLEASE SCROLL DOWN FOR ARTICLE

Full terms and conditions of use: http://www.tandfonline.com/page/terms-and-conditions 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. The publisher does not give any warranty express or implied or make any representation that the contents will be complete or accurate or up to date. The accuracy of any instructions, formulae, and drug doses should be independently verified with primary sources. The publisher shall not be liable for any loss, actions, claims, proceedings, demand, or costs or damages whatsoever or howsoever caused arising directly or indirectly in connection with or arising out of the use of this material. Transactions of the American Fisheries Society 141:872–880, 2012 C American Fisheries Society 2012 ISSN: 0002-8487 print / 1548-8659 online DOI: 10.1080/00028487.2012.675907

ARTICLE

Evidence of Local Adaptation in Westslope Cutthroat Trout

Daniel P. Drinan*1 Montana Cooperative Fishery Research Unit, Department of Ecology, Montana State University, Bozeman, Montana 59717, USA Alexander V. Zale U.S. Geological Survey, Montana Cooperative Fishery Research Unit and Department of Ecology, Montana State University, Bozeman, Montana 59717, USA Molly A. H. Webb U.S. Fish and Wildlife Service, Bozeman Fish Technology Center, Bozeman, Montana 59715, USA Mark L. Taper Department of Ecology, Montana State University, Bozeman, Montana 59717, USA Bradley B. Shepard2 Montana Cooperative Fishery Research Unit, Department of Ecology, Montana State University, Bozeman, Montana 59717, USA Steven T. Kalinowski Department of Ecology, Montana State University, Bozeman, Montana 59717, USA

Abstract An understanding of the process of local adaptation would allow managers to better protect and conserve species. Many salmonids are in need of such efforts, and because they often persist in differing, isolated environments, they are useful organisms for studying local adaptation. In addition, the temperature sensitivity of salmonids provides a likely target for natural selection. We studied thermal adaptation in four wild populations and one hatchery stock of westslope cutthroat trout Oncorhynchus clarkii lewisi. The mean summer temperatures of source streams ranged from 6.7◦C to 11.2◦C. Embryos were collected from the wild, and embryonic development, embryonic survival, and juvenile growth were determined. A signiﬁcant relationship between median embryonic survival and source stream temperature was detected. Based on a rank test, populations from colder streams had a greater decline in median embryonic survival at warm temperatures than populations from warmer streams. Embryonic development and

Downloaded by [Department Of Fisheries] at 20:25 25 September 2012 juvenile growth did not appear to be inﬂuenced by source. These ﬁndings suggest that populations are thermally adapted to their source streams and this should be considered by managers. However, further study is necessary to sort out the potential confounding factors, whether genetic or epigenetic.

Salmonids often persist in disparate and isolated environ- 1993; Quinn et al. 2000). This variation in habitat across their ments, and their strong homing behavior and temporal differ- range may lead to concomitant variation in traits that confer ences in reproduction often further isolate populations (Quinn high ﬁtness. If these traits are heritable and isolation exists,

*Corresponding author: [email protected] 1Present address: School of Aquatic and Fishery Sciences, University of Washington, Box 355020, Seattle, Washington 98195, USA. 2Present address: Wildlife Conservation Society, 301 North Willson Avenue, Bozeman, Montana 59715, USA. Received July 13, 2011; accepted December 12, 2011 Published online June 11, 2012

872 LOCAL ADAPTATION IN WESTSLOPE CUTTHROAT TROUT 873

populations will probably become locally adapted through nat- how climate change may affect populations. The Intergovern- ural selection. Understanding whether populations are locally mental Panel on Climate Change estimates that by 2100 global adapted would provide managers with valuable information, as air temperature will increase by 1.4–5.8◦C (IPCC 2003). many salmonids are currently imperiled. The westslope cut- Warming trends have already been seen (Stewart et al. 2005), throat trout Oncorhynchus clarkii lewisi is one such species. and extinctions are predicted to increase (Thomas et al. 2004; Westslope cutthroat trout are native to the northwestern Wenger et al. 2011). How future changes will affect already United States and southwestern Canada. Historically, their range imperiled ecosystems and species is of great concern. was the largest of any interior cutthroat trout subspecies and in- Our objective was to investigate whether westslope cutthroat cluded the Missouri, Columbia, and Saskatchewan River basins trout were thermally adapted so as to better understand how (Behnke 1992). Populations are declining rangewide (Liknes natural selection is operating, to provide managers with addi- and Graham 1988; Shepard et al. 1997, 2005) and are estimated tional information for conservation, and to discuss the potential to inhabit only 59% of their historical range (May 2009). Habitat effects of global climate change. We explored both embryonic alterations, competition with nonnative fishes, overexploitation, and juvenile responses to temperature by investigating embry- and hybridization are all factors that have been implicated in onic survival and development rates and juvenile growth and their decline (Allendorf and Leary 1988; Liknes and Graham composition. 1988; Rieman and Apperson 1989; Van Eimeren 1996). Of If westslope cutthroat trout populations are locally adapted, these potential causes, genetic introgression (mainly from rain- then natural selection has preferentially selected traits that result bow trout O. mykiss, Yellowstone cutthroat trout O. c. bouvieri, in a relative increase in reproductive success (Taylor 1991). and golden trout O. m. aguabonita) is believed to have had the Therefore, we would expect that populations would perform most detrimental effect (Allendorf and Leary 1988; Liknes and best in environments most similar to those of their native habitat, Graham 1988). Only 43% of the total stream length currently or temperature in the case of thermal adaptation. occupied by westslope cutthroat trout is believed to be inhabited by genetically pure populations (May 2009). This equates to about 25% of the species’ historical range (May 2009). METHODS Many of the remaining genetically pure populations are small Five populations of westslope cutthroat trout, four wild and and isolated above either natural or constructed barriers. These one hatchery, were investigated. The wild populations occupied barriers prevent upstream fish migrations and make the transfer Chamberlain, Cottonwood, O’Brien, and Ray creeks, all in the of genetic material between isolated populations impossible. Missouri River drainage in Montana (Figure 1). Cottonwood Small, isolated populations such as these are at a greater risk of Creek is the warmest of the four, with an average summer (June inbreeding, which has had a detrimental effect in a number of 21–September 20) temperature of 11.2◦C in 2009, followed by taxa (Darwin 1892; Ralls et al. 1979; Gjerde et al. 1983; Berger O’Brien (8.2◦C), Chamberlain (7.4◦C), and Ray (6.7◦C) creeks 1990; Frankham 1995; Saccheri et al. 1998). To counter the ef- (Figure 2). fects of inbreeding, artificial immigration projects (i.e., translo- The hatchery population came from the Montana Department cations to reestablish and the genetic rescue of critically small of Fish, Wildlife and Parks’ Washoe Park Trout Hatchery at populations) have been considered by managers. Modeling Anaconda. Fish housed in this facility are of the MO12 strain, simulations suggest that artificial immigration projects can be which was developed in 1983 and 1984 from populations in successful (Hilderbrand 2002), and they have been shown to the Flathead and Clark Fork river drainages. The broodstock have positive effects on reproductive fitness and population was infused with additional wild fish from the Flathead River growth in various taxa (Westemeier et al. 1998; Frankham et al. drainage from 2003 to 2005. Embryos at the Washoe Park Trout

Downloaded by [Department Of Fisheries] at 20:25 25 September 2012 2006; Johnson et al. 2010). Before any translocation projects Hatchery are generally, but not always, incubated and reared at are initiated, it would be prudent to better understand the about 13.3◦C. role of local adaptation. The introduction of individuals from Potential differences in embryonic survival and development a population poorly adapted to a translocated habitat could due to thermal adaptation were investigated using offspring of result in further population depression through the introduction the five study populations. Broodstock from all wild populations of poorly adapted genotypes. To date, no studies of local were collected starting in June 2009 and continuing throughout adaptation have been performed on westslope cutthroat trout. the spawning period (until about the end of July) using a back- Because fish are poikilothermic, temperature affects nearly pack electrofisher. Captured westslope cutthroat trout were re- every aspect of their lives (Brett 1970). Survival, development, tained and housed in 110-L live-cars in their respective creeks. growth, migration, and reproduction are all influenced by tem- Female broodstock were checked every 2 d for signs of ripeness. perature (Embody 1934; Pankhurst et al. 1996; Swanberg 1997; A female was designated as ripe when gentle ventral pressure re- Bear et al. 2007). As such, natural selection probably operates leased eggs. Eggs from each ripe female were manually stripped on these traits, and temperature could be an environmental and separated into two lots. Each lot was fertilized with the milt factor to which populations are locally adapted. In addition, the of one male, and each pairing was considered one family (half- study of thermal adaptation allows for additional inquiry into sibs; Table 1). If a female had fewer than 50 eggs (estimated by 874 DRINAN ET AL.

TABLE 1. Numbers of females, males, and families contributing to data analyses.

Population Females Males Families MO12 8 9 9 Cottonwood Creek 16 23 24 O’Brien Creek 18 24 24 Chamberlain Creek 11 13 13 Ray Creek 6 6 7

observation), she was spawned with only one male. Eggs were exposed to milt for 60 s, after which excess milt was removed by rinsing the embryos. Embryos were allowed to water harden for 30 min before being placed in coolers and transported to the U.S. Fish and Wildlife Service’s Bozeman Fish Technol- ogy Center (BFTC) at Bozeman, Montana. Embryos from each family were divided and incubated at one of three target temperatures: 8◦C (mean, 7.99◦C; standard deviation, 0.45◦C), 10◦C (mean, 10.15◦C; 0.33◦C), or 14◦C (mean, 14.60◦C; 0.58◦C) at the BFTC. Two Heath tray batteries (8-Tray Vertical Incuba- tor, MariSource, Milton, Washington) were maintained at each temperature. Each battery contained six trays. Each tray was divided into 48 individual compartments (hereafter “wells”). Upon arrival at the BFTC, embryos were sterilized for 10 min in iodophore Betadine (Piper et al. 1982). Next, embryos were ◦ FIGURE 1. Sources of the wild populations of westslope cutthroat trout stud- acclimated to their treatment temperature at a rate of 1 C per ied. Cottonwood Creek had the warmest average daily summer temperature 15 min. After acclimation, embryos were deposited into their as- (11.2◦C), followed by O’Brien (8.2◦C), Chamberlain (7.4◦C), and Ray creeks signed incubator locations. Families were assigned to two trays ◦ (6.7 C). in each Heath battery to prevent the same family from being present on the same tray. Within a tray, families were randomly assigned to a well location. Embryos were examined every other day, and a record was kept of total embryos, mortalities, and hatching. Embryos that died or hatched were removed from the system. A formalin treatment of about 1,666 mg/L was applied for 15 min every other day to prevent the occurrence of fungus (Piper et al. 1982). Water temperatures were monitored daily by technicians and

Downloaded by [Department Of Fisheries] at 20:25 25 September 2012 hourly with a temperature logger (UA-002-64 and UA-002-08, Onset HOBO, Pocasset, Massachusetts). Differences in growth (weight, length, and condition) and body composition (percent dry protein and percent dry fat) due to thermal adaptation were investigated using the surviving individuals from the embryonic adaptation experiment. Juveniles were raised in water of the same temperature as incubation (8, 10, or 14◦C). At the BFTC, sixty-three 75-L aluminum tanks measuring 120 × 35 × 25 cm were used. Twenty-one tanks were assigned to each of three treatment temperatures: 8◦C (mean, 8.34◦C; standard deviation, 0.19), 10◦C (mean, 10.16◦C; 0.18), and 14◦C (mean, 14.24◦C; 0.15). Within each temperature group, three to six tanks were randomly assigned to each of the FIGURE 2. Temperature proﬁles for the wild populations of westslope cut- ﬁve populations. Embryos were incubated at about the same throat trout. temperature as the treatment groups (see temperatures from LOCAL ADAPTATION IN WESTSLOPE CUTTHROAT TROUT 875

embryonic adaptation experiment), and upon hatching were ran- perature. Using those data, the median survival of each pop- domly assigned to a tank for their population. When fry in a tank ulation at 8, 10, and 14◦C was estimated. We then estimated began actively seeking food, a 12-h automatic belt feeder was the proportional change that each population had between 8◦C used to provide BioVita Starter pellet (Bio-Oregon, Longview, and 10◦C and 10◦C and 14◦C. The proportional changes were Washington). Fish were fed to excess, defined as the presence then compared among populations. Changes from 8◦Cto14◦C of food remaining in the tank 24 h after feeding. If no food re- were not assessed, as they masked survival at 10◦C. This type of mained, rations were increased daily until excess was reached. analysis was necessary for two reasons. First, the underlying dis- Tanks were cleaned daily. Fish were taken off feed 89 d after the tributions were not normally distributed. Second, because only date of modal hatch for each tank. Individuals were euthanized four wild populations were included in this study, it is possible the following day using tricaine methanesulfonate (MS-222; that one (or more) populations simply performed poorly. The Argent Laboratories, Redmond, Washington). The lengths and hatchery population was not included in this analysis because weights of all fish were measured immediately after euthaniza- their incubation temperature is not always 13.3◦C but instead tion. Fish were stored at –80◦C until proximate analysis of their is changed depending on the hatchery needs and it was unclear carcasses. Proximate analysis was performed to determine the how this would influence natural selection. mean proportions of dry protein and dry fat among the fish in Individuals surviving the embryonic portion of the study were each tank. included in the growth study. All individuals were maintained at Standard proximate analysis protocols were used. All indi- the same temperature for growth as for embryogenesis. Growth viduals from a tank were homogenized together with a known metrics (weight, length, and Fulton condition factor [K]) were amount of deionized water. Samples were immediately placed analyzed using both nested ANOVAs in which each tank was into a freezer. After freezing, samples were freeze-dried using a random effect (package nlme) and each individual was the a Labconco FreeZone 12 (Labconco Corporation, Kansas City, experimental unit and ANOVAs using averages from each tank, Missouri). Samples remained in the freeze dryer until no weight where the tank was the experimental unit. Both analyses resulted change occurred over a 24-h period. Protein analysis was per- in the same conclusions, and ANOVAs using tank averages as the formed using a LECO TruSpec CN (LECO Corporation, St. experimental units are reported here because of their simplicity. Joseph, Michigan). Samples were combusted at 950◦C. The re- Across temperatures, a series of two-way ANOVAs was used sulting product was measured for traces of nitrogen, which were to assess possible population × treatment temperature interac- converted to protein content by multiplication by a protein factor tions after accounting for population and treatment temperature of 6.25 (Lawrence 2006). Protein was measured in duplicate for (Ramsey and Schafer 2002). each tank using samples of about 0.15 g. Fat content was ana- Both percent dry protein and percent dry fat were investigated lyzed using the AnkomXT10 Extractor (ANKOM Technology, only for fish raised at 8◦C and 10◦C. Fish raised at 14◦Cwere Macedon, New York). The extraction system was a modified used for an additional study and were unavailable for proximate soxhlet extractor that uses petroleum ether as the solvent. Du- analysis. Using homogenized fish, duplicate samples from each plicate samples of about 0.1 g from each tank were weighed and tank (the experimental unit) were analyzed for both dry protein placed into the extractor. After extraction, samples were dried and dry fat, and the average proportions of protein and fat were for 30 min at 135◦C. Next, they were cooled in a vented desic- calculated for each tank. The proportion of protein was analyzed cator for 20 min. Weights were measured again, and percent fat using one-way ANOVAs with population as the explanatory pa- was calculated by multiplying the estimated percent fat in the rameter at each treatment temperature. Across temperatures, a subsample by the percent dry matter. two-way ANOVA was used to assess possible population × All calculations were performed using R version 2.11.1 temperature interactions. Outliers existed in the proportion dry

Downloaded by [Department Of Fisheries] at 20:25 25 September 2012 (R Development Core Team 2010). The number of days to fat data. Therefore, a Kruskal–Wallis nonparametric ANOVA hatch was analyzed using a Kruskal–Wallis nonparametric anal- test was used at each treatment temperature. If significant dif- ysis of variance (ANOVA) test for each incubation temperature ferences were detected among treatments, pairwise compar- (Ramsey and Schafer 2002). A nonparametric test was necessary isons were made using a Mann–Whitney–Wilcoxon test with because the variance in the number of days to hatch decreased a Bonferroni adjustment for multiple comparisons (Ramsey and with an increase in incubation temperature. A family average Schafer 2002). for days to hatch, the experimental unit, was calculated for each family at each temperature. If significant differences were detected among populations, pairwise comparisons were made RESULTS using a Mann–Whitney–Wilcoxon test with a Bonferroni ad- A significant, positive correlation existed between source justment for multiple comparisons (Ramsey and Schafer 2002). stream temperature and embryonic survival at warm incuba- Embryonic survival was assessed by comparing the proportion temperatures (Figure 3). The proportional decline in me- tional changes in survival across incubation temperatures based dian embryonic survival from incubation temperatures of 10– on source population. That is, we calculated a pooled survival 14◦C was significantly greater for populations from colder for each family, the experimental unit, at each incubation tem- source streams (randomization test based on ranks: P = 0.04). 876 DRINAN ET AL.

TABLE 2. Median embryonic survival of each population at each incubation temperature. Cottonwood Creek had the warmest average daily summer temperature (11.2◦C), followed by O’Brien (8.2◦C), Chamberlain (7.4◦C), and Ray creeks (6.7◦C); MO12 was the hatchery population.

Population 8◦C10◦C14◦C MO12 0.62 0.47 0.41 Cottonwood Creek 0.69 0.61 0.55 O’Brien Creek 0.77 0.79 0.62 Chamberlain Creek 0.82 0.83 0.62 Ray Creek 0.86 0.79 0.53

For all populations, there was a reduction in the mean number of days to hatch with an increase in incubation temperature (Table 3). However, populations did not differ in number of ◦ ◦ ◦ days to hatch when incubated at 8 C (Kruskal–Wallis test: FIGURE 3. Change in median embryonic survival between 10 C and 14 C KW = 6.262, df = 4, P = 0.18) or 10◦C(KW = 3.038, df = 4, for each population. Populations from warmer streams had smaller proportional = ◦ changes. P 0.55). When incubated at 14 C, MO12 embryos hatched in significantly fewer days than those from Cottonwood Creek, the population with the warmest source stream temperature ◦ = < Observed changes in median embryonic survival between 8 C (Mann–Whitney–Wilcoxon test: W 31, P 0.01). Embryos and 10◦C were not in agreement with a predictable pattern of from Cottonwood Creek were estimated to take 0.8 more local adaptation (Figure 4). days to hatch than MO12 embryos with a bootstrapped 95% Median embryonic survival was higher for all populations at confidence interval (resampling with replacement) for the difference of 0.2–1.6. No other significant differences in days colder incubation temperatures (Table 2). The median survival ◦ of Ray, Cottonwood, and hatchery embryos was best at incu- to hatch were detected at 14 C. bation temperature 8◦C. The median survival of Chamberlain For all populations, mean weight and length increased with an and O’Brien Creek embryos was slightly higher at 10◦C than increase in temperature (Table 4). Fulton condition factor (K)did 8◦C. The embryonic survival of all populations was lowest at not differ among treatment temperatures (Table 4). Significant ◦ ◦ = = = incubation temperature 14 C. The survival of MO12 embryos differences in weight (ANOVA; 8 C: F 3.82, df 4, 12, P ◦ = = < ◦ was lower than that of all other populations at all treatment 0.03; 10 C: F 17.10, df 4, 15, P 0.01), length (10 C: = = < temperatures. F 8.61, df 4, 15, P 0.01), and K (data pooled across temperatures, ANOVA: F = 12.93, df = 4, 42, P < 0.01) among populations were detected (Table 4). However, these differences did not correspond to a pattern of local adaptation predictable by source stream temperature. No evidence of any population × rearing temperature interactions existed for weight (ANOVA: F = 1.7436, df = 4, 37, P = 0.16), length (ANOVA: F = 1.337, Downloaded by [Department Of Fisheries] at 20:25 25 September 2012 df = 4, 37, P = 0.27), or condition (ANOVA: F = 1.7436, df = 4, 37, P = 0.16). The proportions of dry protein (ANOVA: F = 2.35, df = 4, 32, P = 0.14) and fat (KW = 0.7442, df = 1, P = 0.38) were not significantly different among rearing temperatures (Table 4). Differences existed in proportions of dry protein (ANOVA: F = 6.98, df = 4, 29, P < 0.01) and fat (KW = 19.3648, df = 4, P < 0.01) among the populations after pooling across the temperatures for each population. However, these differences did not correlate with a predictable pattern of local adaptation.

DISCUSSION FIGURE 4. Change in median embryonic survival between 8◦C and 10◦Cfor A thermal cline over natal temperatures existed in embryonic each population. No detectable pattern was observed. survival at warm incubation temperatures. Survival responses LOCAL ADAPTATION IN WESTSLOPE CUTTHROAT TROUT 877

TABLE 3. Average days to hatch for each population at each incubation temperature, with 95% conﬁdence intervals in parentheses. Different lowercase letters indicate signiﬁcant differences between populations at a given temperature.

Population 8◦C10◦C14◦C MO12 38.19 (37.37, 38.90) 27.77 (27.08, 28.50) 17.58 (16.77, 18.17) y Cottonwood Creek 38.80 (38.44, 39.17) 28.35 (28.03, 28.70) 18.72 (18.44, 19.00) z O’Brien Creek 38.68 (38.27, 39.10) 28.08 (27.59, 28.46) 18.57 (18.15, 18.93) zy Chamberlain Creek 38.00 (37.54, 38.35) 28.14 (27.60, 28.54) 18.16 (17.89, 18.40) zy Ray Creek 38.50 (38.01, 38.96) 27.96 (27.56, 28.33) 19.15 (18.39, 19.96) zy

TABLE 4. Growth metrics and 95% confidence intervals for westslope cutthroat trout. Within temperature categories, different lowercase letters denote significant differences in means (weight, length, condition, and protein) or medians (fat). Confidence intervals (95%) of the medians were calculated by bootstrapping with replacement. Data were combined across temperatures for condition (K), protein, and fat because differences among temperature categories were not observed.

Metric MO12 Ray Creek Chamberlain Creek O’Brien Creek Cottonwood Creek Weight (g) 8◦C0.74zy0.60 zy 0.64 zy 0.58 y 0.79 z (–0.70, 2.18) (0.56, 0.65) (0.45, 0.83) (0.50, 0.65) (0.71, 0.86) 10◦C1.14zy1.03 y 0.86 y 0.87 y 1.45 z (0.92, 1.35) (0.86, 1.20) (0.63, 1.08) (0.78, 0.96) (1.36, 1.55) 14◦C1.85 1.90 1.74 1.45 2.11 (–3.66, 7.36) (1.78, 2.01) (–1.60, 5.09) (0.79, 2.10) (1.15, 3.06) Length (mm) 8◦C 45.47 42.63 43.82 42.11 45.51 (18.82, 72.12) (40.55, 44.70) (39.82, 47.83) (40.67, 43.54) (44.32, 46.69) 10◦C 52.24 zy 50.13 zy 47.33 y 48.33 y 54.64 z (50.49, 53.99) (47.15, 53.12) (43.49, 51.17) (46.52, 50.14) (53.45, 55.82) 14◦C 59.25 62.07 60.95 55.58 62.12 (–1.10, 119.60) (61.16, 62.98) (12.00, 109.90) (48.66, 62.51) (53.34, 70.91) Condition (K) 8◦C0.760.76 0.73 0.75 0.80 (0.53, 1.00) (0.72, 0.80) (0.72, 0.74) (0.73, 0.77) (0.78, 0.83) 10◦C0.76 0.79 0.77 0.74 0.85 (0.69, 0.83) (0.73, 0.85) (0.74, 0.81) (0.71, 0.76) (0.82, 0.87) 14◦C0.83 0.78 0.72 0.78 0.84 (0.38, 1.29) (0.77, 0.79) (0.13, 1.32) (0.67, 0.90) (0.81, 0.88) Combined 0.78 y 0.77 y 0.75 y 0.75 y 0.83 z (0.71, 0.86) (0.76, 0.79) (0.73, 0.76) (0.73, 0.77) (0.81, 0.85)

Downloaded by [Department Of Fisheries] at 20:25 25 September 2012 Protein (% dry) 8◦C 65.96 64.07 66.16 66.02 65.27 (61.91, 70.00) (62.74, 65.41) (64.81, 67.52) (64.89, 67.16) (65.00, 65.54) 10◦C 65.86 63.63 65.29 65.64 63.43 (63.30, 68.41) (61.84, 65.41) (61.69, 68.89) (64.88, 66.40) (62.35, 64.50) Combined 65.90 zy 63.85 y 65.73 zy 65.81 z 64.35 zy (64.56, 67.23) (63.20, 64.50) (64.59, 66.87) (65.28, 66.35) (63.25, 65.45) Fat (% dry) 8◦C 19.81 21.02 19.87 19.91 22.23 (7.92, 31.69) (19.16, 22.88) (18.72, 21.02) (19.60, 20.23) (21.22, 23.24) 10◦C 19.56 21.80 20.51 19.72 23.32 (13.20, 25.92) (19.46, 24.13) (19.39, 21.63) (18.41, 21.02) (22.67, 23.97) Combined 19.88 yx 21.36 zy 20.25 yx 19.80 x 22.81 z (16.86 21.95) (20.54, 22.32) (19.64, 20.68) (19.71, 20.34) (22.07, 23.45) 878 DRINAN ET AL.

were different depending on the population, with fish from of wild populations (Srivastava and Brown 1991). These factors the coolest creek (Ray Creek) having a proportional reduction probably contributed to the lower survival effect we observed. in median survival that was over three times that of fish from It was not surprising that juveniles of all populations in- the warmest creek (Cottonwood Creek). In addition, warmwa- creased in size with an increase in temperature between 8◦C ter populations survived more consistently (exhibited less and 14◦C. Hatchery westslope cutthroat trout grow optimally proportional change) at all incubation temperatures, whereas between 13◦C and 15◦C (Bear et al. 2007). This suggests that coldwater populations survived relatively better at colder both hatchery and wild westslope cutthroat trout grow best at incubation temperatures. The proportional change in median a similar temperature range, but without more extensive study survival of Cottonwood Creek embryos was nearly constant this cannot be confirmed. when changing from 8◦Cto10◦Cor10◦Cto14◦C (–11% and Although simulations verify that the rank correlation test –10%), whereas a considerable difference existed for Ray Creek accurately maintains its nominal size even with only four pop- embryos (–8% and –34%). Embryos from the intermediate ulations, the small number of wild populations studied does Chamberlain and O’Brien creeks had intermediate changes suggest caution in making generalizations. In addition, the use ( +1% and –26% and +2% and –21%). These results suggest of wild broodstock may have introduced unknown bias into the that populations have adapted as thermal generalists and special- observed results by including environmental effects. Our anal- ists. Most Rocky Mountain streams experience near-freezing ysis of embryonic survival attempted to reduce such bias, but it temperatures in winter, but their summer high temperatures are was not possible to eliminate it altogether. For example, favor- dependent on a wide range of factors (Poole and Berman 2001), able environmental conditions during oogenesis could result in and the temperatures of neighboring streams can be quite differ- gametes which perform well at a wide range of temperatures, ent. Thus, populations in streams with high peak temperatures while less favorable conditions could result in gametes which experience a wider range of temperatures annually than popula- perform differently. The use of second-generation individuals tions in cooler streams. Consequently, natural selection would would better remove these environmental effects, but domesti- favor coolwater specialists for colder streams and thermal gen- cation could be of concern. These issues highlight the difficulty eralists for warmwater streams. During the westslope cutthroat in determining genetic versus environmental contributions to trout incubation period in spring and summer, the temperatures observed phenotypes. in Cottonwood Creek reached a maximum of 19.9◦C, whereas Our findings have both short- and long-term management im- those in Ray Creek only reached 9.5◦C. It is of note that the plications. In the short term, managers are wrestling with how hatchery strain MO12 was created using broodstock from 20 to properly manage isolated populations of westslope cutthroat different populations and is generalist by design. MO12 had trout, including the translocation of fish for supplementation and a proportional change in median survival from 10◦Cto14◦C recolonization. Our study provides evidence that local adapta- of –13%, which is nearly identical to that of the embryos in tion should be considered before translocation projects are initi- Cottonwood Creek, the warmest stream in the study. ated. Embryonic survival at warm temperatures was greater for The existence of the thermal effect only during embryoge- populations from warm source streams, and the greatest chance nesis is not surprising. Individuals are most sensitive to their of translocation success would probably result from matching environment during embryonic development (McKim 1977), donor and recipient habitats, as populations could be locally and selection is believed to have a strong effect during this life adapted to a number of habitat features. That being said, addi- stage. Throughout embryogenesis, cells are forming the precur- tional work exploring other possible adaptations, as well as any sors of the adult body (Kunz 2004), and abnormalities during effects from outbreeding two distantly related populations, is gastrulation or neurulation could result in death or severe de- needed before any translocation projects are undertaken.

Downloaded by [Department Of Fisheries] at 20:25 25 September 2012 formity. Individuals with genotypes that are poorly adapted to In the long term, given the threat of climate change, the an incubation temperature would probably be removed, leav- persistence of westslope cutthroat trout must be considered. ing only individuals with adapted genotypes at later life stages. Global models predict an air temperature increase of 1.4–5.8◦C This could explain why no patterns of thermal adaptation were by 2100 (IPCC 2003), and our findings suggest that the colder present in postembryonic metrics. It also highlights the necessity the source stream, the greater the mortality at warm temper- for future work to include embryogenesis in study designs. atures. However, the amount of temperature increase will be An anomaly in our study was the poor embryonic survival key. If water temperature increases are small, temperatures for of the hatchery population. The survival of MO12 embryos was most populations will probably stay within an acceptable range. the poorest of all populations at all incubation temperatures. Most changes in embryonic survival would be minimal, similar Most hatchery broodstock used in this study were age 2, which to the change in embryonic survival from 8–10◦C observed in was their first spawning season. Gamete quality for the first this study. In addition, a slight increase in temperature would spawning season is often poor, resulting in reduced survival probably result in increased fish growth and could be benefi- (Bromage and Cumaranatunga 1988). In addition, the gamete cial, especially for populations now at high elevations (Cooney quality of hatchery populations is often poor compared with that et al. 2005; Bear et al. 2007). However, the changes in flow and LOCAL ADAPTATION IN WESTSLOPE CUTTHROAT TROUT 879

community assemblage accompanying even small temperature Johnson, W. E., D. P. Onorato, M. E. Roelke, E. D. Land, M. Cunningham, R. C. changes could be detrimental (Wenger et al. 2011). Belden, R. McBride, D. Jansen, M. Lotz, D. Shindle, J. Howard, D. E. Wildt, If temperature increases are large, survival could decrease as L. M. Penfold, J. A. Hostetler, M. K. Oli, and S. J. O’Brien. 2010. Genetic restoration of the Florida panther. Science 329:1641–1645. observed in this study. Long-term studies of the evolutionary Kunz, Y.W. 2004. Developmental biology of teleost fishes. Springer, Dordrecht, potential of these populations could help determine whether The Netherlands. such a decline would continue through future generations or Lawrence, J. 2006. Meat and meat products. Chapter 39 in W. Horwitz and they would rebound after the deleterious genes were purged. G. W. Latimer, editors. Official methods of analysis of AOAC International, 18th edition. AOAC International, Gaithersburg, Maryland. Liknes, G. A., and P. J. Graham. 1988. Westslope cutthroat trout in Montana: ACKNOWLEDGMENTS life history, status, and management. Pages 53–60 in R. E. Gresswell, edi- This project was funded by the National Science Founda- tor. Status and management of interior stocks of cutthroat trout. American Fisheries Society, Symposium 4, Bethesda, Maryland. tion under grant DEB 0717456. Thanks to Chris Clancy, Lee May, B. E. 2009. Westslope cutthroat trout status update summary. Wild Trout Nelson, and Dave Moser, Montana Fish, Wildlife and Parks. Enterprises, Bozeman, Montana. Thanks to Chris Hooley, Jason Ilgen, Aaron Nistler, and Matt McKim, J. M. 1977. Evaluation of tests with early life stages of fish for predict- Toner, U.S. Fish and Wildlife Service. Thanks to Tessa Andrews, ing long- term toxicity. Journal of the Fisheries Research Board of Canada Mathew Gjukis, Jennifer Ard, Clinton Smith, and Alex Hopkins, 34:1148–1154. Pankhurst, N. W., G. J. Purser, G. Van Der Kraak, P. M. Thomas, and G. N. R. Montana State University. The use of trade names or products Forteath. 1996. Effect of holding temperature on ovulation, egg fertility, does not constitute endorsement by the U.S. Government. plasma levels of reproductive hormones and in vitro ovarian steroidogenesis in the rainbow trout Oncorhynchus mykiss. Aquaculture 146:277–290. Piper, R. G., I. B. McElwain, L. E. Orme, J. P. McCraren, L. G. Fowler, and J. R. REFERENCES Leonard. 1982. Fish hatchery management. U.S. Fish and Wildlife Service, Allendorf, F. W., and R. F. Leary. 1988. Conservation and distribution of genetic Washington, D.C. variation in a polytypic species, the cutthroat trout. Conservation Biology Poole, G. C., and C. H. Berman. 2001. An ecological perspective on in-stream 2:170–184. temperature: natural heat dynamics and mechanisms of human-caused ther- Bear, E. A., T. E. McMahon, and A. V.Zale. 2007. Comparative thermal require- mal degradation. Environmental Management 27:787–802. ments of westslope cutthroat trout and rainbow trout: implications for species Quinn, T. P. 1993. A review of homing and straying of wild and hatchery- interactions and development of thermal protection standards. Transactions produced salmon. Fisheries Research 18:29–44. of the American Fisheries Society 136:1113–1121. Quinn, T. P., M. J. Unwin, and M. T. Kinnison. 2000. Evolution of temporal Behnke, R. J. 1992. Native trout of western North America. American Fisheries isolation in the wild: genetic divergence in timing of migration and breeding Society, Monograph 6, Bethesda, Maryland. by introduced Chinook salmon populations. Evolution 54:1372–1385. Berger, J. 1990. Persistence of different-sized populations: an empirical assess- R Development Core Team. 2010. R: a language and environment for statistical ment of rapid extinctions in bighorn sheep. Conservation Biology 4:91–98. computing. R Foundation for Statistical Computing, Vienna. Brett, J. R. 1970. Temperature, animals, fishes. Pages 515–573 in O. Kinne, Ralls, K., K. Brugger, and J. Ballou. 1979. Inbreeding and juvenile mortality in editor. Marine ecology: a comprehensive, integrated treatise on life in oceans small populations of ungulates. Science 206:1101–1103. and coastal waters, volume I, part I. Wiley-Interscience, London. Ramsey, F., and D. Schafer. 2002. The statistical sleuth: a course in methods of Bromage, N. R., and R. Cumaranatunga. 1988. Egg production in the rainbow data analysis. Duxbury/Thomson Learning, Pacific Grove, California. trout. Pages 63–138 in J. F. Muir and R. Roberts, editors. Recent advances in Rieman, B. E., and K. A. Apperson. 1989. Status and analysis of salmonid fish- aquaculture. Croom Helm, London. eries: westslope cutthroat trout synopsis and analysis of fishery information. Cooney, S. J., A. P. Covich, P. M. Lukacs, A. L. Harig, and K. D. Fausch. Idaho Department of Fish and Game, Project F-73-R-11, Boise. 2005. Modeling global warming scenarios in greenback cutthroat trout Saccheri, I., M. Kuussaari, M. Kankare, P. Vikman, W. Fortelius, and I. (Oncorhynchus clarki stomias) streams: implications for species recovery. Hanski. 1998. Inbreeding and extinction in a butterfly metapopulation. Nature Western North American Naturalist 65:371–381. (London) 392:491–494. Darwin, C. 1892. The effects of cross- and self-fertilization in the vegetable Shepard, B. B., B. E. May, and W. Urie. 2005. Status and conservation of kingdom. D Appleton and Company, New York. westslope cutthroat trout within the western United States. North American

Downloaded by [Department Of Fisheries] at 20:25 25 September 2012 Embody, G. C. 1934. Relation of temperature to the incubation periods of Journal of Fisheries Management 25:1426–1440. eggs of four species of trout. Transactions of the American Fisheries Society Shepard, B. B., B. Sanborn, L. Ulmer, and D. C. Lee. 1997. Status and risk 64:281–292. of extinction for westslope cutthroat trout in the upper Missouri River basin, Frankham, R. 1995. Inbreeding and extinction: a threshold effect. Conservation Montana. North American Journal of Fisheries Management 17:1158–1172. Biology 9:792–799. Srivastava, R. K., and J. A. Brown. 1991. The biochemical characteristics and Frankham, R., J. D. Ballou, and D. A. Briscoe. 2006. Introduction to conserva- hatching performance of cultured and wild Atlantic salmon (Salmo salar) tion genetics. Cambridge University Press, Cambridge, UK. eggs. Canadian Journal of Zoology 69:2436–2441. Gjerde, B., K. Gunnes, and T. Gjedrem. 1983. Effect of inbreeding on survival Stewart, I. T., D. R. Cayan, and M. D. Dettinger. 2005. Changes toward ear- and growth in rainbow trout. Aquaculture 34:327–332. lier streamflow timing across western North America. Journal of Climate Hilderbrand, R. H. 2002. Simulating supplementation strategies for restoring 18:1136–1155. and maintaining stream resident cutthroat trout populations. North American Swanberg, T. R. 1997. Movements of and habitat use by fluvial bull trout in the Journal of Fisheries Management 22:879–887. Blackfoot River, Montana. Transactions of the American Fisheries Society IPCC (Intergovernmental Panel on Climate Change). 2003. Climate change 126:735–746. 2001: impacts, adaptation, and vulnerability—IPCC third assessment re- Taylor, E. B. 1991. A review of local adaptation in Salmonidae, with particular port. Cambridge University Press, Cambridge, UK. Available: www.grida. reference to Pacific and Atlantic salmon. Aquaculture 98:185–207. no/publications/other/ipcc tar/?src=/climate/ipcc tar/wg2/index.htm. (July Thomas, C. D., A. Cameron, R. E. Green, M. Bakkenes, L. J. Beaumont, 2011). Y. C. Collingham, B. F. N. Erasmus, M. F. de Siqueira, A. Grainger, 880 DRINAN ET AL.

L. Hannah, L. Hughes, B. Huntley, A. S. van Jaarsveld, G. F. Midgley, Wenger, S. J., D. J. Isaak, C. H. Luce, H. M. Neville, K. D. Fausch, J. B. Dunham, L. Miles, M. A. Ortega-Huerta, A. T. Peterson, O. L. Phillips, and S. E. D. C. Dauwalter, M. K. Young, M. M. Elsner, B. E. Rieman, A. F. Hamlet, Williams. 2004. Extinction risk from climate change. Nature (London) 427: and J. E. Williams. 2011. Flow regime, temperature, and biotic interactions 145–148. drive differential declines of trout species under climate change. Proceedings Van Eimeren, P. 1996. Westslope cutthroat trout Oncorhynchus clarki lewsi. of the National Academy of Sciences of the USA 108:14175–14180. DOI: Pages 1–10 in D. A. Duff, editor. Conservation assessment for inland cut- 10.1073/pnas.1103097108. throat trout: distribution, status and habitat management implications. U.S. Westemeier, R. L., J. D. Brawn, S. A. Simpson, T. L. Esker, R. W. Jansen, J. W. Department of Agriculture, Forest Service, Intermountain Region, Ogden, Walk, E. L. Kershner, J. L. Bouzat, and K. N. Paige. 1998. Tracking the long- Utah. term decline and recovery of an isolated population. Science 282:1695–1698. Downloaded by [Department Of Fisheries] at 20:25 25 September 2012 This article was downloaded by: [Department Of Fisheries] On: 25 September 2012, At: 20:26 Publisher: Taylor & Francis Informa Ltd Registered in England and Wales Registered Number: 1072954 Registered office: Mortimer House, 37-41 Mortimer Street, London W1T 3JH, UK

Transactions of the American Fisheries Society Publication details, including instructions for authors and subscription information: http://www.tandfonline.com/loi/utaf20 Passage of Spawning Alabama Shad at Jim Woodruff Lock and Dam, Apalachicola River, Florida Shawn P. Young a , Travis R. Ingram b , Josh E. Tannehill b & J. Jeffery Isely c a Department of Forestry and Natural Resources, Clemson University, 115 Lehotsky Hall, Clemson, South Carolina, 29634-0317, USA b Georgia Department of Natural Resources, 2024 Newton Road, Albany, Georgia, 31701, USA c National Oceanic and Atmospheric Administration Fisheries Service, 75 Virginia Beach Drive, Miami, Florida, 33149, USA

Version of record first published: 13 Jun 2012.

To cite this article: Shawn P. Young, Travis R. Ingram, Josh E. Tannehill & J. Jeffery Isely (2012): Passage of Spawning Alabama Shad at Jim Woodruff Lock and Dam, Apalachicola River, Florida, Transactions of the American Fisheries Society, 141:4, 881-889 To link to this article: http://dx.doi.org/10.1080/00028487.2012.675917

PLEASE SCROLL DOWN FOR ARTICLE

Full terms and conditions of use: http://www.tandfonline.com/page/terms-and-conditions 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. The publisher does not give any warranty express or implied or make any representation that the contents will be complete or accurate or up to date. The accuracy of any instructions, formulae, and drug doses should be independently verified with primary sources. The publisher shall not be liable for any loss, actions, claims, proceedings, demand, or costs or damages whatsoever or howsoever caused arising directly or indirectly in connection with or arising out of the use of this material. Transactions of the American Fisheries Society 141:881–889, 2012 C American Fisheries Society 2012 ISSN: 0002-8487 print / 1548-8659 online DOI: 10.1080/00028487.2012.675917

ARTICLE

Passage of Spawning Alabama Shad at Jim Woodruff Lock and Dam, Apalachicola River, Florida

Shawn P. Young Department of Forestry and Natural Resources, Clemson University, 115 Lehotsky Hall, Clemson, South Carolina 29634-0317, USA Travis R. Ingram* and Josh E. Tannehill Georgia Department of Natural Resources, 2024 Newton Road, Albany, Georgia 31701, USA J. Jeffery Isely National Oceanic and Atmospheric Administration Fisheries Service, 75 Virginia Beach Drive, Miami, Florida 33149, USA

Abstract In 2005, a pilot study was initiated to evaluate the potential use of the navigation lock at Jim Woodruff Lock and Dam (JWLD), Florida, as a means for ﬁsh passage. The focal species was the Alabama shad Alosa alabamae. The Apalachicola River population is one of the last-remaining, extant self-sustaining populations, estimated adult spawning returns ranging from lows of 5,211 to 14,674 individuals in 2007 to highs of 51,417 to 127,251 individuals in 2010. We estimated the passage of migrating Alabama shad during spawning migrations during March–May 2005, 2007, 2010, and 2011. During 2005, 23 of 36 Alabama shad that were implanted with transmitters successfully passed from the lock into the reservoir 1–7 d after release, for a passage rate of 64% (95% CI = 48–80%). Advancing on the promising results from 2005, voluntary passage was evaluated during spring 2007, 2010, and 2011. During these years, 63–100% of implanted shad were relocated at the lock at least once, and voluntary passage ranged from 33% to 45%. Voluntary passage occurred 3–39 d postimplanting, most shad passing < 28 d after initial capture. Implanted Alabama shad were subsequently relocated upstream in both the Flint and Chattahoochee rivers. Based on these results, the navigation lock at JWLD was an effective means to pass migrating Alabama shad. Increased passage could be achieved by maximizing attraction ﬂow near the lock entrance and increasing the time the upper gates are open during an afternoon locking cycle. By coupling passage and population estimates, the total number of shad that migrated through JWLD ranged from 2,137 in 2007 to 57,262 in 2010. Downloaded by [Department Of Fisheries] at 20:26 25 September 2012

The Alabama shad Alosa alabamae is an anadromous clu- Mettee and O’Neil 2003). There is evidence that Alabama shad peid that lives in the northern Gulf of Mexico and migrates may have multiple spawning periods during a spawning mi- into freshwater rivers to spawn (Mettee and O’Neil 2003; Ely gration due to the ﬂuctuating abundances of mature versus im- et al. 2008). Spawning of Alabama shad in the Apalachicola mature ova in female shad (Mills 1972; Mettee and O’Neil and Choctawhatchee rivers occurs when temperatures reach 2003). Although little is known about the spawning habits and 19–23◦C (Laurence and Yerger 1967; Mills 1972; Mettee and requirements of Alabama shad, they are likely similar to the O’Neil 2003). During the spawning migration, Alabama shad requirements of the American shad A. sapidissima. inhabit large rivers and generally reproduce in areas of moder- The historical spawning range of Alabama shad included ate current composed of sand and gravel substrates (Mills 1972; the Mississippi River and drainages eastward to the Suwannee

*Corresponding author: [email protected] Received October 5, 2011; accepted January 25, 2012 Published online June 13, 2012 881 882 YOUNG ET AL.

River in Florida (Mettee and O’Neil 2003; Ely et al. 2008). Al- 2000; Mettee and O’Neil 2003). Concurrent research along with though Alabama shad populations were once abundant enough the study described herein has estimated annual spawning adult to possibly support commercial fisheries (Coker 1929; Mills abundance of the Apalachicola River population since 2005. 1972), research now suggests that they are small and declining Spawning population estimates from 2005 to 2007 were pre- throughout their range (Laurence and Yerger 1967; Mills 1972; viously reported by Ely et al. (2008), and the same population Rulifson et al. 1982; Mettee and O’Neil 2003). This decline estimation method has been employed from 2005 to 2011. Ely in abundance has been primarily attributed to the construction et al. (2008) used mark–recapture tagging methods at JWLD of navigational locks and dams that block upstream migration from 2005 to 2007 to estimate spawner abundance that ranged to historical spawning grounds (Coker 1929; Hildebrand 1963; from a low of 2,767 shad in 2006 to a high of 25,935 individuals Barkuloo et al. 1993). Dams have also contributed to the de- in 2005. The objective of this study was to evaluate and improve cline of a number of other Gulf and Atlantic coast anadromous the effectiveness of the navigation lock at JWLD for upstream species such as American shad and other river herring Alosa passage of spawning Alabama shad from the Apalachicola River spp., striped bass Morone saxatilis, and sturgeon Acipenser spp. into the Flint and Chattahoochee rivers. (Richkus and DiNardo 1984; Zehfuss et al. 1999; Beasley and Hightower 2000; Schmidt et al. 2003). METHODS Fishways, fish lifts, and navigation locks may assist in up- Study area.—The JWLD is a run-of-the-river facility located stream passage of migrating species (Clay 1995; Moser et al. in the central panhandle of northwest Florida along the Georgia 2000). Intended use of many existing navigation locks has de- border and impounds the waters of the Chattahoochee and Flint clined in recent decades due to other means of transportation. rivers, forming Lake Seminole (Figure 1). The Apalachicola Navigation locks represent a cost-effective alternative for fish River originates below JWLD and flows without obstruction for passage at these facilities. However, the potential for fish pas- 171 km to Apalachicola Bay, an inlet of the Gulf of Mexico. sage at many navigation locks has not been assessed (Nichols and Louder 1970; Clay 1995). Further, the voluntary use of navigation locks by a multitude of fish species is still not known. The American shad, sister species of the Alabama shad (Berry 1964), and other migratory clupeids are known to utilize locks, although passage success has been variable (Chappelear and Cooke 1994; Moser et al. 2000; Petersen et al. 2003; Bailey et al. 2004). Overall, increased fish passage has proven to be beneficial in increasing abundances of alosine stocks (Cooke and Leach 2003). Passage may be related to retention time in the vicinity of the passage structure, discharge, current velocity, and the seasonal and diel timing of the locking schedule (Barry and Kynard 1986; Moser et al. 2000; Bailey et al. 2004). In 1997, the National Marine Fisheries Service identified Alabama shad as a candidate for listing under the Endangered Species Act. In 2004, the status of Alabama shad was reclas- sified as a species of concern by the National Oceanic and Atmospheric Administration (NOAA; 2004) In 2010, Alabama

Downloaded by [Department Of Fisheries] at 20:26 25 September 2012 shad were again petitioned for listing under the Endangered Species Act, but no change in federal status resulted from this petition. Ongoing research at Jim Woodruff Lock and Dam (JWLD) was cited by NOAA as one of the reasons to not list this species (NOAA 2011). Although few studies have been conducted concerning species life history, distribution, and historical abundance, research suggests populations of Alabama shad in Gulf Coast rivers are relatively small and declining (Laurence and Yerger 1967; Mills 1972; Rulifson et al. 1982; Mettee and O’Neil 2003). Self-sustaining populations still inhabit the Choctawhatchee and Apalachicola River systems in Alabama and Florida (Barkuloo et al. 1993; Mettee and O’Neil 2003). Currently, the Apalachicola River below JWLD in northwest Florida supports the largest-known extant spawning pop- FIGURE 1. Location of study area on the Apalachicola River, Florida, below ulation of Alabama shad (Laurence and Yerger 1967; McBride JWLD. PASSAGE OF SPAWNING ALABAMA SHAD 883

Completed in 1957 to aid navigation and generate hydroelectric power, JWLD is operated by the U.S. Army Corps of Engineers (USACE), and consists of a powerhouse, 16 independent, 12.2- m-wide spill gates, a connected navigation lock on the west riverbank, and a fishing pier on the east riverbank (Figure 2). The lock measures 25 m wide by 137.2 m long, and has a lift capacity of 10.1 m. The navigation lock at JWLD was operated twice a day from March to May during the study years. In 2004, the USACE agreed to operate the lock 2 times/d during spring for the expressed purpose of fish passage. We sought to evaluate the efficiency of the lock for fish passage and initially emphasized Alabama shad. Two fish locking cycles (lockages) were performed each day (between 0800 and 1600 hours) during the study period. At the beginning of a cycle, water was released through valves from the lock chamber over the course of about 15 min (maximum flow = 3,930 cfs). Once the lock elevation reached river level, only one lower gate was opened in order to help retain circulating fish in the lock chamber (Moser et al. 2000). A gas-powered water pump with a 3-in line was used to create an attraction flow that was discharged from the top of the west lock wall into the river just below the lower west gate in order to draw fish into the lock. After 1 h, the lower gate was closed. In order to attract fish from the chamber to the upper gates, the water pump discharge was diverted to another outlet just below the upper west gate. With the upper discharge flowing, the lock was then raised to lake level over the course of 20 min, and the upper gates opened. After 45 min, the upper gates closed and the lock was drained to the downstream elevation. Field techniques: pilot study.—During March–April 2005, Alabama shad (n = 45; 385.1 ± 29.8 mm, mean length ± SD; sex not determined) were implanted with acoustic transmitters (Hydroacoustic Technology, Seattle, Washington; Table 1). The transmitters measured 6.8 mm in diameter by 21 mm in length, weighed 0.8 g in water, and possessed a minimum battery life of 25 d and a unique frequency. Transmitters were orally inserted into the stomach. To facilitate gastric implantation, transmitters were affixed to a hollow flexible tube and were coated in

Downloaded by [Department Of Fisheries] at 20:26 25 September 2012 a nontoxic, water-soluble lubricant to alleviate trauma during insertion. In an effort to decrease fallback, mortality, or both following release, transmitters represented less than 1% of the body weight of the smallest study fish and were only inserted into shad that appeared to be in excellent condition. Fish were also marked with one T-bar internal anchor tag (Floy Tag, Seat- tle, Washington) imprinted with a unique identification number. All T-bar tags were inserted into the musculature just below the dorsal fin perpendicular to the pterygiophores. Total handling time of shad required less than 2 min each. Shad collection was conducted between 0800 and 1700 hours using boat electrofishing and hook-and-line fishing gear. All FIGURE 2. A schematic depicting main structures of JWLD and Apalachicola Alabama shad were captured in the Apalachicola River < 2km River immediately downstream (2.5 km). Black stars denote Alabama shad downstream of JWLD (Figure 2). Shad collected by hook and capture and release areas; solid arrow denotes turbine discharge; dashed arrows denote spill gate and navigation lock discharge. line were captured along the fishing wall located below the 884 YOUNG ET AL.

TABLE 1. Summary of data for Alabama shad passage at JWLD from 2005 to 2011. Study year, number of transmitter-implanted fish, telemetry manufacturer used, number of fish deemed viable after transmitter implantation, number and overall percent of fish attracted to the navigation lock, estimated spawning population at JWLD (see Ely et al. 2008 for methods), and number and overall percent of fish that passed JWLD are shown. An asterisk indicates bias due to fish being transported to the lock, unlike subsequent years when fish voluntarily passed via the lock.

Number of transmitter- Number implanted of viable Lock Spawning Total number Year ﬁsh Telemetry ﬁsh attraction Passage population passed 2005 45 HTI 36 * 23 (64%)* 17,715–39,535 * 2007 59 HTI 41 26 (63%) 17 (41%) 5,211–14,674 2,137–6,016 2010 48 Vemco 31 29 (94%) 14 (45%) 51,417–127,251 23,137–57,262 2011 50 ATS 49 49 (100%) 16 (33%) 22,371–43,713 7,382–14,425

powerhouse on the east bank. In 2005, Alabama shad were the same T-bar internal anchor tag as in previous years, and captured and placed in a live well, and transported inside the implanted with a hydroacoustic transmitter (Vemco, Halifax, navigational lock. Once inside, the lower lock gates were closed. Nova Scotia; model V7-2 L). The model V7-2 L transmitters Following transmitter insertion and marking, shad were released measured 7.0 mm in diameter and 20 mm in length, weighed into the lock. The lock was then raised to lake level (20 min) and 0.75 g in water, and possessed a minimum battery life of 95 d. the upper gates were opened. After 30 min, the upper gates were Each transmitter was uniquely coded, programmed to transmit closed and the lock was drained to the downstream elevation. every 40–80 s, and operated at a frequency of 69.0 kHz. All Alabama shad were detected using a fixed-station sonic transmitters were gastrically implanted using the same methods telemetry system (HTI 290; Hydroacoustic Technology). Im- as in 2005 and 2007. The change in telemetry technology was planted shad were continuously monitored over a 55-d period to improve data acquisition via mobile telemetry. following release. Hydrophones were placed above, within, and Sampling to capture Alabama shad for implanting transmit- below the lock. Pulse frequency, time, and date were recorded ters was conducted between 0700 and 2100 hours by hook and when a transmitter was detected at a hydrophone. Direction of line only. All Alabama shad were captured by angling with a movement, emigration from the system, time spent in the lock, sabiki rig, a series of small hooks with sparkle flash or small transmitter failure, and mortality were then determined. lures typically used to catch small prey or plankton-eating fish, Field techniques: 2007.—During March 14–28, adult Al- or a single weighted jig. All shad collected by hook and line were abama shad (n = 59; 371.59 ± 35.17 mm, mean length ± SD; captured along the fishing wall located below the powerhouse 33 females and 26 males) were implanted and tagged using the on the east bank of the tailrace. Alabama shad that received a same technology and methods as in 2005, and were then released transmitter were captured and processed one fish at a time to re- at the approximate location of their initial capture below JWLD duce holding and handling time. Implanted shad were released in order to assess voluntary passage (Figure 2). Implanted shad at the capture site. were monitored over a 60-d period following release and were In 2010, study fish were detected using five data-logging again detected using a fixed-station acoustic telemetry system hydrophones (model VR2W; Vemco) anchored to the naviga- affixed to the navigation lock walls. Hydrophones were placed tion lock below the lower gates, within the lock chamber, and

Downloaded by [Department Of Fisheries] at 20:26 25 September 2012 below lower gates, within the lock chamber, and above the up- above the upper gates to evaluate passage. Also, three additional per gates (Figure 2). In 2007, implanted shad were also detected data-logging hydrophones (model VR2W; Vemco) were used to using three tactically placed remote sensing data loggers. Two detect movement above and below JWLD at the same locations data loggers were placed above JWLD in each of the tributaries as in 2007. Additionally, mobile telemetry by boat was also at the next established barrier to migration. One data logger was conducted during daylight hours above and below JWLD using placed in the Flint River < 2 km below Georgia Power Dam a Vemco VR-100 general-purpose receiver. in Albany, Georgia. A second data logger was placed in the Field techniques: 2011.—On March 31, Alabama shad (n = Chattahoochee River < 2 km below George W. Andrews Lock 50; 360.3 ± 40.3 mm, mean length ± SD; 26 females and and Dam near Columbia, Alabama. The third data logger was 24 males) were captured by electroﬁshing between 0700 and placed < 2 km below JWLD in the Apalachicola River in or- 1600 hours, marked with one T-bar internal anchor tag, and then der to monitor fallback, mortality, and movement of implanted implanted with radio transmitters (Advanced Telemetry Sys- shad. tems, Isanti, Minnesota; model F1030), another change in Field techniques: 2010.—During April 13–15, Alabama shad telemetry technology that was necessary to further improve relo- (n = 48; 327.3 ± 48.7 mm, mean length ± SD; 37 males cation and data acquisition via mobile telemetry in riverine habi- and 11 females) were captured by hook and line, tagged using tats. All implanted ﬁsh were again caught in the Apalachicola PASSAGE OF SPAWNING ALABAMA SHAD 885

River < 2 km from JWLD (Figure 2). The model F1030 trans- may have continued migration into the Flint and Chattahoochee mitter has an internal-coiled antenna. Each transmitter measured rivers. 10.0 mm in diameter and 21 mm in length, weighed 1.9 g in water, and possessed a minimum battery life of 47 d. Each transmit- RESULTS ter had a unique frequency ranging from 148 to 149.999 mHz, and was programmed to transmit every 2 s. All transmitters were Pilot Study implanted using the same methods as previous years. Four of the 45 implanted Alabama shad died within 2 d of In 2011, study fish were detected using two data-logging implantation as a result of tagging, handling, or both, and were receivers (model R4500S; Advanced Telemetry Systems). One removed from the data set. Another five fish were removed receiver was placed at each of the two lock gates to evaluate from the data set due to transmitter malfunction. Of 36 viable passage. Mobile telemetry by boat was also conducted during Alabama shad that were released in the lock, 23 successfully daylight hours above and below JWLD using an ATS R2000 passed from the lock into the reservoir, for a passage estimate of general-purpose receiver with a four-element, Yagi-style an- 64% (95% CI: 48–80%; Table 1). Of the 23 shad that passed, six tenna, another change in telemetry technology that was again passed during initial release, and the other 17 fish passed 1–7 d needed to further improve data acquisition via mobile telemetry after release (Table 2). Four shad emigrated out of the study to assess tributary spawning habitat selection and behavior of area and never returned, and two fish emigrated out of the study those shad spawning below JWLD. area but returned approximately 20 d later. One of the fish that Field techniques: all years.—Shad that died, permanently exited downstream was recaptured in good condition < 2km emigrated, or experienced transmitter failure were considered below JWLD 33 d after transmitter implantation. Three fish that nonviable and excluded from passage analysis. Shad that tem- initially passed were later detected below the dam during a pe- porarily emigrated were designated as “fallbacks.” Passage suc- riod of high water, indicating that they emigrated downriver via cess was determined from the viable proportion of shad that spill gates. One was detected below JWLD the day after pas- were released into the river below JWLD and then successfully sage, and the other two were detected within 3–4 d. No fish that passed into Lake Seminole reservoir using the lock. The effec- passed upstream reentered the lock from the reservoir and exited tiveness of attracting shad to the lock was calculated from the downstream. Because study fish were released inside the lock number of viable fish that entered the lock chamber from the during this pilot evaluation, unbiased attraction effectiveness river at least once regardless of whether the fish passed on that and voluntary passage success could not be determined. event. The concurrent mark–recapture study estimated annual 2007 spawning adult abundance of the Apalachicola River popula- During 2007, Alabama shad were successfully implanted tion during 2005–2011. During this study, shad were sampled and released at the capture site to evaluate voluntary passage weekly each year from March to May. Shad were marked with (Table 1). Eighteen of 59 (31%) Alabama shad were never relo- dorsal T-bar internal anchor tags to identify recaptured fish. cated in the study area following release into the river and were A goal of tagging 1,000 individuals annually was attempted. removed from the data set. These fish were considered nonvi- Mark–recapture estimates were accomplished using Schnabel’s able, indicating they suffered mortality, experienced transmitter multiple-census method for a randomly mixed population as failure, or abandoned their spawning migration. All of the re- modified by Chapman for small number of recaptures (Ricker maining shad were located at least once in the vicinity of JWLD. 1975). A detailed explanation is found in Ely et al. (2008). Two study fish were recaptured in good condition < 2 km below Spawning population estimates for 2010 and 2011 followed the dam approximately 32 d after transmitter implantation.

Downloaded by [Department Of Fisheries] at 20:26 25 September 2012 the same mark–recapture and tagging methods as Ely et al. Of 41 viable Alabama shad, 26 entered the lock from the (2008). For the purpose of this manuscript, the spawning pop- river, for an attraction effectiveness of 63% (95% CI: 49–78%), ulation estimates were coupled with the passage success rates and 17 subsequently passed into the reservoir, for a passage in order to estimate the number of adult Alabama shad that success of 41% (95% CI: 26–57%; Table 1). Alabama shad

TABLE 2. Summary of the number of days until implanted Alabama shad passed JWLD via the navigation lock. Study year, date transmitter implanted, and days until passage event are shown. Asterisk signifies bias due to fish being transported to the lock, unlike subsequent years when fish voluntarily passed via the lock.

Year Date ≤7 d 8–14 d 15–21 d 22–28 d 29–35 d >35 d 2005 Mar 15–Apr 30 23* 0 0 0 0 0 2007 Mar 15–28 2 6 7 1 0 1 2010 Apr 13–15 6 4 2 2 0 0 2011 Mar 31 3 2 0 10 1 0 886 YOUNG ET AL.

voluntarily passed using the navigation lock 4–36 d after capture a coordinated behavior pattern exhibited by all of the adult Al- and telemetering (Table 2). Fifteen of 17 shad passed within the abama shad in the vicinity of JWLD. Instead, the spawning same day of entering the lock. Fifteen of 26 shad entered the lock population appears to comprise multiple groups of shad that at chamber only once, while 11 entered an average of 2.5 times times behaved singularly and at other times merged with other (range = 2–4). All Alabama shad passed during the afternoon groups, some individuals moving between groups. Also, during locking cycle. Of 17 shad that passed, 15 entered and passed hook-and-line sampling, sexes were frequently separated across in the afternoon, and the remaining two shad entered in the time in daily catch records. This indicated shad grouped by sex morning and passed in the afternoon. or, possibly, each group has a spatial schooling structure based Combining passage rate with spawning population estimate, on sex and body size. as many as 6,016 shad migrated past JWLD via the navigation lock (Table 1). Alabama shad were relocated at the next 2011 upstream impediments to spawning migrations. One shad that After release, 98% (49 of 50) of implanted Alabama shad passed was detected at the Georgia Power Dam in the Flint were relocated in the area near JWLD on at least one occasion. River 9 d following passage, and one was detected at George W. After accounting for negligible fallback (one fish censored), Andrews Lock and Dam in the Chattahoochee River 8 d after 33% (16 of 49; 95% CI: 20–46%) of implanted Alabama shad passage. passed through the navigation lock. Of the 16 shad that passed upstream, 11 were females and five were males. Voluntary pas- 2010 sage occurred 5–29 d after capture and telemetering (Table 2). Alabama shad were captured, implanted, and successfully Combining passage rate with a spawning population esti- released in the tailrace of JWLD to evaluate voluntary passage mate, as many as 14,425 shad may have migrated past JWLD (Table 1). After release, 35% (17 of 48) of implanted Alabama via the navigation lock (Table 1). Of the 16 fish that passed shad vacated the area near JWLD and were not relocated. Of the volitionally, four were later relocated in the Flint River and one 17 fish never relocated, 14 were males (the majority measuring fish was relocated in the Chattahoochee River. < 300 mm) and three were females. Thus, 38% of the males and In 2011, Alabama shad again exhibited the behavior of drop- 27% of females were classified as fallbacks and were removed ping downriver of JWLD for variable amounts of time. Some from the data set. fish fell back for up to 39 d before returning to JWLD, while After accounting for fallback, 45% (14 of 31; 95% CI: 36– some individuals made almost daily excursions. Also, high site 54%) of viable, implanted Alabama shad passed JWLD via the fidelity for holding positions below the main study area was navigation lock (Table 1). Of the 14 shad that passed upstream, observed during 2011 to the extent that fish were designated as 12 were males (the majority measuring > 300 mm) and two mortalities at one point, only to be relocated upriver thereafter. were females. Of the 31 fish that were relocated in the vicinity of This behavior, consisting of migrating up to JWLD and then JWLD, 29 fish (94%) were recorded by a stationary hydrophone dropping downriver again, did not appear to be a coordinated below the lower lock gates at least one time, 23 fish (74%) behavior pattern exhibited by all adult Alabama shad in the entered the inner lock chamber at least one time, and 14 fish vicinity of JWLD at the same time but rather by smaller groups (45%) exited the upper gates into Lake Seminole. Voluntary pas- consisting of variable numbers over time. Multiple implanted sage occurred 3–28 d after capture and telemetering (Table 2). shad, up to 14 individuals, but not always the same individuals, Alabama shad again exhibited a strong preference to pass up- were found together in holding positions below JWLD. In late stream during the afternoon locking (13 of 14, or 93%). How- April, as many as 31 Alabama shad were detected by data loggers at the lock on the same day. Given the water temperature ever, implanted shad did enter the lock during both morning and ◦

Downloaded by [Department Of Fisheries] at 20:26 25 September 2012 afternoon locking. was 22.5 C, this may have been a precursor to a spawning event. Combining passage rate with a spawning population esti- All Years mate, as many as 57,262 shad may have migrated past JWLD The number of shad per hour of sampling over the project via the navigation lock (Table 1). Of the 14 fish that passed vo- ranged from a low of 7.17 to a high of 72.9 (Table 3). The large litionally, five (36%) were later relocated in the Flint River and range of catch rates can be attributed to a host of factors, in- no fish were relocated in the Chattahoochee River (Figure 2). cluding environmental conditions, crew experience, and pulses No fish were detected by data loggers at the next upstream im- of shad migrating into the river system. Total number of tagged pediments. Instead, shad were located between Bainbridge and fish ranged from a low of 599 in 2007 to a high of 1,384 in 2005 Albany, Georgia, in the Flint River. (Table 3). The total annual number of shad tagged represented Most implanted shad exhibited the behavior of dropping 1.5–6.3% of the estimated population. downriver of JWLD for variable amounts of time and then returning to JWLD on multiple occasions. Upon examining the telemetry data, there are similarities in behaviors exhibited by DISCUSSION individuals. However, this behavior, consisting of migrating up The effectiveness of the navigation lock at attracting Alabama to JWLD and then dropping downriver, does not appear to be shad (63–100%) is consistent with studies involving American PASSAGE OF SPAWNING ALABAMA SHAD 887

TABLE 3. Summary of catch rates and total number of shad tagged annually below JWLD.

Number of shad Total number of Year per hour shad tagged 2005 20.47 1,384 2007 13.17 599 2010 7.17 1,116 2011 72.92 931

shad (Moser et al. 2000; Normandeau Associates 2003). Moser et al. (2000) reported a percentage of 76% (range = 54–83%) for American shad returns that entered a lock from 1996 to 1998. In 2002, American shad in the Cooper River, South Carolina, yielded an effectiveness of 66% for study fish that entered Pinopolis Lock (Normandeau Associates 2003). Attrac- tion of Alabama shad into the lock appeared to be correlated with low to zero discharge from the spill gates (Figure 3). Annual passage success ranged from 33% to 45% of Al- abama shad at JWLD, and is comparable to passage estimates for American shad in southeastern United States rivers using navigation locks. Studies for American shad have shown a wide fluctuation between years. In 1996–1997, passage for American shad in the Cape Fear River, North Carolina, was an estimated 33% but improved to 61% in 1998 (Moser et al. 2000). Passage success for American shad in the Savannah River, Georgia, was approximately 53% in 2001 but significantly decreased to 9% in 2002 (Bailey et al. 2004). That same year at Pinopolis Lock in the Cooper River, passage success for American shad was estimated at 59%; in 2003, passage increased to 88% (Norman- deau Associates 2003). Fluctuations in passage could be related to locking schedule as well as unfavorable environmental conditions, such as a decrease in attraction flow near the lock entrance and extreme water temperatures. Passage of Alabama shad at JWLD may vary similarly over time. Changes to telemetry equipment occurred throughout the project, possibly biasing passage results. However, it is unlikely FIGURE 3. Timing of Alabama shad passage events compared with discharge

Downloaded by [Department Of Fisheries] at 20:26 25 September 2012 any large bias occurred from these changes. The acoustic and and temperature at JWLD during March–April 2007, 2010, and 2011. Solid line ◦ radio transmitters used were essentially the same dimensions. denotes discharge (cfs); dashed line denotes water temperature ( C); solid circles The change in 2011 to internal-antenna radio transmitters was indicate shad passage events occurring on that day. essential to follow the scope of the project. Implanted shad were tracked upstream after passage, and this was not feasible During all years at JWLD, most fish passed during periods of in riverine environments with acoustic transmitters. All trans- lower overall discharge, particularly low levels of spill gate dis- mitters were similar in weight and design, and reflected < 1% of charge (Figure 3). Moser et al. (2000) reported similar results for the weight of the implanted shad. Radio transmitters were 2 mm American shad passage at a navigation lock. Turbine discharge wider than hydroacoustic transmitters but were still easily in- appeared to have no effect on passage of Alabama shad and re- serted into the same size range of adults as the other transmitters. mained relatively constant for electric power production during The use of the somewhat larger radio transmitters actually re- the study period. Water releases from the spill gates can create a duced fallback but still resulted in passage comparable to 2007 turbulent discharge that may disorient fish, subsequently mak- and 2010. All transmitter types had sufficient range to be ef- ing it harder for them to locate the lock entrance. Other authors fective within the relatively small area occupied by JWLD and have suggested that American shad are repelled by turbulent covered by fixed-station arrays. flow, which negatively affects their ability to locate fish passage 888 YOUNG ET AL.

facilities (Barry and Kynard 1986; Moser et al. 2000). Addi- Kynard 1986; Moser et al. 2000). Alabama shad were found tional steps to increase attraction during high river discharge in the vicinity of JWLD from mid-February to mid-May when may be needed to increase passage success during high water water temperatures were 12–25◦C (Figure 3). As water tempera- years. tures increased to 21–22◦C, shad that had not passed were found In 2005, about 20% of study fish were considered nonviable further downstream of JWLD, occupying the best-available due to mortality, transmitter failure, or emigration. Shad that shoal and gravel bar habitats. This habitat shift may have corre- emigrated from the study area were assumed to have abandoned sponded to spawning. their spawning migration. This response is referred to as fallback Passage studies were conducted concurrent to mark– (Moser and Ross 1993) and is a common result of stress caused recapture studies to estimate the annual abundance of spawning by handling or transmitter implantation as well as unfavorable adults reaching JWLD. Utilizing the point estimates and 95% environmental conditions. Fallback in this study ranged from CI for both passage and spawner abundance, we were able esti- 2% to 35%. The total proportion of Alabama shad loss for all mate the number of Alabama shad that passed upstream to the 3 years was approximately 23%. Losses of Alabama shad in this Flint and Chattahoochee rivers (Table 2). Both the population study are comparable to previous estimates for American shad. estimate and percent passage were highest during 2010, when Moser et al. (2000) observed a similar estimate of 27% (range = as many as 57,262 spawning adults used the navigation lock to 5–61%) for American shad in the Cape Fear River, North Car- migrate upstream, predominantly to the Flint River. The low- olina, as did a study in the Cooper River, South Carolina, where est number of shad passed occurred during 2007, an estimated 22% (range = 18–26%) never returned to the dam after trans- low of only 2,137 spawners migrating up the Flint and Chat- mitter implantation (Normandeau Associates 2003). However, tahoochee rivers. This estimate was further verified by an in- American shad in the Savannah River, Georgia–South Carolina, dependent mark–recapture study that yielded a similar estimate observed higher losses, averaging 61% (range = 39–80%; Bai- of 875 Alabama shad in the Flint River (Ingram, unpublished ley et al. 2004). Sampling gear, water temperatures, transmitter data). type, condition of individual fish, and duration of time spent Based on our results, we conclude that operation of the nav- near migration obstruction may influence postrelease behavior. igation lock at JWLD was an effective means to pass migrating Most shad entered the lock chamber and passed during the Alabama shad. Increased passage could be achieved by max- afternoon locking cycle. Our results indicate that the afternoon imizing attraction flow near the lock entrance and increasing locking cycle could be the most critical for passage of Alabama the time the upper gates are open during an afternoon locking shad. It is also apparent that passage efficiency is positively cycle. Further study of Alabama shad behavior during various correlated to the time the upper lock gates are open. Therefore, in water discharge levels and the various means to release water order to increase passage efficiency, it is important for shad that through JWLD may also aid in successful passage of migrating have found their way into the lock chamber to have sufficient Alabama shad. Future population and passage monitoring will time to pass upstream before closing the upper gates. Filling better determine the consequences of renewed access to more the lock chamber too quickly can create turbulent water, thus and potentially higher-quality spawning habitat in the tributaries disorienting fish and hampering their ability to pass. Therefore, upstream of JWLD. Preliminary results indicate an increase in in order to prevent disorientation, the lock chamber should be population, but annual variation is still present. The popula- filled as slowly as possible. tion remains self-sustaining, some spawning year-classes of During our study, the lock was in operation 7 d/week be- >25,000 spawning adults reaching JWLD. Future research tween 0800 and 1600 hours. As a result, all Alabama shad were should focus on behavior, fate, and success of reproduction for passed during these daytime operation hours. Analogous stud- spawning adults passing upriver of JWLD and subsequent juve-

Downloaded by [Department Of Fisheries] at 20:26 25 September 2012 ies of American shad illustrate the diel timing of fish lockages nile recruitment in the Flint and Chattahoochee rivers in order is critical for successful passage and that highest passage has to determine if fish passage has significant long-term benefits been observed during the day (Moser et al. 2000; Bailey et al. for this population. 2004). It is not known if Alabama shad migrate at night or if night lock operation would increase passage, but new evidence suggests the contrary (T. R. Ingram, unpublished data). Night- ACKNOWLEDGMENTS time lock operation is also hindered by the work schedule of Cooperating agencies were the South Carolina Cooperative lock operators. During mobile telemetry, Alabama shad were Fish and Wildlife Research Unit; the U.S. Geological Survey more often found during daytime hours occupying the tailrace Biological Resources Division, Clemson University; Georgia on the east bank opposite the lock chamber (Figure 2). Alabama Department of Natural Resources; Florida Fish and Wildlife shad occupied the area near the lock entrance more during the Conservation Commission; The Nature Conservancy; USACE; night and early morning. This may be because of the channel and U.S. Fish and Wildlife Service. Funding was provided by depth and minimal water velocities near the lock entrance. Other the Georgia Department of Natural Resources and the National studies have shown American shad to use higher water velocity Marine Fisheries Service. Reference to trade names does not during the day and deeper, quieter areas at night (Barry and imply endorsement by the U.S. Government. PASSAGE OF SPAWNING ALABAMA SHAD 889

REFERENCES Mills, J. G., Jr. 1972. Biology of the Alabama shad in northwest Florida. Florida Bailey, M. M., J. J. Isely, and W. C. Bridges Jr. 2004. Movement and population Department of Natural Resources Marine Research Laboratory Technical size of American shad near a low-head lock and dam. Transactions of the Series 68. American Fisheries Society 133:300–308. Moser, M. L., A. M. Darazsdi, and J. R. Hall. 2000. Improving passage efficiency Barkuloo, J. M., M. Mettee, and L. Jenkins. 1993. Systematic and population of adult American shad at low-elevation dams with navigation locks. North status of Alabama shad in rivers tributary to the Gulf of Mexico. Report to American Journal of Fisheries Management 20:376–385. the U.S. Fish and Wildlife Service, Panama City, Florida. Moser, M. L., and S. W. Ross. 1993. Distribution and movements of anadromous Barry, T., and B. Kynard. 1986. Attraction of adult American shad to fish lifts fishes of the lower Cape Fear River, North Carolina. U.S. Army Corps of at Holyoke Dam, Connecticut River. North American Journal of Fisheries Engineers, Final Report, Wilmington, North Carolina. Management 6:233–241. Nichols, P. R., and D. E. Louder. 1970. Upstream passage of anadromous fish Beasley, C. A., and J. E. Hightower. 2000. Effects of a low-head dam on the through navigation locks and use of the stream for nursery and spawning distribution and characteristics of spawning habitat used by striped bass and habitat, Cape Fear River, North Carolina, 1962–1966. U.S. Fish and Wildlife American shad. Transactions of the American Fisheries Society 129:1316– Service Circular 352. 1330. NOAA (National Oceanic and Atmospheric Administration). 2004. Endangered Berry, F. H. 1964. Review and emendation of family Clupeidae. Copeia and threatened species; establishment of species of concern list, addition 1964:720–730. of species to species of concern list, description of factors for identifying Chappelear, S. J., and D. W. Cooke. 1994. Blueback herring behavior in the species of concern, and revision of candidate species list under the Endangered tailrace of the St. Stephen dam and fish lock. Pages 108–112 in J. E. Cooper, Species Act. Federal Register 69:73(15 April 2004):19975–19979. R. T. Eades, R. J. Klauda, and J. G. Loesch, editors. Anadromous Alosa NOAA (National Oceanic and Atmospheric Administration). 2011. Endangered symposium. American Fisheries Society, Tidewater Chapter, Bethesda, and threatened wildlife; 90-day finding on a petition to list Alabama shad as Maryland. threatened or endangered under the Endangered Species Act (ESA). Federal Clay, C. H. 1995. Design of fishways and other fish facilities, 2nd edition. CRC Register 76:33(17 February 2011):9320–9327. Press, Boca Raton, Florida. Normandeau Associates. 2003. Effectiveness of the Pinopolis Lock at attracting Coker, R. E. 1929. Studies of common fishes of the Mississippi River at Keokuk. adult blueback herring and American shad. Normandeau Associates, Project U.S. Bureau of Fisheries Bulletin 45:141–225. 19409.000, Westmoreland, New Hampshire. Cooke, D. W., and S. D. Leach. 2003. Beneficial effects of increased river flow Petersen, J. H., R. A. Hinrichsen, D. M. Gadomski, D. H. Feil, and D. W. and upstream fish passage on anadromous alosine stocks. Pages 331–338 in Rondorf. 2003. American shad in the Columbia River. Pages 141–155 in K. E. Limburg and J. R. Waldman, editors. Biodiversity, status, and con- K. E. Limburg and J. R. Waldman, editors. Biodiversity, status, and conservation of the world’s shads. American Fisheries Society, Symposium 35, servation of the world’s shads. American Fisheries Society, Symposium 35, Bethesda, Maryland. Bethesda, Maryland. Ely, P. C., S. P. Young, and J. J. Isely. 2008. Population size and relative abun- Richkus, W. A., and G. DiNardo. 1984. Current status and biological characteris- dance of Alabama shad reaching Jim Woodruff lock and dam, Apalachicola tics of the anadromous alosid stocks of eastern United States: American shad, River, Florida. North American Journal of Fisheries Management 28: hickory shad, alewife, and blueback herring. Martin Marietta Environmental 827–831. Center, Baltimore, Maryland. Hildebrand, S. F. 1963. Family Clupeidae. Pages 257–454 in Y.H. Olsen, editor. Ricker, W. E. 1975. Computation and interpretation of biological statistics of Fishes of the western North Atlantic, part three. Sears Foundation for Marine fish populations. Fisheries Research Board of Canada Bulletin 191. Research, Yale University, New Haven, Connecticut. Rulifson, R. A., M. T. Huish, and R. W. Thoesen. 1982. Anadromous fish in Laurence, G. C., and R. W. Yerger. 1967. Life history studies of the Alabama the southeastern United States and recommendations for development of a shad, Alosa alabamae, in the Apalachicola River, Florida. Master’s thesis. management plan. U.S. Fish and Wildlife Service, Fishery Resources, Region Florida State University, Tallahassee. 4, Atlanta, Georgia. McBride, R. S. 2000. Florida’s shad and river herrings (Alosa species): a review Schmidt, R. E., B. M. Jessop, and J. E. Hightower. 2003. Status of river herring of population and fishery characteristics. Florida Marine Research Institute stocks in large rivers. Pages 171–182 in K. E. Limburg and J. R. Waldman, Technical Report TR-5. editors. Biodiversity, status, and conservation of the world’s shads. American Mettee, M. F., and P. E. O’Neil. 2003. Status of Alabama shad and skipjack Fisheries Society, Symposium 35, Bethesda, Maryland. herring in Gulf of Mexico drainages. Pages 157–170 in K. E. Limburg and Zehfuss, K. P., J. E. Hightower, and K. H. Pollock. 1999. Abundance of gulf J. R. Waldman, editors. Biodiversity, status, and conservation of the world’s sturgeon in the Apalachicola River, Florida. Transactions of the American shads. American Fisheries Society, Symposium 35, Bethesda, Maryland. Fisheries Society 128:130–143. Downloaded by [Department Of Fisheries] at 20:26 25 September 2012 This article was downloaded by: [Department Of Fisheries] On: 25 September 2012, At: 20:29 Publisher: Taylor & Francis Informa Ltd Registered in England and Wales Registered Number: 1072954 Registered office: Mortimer House, 37-41 Mortimer Street, London W1T 3JH, UK

Transactions of the American Fisheries Society Publication details, including instructions for authors and subscription information: http://www.tandfonline.com/loi/utaf20 Factors Affecting Migration Timing, Growth, and Survival of Juvenile Coho Salmon in Two Coastal Washington Watersheds Philip Roni a , Todd Bennett a , Ranae Holland a , George Pess a , Karrie Hanson a , Raymond Moses b , Mike McHenry b , William Ehinger c & Jason Walter d a Watershed Program, Fisheries Ecology Division, Northwest Fisheries Science Center, 2725 Montlake Boulevard East, Seattle, Washington, 98112, USA b Lower Elwha Klallam Tribe, Fisheries Department, 51 Hatchery Road, Port Angeles, Washington, 98363, USA c Environmental Assessment Program, Washington Department of Ecology, Post Office Box 47710, Olympia, Washington, 98504-7710, USA d Weyerhaeuser Company, WTC-1B10, 33663 Weyerhaeuser Way South, Federal Way, Washington, 98001, USA Version of record first published: 21 Jun 2012.

To cite this article: Philip Roni, Todd Bennett, Ranae Holland, George Pess, Karrie Hanson, Raymond Moses, Mike McHenry, William Ehinger & Jason Walter (2012): Factors Affecting Migration Timing, Growth, and Survival of Juvenile Coho Salmon in Two Coastal Washington Watersheds, Transactions of the American Fisheries Society, 141:4, 890-906 To link to this article: http://dx.doi.org/10.1080/00028487.2012.675895

PLEASE SCROLL DOWN FOR ARTICLE

Full terms and conditions of use: http://www.tandfonline.com/page/terms-and-conditions 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. The publisher does not give any warranty express or implied or make any representation that the contents will be complete or accurate or up to date. The accuracy of any instructions, formulae, and drug doses should be independently verified with primary sources. The publisher shall not be liable for any loss, actions, claims, proceedings, demand, or costs or damages whatsoever or howsoever caused arising directly or indirectly in connection with or arising out of the use of this material. Transactions of the American Fisheries Society 141:890–906, 2012 C American Fisheries Society 2012 ISSN: 0002-8487 print / 1548-8659 online DOI: 10.1080/00028487.2012.675895

ARTICLE

Factors Affecting Migration Timing, Growth, and Survival of Juvenile Coho Salmon in Two Coastal Washington Watersheds

Philip Roni,* Todd Bennett, Ranae Holland, George Pess, and Karrie Hanson Watershed Program, Fisheries Ecology Division, Northwest Fisheries Science Center, 2725 Montlake Boulevard East, Seattle, Washington 98112, USA Raymond Moses and Mike McHenry Lower Elwha Klallam Tribe, Fisheries Department, 51 Hatchery Road, Port Angeles, Washington 98363, USA William Ehinger Environmental Assessment Program, Washington Department of Ecology, Post Ofﬁce Box 47710, Olympia, Washington 98504-7710, USA Jason Walter Weyerhaeuser Company, WTC-1B10, 33663 Weyerhaeuser Way South, Federal Way, Washington 98001, USA

Abstract Recent improvements in tagging technology allow for the examination of the migration of individual fish, the detection of previously unidentified life histories, and the detailed examination of factors affecting growth, migration, and survival. Using passive integrated transponder tags and instream readers installed near tidewater, we examined the migration, growth, and survival of 18,642 juvenile coho salmon Oncorhynchus kisutch in two small western Washington rivers from 2005 to 2009. In most years, more than 50% of the juvenile coho salmon from a given brood year migrated to sea between 1 October and 31 December (fall migrants). These fall migrants were significantly smaller at tagging than fish that migrated between 1 January and 30 June (spring migrants) but were similar in size to fish that were never detected after tagging and assumed to have died. Annual coho salmon survival estimates from tagging to out-migration ranged from 31% to 40% for fall and spring migrants combined but from 5% to 15% for spring migrants only. The best fitting regression models indicated that survival differed by river and year and was negatively correlated with tagging location (river kilometer) and positively correlated with fish length: larger fish and Downloaded by [Department Of Fisheries] at 20:29 25 September 2012 those tagged lower in the watershed were more likely to survive. The number of days juvenile coho salmon spent in freshwater before migrating to sea was positively correlated with tagging location, fish length (mm), and habitat depth (m) and negatively with density (coho salmon/m2). Our results suggest that fall or early winter migration is a common life history for juvenile coho salmon that is driven in part by fish size and location in the watershed. The exclusion of fall migrants may lead to underestimates of the total number of migrants and parr-to-smolt survival.

Estimates of ﬁsh movement, growth, and survival at vari- The degradation of stream habitat can decrease juvenile coho ous life stages are critical to understanding the effects of habi- salmon Oncorhynchus kisutch egg-to-fry, fry-to-parr (sum- tat changes at reach (∼100–1,000 m) and watershed scales. mer), and parr-to-smolt (overwinter) survival (Tschaplinski and

*Corresponding author: [email protected] Received March 12, 2011; accepted January 9, 2012 Published online June 21, 2012 890 MIGRATION, GROWTH, AND SURVIVAL OF JUVENILE SALMON 891

Hartman 1983; Cederholm and Reid 1987). Habitat improve- marine survival (Holtby et al. 1990; Quinn and Peterson 1996) ment and restoration for coho salmon is often designed to in- and appears to influence movement and migration timing (Bilby crease overwinter survival (Cederholm et al. 1997; Solazzi et al. and Bisson 1987; Bennett 2006). Within a given cohort, larger 2000). Thus, accurately quantifying movement, growth, and sur- coho salmon parr generally have higher overwinter survival vival is critical for understanding why fish numbers increase (Quinn and Peterson 1996), while smaller fish appear more or decrease in response to various habitat changes, be it from likely to migrate or be forced downstream in the fall (Bilby and restoration or natural changes within a watershed. Moreover, to Bisson 1987; Bennett 2006). A large size at smolting can be adequately estimate survival, we must understand basic infor- particularly important for early marine survival in years when mation about fish life history and ecology, including local move- marine productivity is poor (Holtby et al. 1990). Some studies ment (reach scale) as well as seasonal migrations patterns—all have reported that fish moving into tributary, off-channel, and factors that can be influenced by changes in habitat quality and estuarine habitats have higher growth than those remaining in quantity. mainstream reaches (Peterson 1982; Miller and Sadro 2003; Evidence for the effects of habitat alteration on survival Wigington et al. 2006). However, growth and survival are not comes largely from studies comparing summer and winter parr always positively correlated and are inversely correlated in or smolt counts in disturbed versus undisturbed habitats, simple some instances (Peterson 1982). versus complex habitats, or before versus after restoration Recent advances in tagging technology have allowed for de- activities (Tschaplinski and Hartman 1983; Martin et al. 1984; ployment of passive integrated transponder (PIT) tag readers in Cederholm et al. 1997; Solazzi et al. 2000). Other studies have remote locations and for placement of antenna arrays (rows of directly measured the survival of fish in individual habitats antennas) across an entire stream. These systems provide the (Ebersole et al. 2006; Brakensiek and Hankin 2007; Pess means to accurately quantify and compare not only survival, but et al. 2011). For example, Quinn and Peterson (1996) marked also growth, movement, or migration among habitat types, sub- coho salmon parr in different habitats within Big Beef Creek, basins, or watersheds (Zydlewski et al. 2006). Detection data Washington, and found that overwinter survival ranged from from PIT tags also allow for the observation of unique life his- 25% to 46% among years. Coho salmon overwinter survival tories that would not be discernible with traditional tagging and estimates from Alaska to northern California range from as trapping techniques. We used PIT tags to monitor the migra- low as 11% to as high as 78% (Brakensiek and Hankin 2007). tion, growth, and survival of individual juvenile coho salmon in A common assumption of survival and movement studies is two watersheds along the Strait of Juan de Fuca in northwest that fish that migrate out of the study area before recapture Washington State from 2005 to 2009. Specific objectives were do not survive (Gowan et al. 1994). This assumption does to examine coho migration, growth, and survival at a watershed not hold true in many populations, as several studies have and stream reach scale and to examine the relationship between documented extensive upstream or downstream movement in these factors and physical habitat conditions (flow, temperature, fall to overwinter in off-channel, tributary, or other main-stem location within a watershed, habitat type, depth) as well as fish habitats (Skeesick 1970; Scarlett and Cederholm 1984; Murphy condition (length) and competition (fish density). et al. 1997; Bramblett et al. 2002). A complete understanding of migration timing and potentially different juvenile life histories also requires more than METHODS just sampling during spring months. As early as 1914, Gilbert Study sites.—The two study streams, East Twin and West reported that a portion of coho salmon fry migrate to sea in Twin rivers, are located on the north coast of the Olympic Penin- the first year of life, particularly in Alaska, but that this phe- sula in Washington State and flow directly into the Strait of Juan

Downloaded by [Department Of Fisheries] at 20:29 25 September 2012 nomenon appeared to be rare in the southern part of their range de Fuca (Pacific Ocean) (Figure 1). The drainage area is 35 and (Washington, Oregon, and California). More recent studies have 33 km2, respectively for East and West Twin, and the elevation indicated that large numbers of coho salmon fry or parr migrate in these basins ranges from sea level to 915 m in the head- to estuarine habitats shortly after emergence or during summer waters. Precipitation occurs primarily as rain between October months (Crone and Bond 1976; Tschaplinski 1982; Miller and and May and averages 190 cm per year (Olympic National For- Sadro 2003; Koski 2009). There is also evidence that suggests est 2002). Mean instantaneous flow from September 2005 to some coho salmon parr migrate to estuarine or marine environ- 2009 (study period) was 1.7 m3/s (range, 0.1–29.6) in East Twin ments in the fall (Bennett 2006; Bennett et al. 2011), but it is and 1.4 m3/s (range, 0.1–41.6) for the West Twin (B. Ehinger, not known if this is a common strategy. Both Gilbert (1914) and Washington Department of Ecology, unpublished data). Water modern authors have noted the need for more detailed investi- temperature during the same period ranged from 0◦ to 16.7◦C gations to understand the frequency of fry and parr migrants and for both rivers and averaged 8.1◦C and 7.8◦C for East and West whether the proportion of the migrants varies among years or Twin, respectively (Ehinger, unpublished data). The geology of streams, or is influenced by other physical or biological factors. these watersheds is characterized by Crescent Formation vol- Fish size and growth are also important determinants of canic rock in the upper watershed, marine sedimentary rock in habitat quality as size influences both overwinter and early the lower watershed, and terraces of glacial deposits in the lower 892 RONI ET AL.

FIGURE 1. Map of East and West Twin rivers in northwest Washington State and their conﬂuence with the Strait of Juan de Fuca. Dots indicate distance in kilometers from the river mouth. Bars at the upper end of each stream indicate the approximate location of anadromous barriers. Stationary PIT tag readers and smolt traps were located approximately 300–500 m from tidewater in East and West Twin (below rkm 0.5).

watershed floodplain (Olympic National Forest 2002). Three throat trout O. clarki, chum salmon O. keta, Pacific lamprey vegetation zones are found in the watershed: the Sitka Spruce Lampetra tridentata, Western brook lamprey L. richardsoni,tor- Picea sitchensis zone in the valley bottom, the Western Hem- rent sculpin Cottus rhotheus and reticulate sculpin C. perplexus. lock Tsuga heterophylla zone in the low to mid elevations, and Coho salmon and other anadromous fishes are found below river the Silver Fir Abies amabilis zone in the headwaters (Olympic kilometer (rkm) 5.8 on East Twin and at approximately rkm 6.3 National Forest 2002). on the West Twin (Figure 1). Sadie Creek, a major tributary of The primary land use within both basins for the last 100 years the East Twin River at rkm 3.3, is an important spawning and has been forestry (Olympic National Forest 2002; Bilby et al. rearing area for coho salmon (M. McHenry, unpublished data). 2005). Logging, removal of in-channel woody debris, and During the study period, estimates of coho salmon smolt pro-

Downloaded by [Department Of Fisheries] at 20:29 25 September 2012 construction of logging roads on steep slopes have increased duction by the Lower Elwha Klallam Tribe ranged from 11,288 landslide frequency and have simplified and degraded channel to 22,026 for the East Twin and from 5,272 to 8,103 for the West habitat conditions (Bilby et al. 2005). Both streams are part Twin. Coho salmon adult escapement estimates ranged from 66 of the Washington State Intensively Monitored Watershed to 320 in the East Twin and 57 to 311 in the West Twin from Program (IMW), which began in 2005 (see Bilby et al. 2005 2003 to 2008. Estimates of smolt-to-adult survival ranged from for details of the program). A number of restoration actions 0.3% to 3.4% in the East Twin and 0.7% to 5.8% in the West have been implemented in the East Twin River including road Twin (McHenry, unpublished data). removal, off-channel connection and construction, large woody The PIT tag readers and field methods.—Stationary mul- debris (LWD) placement, riparian planting, and removal of fish tiplex PIT tag readers were installed approximately 300 and migration barriers (impassible culverts). West Twin River is the 400 m from tidewater in the East Twin and West Twin rivers, “control” watershed for the IMW study, and other than removal respectively (Figures 1, 2). To maximize our probability of de- of some forest roads in the upper watershed, no restoration has tecting PIT-tagged fish, each reader included two antenna arrays occurred there (Bilby et al. 2005). each composed of three antennas that spanned the stream un- The fish species present in the two basins include coho der most flows (Figure 2). In each stream the downstream array salmon, steelhead or rainbow trout Oncorhynchus mykiss, cut- was positioned approximately 3–5 m below the upstream array. MIGRATION, GROWTH, AND SURVIVAL OF JUVENILE SALMON 893

lating survival to fish density or physical habitat (see statistical methods below). These data were, however, used to examine Multiplex PIT Tag watershed-scale patterns of emigration, survival, and growth. Reader Upstream Antenna Array To increase the numbers of juvenile coho salmon tagged, and to collect more detailed density and habitat information, we conducted additional sampling in the lower 3,000 m of both rivers using single-pass electrofishing in 2005 and three-pass electrofishing during 2006–2008. Fish abundance in each habitat FLOW

High Bank unit was calculated using a multiple removal estimator (2006–

~ 5 m ~ 5 2008; Carle and Strub 1978) and densities were estimated by dividing abundance by total wetted area of the habitat unit. We Gravel Bar Bar Gravel sampled each habitat unit separately and recorded the habitat type (pool, riffle, or glide) and reach (rkm) where each fish Downstream Antenna Array was tagged. During 2006–2008, we also measured pool depth (maximum and depth of tailout) and riffle depth and visually ~ 10 m estimated dominant and subdominant substrate size and percent wood cover (percent of habitat covered, in- or above-water LWD) in each habitat unit. Depth measurements were used to calculate residual pool depth and mean pool and riffle depth. FIGURE 2. Diagram of typical PIT tag antenna arrays in the West Twin River. In 2007 and 2008, we counted and measured LWD (>10 cm The configuration was modified periodically due to channel migration. Three diameter and 1 m in length) and estimated total LWD volume. 3–5-m-wide antennas comprised each array and were attached to the stream For each stream and year, survival from tagging in August and bottom. The East Twin configuration differed slightly due to a high gravel bar in the middle of channel. [Figure available online in color.] September to out-migration was estimated in two steps. First, we calculated the total number of tagged juvenile coho salmon that out-migrated each month based on the last detection date This configuration allowed for the detection of PIT-tagged fish from September through June. Then we corrected those numbers emigrating from the watersheds to the marine environment. The by the PIT tag reader efficiency. Because each PIT tag reader PIT tag readers ran continuously from 1 August 2005 until 31 included two antenna arrays in each stream, we calculated the December 2009. combined efficiency of both arrays (detected fish ÷ combined Juvenile coho salmon were collected by electrofishing and efficiency; Zydlewski et al. 2006). The in situ PIT tag detection then were anesthetized, measured, weighed, tagged, and placed efficiency of individual arrays and the combined efficiency of in a recovery tub in the stream for at least 15 min before being re- both arrays were estimated monthly for each reader using the leased into the same habitat from which they were captured. All methods below. juvenile coho salmon 55 mm or larger were tagged with 12-mm i = d ÷ PIT tags. A total of 18,642 fish were tagged in both rivers be- Efficiency of array ( common to both arrays) [(dunique to array i ) tween mid-August and mid-September from 2005 to 2008. Ap- + (d )] and proximately one-third of these fish were tagged while working common to both arrays with the Weyerhaeuser IMW crew, who were conducting single- Combined efficiency of both arrays pass and occasionally three-pass electrofishing surveys in 10 = 1 − [(1 − Earray 1) × (1 − Earray 2)],

Downloaded by [Department Of Fisheries] at 20:29 25 September 2012 reaches of each watershed as part of the larger IMW study (Bilby et al. 2005). Each sample reach was approximately 50 m long, where d is the number of tags detected and E is the efficiency and reaches were randomly selected throughout each water- (Zydlewski et al. 2006). shed using the Environmental Protection Agency Environmental The combined efficiency was used to correct monthly rates Monitoring and Assessment Program’s generalized random tes- of detection and survival for each stream. The annual survival sellation stratified sample procedure (Stevens and Olsen 2004; from tagging to out-migration was calculated by summing the Bilby et al. 2005). However, this approach resulted in low num- total monthly corrected detections by the total number of fish bers of fish being collected. Previous surveys have documented tagged that year. We examined each tagging cohort separately that the majority of juvenile and adult coho salmon are found from 1 September to 30 June because all tagged fish were last in the lower 3 km of both rivers, and most IMW sites were in detected during this period, few or no fish emigrated in July the upper watershed in areas of low juvenile coho salmon abun- and August, and we detected no 2-year-old juvenile coho. In dance (McHenry, unpublished data). Detailed habitat-specific addition, we classified fish as fall migrants if they emigrated information on fish densities and physical habitat was not col- before or on 31 December and spring migrants if they emigrated lected as part of the Weyerhaeuser IMW surveys. Therefore, between 1 January and 30 June. The end of December was data from these fish were not used in our regression models re- used to distinguish between fall and spring migrants because 894 RONI ET AL.

TABLE 1. Independent variables used in regression models to examine correlation between survival, residence time in freshwater, and growth of coho salmon.

Tagging years Independent variable Description measured River East or West Twin River 2005–2008 Habitat type Slow (pool or glide) or fast (riffle) 2006–2008 Maximum habitat depth Maximum depth of pool or riffle 2006–2008 Large woody debris volume Total volume (length × width) of wood within habitat unit/total 2007–2008 habitat unit area River kilometer (rkm) Distance in thousands of meters from stream mouth 2005–2008 Length Fish length in millimeters 2005–2008 Relative length Fish length (mm) at tagging minus the mean length of all fish in habitat 2006–2008 unit divided by the mean length of all fish Density Total fish number (coho salmon and trout) per unit area (m2) 2006–2008 Days in freshwater Total days from date of tagging to last detection 2005–2008 Growth (Fish length at smolt trap – fish length at initial tagging)/(number of 2005–2008 days between tagging and capture at smolt trap)

most fall and early winter emigration occurred in October and dence, and growth) and independent variables (Table 1) for the November and migration tapered off to nearly zero by the end years for which we had detailed habitat-specific physical and of December. The peak spring migration typically took place biological data (2006 to 2008). To determine which model best during April or May, with few fish emigrating before March fit the data, we used an information-theoretic approach with the or after mid-June. Survival from tagging to out-migration was Akaike information criterion adjusted for small sample sizes estimated by dividing the total number of spring or fall migrants (AICc) (Burnham and Anderson 2002). The difference between by the total number of fish tagged in that year. the AICc of a candidate model and the model with the lowest High flows and occasional equipment failure rendered our AICc provided the ranking metric (AICc). A AICc between PIT tag readers inoperable on the following dates or periods: 0 and 3 generally indicates substantial support for a model be- 4 May 2005, 7–9 November 2006 in East Twin; and 6–9 Novem- ing as good as the best approximating model, AICc between ber 2005, 27 December 2006, 3–11 December 2007, and 13–14 4 and 7 represented less support, and values of greater than 7 February 2008 in West Twin. During extreme high flows there indicates very little support for a candidate model relative to the were also short periods (a few days) where one or more anten- best model (Burnham and Anderson 2002). nas within an array were inoperable which reduced detection Akaike weights (wi) were calculated to represent the strength efficiency. However, these reductions in efficiency are reflected of evidence in favor of model i being the best model and model in our calculations of combined array efficiency. Because mark– j being the next best model. The ratio of Akaike weights (wi/wj) recapture studies such as ours cannot account for undetected in- indicates the plausibility of the best-fitting model compared to dividuals, our estimates of survival are more accurately termed other models (Burnham and Anderson 2002). Models with evi- apparent survival (Burnham et al. 1987). However, for the sake dence ratios of 10 or less were considered plausible (Burnham

Downloaded by [Department Of Fisheries] at 20:29 25 September 2012 of readability we refer to them as survival rather than apparent and Anderson 2002). If models were not clearly “best” based survival. on the preceding criteria, then those having a score of 3 AICc The Lower Elwha Klallam Tribe operated fence-weir smolt or less were considered the best suite of models. Each model traps downstream of the PIT tag arrays on both streams from was limited to four independent variables and all subsets were April to June in each study year. Coho salmon captured in the examined during model selection. smolt traps were examined for PIT tags, and fork length was The response variable survival had a binomial distribution recorded for any tagged fish encountered. We calculated total (i.e., yes or no), necessitating logistic regression (Agresti 1996). growth for all tagged coho salmon captured in the smolt trap Therefore, we developed multiple logistic regressions and used a by taking the difference between the fish length recorded at model selection approach to determine the relative importance the smolt trap and the length at initial tagging. Daily growth of independent variables: tag year, river basin, river location was estimated by dividing total growth by the number of days (reach or river kilometer), habitat depth, fish size (length), and between tagging and recapture. The daily migration timing of competition (coho salmon density) on juvenile coho salmon fish was examined graphically. survival (Pess et al. 2011). We modeled tag year as a fixed effect Statistical analysis.—We developed multiple regression because with only 4 years of data, the variance component for models (i.e., single, multiple, and logistic) and selected among tag year would be poorly estimated and using a fixed effect only models with key dependent variables (survival, freshwater resi- requires three degrees of freedom. MIGRATION, GROWTH, AND SURVIVAL OF JUVENILE SALMON 895

Using a similar approach but with linear regression, we ex- (Table 2). Despite efforts to tag coho salmon in riffles as well amined whether fish out-migration timing or growth was corre- as pools and glides, the vast majority of fish were found and lated with the same independent variables (Pess et al. 2011). We tagged in pools, and fish densities were consistently higher in used freshwater residence time (tag date – last date detected at pool habitats. The average depths of pools varied among years PIT tag reader) for out-migration timing, as this was a continu- and streams and ranged from 0.58 (SD = 0.18) to 0.80 m (SD = ous variable and more easily compared among years. We used 0.31) (Table 2). relative rather than absolute length as a dependent variable in growth models because it was thought to better reflect competi- Out-migration Timing tion that might affect growth (Pess et al. 2011; Table 1). Growth The migration timing of tagged juvenile coho salmon showed was standardized to daily growth rate (mm/d) to account for a consistent pattern among years. Distinct peaks occurred in fall small differences among days from tagging to capture in the and again in spring during all years (Figure 4). The median smolt trap. Because 95% of coho salmon were tagged in pools fall migration date ranged from 23 October to 8 November or glides and glides are shallow pools, rather than habitat type, for both rivers, while the median spring migration date ranged habitat depth was used as a continuous indicator of habitat type from 6 to 25 May for East Twin and 17 to 27 May for West Twin and quality. We limited the independent variables in our models (Table 3). Daily numbers of out-migrating juvenile coho salmon to four or fewer to focus on those that were the most reason- occasionally tracked mean daily flow, but the initiation of fall able predictors of survival, freshwater residence or growth, and emigration did not appear to be directly related to either flow to also focus on independent variables that were not strongly or temperature. Temperature in fall and winter closely tracked correlated (Pearson correlation <0.50) (Pess et al. 2011). flow. We more closely examined the correlation between daily out- migrant numbers and mean daily flows in October and Novem- RESULTS ber for each river and year separately using linear regression. In A total of 18,642 juvenile coho salmon were tagged in the the East Twin River, numbers of coho salmon out-migrants and East and the West Twin rivers from 2005 to 2008. Tagged fish mean daily flow were significantly correlated in 2005, 2007, and ranged in size from 55 mm to more than 100 mm, with most 2008 (P < 0.01, r2 = 0.26, 0.29, and 0.38 respectively) but not fish being less than 80 mm (Figure 3). Mean fish size, ranging in 2006 (P = 0.09). In the West Twin River, flow and number of from 60 mm to more than 74 mm, varied among habitat types, out-migrants were correlated in 2005 (P = 0.04, r2 = 0.08) and streams, and years and was negatively related to fish density 2008 (P < 0.01, r2 = 0.39) but not in 2006 or 2007 (P > 0.40). Downloaded by [Department Of Fisheries] at 20:29 25 September 2012

FIGURE 3. Size frequency of all tagged and detected juvenile coho salmon and their survival for all years and streams combined. 896 RONI ET AL.

TABLE 2. Summary of habitat and ﬁsh surveys from 2006 to 2008, including the number of habitat units sampled, area, depth, large woody debris (LWD) volume and coho salmon density (ﬁsh/m2), biomass, and fork length for habitats sampled. All surveys occurred in August and September.

Habitat area Max. depth LWD volume Coho biomass Coho fork (m2) (m) (m3) Coho density Fish density (g) length (mm)

No. No. Year River Habitat type units Mean SD Mean SD Mean SD Mean SD Mean SD Mean SD units Mean SD

2006 East Twin Riffle 8 71.3 35.9 0.19 0.08 0.30 0.26 1.61 0.92 64.3 54.6 7 63.0 4.8 Pool/glide 26 84.3 42.3 0.58 0.18 1.57 0.72 2.61 0.97 312.5 159.0 26 60.2 2.8 West Twin Riffle 7 101.7 43.8 0.16 0.03 0.11 0.15 0.91 0.43 36.9 52.8 6 65.6 6.9 Pool/glide 30 88.3 57.7 0.72 0.41 1.17 0.85 l.88 1.11 243.2 174.7 30 61.6 3.5 2007 East Twin Riffle 6 94.6 44.3 0.31 0.11 2.16 2.90 0.03 0.03 0.71 0.25 12.9 15.1 4 71.2 5.3 Pool/glide 37 85.8 49.3 0.72 0.25 2.98 2.60 0.38 0.32 1.20 0.79 135.5 111.9 35 73.5 3.6 West Twin Riffle 7 100.9 54.6 0.23 0.04 5.43 5.48 0.01 0.04 0.80 0.25 6.2 16.4 1 70.8 Pool/glide 41 87.4 46.3 0.72 0.27 3.59 4.30 0.37 0.36 1.21 0.85 117.7 108.8 41 74.1 3.6 2008 East Twin Riffle 8 114.8 64.3 0.23 0.07 2.88 2.84 0.49 0.41 1.75 0.66 173.2 245.7 8 59.5 1.3 Pool/glide 18 112.1 57.9 0.78 0.33 20.65 55.07 1.94 1.11 2.94 1.33 435.4 218.6 18 58.4 2.4 West Twin Riffle 5 63.3 34.3 0.28 0.13 1.83 1.99 0.15 0.15 0.99 0.37 31.5 35.2 5 62.7 7.7 Pool/glide 23 113.4 56.0 0.80 0.31 3.91 4.14 0.96 0.53 1.78 0.60 316.2 209.2 23 63.9 4.5

The percentage of fish that out-migrated in fall or early winter they moved from one stream to another and then out to sea in (before 31 December) varied among the 4 years and ranged from the fall). 57% to 84% in the East Twin River and 44% to 69% in the West Twin River (Table 4). The percentage of fish last detected out- Out-migration Timing and Length at Tagging migrating in the spring that were also detected out-migrating in Mean length at tagging differed by river, year tagged, and fall (i.e., fish that went to sea in fall but returned to overwinter whether fish were detected emigrating in the fall or spring or in freshwater) ranged from 4% to 12% each year for both rivers. were never detected after tagging (analysis of variance: P = Little difference in migration timing was apparent between 0.00, F = 125.0; Figure 5). Multiple comparisons revealed that slow (pools and glides) and fast-water (riffle) habitats. How- undetected fish and those detected out-migrating in the fall were ever, less than 10% of all juvenile coho salmon were captured not significantly different in length at tagging (mean length 64.7 or tagged in riffle habitats, making comparisons between habi- versus 65.1 mm; Tukey multiple comparison: P = 0.08), while tat types difficult. Due to the relatively low numbers of fish fish detected out-migrating in the spring were significantly larger tagged in riffles, no further comparison between habitat types at tagging than fall migrants or nondetected fish (P < 0.01; was conducted. In all 4 years, a small number of fish (20–50) 68.6 mm). were tagged in one stream and later detected emigrating from another (Table 5). The majority of these (86–90%) were fish originally tagged in East Twin and then detected in West Twin. Watershed-Scale Survival Depending upon the year, 20–68% of the fish moving from one Among all years, survival to out-migration (corrected for stream to another were last detected emigrating in the fall (i.e., efficiency) ranged from 31% to 37% for the East Twin and 28% to 40% among years for the West Twin River (Table 4). Downloaded by [Department Of Fisheries] at 20:29 25 September 2012 TABLE 3. Median coho salmon migration date for fall (1 October to 31 When only spring migrants were considered, survival ranged December), winter–spring (1 January to 30 June), and all migrants (1 October from 5% to 15% in the East Twin and 8% to 15% for the to 30 June) for East and West Twin rivers from October 2005 to June 2009. West Twin River. Spring migrants made up less than 50% of all fish detected in all years except 2007 in the West Twin River Season 2005–2006 2006–2007 2007–2008 2008–2009 (Table 4). Monthly combined efficiency of antenna arrays varied East Twin River from 80% to 100% in East Twin for all years and from 37% to Oct to Dec 4 Nov 5 Nov 23 Oct 8 Nov 92% from 2005 to early 2007 for the West Twin. After moving Jan to Jun 23 May 6 May 25 May 17 May the West Twin arrays upstream a short distance to a more stable, Oct to Jun 21 Dec 5 Nov 4 Dec 10 Nov less flood-prone site in late March 2007, the combined efficiency exceeded 82% and was often 95% or higher (Table 6). Despite West Twin River some differences in array efficiency, no significant difference Oct to Dec 30 Oct 5 Nov 23 Oct 8 Nov was found between rivers in corrected (paired t-test: t = 0.44, Jan to Jun 23 May 20 May 27 May 17 May P = 0.68) or uncorrected (t = 1.73, P = 0.18) annual survival Oct to Jun 18 Nov 6 Nov 25 Jan 15 Nov estimates. MIGRATION, GROWTH, AND SURVIVAL OF JUVENILE SALMON 897 Downloaded by [Department Of Fisheries] at 20:29 25 September 2012

FIGURE 4. Number of tagged juvenile coho salmon detected out-migrating by day (black bars) and mean daily ﬂow (dashed line) from September through June in East Twin and West Twin rivers from 2005 to 2009. [Figure available online in color.]

The Relative Importance of Body Size, Competition, and nile coho salmon density instead of length (wAICc of 1.00 and Habitat on Survival, Residence Time, and Growth AICc of 11.43) (Table 7). The two best fitting models revealed Survival from tagging to out-migration varied among years a negative correlation between river kilometer and juvenile coho and as a function of tag year, river, river kilometer, and fish length salmon survival (detection) while tag year and fish length were at tagging (Table 7; Figures 5, 6). All variables had significance positively correlated with survival (Table 8). River and tag year values less than 0.001. The best fitting model for survival was were also significant factors in the models with survival varying significantly better than the next model, which included juve- among years and East Twin River having higher survival. 898 RONI ET AL.

TABLE 4. Number of tagged coho salmon detected (Det.) out-migrating during the fall (October to December) and spring (January through June) from East and West Twin rivers between 2005 and 2009, total number out-migrating after correcting for antenna array efficiency (Cor.), and estimated survival from tagging until final detection (parr-to-smolt survival). Fall and spring survival represent the total proportion of all tagged fish for that year that emigrated in the fall or spring.

Number of ﬁsh Survival Tag year Detected or corrected Fall Spring All Number tagged Fall Spring All East Twin River 2005 Det. 507 382 889 Cor. 570 393 962 3,120 0.18 0.13 0.31 2006 Det. 655 125 780 Cor. 718 138 856 2,511 0.29 0.05 0.34 2007 Det. 324 235 559 Cor. 342 242 584 1,657 0.21 0.15 0.35 2008 Det. 610 181 791 Cor. 669 185 854 2,298 0.29 0.08 0.37 West Twin River 2005 Det. 319 227 546 Cor. 743 330 1,073 3,033 0.24 0.11 0.35 2006 Det. 272 183 455 Cor. 509 204 713 2,496 0.20 0.08 0.29 2007 Det. 151 195 346 Cor. 166 198 364 1,286 0.13 0.15 0.28 2008 Det. 602 266 868 Cor. 629 271 900 2,271 0.28 0.12 0.40

Freshwater Residence fish and those in deeper habitats having a longer residence time. The number of days spent in freshwater before last detection Coho salmon density was negatively related to residence time, (emigration to sea), which we refer to as freshwater residence though it appeared in only one of the best models (Tables 7, 8). time, ranged from 2 to 305 d. The best fitting models revealed that river, river kilometer, coho length, habitat depth and coho Growth density were all related to freshwater residence (Tables 7, 8). Average growth from tagging to capture in smolt traps the Location at tagging (rkm) and length were factors in all the following spring ranged from 0.10 to 0.12 mm/d in the East Twin models, while river, depth, and density could be substituted for and 0.10 to 0.14 mm/d in the West Twin River (Table 9). The one another (Table 7). Residence time was positively related to daily coho salmon growth rate was primarily a function of tag tagging location, with fish further upstream in the watershed year, river, habitat depth, and relative length (Table 7). The best having a longer residence time. Body size (length) and habitat fitting model explained a large portion of the variation (wAICc depth were also positively related to residence time, with larger of 0.90) relative to the next best models (wAICc of 0.04 and 0.02,

Downloaded by [Department Of Fisheries] at 20:29 25 September 2012 respectively) (Table 7). Tag year was an important determinant TABLE 5. Number of tagged juvenile coho salmon that moved from one of growth rate, with later years having higher growth rates. The stream to another, the proportion of those that moved from East Twin to West relative length of juvenile coho at tagging was negatively related Twin River, and the proportion that were last detected out-migrating from either to growth, suggesting that longer fish grew more slowly than stream in the fall (fall migrants). their conspecifics. Habitat depth was also negatively correlated Proportion of with growth rate, suggesting that fish in deeper habitats grew fall migrants more slowly (Table 8). Proportion moving from East West Year Number East to West Twin Twin Twin DISCUSSION 2005 25 0.88 0.23 0.00 Out-migration 2006 21 0.90 0.68 0.50 Our results confirm that large numbers of juvenile coho 2007 29 0.86 0.68 0.00 salmon migrate to sea in the fall. The tagging and continu- 2008 50 0.86 0.70 0.57 ous monitoring of more than 18,500 juvenile coho salmon in East and West Twin rivers over a 4-year period demonstrated MIGRATION, GROWTH, AND SURVIVAL OF JUVENILE SALMON 899

to overwinter. While we are unsure of the eventual fate of these fall migrants, initial adult returns from our ongoing tagging efforts in our study streams suggest that more than 20% of the returning jacks and adults were fall migrants.

Size and Out-Migration Our results indicate that juvenile coho salmon size in late summer determines in part whether fish survive and whether they migrate to sea in the fall or spring. This is consistent with other studies that have reported juvenile coho salmon size in late summer or fall is an important determinant of overwinter survival and also of whether fish move from summer to winter habitats in other parts of the watershed (Quinn and Peterson 1996; Ebersole et al. 2006; Pess et al. 2011). Our results consistently showed that fish that emigrated in fall were on average smaller at tagging than those remaining in the stream and emigrating in the spring as smolts. In contrast, earlier work on coho salmon movement during the summer indicates that the largest fish are the most mobile and obtain a growth advantage from this movement (Kahler et al. 2001). Location within the watershed was also an important determinant of timing; fish tagged higher in the watershed emigrated at a later date. Both Quinn and Peterson (1996) and Ebersole et al. (2006) found location in the watershed to be an important determinant of overwinter survival. However, Ebersole et al. (2006) found increased survival in tributaries and Quinn and Peterson (1996) found higher survival in or near a lake. The main driver of fall emigration is unclear as the daily number of migrants in our study was not strongly correlated with flow in all years. Evidence from other studies suggests that downstream movement of fry and parr is forced migration related to increased flow or a change in water temperature (Bilby FIGURE 5. Mean size at tagging and 95% confidence intervals of undetected and Bisson 1987; Pearsons et al. 1992; Giannico and Healey and detected PIT-tagged juvenile coho salmon out-migrating in the fall and fish detected out-migrating in the spring for tagging years 2005 to 2008 in East Twin 1998). However, in other cases coho salmon fry and smolt mi- and West Twin rivers. gration timing was only partially related to stream discharge (Hartman et al. 1982; Sandercock 1991). Hartman et al. (1982) that 44–84% of coho salmon parr migrate to sea in fall and early found that falling temperatures, rainfall, or both initiated mi- winter each year. Bennett (2006) first observed this phenomenon gration of coho salmon fry to sea in Carnation Creek, British in a pilot study in East Twin River in 2004, but it was not clear Columbia. Out-migration in our study did not appear to be con-

Downloaded by [Department Of Fisheries] at 20:29 25 September 2012 if it was an anomaly or common in other nearby streams. The sistently influenced by falling temperatures. For example, in fact that we observed this over multiple years and in two rivers some years fish began emigrating when the temperature was suggests that it may be a common juvenile coho salmon life his- increasing while in other years fish began migrating when the tory in the region. Moreover, tagging in 2010 in another nearby temperature was decreasing or was stable. Moreover, tempera- stream showed the same pattern of fall out-migration. Several ture tracked flow in both study streams. The initiation and peak studies have reported coho salmon fry migrating to the marine of fall migration within our study streams was consistent among environment in the spring (see Koski 2009 for a detailed re- years with about half the fish emigrating by the first week of view) but have not examined or reported fall migration to sea. November. This suggests some factors other than flow or tem- Other studies have documented juvenile coho salmon migrat- perature are influencing emigration timing. Given the consistent ing to estuarine habitats in summer or fall and then returning to timing from year to year, it may be the photoperiod, which is freshwater to overwinter before emigrating as smolts (Crone and an important factor affecting the onset of smolt migrations for Bond 1976; Tschaplinski 1982; Miller and Sadro 2003; Koski many salmon (Quinn 2005). 2009). While a small number of fish in our study demonstrated Others have suggested that fry or parr migration results from this movement pattern or life history, the vast majority of fish limitations on or competition for food or space (Chapman 1962; that emigrated to sea in the fall did not ascend our study streams Koski 2009). The fact that coho salmon density was negatively 900 RONI ET AL.

TABLE 6. Monthly in situ combined efficiency of PIT tag arrays by stream and tag year. The location of antenna arrays on West Twin was changed in the spring of 2007 because of flood damage and poor read efficiency in the fall of 2005. Blank cells indicate that no fish were detected or that there were not enough detections to calculate combined efficiency.

East Twin West Twin Month 2005–2006 2006–2007 2007–2008 2008–2009 2005–2006 2006–2007 2007–2008 2008–2009 Sep 0.98 1.00 0.75 0.59 1.00 Oct 0.88 1.00 0.94 1.00 0.39 0.78 0.90 0.91 Nov 0.92 0.91 0.98 0.91 0.45 0.52 1.00 0.95 Dec 0.80 0.86 0.94 0.95 0.67 0.65 0.89 1.00 Jan 0.95 0.92 0.89 0.95 0.38 0.95 0.96 Feb 0.90 0.90 0.89 1.00 0.37 0.44 1.00 1.00 Mar 0.87 1.00 1.00 0.95 0.47 0.82 0.84 1.00 Apr 1.00 1.00 0.91 0.96 0.92 0.92 1.00 0.98 May 0.99 0.84 0.99 1.00 0.84 0.99 1.00 1.00 Jun 1.00 0.97 0.99 1.00 0.90 1.00 1.00 1.00

correlated with residence time in one of our regressions models gration timing. One way of testing the relative influence of these provides some evidence that competition for space is also a fac- factors would be to improve the quality and quantity of over- tor in our study streams. However, density was correlated with winter habitat through restoration or addition of food resources, river kilometer in some years but not in others. Thus, density such as salmon carcasses, or nutrients. Both additional habitat was not a consistent predictor of residence time or of whether a improvement and nutrient additions are proposed for East Twin fish emigrated in the fall or spring in our study. Fish size appears River, which may allow us to test the influence of these factors to be a consistent factor in determining whether a fish emigrates in the near future. in the fall or spring. Fish size is influenced by food, temperature, and flow and is strongly correlated with density (Roni and Survival Quinn 2001; Giannico and Hinch 2003). Thus, it is most likely Estimates of overwinter survival are typically based on some combination of temperature, flow, habitat conditions, and smolt trapping during spring out-migration (typically March competition for space and food that influences fish size and mi- to June). These efforts have either not measured fall or winter

TABLE 7. Model selection results for estimating juvenile coho salmon survival, freshwater residence time, and growth in East Twin and West Twin rivers.

Model Parameters Log-likelihood AICc AICc wAICc Survival Tag year, River, Rkm, Length 4 –5,787.64 11,585.28 0.00 1.00

Downloaded by [Department Of Fisheries] at 20:29 25 September 2012 Tag year, River, Rkm, Coho density 4 –5,793.35 11,596.71 11.43 0.00 Freshwater Residence River, Rkm, Length 4 –17,752.81 35,515.64 0.00 0.30 River, Rkm, Length, Coho density 5 –17,752.13 35,516.27 0.63 0.22 River, Rkm, Length, Habitat depth 5 –17,752.33 35,516.68 1.04 0.18 Tag year, River, Rkm, Length 5 –17,752.81 35,517.64 2.00 0.11 River, Length 3 –17,755.27 35,518.55 2.91 0.07 River, Length, Habitat depth 4 –17,755.00 35,520.01 4.37 0.03 River, Length, Coho density 4 –17,755.25 35,520.51 4.87 0.03 Tag year, River, Length 4 –17,755.26 35,520.54 4.90 0.03 Growth Tag year, Habitat depth, Relative length 4 683.52 –1,356.89 0.00 0.90 River, Tag year, Relative length 4 680.44 –1,350.72 6.17 0.04 Tag year, Relative length 3 678.85 –1,349.61 7.28 0.02 MIGRATION, GROWTH, AND SURVIVAL OF JUVENILE SALMON 901 Downloaded by [Department Of Fisheries] at 20:29 25 September 2012

FIGURE 6. Number of ﬁsh tagged and survival (detected) by location for East Twin and West Twin rivers for all years combined. [Figure available online in color.]

out-migration, have ignored these data, or have assumed fish migrants to the sea. Our overwinter survival estimates using de- out-migrating prior to spring were seeking downstream fresh- tections of fish out-migrating from October to June were well water rearing habitats. Our study is unique in that our PIT tag within the range of 11–78% recorded by other studies (Quinn arrays were located near tidewater and operated continuously, and Peterson 1996; Brakensiek and Hankin 2007; Pess et al. which allowed us to estimate the number of both fall and spring 2011). However, when we exclude all the fish detected before 902 RONI ET AL.

TABLE 8. Maximum-likelihood estimates of parameters for the best approximating models for predicting coho salmon survival, freshwater residence time, and growth in the East Twin and West Twin rivers. Standard errors are in parentheses.

Independent variables Dependent variable Tag year River River kilometer Coho length Coho density Habitat depth Deviance df Survival 1 0.2312 –0.4114 –0.0421 0.01557 11,571 9,309 (0.0253) (0.0468) (0.00381) (0.0029) Residence 1 14.6916 0.5067 2.8037 17,276,086 3,093 (2.8402) (0.2285) (0.1743) Residence 2 14.3215 0.62 2.761 –1.916 17,268,427 3,092 (2.8575) (0.2482) (0.1780) (1.6357) Residence 3 14.2997 0.5309 2.7729 4.2572 17,270,600 3,092 (2.8681) (0.2299) (0.1771) (4.3350) Residence 4 –0.1673 14.7384 0.5063 2.8039 17,276,021 3,092 (1.5550) (2.8737) (0.2286) (0.1743) Residence 5 12.828 2.8053 17,303,547 3,094 (2.7147) (0.1744) Growth 1 0.0154 –0.068a –0.0166 0.30687 326 (0.0023) (0.0125) (0.0054)

a Relative length was used in growth model.

January, overwinter survival of spring migrants ranged from 5% fall migrants as we had no means to recapture fall migrants to 15%. This strongly suggests that monitoring out-migration in saltwater during the spring. Similarly, we could not directly only during the spring may underestimate the number of juve- compare survival of fall and spring migrants without know- nile coho salmon smolts or migrants entering saltwater, under- ing whether fish were fall or spring migrants at the time of estimate overwinter survival, and overestimate smolt-to-adult tagging. survival. It was not possible to calculate overwinter survival of Even in the relatively short coastal streams that we examined, both the watershed and the location within the watershed are important determinants of survival. Location at tagging had a small TABLE 9. Mean growth of coho salmon from tagging to recapture in the but significant negative effect on survival but was positively cor- smolt trap by year and stream. related with residence time, suggesting that fish higher in the Mean daily growth watershed emigrated later and had slightly lower survival. Loca- Mean growth (mm) (mm/d) tions further upstream also may reflect greater exposure to predation as fish travel downstream. Watershed location has been West West shown to be an important determinant of coho salmon survival, Tag year East Twin Twin East Twin Twin though differences in survival have most often been reported between main-stem and off-channel habitats or tributaries (Eber- 2005 sole et al. 2006; Brakensiek and Hankin 2007: Pess et al. 2011). Mean growth 30.5 30.6 0.11 0.11

Downloaded by [Department Of Fisheries] at 20:29 25 September 2012 Some studies have shown a positive relationship between dis- SD 9.5 6.7 0.03 0.02 tance from the estuary and survival (Quinn and Peterson 1996). N 200 168 200 168 Unlike streams examined by Quinn and Peterson (1996), Solazzi 2006 et al. (2000), and Ebersole et al. (2006), the streams we examined Mean growth 31.5 33.5 0.10 0.11 had little off-channel rearing habitat and few tributaries with the SD 12.8 11.1 0.03 0.03 potential to support overwintering juvenile coho salmon. N 73 45 73 45 The fact that our models indicated that survival differed be- 2007 tween East Twin and West Twin rivers was not unexpected, given Mean growth 31.6 29.1 0.11 0.10 results from other studies and some small but important differ- SD 9.9 8.4 0.02 0.03 ences between the two watersheds. First, differences in smolt N 95 101 95 101 production and survival typically exist even between nearby 2008 watersheds (e.g., Sharma 1998; Solazzi et al. 2000; Johnson Mean growth 33.6 38.5 0.12 0.14 et al. 2005) and ﬁsh densities and lengths differed between the SD 8.5 12.8 0.03 0.04 two rivers we examined. Moreover, both rivers are part of a N 66 108 66 108 long-term watershed restoration and monitoring program. Prior MIGRATION, GROWTH, AND SURVIVAL OF JUVENILE SALMON 903

to our study, restoration efforts including riparian planting and migrants that reared in the estuarine habitat during the summer LWD placement occurred in the East Twin but not the West and went to sea in fall. Twin River. Thus, one might expect some differences in habitat There has been considerable debate as to the fate of fry or conditions between the two rivers. Unfortunately, the restora- parr migrants. Koski (2009) reviewed the evidence on the fate of tion work was largely completed before our study began, and spring fry migrants and suggested that fry migrants represented we did not have data on survival before restoration was imple- a successful life history that contribute to returning adults. He mented. In addition, the differences in survival between the two based this primarily on evidence that some age-0 migrants return streams were relatively small and not readily apparent when to their natal streams to overwinter or move into nearby streams measured at a watershed scale (Table 4). Finally, because it was with suitable overwinter habitat. However, most studies have only possible to correct for antenna efficiency for groups of fish reported age-0 coho salmon (fry or parr) migrating to the estu- that migrated over a period of weeks or a month, we could not arine habitat in spring and returning to the stream to overwinter correct for antenna efficiency for the individually tagged fish (Murphy et al. 1997; Miller and Sadro 2003; Koski 2009). In used in our regression models. This could in part explain why contrast to these studies, we detected large numbers of juvenile river was an important determinant of survival in our individual- coho salmon moving to sea in the fall. Unfortunately, we did based regression models, but no differences in annual survival not have data on fry migrants because they are too small to PIT were detected at a watershed-scale. tag. Moreover, few fry were detected in smolt traps operated from April until June. This suggests that fry migration is not a common life history type in East Twin and West Twin rivers or Growth that the smolt traps are not effective at capturing fry or possibly Overwinter growth data were available for only a subsample that fry migrate to sea prior to installation of smolt traps. of fish tagged and varied primarily by year and relative length Similar to that reported by Scrivener et al. (1998) in Carnation and slightly by habitat depth. Relative length provides an Creek, we observed that some fish classified as spring migrants indication of how large the fish are in relation to conspecifics. (4–12%) had actually moved into the tidal areas of the stream Our findings on growth suggest that fish that were large in or into the marine environment in the fall before returning to relation to other fish within the same habitat grew more slowly, freshwater to overwinter. We also found that some juvenile coho which is consistent with work on other salmonids (Brett 1976, salmon moved through the marine environment from East Twin 1979; Achord et al. 2007). Habitat depth may be an indicator to West Twin to overwinter. This is also consistent with other of pool or glide quality, so it is not clear why depth was studies that report some age-0 coho salmon moving into saltwa- negatively associated with growth. However, habitat depth was ter and then ascending nearby streams to overwinter (Hartman positively correlated with fish length (Pearson’s correlation = and Scrivener 1990; Miller and Sadro 2003; Koski 2009). It is 0.16, P < 0.01) and larger fish grew more slowly. Several not clear why fish moved primarily from East Twin to West studies have demonstrated that fish that move into tributaries, Twin, as there appears to be less overwinter rearing habitat in off-channel habitats, or even estuaries have higher growth the latter. It may be that fish are pushed westward with prevail- than fish that do not migrate (e.g., Peterson 1982; Miller and ing currents in the Strait of Juan de Fuca and thus West Twin is Sadro 2003; Ebersole et al. 2006). We had data only for fish the first stream that fish from the East Twin encounter. Another that were recaptured during spring smolt trapping and we did PIT tag reader was recently installed in Deep Creek, a stream not have data on fish out-migrating at other times of the year. flowing into the Strait of Juan de Fuca 4–5 kilometers west of Additional trapping of fish in freshwater, intertidal, and marine the West Twin River. This provides some evidence to support environments is needed to adequately determine whether the this theory, since fish tagged in both East Twin and West Twin fall out-migrants we observed obtain some growth advantage Downloaded by [Department Of Fisheries] at 20:29 25 September 2012 rivers have been found in Deep Creek, but no fish tagged in from entering the marine environment. Deep Creek have been detected in the East Twin or West Twin rivers. Juvenile Coho Salmon Life Histories Two other common juvenile salmon life histories are fall Our data suggest that juvenile coho salmon migration is more movement to off-channel habitats or juveniles spending mul- complex than previously reported. There may be at least four tiple years in freshwater (Sandercock 1991). The movement major juvenile life histories present in the East Twin and West of coho salmon parr upstream or downstream into tributaries Twin rivers: fall migrants that go to sea, fall migrants that go to or floodplain habitats to overwinter has been well documented sea but return to the same stream to overwinter, fall migrants that (e.g., Skeesick 1970; Cederholm and Scarlett 1982; Peterson go to sea only to move to nearby streams to overwinter, and fish 1982; Ebersole et al. 2006). This may also occur in East Twin that remain in freshwater in fall and winter and migrate only as and West Twin rivers, but our study was not designed to monitor smolts in the spring. The presence of four or more juvenile coho movement among habitats within a watershed. Moreover, rela- salmon life histories is supported by other research. Scrivener tively little off-channel habitat exists in our study streams. We et al. (1998) reported five juvenile coho salmon life histories also did not detect any coho salmon 2 years or older. However, in Carnation Creek, British Columbia, including age-0 or fry in more northern latitudes, many juvenile coho salmon rear in 904 RONI ET AL.

freshwater for 2 years before smolting, with some populations tality of less than a tenth of one percent for Chinook salmon in Alaska and Kamchatka rearing 3 or 4 years in freshwater and none for coho salmon after holding fish for 24 h. More- (Sandercock 1991). over, Peterson et al. (1994) found no effect of PIT tags versus When one examines the various combinations of freshwa- coded wire tags on juvenile coho salmon overwinter growth or ter, estuarine, and marine residence observed in this and other survival. Therefore, while these studies and our methods do not studies, it is clear that juvenile coho salmon life history is much rule out the possibility that PIT tagging juvenile coho salmon more complex than previously reported (e.g., Sandercock 1991; may have influenced growth, movement, and survival in our Quinn 2005). It is likely that any one stream or population may study, they suggest that it is very unlikely. include several life histories. Insight into the fate of these different freshwater life histories and their relative contribution to re- Summary turning adults is an important research need, which will require In summary, we found that in most years over 50% of juvenile sampling additional streams and estuarine habitats. While long- coho salmon emigrated to the marine environment in the fall. term monitoring of returning PIT-tagged adults will shed some This life history or migration pattern has not been reported in light on this, the collection of otoliths from both out-migrants other populations. These findings, combined with other studies, and returning adults and the use of strontium or strontium-to- suggest that juvenile coho salmon life history diversity is much calcium ratios may provide additional insight into the fate and greater than commonly believed, and fall emigration to sea is residence patterns of different coho salmon life history types a common behavior in coho salmon. Survival of coho salmon (Zimmerman 2005; Koski 2009). from tagging to migration to the sea in our study was largely dependent upon fish size in late summer, with larger fish having Limitations of PIT Tag Readers and Arrays higher survival and being more likely to emigrate in the spring. Studies using PIT tags and other tagging equipment are lim- Residence time in freshwater was similarly dependent on fish ited by their efficiency at detecting fish under a variety of condi- length but also on distance from the marine environment, with tions. The monthly combined in situ detection efficiency of the larger fish and those further upstream having the longest fresh- East Twin PIT tag arrays ranged from 80% to 92% for all years. water residence time. Long-term research is needed to better Prior to reconfiguration of the West Twin PIT tag arrays, the understand the contribution of fall coho salmon migrants to the monthly efficiency at West Twin ranged from 37% to 90% from adult population and to determine whether fall migrants remain October 2005 to February 2007. Following the reconfiguration, in the marine or estuarine environment or enter other streams to the monthly efficiency ranged from 82% to 100% from March overwinter. If juvenile coho salmon that migrate to sea in peri- 2007 to July 2009. This difference in reader efficiency between ods other than the spring smolt migration survive and contribute streams may have influenced our findings though the detection to the adult population, traditional estimates of survival based efficiency was nearly identical in the two streams from March on spring trapping likely underestimate smolt production and 2007 to the completion of the study in July 2009. While com- overwinter survival. bined efficiency of the two arrays (rows of antennas) at each site was fairly high (Table 6), the efficiency of an individual array ACKNOWLEDGMENTS varied from 30% to 90%. This variation prevented us from de- We thank all those who assisted with fish collection and termining upstream and downstream direction of individual fish tagging; the Washington Department of Ecology and NOAA– movement. Frequent flooding in November and early December, Fisheries for providing funding; and Jason Hall for develop- and the subsequent movement of bed load and stream channels, ing Figure 1. We also thank the three anonymous reviewers, made maintaining our equipment very difficult, which likely who provided helpful comments on earlier versions of this Downloaded by [Department Of Fisheries] at 20:29 25 September 2012 reduced detection efficiency for short periods of time. These manuscript. flow events partially overlapped with fall emigration (Figure 4). Therefore, it is likely that the number of fall out-migrants detected and our estimates of survival are conservative. REFERENCES Our study assumed that growth, movement, and survival of Achord, S., B. P. Sandford, E. E. Hockersmith, M. G. Nesbit, N. D. Dumdei, J. L. PIT-tagged fish were equal to that of untagged fish, which we Lamb, K. W. McIntyre, N. N. Paasch, and J. G. Williams. 2010. Monitoring feel is a reasonable assumption for several reasons. First, we the migrations of wild Snake River spring/summer Chinook salmon juveniles, 2008–2009. Report to the Bonneville Power Administration, Contract had very low tagging and handling mortality: only 14 out of 004105, Project 1991-028-00, Portland, Oregon. the 18,642 coho salmon tagged during our study died during Achord, S., R. W. Zabel, and B. P. Sandford. 2007. Migration timing, growth, tagging. It should be noted that we only held fish for 15–45 min and estimated parr-to-smolt survival rates of wild Snake River spring-summer following tagging before releasing them back into the stream. Chinook salmon from the Salmon River basin, Idaho, to the lower Snake However, we used methods nearly identical to Achord et al. River. Transactions of the American Fisheries Society 136:142–154. Agresti, A. 1996. An introduction to categorical data analysis. Wiley, New York. (2010) and Bennett et al. (2011) who examined delayed mor- Bennett, T. 2006. Movement, growth, and survival of juvenile coho salmon tality of Chinook salmon Oncorhynchus tshawytscha and coho and trout in the East Twin River, Washington. Master’s thesis. University of salmon following PIT tagging and found delayed tagging mor- Washington, Seattle. MIGRATION, GROWTH, AND SURVIVAL OF JUVENILE SALMON 905

Bennett, T., R. Wissmar, and P. Roni. 2011. Fall and spring emigration timing of Gowan, C., M. K. Young, K. D. Fausch, and S. C. Riley. 1994. Restricted juvenile coho salmon from East Twin River, Washington. Northwest Science movement in resident stream salmonids: a paradigm lost? Canadian Journal 85:562–570. of Fisheries and Aquatic Sciences 51:2626–2637. Bilby, R. E., and P.A. Bisson. 1987. Emigration and production of hatchery coho Hartman, G. F., B. C. Andersen, and J. C. Scrivener. 1982. Seaward movement salmon (Oncorhynchus kisutch) stocked in streams draining an old-growth of coho salmon (Oncorhynchus kisutch) fry in Carnation Creek, an unstable and clear-cut watershed. Canadian Journal of Fisheries and Aquatic Sciences coastal stream in British Columbia. Canadian Journal of Fisheries and Aquatic 44:1397–1407. Sciences 39:588–597. Bilby, R. E., W. Ehinger, T. P. Quinn, G. Volkhardt, K. Krueger, D. E. Seiler, Hartman, G. F., and J. C. Scrivener. 1990. Impacts of forestry practices on G. R. Pess, C. E. Jordan, M. McHenry, and D. Poon. 2005. Study evaluates a coastal stream ecosystem, Carnation Creek, British Columbia. Canadian fish response to management actions. Western Forester 50(2):14–15. Bulletin of Fisheries and Aquatic Sciences 223. Brakensiek, K. E., and D. G. Hankin. 2007. Estimating overwinter survival of Holtby, L. B., B. C. Andersen, and R. K. Kadowaki. 1990. Importance of smolt juvenile coho salmon in a northern California stream: accounting for effects of size and early ocean growth to interannual variability in marine survival passive integrated transponder tagging mortality and size-dependent survival. of coho salmon (Oncorhynchus kisutch). Canadian Journal of Fisheries and Transactions of the American Fisheries Society 136:1423–1437. Aquatic Sciences 47:2181–2194. Bramblett, R. G., M. D. Bryant, B. E. Wright, and R. G. White. 2002. Sea- Johnson, S. L., J. D. Rodgers, M. F. Solazzi, and T. E. Nickelson. 2005. Effects sonal use of small tributary and main-stem habitats by juvenile steelhead, of an increase in large wood on abundance and survival of juvenile salmonids coho salmon, and Dolly Varden in a southeastern Alaska drainage basin. (Oncorhynchus spp.) in an Oregon coastal stream. Canadian Journal of Fish- Transactions of the American Fisheries Society 131:498–506. eries and Aquatic Sciences 62:412–424. Brett, J. R. 1976. Scope for metabolism and growth of sockeye salmon, On- Kahler, T. H., P. Roni, and T. P. Quinn. 2001. Summer movement and growth corhynchus nerka, and some related energetic. Journal of the Fisheries Re- of juvenile anadromous salmonids in small western Washington streams. search Board of Canada 33:307–313. Canadian Journal of Fisheries and Aquatic Sciences 58:1947–1956. Brett, J. R. 1979. Environmental factors and growth. Pages 599–675 in W. S. Koski, K. V. 2009. The fate of coho salmon nomads: the story of an estuarine- Hoar, D. J. Randall, and J. R. Brett, editors. Fish physiology, volume 8. rearing strategy promoting resilience. Ecology and Societyy [online serial] Academic Press, New York. 14(1):article 4. Available: www.ecologyandsociety.org/vol14/iss1/art4./ (Jan- Burnham, K. P., and D. R. Anderson. 2002. Model selection and inference: a uary 2011). practical information theoretical approach. Springer-Verlag, New York. Martin, D. J., L. J. Wasserman, R. P. Jones, and E. O. Salo. 1984. Effects of Burnham, K. P., D. R. Anderson, G. C. White, C. Brownie, and K. H. Pol- Mount St. Helens eruption on salmon populations and habitat in the Toutle lock. 1987. Design and analysis methods for fish survival experiments based River. U.S. Department of the Interior, Bureau of Reclamation, Technical on release-recapture. American Fisheries Society, Monograph 5, Bethesda, Completion Report, Grant 14-34-0001-1418, Washington, D.C. Maryland. Miller, B. A., and S. Sadro. 2003. Residence time and seasonal movements Carle, F. L., and M. R. Strub. 1978. A new method for estimating population of juvenile coho salmon in the ecotone and lower estuary of Winchester size from removal data. Biometrics 34:621–630. Creek, South Slough, Oregon. Transactions of the American Fisheries Society Cederholm, C. J., R. E. Bilby, P. A. Bisson, T. W. Bumstead, B. R. Fransen, 132:546–559. W. J. Scarlett, and J. W. Ward. 1997. Response of juvenile coho salmon and Murphy, M. L., K. V. Koski, J. M. Lorenz, and J. F. Thedinga. 1997. Down- steelhead to placement of large woody debris in a coastal Washington stream. stream migrations of juvenile pacific salmon (Oncorhynchus spp.) in a glacial North American Journal of Fisheries Management 17:947–963. transboundary river. Canadian Journal of Fisheries and Aquatic Sciences Cederholm, C. J., and L. M. Reid. 1987. Impact of forest management on 54:2837–2846. coho salmon (Oncorhynchus kisutch) populations of the Clearwater River, Olympic National Forest. 2002. Deep Creek and East and West Twin rivers Washington: a project summary. Pages 373–398 in E. O. Salo and T. W. watershed analysis. Olympic National Forest, Olympia, Washington. Cundy, editors. Streamside management: forestry and fishery interactions. Pearsons, T. N., H. W. Li, and G. A. Lamberti. 1992. Influence of habitat com- Institute of Forest Resources, University of Washington, Seattle. plexity on resistance to flooding and resilience of stream fish assemblages. Cederholm, C. J., and W. J. Scarlett. 1982. Seasonal immigrations of juvenile Transactions of the American Fisheries Society 121:427–436. salmonids into four small tributaries of the Clearwater River, Washington, Pess, G. R., P. M. Kiffney, M. C. Liermann, T. R. Bennett, J. H. Anderson, and 1977–1981. Pages 98–110 in E. L. Brannon and E. O. Salo, editors. Pro- T. P. Quinn. 2011. The influences of body size, habitat quality, and competi- ceedings of the salmon and trout migratory behavior symposium. School of tion on the movement and survival of juvenile coho salmon, Oncorhynchus Fisheries, University of Washington, Seattle. kisutch, during the early stages of stream recolonization. Transactions of the Chapman, D. W. 1962. Aggressive behavior in juvenile coho salmon as a cause American Fisheries Society 140:883–897.

Downloaded by [Department Of Fisheries] at 20:29 25 September 2012 of emigration. Journal of the Fisheries Research Board of Canada 19:1047– Peterson, N. P. 1982. Population characteristics of juvenile coho salmon (On- 1080. corhynchus kisutch) overwintering in riverine ponds. Canadian Journal of Crone, R. A., and C. E. Bond. 1976. Life history of coho salmon, Oncorhynchus Fisheries and Aquatic Sciences 39:1303–1307. kisutch, in Sashin Creek, southeastern Alaska. U.S. National Marine Fisheries Peterson, N. P., E. F. Prentice, and T. P. Quinn. 1994. Comparison of sequential Service Fishery Bulletin 74:897–923. coded-wire and passive integrated transponder tags for assessing overwinter Ebersole, J. L., P. J. Wigington Jr., J. P. Baker, M. A. Cairns, and M. R. Church. growth and survival of juvenile coho salmon. North American Journal of 2006. Juvenile coho salmon growth and survival across stream network sea- Fisheries Management 14:870–873. sonal habitats. Transactions of the American Fisheries Society 135:1681– Quinn, T. P. 2005. The behavior and ecology of Pacific salmon and trout. 1697. American Fisheries Society, Bethesda, Maryland. Giannico, G. R., and M. C. Healey. 1998. Effects of flow and food on winter Quinn, T. P., and N. P. Peterson. 1996. The influence of habitat complexity movements of juvenile coho salmon. Transactions of the American Fisheries and fish size on over-winter survival and growth of individually marked Society 127:645–651. juvenile coho salmon (Oncorhynchus kisutch) in Big Beef Creek, Washington. Giannico, G. R., and S. G. Hinch. 2003. The effect of wood and temperature Canadian Journal of Fisheries and Aquatic Sciences 53:1555–1564. on juvenile coho salmon winter movement, growth, density and survival in Roni, P., and T. P. Quinn. 2001. Density and size of juvenile salmonids in reside-channels. River Research and Applications 19:219–231. sponse to placement of large woody debris in western Oregon and Wash- Gilbert, C. H. 1914. Age at maturity of the Pacific salmon of the genus On- ington streams. Canadian Journal of Fisheries and Aquatic Sciences 58: corhynchus. U.S. Bureau of Fisheries Bulletin 32:1–22. 282–292. 906 RONI ET AL.

Sandercock, F. K. 1991. Life history of coho salmon (Oncorhynchus kisutch). Oregon streams. Canadian Journal of Fisheries and Aquatic Sciences 57:906– Pages 397–445 in C. Groot and L. Margolis, editors. Pacific salmon life 914. histories. University of British Columbia Press, Vancouver. Stevens, D. L., and A. R. Olsen. 2004. Spatially balanced sampling of natural Scarlett, W. J., and C. J. Cederholm. 1984. Juvenile coho salmon fall-winter resources. Journal of the American Statistical Association 99:262–278. utilization of two small tributaries of the Clearwater River, Jefferson County, Tschaplinski, P. J. 1982. Aspects of the population biology of estuary-reared Washington. Pages 227–242 in J. M. Walton and D. B. Houston, editors. Pro- and stream-reared juvenile coho salmon in Carnation Creek: a summary of ceedings of the Olympic wild fish conference (Port Angeles, 1983). Fisheries current research. Pages 289–307 in G. F. Hartman, editor. Proceedings of the Technology Program, Peninsula College, Port Angeles, Washington. Carnation Creek workshop: a ten-year review. Coordinating Committee of Scrivener, J. C., P. J. Tschaplinski, and J. S. Macdonald. 1998. An introduc- the Carnation Creek Watershed Project, Nanaimo, British Columbia. tion to the ecological complexity of salmonid life history strategies and of Tschaplinski, P.J., and G. F. Hartman. 1983. Winter distribution of juvenile coho forest harvesting impacts in coastal British Columbia. Pages 23–27 in D. salmon (Oncorhynchus kisutch) before and after logging in Carnation Creek, L. Hogan, P. J. Tschaplinski, and S. Chatwin, editors. Carnation Creek and British Columbia, and some implications for overwinter survival. Canadian Queen Charlotte Islands fish/forestry workshop: applying 20 years of coast Journal of Fisheries and Aquatic Sciences 40:452–461. research to management solutions. British Columbia Ministry of Forests, Wigington, P. J., Jr., J. L. Ebersole, M. E. Colvin, S. G. Leibowitz, B. Miller, B. Research Branch, Land Management Handbook 41, Victoria. Hansen, H. R. Lavigne, D. White, J. P. Baker, M. R. Church, J. R. Brooks, M. Sharma, R. 1998. Influence of habitat on smolt production in coho salmon A. Cairns, and J. E. Compton. 2006. Coho salmon dependence on intermittent (Oncorhynchus kisutch) in fourteen western Washington streams. Master’s streams. Frontiers in Ecology and the Environment 4:513–518. thesis. University of Washington, Seattle. Zimmerman, C. E. 2005. Relationship of otolith strontium-to-calcium ratios and Skeesick, D. G. 1970. The fall immigration of juvenile coho salmon (On- salinity: experimental validation for juvenile salmonids. Canadian Journal of corhynchus kisutch) into a small tributary. Research Reports of the Fish Fisheries and Aquatic Sciences 62:88–97. Commission of Oregon 2:90–95. Zydlewski, G. B., G. Horton, T. Dubreuil, B. Letcher, S. Casey, and J. Zydlewski. Solazzi, M. F., T. E. Nickelson, S. L. Johnson, and J. D. Rodgers. 2000. Effects 2006. Remote monitoring of fish in small streams: a unified approach using of increasing winter rearing habitat on abundance of salmonids in two coastal PIT tags. Fisheries 31:492–502. Downloaded by [Department Of Fisheries] at 20:29 25 September 2012 This article was downloaded by: [Department Of Fisheries] On: 25 September 2012, At: 20:29 Publisher: Taylor & Francis Informa Ltd Registered in England and Wales Registered Number: 1072954 Registered office: Mortimer House, 37-41 Mortimer Street, London W1T 3JH, UK

Transactions of the American Fisheries Society Publication details, including instructions for authors and subscription information: http://www.tandfonline.com/loi/utaf20 The Effect of Nonnative Salmonids on Social Dominance and Growth of Juvenile Atlantic Salmon Jessica A. van Zwol a , Bryan D. Neff a & Chris C. Wilson b a Department of Biology, University of Western Ontario, 1151 Richmond Street, London, Ontario, N6A 5B7, Canada b Aquatic Research and Development Section, Ontario Ministry of Natural Resources, Trent University, 2140 East Bank Drive, Peterborough, Ontario, K9J 7B8, Canada

Version of record first published: 20 Jun 2012.

To cite this article: Jessica A. van Zwol, Bryan D. Neff & Chris C. Wilson (2012): The Effect of Nonnative Salmonids on Social Dominance and Growth of Juvenile Atlantic Salmon, Transactions of the American Fisheries Society, 141:4, 907-918 To link to this article: http://dx.doi.org/10.1080/00028487.2012.675899

PLEASE SCROLL DOWN FOR ARTICLE

Full terms and conditions of use: http://www.tandfonline.com/page/terms-and-conditions 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. The publisher does not give any warranty express or implied or make any representation that the contents will be complete or accurate or up to date. The accuracy of any instructions, formulae, and drug doses should be independently verified with primary sources. The publisher shall not be liable for any loss, actions, claims, proceedings, demand, or costs or damages whatsoever or howsoever caused arising directly or indirectly in connection with or arising out of the use of this material. Transactions of the American Fisheries Society 141:907–918, 2012 C American Fisheries Society 2012 ISSN: 0002-8487 print / 1548-8659 online DOI: 10.1080/00028487.2012.675899

ARTICLE

The Effect of Nonnative Salmonids on Social Dominance and Growth of Juvenile Atlantic Salmon

Jessica A. Van Zwol and Bryan D. Neff* Department of Biology, University of Western Ontario, 1151 Richmond Street, London, Ontario N6A 5B7, Canada Chris C. Wilson Aquatic Research and Development Section, Ontario Ministry of Natural Resources, Trent University, 2140 East Bank Drive, Peterborough, Ontario K9J 7B8, Canada

Abstract Nonnative species have been shown to negatively impact the native community in which they are introduced. In the Great Lakes, competition with nonnative salmonids may be hindering the restoration efforts of Atlantic salmon Salmo salar, a once-native top predator in Lake Ontario. We examined the effects of brown trout S. trutta and rainbow trout Oncorhynchus mykiss, two nonnative ﬁshes in Lake Ontario, on the social dominance and growth rate of juvenile Atlantic salmon from three strains being used for reintroduction efforts in Lake Ontario. Using seminatural stream channels, we found that the presence of either rainbow trout or brown trout reduced aggression, dominance, and food consumption of the Atlantic salmon. Brown trout had the strongest effect, increasing aggression levels in the channels by a factor of two and sharply reducing the dominance of Atlantic salmon. When both nonnatives were present, initiated aggression by Atlantic salmon decreased by a factor of three and food consumption halved as compared with when the salmon were alone. Consequently, over a 7-d time period, standard growth rate of the Atlantic salmon dropped from no change in mass when alone to a value of –0.3% per day when with the nonnative species. Of the three strains tested, one strain was least affected by the nonnative trouts, implicating genetic differences among the strains and suggesting that one strain may have greater poststocking success in Lake Ontario tributaries with naturalized populations of brown trout and rainbow trout.

The introduction of nonnative species can negatively affect leaving native species to cope or risk extirpation (Ricciardi et al.

Downloaded by [Department Of Fisheries] at 20:29 25 September 2012 individuals through competition and displacement (Hamilton 1998). et al. 1999), as well as entire communities or ecosystems Among fishes, salmonids are among the most widely in- by altering productivity and nutrient cycling (D’Antonio and troduced species around the world (Crawford and Muir 2008). Vitousek 1992). Reductions in biodiversity and abundance of Indeed, Edge et al. (1993) and Dewald and Wilzbach (1992) biota within the ecosystem typically follow (D’Antonio and showed that native fishes fed less in the presence of brown trout Vitousek 1992; Olden et al. 2004). For example, establishment Salmo trutta, while Nakano et al. (1998) found that introduced of nonnative species like the zebra mussel Dreissena polymor- brook trout Salvelinus fontinalis shifted foraging frequency, pha and Eurasian water milfoil Myriophyllum spicatum have microhabitat selection, and reaction distances of native bull sharply decreased species diversity in impacted native commu- trout Salvelinus confluentus. Similarly, Kitano (2004) found nities with dramatic species loss (Madsen et al. 1991; Ricciardi that rainbow trout Oncorhynchus mykiss, brown trout, and et al. 1998). Nonnative species often overwhelm ecosystems brook trout have reduced native populations of whitespotted

*Corresponding author: [email protected] Received April 26, 2011; accepted January 18, 2012 Published online June 20, 2012 907 908 VANZWOLETAL.

char Salvelinus leucomaenis, Dolly Varden Salvelinus malma, juvenile life stage. Using seminatural stream environments, we masu salmon (also known as cherry salmon) Oncorhynchus observed agonistic and feeding behaviors of juvenile Atlantic masou, and Sakhalin taimen (also known as Japanese huchen) salmon in the presence of juvenile brown trout and rainbow Hucho perryi, either directly by predation, or indirectly by trout to determine if their presence hindered Atlantic salmon ag- competition for resources. In fact, rainbow trout and brown gression, food consumption, or growth. Although rainbow trout trout have had such widespread negative effects on native and brown trout represent taxa that are subdivided into major ecosystems that they have been listed among the top 100 of the lineages (e.g., Bernatchez 2001; Blankenship et al. 2011), our world’s worst invasive alien species (Lowe et al. 2000). study focuses on the impact of the naturalized populations of In the Great Lakes, nonnative salmonids may also be these species in the Laurentian Great Lakes watershed. We ex- impacting the restoration efforts of native fishes (Crawford amined the comparative performance of three different strains 2001). Specifically, Atlantic salmon Salmo salar was once a of Atlantic salmon, (LaHave River, Riviere` aux Saumons [Lac native top predator in Lake Ontario but was extirpated at the Saint-Jean], and Sebago Lake) that are being used as part of a end of the 19th century through a combination of habitat loss large-scale effort to reestablish Atlantic salmon in Lake Ontario and exploitation (MacCrimmon 1977). During the past century, (Grieg et al. 2003). Thus, we were able to look at population- there have been numerous attempts to restore this species, but specific genetic differences in Atlantic salmon behavior and a naturally reproducing population has yet to be established performance when in competition with the nonnative salmonids (Stanfield and Jones 2003). Conversely, nonnative rainbow and assess the potential importance of performance differences trout, brown trout, chinook salmon O. tshawytscha, coho salmon within and among strains for reestablishing this formerly native O. kisutch, and sockeye salmon O. nerka, have been routinely species in Lake Ontario. stocked in Lake Ontario tributaries to enhance recreational fisheries (Crawford 2001; Stewart and Schaner 2002). Rainbow trout and brown trout were both introduced starting in the METHODS early 1900s from different populations across North America Study species.—In this study, brood stocks were used from and developed naturalized populations (MacCrimmon 1977; three Atlantic salmon populations. Past restoration efforts have Crawford 2001; Kerr 2006). The presence of these nonnative focused on stocking only one Atlantic salmon strain originating salmonid species could be adversely affecting Atlantic salmon from the LaHave River in Nova Scotia (Stanfield and Jones restoration efforts (Grieg et al. 2003). Indeed, Scott et al. (2005) 2003; Dietrich et al. 2008). This strain was chosen primarily examined brief interactions (<1 d) between some of these because of its availability as a broodstock, rather than specific nonnative salmonids and Atlantic salmon and noted adverse ecological considerations (Grieg et al. 2003). The LaHave strain impacts on social behavior of juvenile Atlantic salmon (also of Atlantic salmon is anadromous, a life cycle thought to be see Scott et al. 2003). Other research revealed that interactions different from the original strain that inhabited Lake Ontario, with rainbow trout may heighten aggression, territoriality, and which may have spent its entire life cycle in freshwater (Blair competition for resources in stream because of niche overlap 1938; Parsons 1973). The freshwater Sebago Lake strain from with Atlantic salmon and, in some cases, these two species have Maine is both physiologically and (now) physically landlocked been considered ecological equivalents (Gibson 1981; Hearn (Ward 1932; Watts 1999). Finally, the Lac Saint-Jean strain and Kynard 1986). In Europe, brown trout coexist with Atlantic from Quebec lives entirely in freshwater much like the original salmon (Armstrong et al. 2003) and are the more aggressive and extirpated strain of Lake Ontario was believed to have been socially dominant of the two species (Stradmeyer et al. 2008). (Blair 1938; Gage 1963). Dominant individuals typically have preferential access to As juvenile mortality among salmonids is high (Elliott 1990;

Downloaded by [Department Of Fisheries] at 20:29 25 September 2012 resources, which can lead to increased growth and survivorship Good et al. 2001), restoration efforts in Lake Ontario stock (e.g., Ens and Goss-Custard 1984). In addition, in Lake Ontario various age-groups of Atlantic salmon including 1.5-year-old the presence of rainbow trout, brown trout, and other nonnative individuals. Our behavioral trials involved 1.5-year-old Atlantic salmonids increases species richness in the streams and lake salmon (N = 504), brown trout (N = 180), and rainbow trout which can increase competition for the same resources and (N = 180). All ﬁsh were reared from brood stocks established likely alters the carrying capacity for Atlantic salmon simply by the Ontario Ministry of Natural Resources (OMNR). LaHave through density effects (Crawford 2001). Atlantic salmon (N = 168) and brown trout were obtained from Understanding variation in behavior within a species is cru- the OMNR Harwood Fish Culture Station (Harwood, Ontario), cial for determining the role phenotypic differences play in while Lac Saint-Jean (N = 168) and Sebago Lake (N = 168) restoration efforts and for understanding the impact of non- Atlantic salmon and rainbow trout came from the OMNR Nor- native species (Curio 1996; Caro 1999). Knowledge derived mandale Fish Culture Station (Normandale, Ontario). Fish were from such behavioral studies can aid in creating effective man- of the same age and culture history as those routinely stocked in agement strategies for the establishment of a species. As such, streams feeding Lake Ontario. As such, the yearlings of the three we examined the potential effect that nonnative and ecolog- species differed in size (see below) as they do under local natural ically similar salmonids have on Atlantic salmon during the conditions. Prior to the start of the experiment, ﬁsh were held for EFFECTS OF NON-NATIVE SALMONIDS ON ATLANTIC SALMON 909

one month at the OMNR Codrington Fisheries Research Facil- of 0.95 cm. Tags were applied to either the left or right side of ity (Codrington, Ontario) in flow-through tanks with an average the fish just below the dorsal fin to ensure all the fish within each density of 0.6 fish/L, exposed to a natural light cycle, and fed channel could be uniquely identified. Between fish, the needle trout chow (Corey Aquafeeds, Fredericton, New Brunswick). was disinfected with hydrogen peroxide and rinsed with water. Experimental setup.—Seminatural streams were used to per- The fish were released into a flow-through holding tank to recu- form six behavioral trials in blocks between May and July 2009 perate before being placed in the appropriate stream channel. A at the Codrington hatchery. The streams were designed to pro- random number generator was used to determine the placement vide substrate and flow conditions similar to those used by At- of each treatment in the 12 channels for each trial block. lantic salmon and trout found in southern Ontario and area Behavioral observations began the day after the fish were streams (Gibson 1973; Hearn and Kynard 1986). Each stream tagged (day 1) and continued for 7 d. Behaviors were moni- channel had an overall length of 2.4 m with a riffle and pool sec- tored each day in both a morning and afternoon session using tion. The upstream riffle section was 1.6 m long, 0.4 m high, and a rig made up of three high definition camcorders (Sony HDR- 0.5 m wide with a water depth of approximately 0.2 m and flow XR200V) set up above a stream channel: one camera above the velocity of 0.18 ± 0.05 m/s. Substrate in the riffle consisted of pool and two equally spaced out above the riffle section. The 7–10 cm river rock and two 15–18 cm rocks to provide potential camcorder rig could easily be moved from channel to channel cover. The riffle section was followed by a pool section measur- and was situated approximately 1 m above the water. Two rigs ing 0.8 m long, 0.8 m high, and 0.5 m wide. The pool water depth were constructed (six cameras total), which enabled two stream was 0.6 m with a surface current of 0.027 ± 0.025 cm/s. Pool channels to be simultaneously recorded before moving the rigs substrate consisted of river rock ranging in size from 2 to 10 cm. to the next pair of channels. The fish were given 15 min to ac- Water from the hatchery’s surface water head pond (gravity- climate to the presence of the camcorder rig before recording fed system) was piped to the stream channels through a headbox began. Aggressive and feeding behaviors were then recorded inside the hatchery, which ensured equal flow to all stream for 30 min. channels. Water temperature was 9.8 ± 1.4◦C (mean ± SD). In the morning session (0800–1230 hours), each day for Stream channels were set up in two parallel series of six 7 days, fish were fed trout chow (Corey Aquafeeds) and frozen channels each. Water flowed from the headbox through the bloodworms (Chironomidae; Hikari, Japan). Specifically, every first two channels and then into subsequent channels in both minute for the first 10 min of a recording session, either 50–100 series. Channels were connected using two 10-cm PVC pipes, bloodworms or 1 g of trout chow were alternately released at the which were covered with wire mesh on one end to prevent the top and middle of the stream channel, with the current carrying movement of fish between channels. food items through the channel to simulate natural invertebrate Each trial block was composed of 12 treatments, with 12 fish drift (∼2% of biomass in each stream channel). Care was taken per treatment. Each Atlantic salmon strain underwent four treat- to avoid being seen by the fish. The afternoon recordings (1400– ments: Atlantic salmon alone (12 fish), Atlantic salmon with 1830 hours) did not involve food. The order that channels were brown trout (+BT; 6 salmon, 6 trout), Atlantic salmon with rain- filmed was randomized using a random number generator for bow trout (+RT; 6 salmon, 6 trout), and finally, Atlantic salmon each day. with both brown trout and rainbow trout (+BTRT; 4 salmon, 4 On day 8 of each trial, fish were collected from the stream of each trout species). Density in the stream channels was 10 channels for final mass and length measurements. Collection of fish/m2, which is the upper end of densities found in the field fish began at the channels farthest from the headbox to prevent (Fransen et al. 1993), but by holding density constant, we were disturbance. Netted fish were sedated with MS-222 before final still able to determine the relative strengths of intraspecific and masses and lengths were recorded. The initial and final mass

Downloaded by [Department Of Fisheries] at 20:29 25 September 2012 interspecific competition among ecologically similar species measurements were used to calculate standard growth rate (%/d) (Fausch 1998). The three Atlantic salmon strains were consid- using the following equation (Bernier et al. 2004): ered separately in all trials (not mixed) in order to independently evaluate their comparative performance. There were seven trial start dates (one trial block had a pair of dates, due to logistical Standard growth rate = 100 × [loge(final mass) constraints at the onset of the experiment, with the commence- − loge(initial mass)]/dfed. ment of four treatments followed by eight treatments). At the beginning of each trial, fish were randomly selected with similar catch effort and anesthetized with MS-222 (tri- Video analysis.—Analysis of the videos focused on aggres- caine methanesulfonate). Once sedated, the initial mass and sive and feeding behaviors. Aggressive behaviors monitored total length of each fish were recorded. In order to observe and comprised chasing, charging, and nipping (see Keenleyside and record individual behavior and feeding, each fish was tagged Yamamoto 1962 for definitions of behaviors). Feeding obser- with a colored 2-cm vinyl anchor tag (Floy Tag & Mfg., Seattle, vations were of the number of items consumed. Aggressive and Washington). Tags were applied using a fine fabric gun (Avery feeding behavior data from 4 days of each trial were analyzed, Mark II Fine Fabric Pistol Grip) with a maximum insertion depth comprising day 1, 3, 5, and 7. Approximately 864 h of video 910 VANZWOLETAL.

were observed in real time and paused every time an action Atlantic salmon strains clustered when alone or with nonna- occurred, with actor, act, and recipient recorded. tive trout species by grouping each strain by the presence or Statistical analysis.—Agonistic and feeding behaviors across absence of nonnative trout species (e.g., LaHave individuals the 4 days were summed and converted to a rate by dividing by alone or LaHave individuals with nonnatives). All nonnative the total time of observation. Dominance was calculated using treatments were grouped together for this analysis. All five de- David’s score, which creates an index for individuals within pendent variables were included in the analysis as predictors a social hierarchy based on an individual’s initiated and re- and the pooled within-group structure matrix was analyzed to ceived aggressive acts, while accounting for repeated interac- determine which variables most strongly correlated with the tions among group members (David 1988; see Gammell et al. discriminant functions. A two-way ANOVA was then used to 2003 for details of the calculation). examine the effects of treatment (alone versus nonnative) and Differences in initial mass and total length of the Atlantic strain on the first two DFA axes. All statistics were performed salmon strains were analyzed using one-way analysis of vari- using JMP 4 (version 4.0.2, SAS Institute Inc., Cary, North ance (ANOVA) models, with strain as the main effect. Student’s Carolina), SPSS 16.0 (SPSS Inc., Chicago), or Microsoft Office t-tests were used post hoc to determine differences between Excel 2003 (Microsoft Corporation, Redmond, Washington). pairs of strains. Data of initiated aggression, received aggres- Presented P-values are two-tailed probabilities. sion, and food consumption were normalized using logarith- mic (x + 1) transformation. Next, we conducted linear mixed models to test the effects of strain and treatment on initiated and received aggression, David’s scores, food consumption, and RESULTS standard growth rate. The interaction between strain and treat- The strains of Atlantic salmon differed significantly from ment was included while initial mass was entered as a covariate one another in initial mass and total length (mass: F2, 297 = in the models. Trial block and channel number were entered 71.5, P < 0.001; total length: F2, 297 = 37.8, P < 0.001, Table as random effects. We used a variance components covariance 1). Atlantic salmon from the Sebago Lake strain were the structure and denominator degrees of freedom were calculated largest, followed by Lac Saint-Jean fish, while those from the using a Satterthwaite approximation (Satterthwaite 1946). When LaHave strain were the smallest. Overall, the average mass of main effects were significant or a significant interaction existed Atlantic salmon was 40 ± 16 g (mean ± SD), while the average between strain and treatment, Student’s t-tests were used post length was 164 ± 22 mm. Brown trout had an average mass hoc to determine differences in variables. To test the effect of of 39 ± 14 g and length of 151 ± 18 mm, while rainbow trout dominance on growth parameters, linear regression analysis was were on average 21 ± 10 g and 126 ± 19 mm in length. Both used to compare David’s score and food consumption or stan- mass and length differed among the three species (initial mass: dard growth rate. F2,861 = 130.7, P < 0.001; length: F2,861 = 212.5, P < 0.001), To examine the effect of nonnative trout species on each with Atlantic salmon being longer but not heavier than brown Atlantic salmon strain in multivariate space, we used direct dis- trout (length: t401 = 7.85, P < 0.001; mass: t365 = 0.72, P = criminant function analysis (DFA; Dunteman 1984). The DFA 0.47), while both species were longer and heavier than rainbow examined the variation in the five aggression and growth vari- trout (Atlantic salmon length: t370 = 21.9, Atlantic salmon ables (initiated and received aggression, David’s score, food mass: t520 = 19.5, brown trout length: t358 = 13.0, brown trout consumption, and standard growth rate) to assess how the three mass: t322 = 15.1; P < 0.001 for all comparisons). Downloaded by [Department Of Fisheries] at 20:29 25 September 2012 TABLE 1. Summary of phenotypic and behavioral characteristics of three strains of juvenile Atlantic salmon in four treatments in seminatural stream behavioral trials. Means ± SDs across all four treatments to which the strains were exposed (N = 168 in each strain) are shown. Different lowercase letters indicate significant differences among strains (Student’s t-test; P < 0.05).

Atlantic salmon strains Characteristic LaHave Lac Saint-Jean Sebago Lake Initial mass (g) 30.9 ± 11.7 x 36.6 ± 9.7 y 53.2 ± 16.1 z Total length (mm) 150 ± 22 x 162 ± 15 y 181 ± 18 z Initiated aggression/h 2.8 ± 4.3 z 3.6 ± 4.4 z 1.3 ± 2.8 y Received aggression/h 6.8 ± 6.2 z 6.6 ± 5.6 z 3.1 ± 3.2 y David’s score −4.7 ± 13.5 −3.7 ± 14.8 −1.5 ± 7.8 Food consumption (items/h) 10.7 ± 10.5 z 8.8 ± 7.0 zy 7.4 ± 7.3 y Standard growth rate (%/d) −0.12 ± 0.99 −0.06 ± 0.85 −0.15 ± 0.64 EFFECTS OF NON-NATIVE SALMONIDS ON ATLANTIC SALMON 911

TABLE 2. Summary of linear mixed model results for the frequency of agonistic, foraging, and growth characteristics of three strains of juvenile Atlantic salmon in four treatments in seminatural stream behavioral trials. Strain and treatment were coded as main factors; initial mass was treated as a covariate.

Dependent variable Independent Degrees of freedom F-statistic P-value Initiated aggression/h Treatment 3,380.0 8.60 <0.001 Strain 2,355.1 28.1 <0.001 Initial mass 1,482.7 7.41 0.007 Strain × treatment 6,168.4 6.89 <0.001 Received aggression/h Treatment 3,474.3 27.6 <0.001 Strain 2,462.9 26.0 <0.001 Initial mass 1,488.8 9.90 0.002 Strain × treatment 6,366.9 6.67 <0.001 David’s score Treatment 3,491.0 26.0 <0.001 Strain 2,491.0 2.61 0.07 Initial mass 1,491.0 0.63 0.43 Strain × treatment 6,491.0 2.22 0.04 Food consumption (items/h) Treatment 3,466.7 13.29 <0.001 Strain 2,451.7 6.76 0.001 Initial mass 1,489.8 0.61 0.43 Strain × treatment 6,334.7 3.64 0.002 Standard growth rate (%/d) Treatment 3,472.7 5.16 0.002 Strain 2,472.8 0.21 0.81 Initial mass 1,486.6 18.6 <0.001 Strain × treatment 6,368.4 1.94 0.07

Agonistic Interactions and David’s Score another (Table 1). An interaction between strain and treatment Treatment significantly influenced initiated and received ag- was also found for both initiated and received aggression and gression and David’s score of Atlantic salmon (Table 2). Across David’s score (Table 2, Table A.1 in the appendix). Sebago Lake all strains, Atlantic salmon juveniles initiated more aggression initiated and received the least aggression when alone or with when alone than in either +BT and +BTRT treatments (+BT, brown trout, whereas in the + RT treatment, Lac Saint-Jean Student’s t-test, t258 = 2.28, P = 0.02; +BTRT, Student’s initiated significantly more aggression than either LaHave or t-test, t284 = 5.40, P < 0.001) and were significantly more Sebago Lake (Figure 1). There was no difference among the aggressive in the + RT treatment than in the +BTRT treatment strains in either initiated or received aggression when they were (Figure 1a). Initiated aggression by Atlantic salmon varied with both brown trout and rainbow trout (Figure 1). Sebago Lake between the +BT and +BTRT treatment with aggression individuals scored significantly higher David’s scores than ei- observed to be higher in the +BT treatment (Figure 1a). ther of the two other strains in the +BT treatment and in the Atlantic salmon also received much less aggression when alone +BTRT treatment, this strain scored significantly higher than or in the + RT treatment compared with either +BT (alone, LaHave individuals (Figure 1c). Downloaded by [Department Of Fisheries] at 20:29 25 September 2012 Student’s t-test, t132 = 5.51; + RT, Student’s t-test, t165 = As a covariate, initial mass of Atlantic salmon influenced 4.37, P < 0.001 for both comparisons) or +BTRT treatments agonistic interactions: heavier individuals both initiated more, (alone, Student’s t-test, t106 = 5.48; + RT, Student’s t-test, t178 and received fewer, aggressive acts (Table 2). However, initial = 3.90, P < 0.001 for both comparisons; Figure 1b). When mass did not influence David’s score (Table 2). Atlantic salmon were alone or in the + RT treatment, they scored higher David’s scores than in the +BT and +BTRT Food Consumption and Standard Growth Rate treatments; David’s scores in the +BT treatment were lower Food consumption and standard growth rate of Atlantic than the +BTRT treatment (Table 2; Figure 1c). salmon individuals were significantly influenced by treatment The strain of Atlantic salmon also influenced initiated and (Table 2). Food consumption was highest when Atlantic salmon received aggression (Tables 1, 2). The LaHave and Lac Saint- individuals were with conspecifics, followed by consumption Jean strains both initiated significantly more aggression than in the + RT treatment, and was lowest in the two treatments Sebago Lake, whereas there was no difference in aggression containing brown trout (+BT and +BTRT; Table 2; Figure 2a). between Lac Saint-Jean and LaHave strains (Table 1). Analo- Standard growth rate largely mirrored the food consumption gously, Sebago Lake received significantly less aggression than data: it was highest in the alone and + RT treatments and was both LaHave and Lac Saint-Jean, which did not differ from one thelowestinthe+BT and +BTRT treatments (Figure 2b). 912 VANZWOLETAL.

a) 6 AAB BC C a 5 b a LaHave Lac Saint-Jean 4 a Sebago Lake

a 3

a a 2 b b

Initiated aggressive acts/hour 1

0 b) AS +RT +BT +BTRT

14 A A a B B

12 a 10

a 6 a b 4

b Received aggressive acts/hour aggressive Received 2

0 AS +RT +BT +BTRT c) 4 AABC 2 aaba ab b 0

-2

-4

-6

-8 David's score David's -10

-12

-14

-16 AS +RT +BT +BTRT Treatment Downloaded by [Department Of Fisheries] at 20:29 25 September 2012 FIGURE 1. Agonistic interactions and dominance of three strains of juvenile FIGURE 2. Feeding behaviors and growth of three strains of juvenile At- Atlantic salmon (LaHave, Lac Saint-Jean, and Sebago Lake) in four treatments lantic salmon (LaHave, Lac Saint-Jean, and Sebago Lake) in four treatments in seminatural stream behavioral trials showing (a) the number of initiated in seminatural stream behavioral trials showing (a) the number of food items aggressive acts per hour, (b) the number of received aggressive acts per hour, consumed per hour and (b) the standard growth rate. The four experimental and (c) David’s score. The four experimental treatments include Atlantic salmon treatments include Atlantic salmon alone (AS), Atlantic salmon with rainbow alone (AS), Atlantic salmon with rainbow trout (+RT), Atlantic salmon with trout (+RT), Atlantic salmon with brown trout (+BT), and Atlantic salmon with brown trout (+BT), and Atlantic salmon with both brown trout and rainbow trout both brown trout and rainbow trout (+BTRT; N = 6). Behavioral observations (+BTRT). Behavioral observations were summed for individual fish and then were summed for individual fish and then converted to a rate by dividing by converted to a rate by dividing by the total observation time for a given channel. the total observation time for a given channel. Bars denote mean ± SE for each Bars denote mean ± SE for each of the three strains, while dashed lines indicate of the three strains, while dashed lines indicate the mean of the three strains the mean of the three strains for each treatment. Different uppercase letters for each treatment. Different uppercase letters indicate significant differences indicate significant differences between treatments, while different lowercase between treatments, while different lowercase letters indicate significant dif- letters indicate significant differences among Atlantic salmon strains within a ferences among Atlantic salmon strains within a specific treatment (at P < specific treatment (at P < 0.05). 0.05). EFFECTS OF NON-NATIVE SALMONIDS ON ATLANTIC SALMON 913

The three strains also differed significantly in food consump- between strain and the presence of nonnatives was not significant tion but not standard growth rate (Tables 1, 2). LaHave con- (F1,498 = 0.39, P = 0.68). sumed significantly more food than Sebago Lake and more than Lac Saint-Jean, albeit the latter effect was marginally nonsignificant (post hoc t-test: t292 = 1.96, P = 0.052). An interaction DISCUSSION between strain and treatment revealed that Sebago Lake con- Although we cannot fully rule out the effects of body size sumed significantly fewer food items than either Lac Saint-Jean among species, our data suggest that the presence of nonnative or LaHave in all but the + RT treatment (Figure 2a, Table A.1 salmonids affects the aggressive and foraging behavior of in the appendix). Conversely, the LaHave strain consumed more juvenile Atlantic salmon. When juvenile Atlantic salmon were food or equivalent amounts of food as compared to the other two with conspecifics only, the level of aggression received by strains across the four treatments (Figure 2a). Despite these dif- individual fish was lowest and the level of food consumption ferences in food consumption, however, there was no observed was highest. Additionally, Atlantic salmon were most aggres- difference in standard growth rate among the strains in any of sive in the conspecific treatment, as the presence of nonnative the treatments during the 7-d trials (Figure 2b). trout suppressed the amount of aggression Atlantic salmon David’s score was positively related to both food consump- initiated. Specifically, brown trout exerted a stronger influence tion and standard growth rate of the Atlantic salmon (food con- on Atlantic salmon than rainbow trout. These data mirror sumption, linear regression: R2 = 0.008, β = 0.09, N = 504, those of other researchers including Stradmeyer et al. (2008), P = 0.05; standard growth rate, linear regression: R2 = 0.01, β who found that juvenile brown trout were always dominant to = 0.11, N = 504, P = 0.01). Atlantic salmon. Using both stream channels and field surveys, Hearn and Kynard (1986) found that wild rainbow trout and juvenile Atlantic salmon compete and Blanchet et al. (2009) Discriminant Function Analysis found that food consumption of juvenile Atlantic salmon was Differences in aggression and growth predictors among At- lowered in the presence of rainbow trout. Collective evidence 2 lantic salmon strain groupings were detected by the DFA (χ (25) now suggests that nonnative salmonids, particularly brown = 267.7, P < 0.001, Figure 3). The second function was also trout, can have strong behavioral effects on Atlantic salmon. 2 2 significant (χ (16) = 109.1, P < 0.001), as were the third (χ (9) = Body size is an important factor in determining the outcome 2 30.1, P < 0.001) and fourth functions (χ (4) = 12.1, P = 0.02). of contests among conspecifics. Many studies have shown The first and second discriminant functions of the analysis ac- that dominance in fish is directly linked to larger body size counted for 62% and 28% of the variation, respectively, and (e.g., Abbott et al. 1985; Beaugrand et al. 1996). However, this were the focus of our analysis. The first discriminant function relationship between body size and dominance did not exist (DFA 1) was positively correlated with initiated and received for Atlantic salmon in our study. Consistent with our data, aggression and negatively with, to a lesser extent, David’s score Huntingford et al. (1990) examined dominance competitions (Table 3; Figure 3). between pairs of juvenile Atlantic salmon in spring and summer The two-way ANOVA revealed that for DFA 1, all three (when we conducted our trials) and found no evidence that strains differed significantly from one another (F2,498 = 82.8, dominance tests were won by larger fish, regardless of the size P < 0.001) with the Lac Saint-Jean strain scoring the high- difference between a pair. Interestingly, when the experiment est, followed by LaHave and then Sebago Lake individuals was conducted in September, the relationship did exist with (Figure 3). Treatment also influenced DFA 1 (F1,498 = 7.30, 72% of the dominance tests being won by the larger fish of a P = 0.007), with higher scores generally observed in the nonna- pair (Huntingford et al. 1990). The study found, however, that

Downloaded by [Department Of Fisheries] at 20:29 25 September 2012 tive treatments. There was also, however, an interaction between aggression levels strongly influenced the social dominance of strain and presence of nonnatives (F2,498 = 11.6, P < 0.001): Se- an individual. These data suggest that dominance is a function bago Lake and Lac Saint-Jean, but not LaHave individuals had of behavior and that large body size may be a consequence, not higher DFA 1 values in the nonnative versus alone treatments a cause, of dominance, at least in some salmonids. (Figure 3). It is well known that subordinate fish exhibit less growth as The second discriminant function (DFA 2) was positively a result of the behaviors of dominant fish. This pattern has been correlated with initiated aggression, food consumption, growth shown in a number of salmonids (e.g., Atlantic salmon and rate, and David’s score and negatively with received aggression brook trout; Gibson 1973). Consistent with these studies, we (Table 3). For this function, a two-way ANOVAfound that while found that the presence of brown trout suppressed the growth the strains did not vary (F2,498 = 2.08, P = 0.12), the presence of rate of Atlantic salmon, which were typically subordinate to nonnative trout species significantly influenced canonical scores the brown trout. We also found that in the brown trout and At- (F1,498 = 80.6, P < 0.001), with strains initiating less aggression, lantic salmon treatment, subordinate fish grew at rates much consuming less food, growing less and having lower dominance lower than dominant fish, and food consumption of Atlantic scores, but receiving more aggression in the presence rather salmon significantly declined as compared with when Atlantic than absence of the nonnative trout species. The interaction salmon were alone. While dominant brown trout are known to 914 VANZWOLETAL.

1.0 Increasing initiated aggression, food consumption, growth rate, and David's score Decreasing received aggression

0.5 LSJalone

LHalone SLalone

0.0 DFA 2

LSJNN -0.5 SLNN LHNN

Increasing initiated and received aggression Decreasing David's score -1.0 -1.5 -1.0 -0.5 0.0 0.5 1.0 DFA 1

FIGURE 3. Canonical plot of the ﬁrst two functions of the discriminant function analysis (DFA 1, DFA 2) examining the variation of aggression and growth measurements among three strains of juvenile Atlantic salmon in four treatments in seminatural stream behavioral trials. Treatments with nonnative species present (NN) were grouped together and compared with the Atlantic salmon alone treatment for the LaHave (LH), Lac Saint-Jean (LSJ), and Sebago Lake (SL) strains. The symbols represent strain centroids (with 95% conﬁdence intervals).

monopolize feeding areas, reducing feeding opportunities of tify the behavioral tactics used to acquire food, our results are subordinates (Hojesj¨ o¨ et al. 2005), Metcalfe (1986) has postu- consistent with Metcalfe (1986). For example, Sebago Lake lated that regardless of the actions of the dominant fish, it is salmon appeared to choose a growth strategy that minimized better for subordinate fish to minimize energetic costs, rather energy expenditure, opting out of the competition and conse- than maximize food intake. Although we did not directly quan- quently consuming the least amount of food and losing the most Downloaded by [Department Of Fisheries] at 20:29 25 September 2012 TABLE 3. Summary of discriminant function analysis (DFA) of agonistic and growth measurements of three strains of juvenile Atlantic salmon in four treatments in seminatural stream behavioral trials. The DFA was performed on five agonistic and growth measurements of the LaHave, Lac Saint-Jean, and Sebago Lake strains. Treatments with nonnative species present were grouped together and compared with the Atlantic salmon alone treatment for each strain. Values represent pooled within-group correlations of canonical roots and standardized canonical discriminant function coefficients.

Correlation of variables with discriminant Standardized canonical discriminant functions function coefﬁcients Variable DFA 1 DFA 2 DFA 3 DFA 4 DFA 1 DFA 2 DFA 3 DFA 4 Initiated aggression/h 0.641 0.632 –0.05 0.395 1.26 0.706 −0.460 −0.679 David’s score −0.161 0.564 −0.463 0.587 −0.797 −0.81 −0.169 1.26 Received aggression/h 0.640 −0.397 0.462 0.461 0.157 −0.792 0.344 0.962 Food consumption (items/h) 0.067 0.580 0.786 0.198 −0.493 0.565 0.788 0.271 Standard growth rate (%/d) −0.022 0.263 0.490 −0.137 0.16 −0.164 0.247 −0.220 EFFECTS OF NON-NATIVE SALMONIDS ON ATLANTIC SALMON 915

mass of the three strains in the treatments with nonnative trout. Lac Saint-Jean strain. Differences observed here suggest that Regardless of the actual feeding tactics used by Atlantic salmon, stocking the Lac Saint-Jean strain, the strain believed to be the our data clearly show that Atlantic salmon feed less in the pres- closest geographically and genetically of the three strains to the ence of dominant brown trout and consequently display reduced original Lake Ontario population (Dimond and Smitka 2005), growth. will achieve greater restoration success as they are better com- Community ecology studies have long shown that compe- petitors against brown trout and rainbow trout. Indeed, the fact tition among ecologically similar species can lead to spatial that LaHave strain was the most significantly affected by the separation or shifts in resource use if the species continue to presence of the nonnative trout may explain the previous failed live in sympatry (e.g., Werner and Hall 1977; Langeland et al. attempts of restoring Atlantic salmon with this strain. 1991). Brown trout and Atlantic salmon have historically co- High species richness can lead to competition for resources, existed in rivers in Europe (Hojesj¨ o¨ et al. 2005) but tend to resulting in declines in growth rates of the competing species. spatially separate in streams, largely driven by the aggressive This effect has been shown in, for example, sunfish (Centrar- behavior of brown trout (Armstrong et al. 2003). Our study con- chidae; Mittelbach 1988), Daphnia spp. (Bengtsson 1993), and firmed the dominance of brown trout over Atlantic salmon as desert annuals, where competition among the plants leads to de- has been shown by Stradmeyer et al. (2008). Additionally, we creases in growth, biomass, and fecundity (Inouye et al. 1980). found that the food consumption and growth of Atlantic salmon We found that the presence of multiple salmonid species led declined in the presence of brown trout. Rainbow trout and to increases in aggression received and, in the case of Atlantic Atlantic salmon, however, have not historically coexisted, yet salmon, reductions in food consumption and growth. Such inter- studies have shown there is a degree of niche overlap (Gibson actions often lead to partitioning of habitat and resources among 1981; Hearn and Kynard 1986), which we expected would in- the competing species allowing the individuals to coexist (e.g., fluence the agonistic interactions and growth of Atlantic salmon Robertson and Gaines 1986; Young 2001). Because of our ex- in our study. Similar to research by Blanchet et al. (2008), the perimental setup, we could not easily assess potential habitat or presence of rainbow trout did not affect the food consumption resource partitioning. Nevertheless, we found no evidence that or growth rate of Atlantic salmon. Atlantic salmon received no Atlantic salmon shifted habitat use across the pool and riffle more aggression in the presence of rainbow trout than they did sections when alone versus with either or both of the nonnative in the conspecifics treatment. These data support a study by species. Regardless, our data suggest that high salmonid species Volpe et al. (2001) that found that although rainbow trout were richness could be detrimental for Atlantic salmon during the much more aggressive than Atlantic salmon, agonistic interac- stream stage of life. Assessing the species community of tar- tions were largely between rainbow trout conspecifics and not geted streams and rivers for Atlantic salmon restoration may Atlantic salmon. Hence, it is conceivable that although there is also help to alleviate competition for Atlantic salmon. niche overlap between these two species, rainbow trout largely In conclusion, our data point to some considerations that may ignore Atlantic salmon, at least during agonistic interactions. help to direct restoration of Atlantic salmon among the natural- Thus, density issues aside, these data suggest that brown trout, ized populations of brown trout and rainbow trout in the Lauren- more than rainbow trout, influence Atlantic salmon agonistic tian Great Lakes watershed. First, the three strains are predicted and feeding behaviors, and unless spatial separation is possible to have differential poststocking ecological success in tribu- for brown trout and Atlantic salmon, competition between these tary environments, with Atlantic salmon originating from Lac two species poses a threat to Atlantic salmon establishment in Saint-Jean outperforming the LaHave and Sebago Lake strains. Lake Ontario streams. Whether these differences would similarly extend to increased Behavioral differences among populations or strains within performance in Lake Ontario in terms of growth, survival, and

Downloaded by [Department Of Fisheries] at 20:29 25 September 2012 a species have been observed across many taxa (e.g., Jones adult returns still needs to be determined. Second, the success- 1977; Rex et al. 1996; Moretz et al. 2007), and comparing ful establishment of juvenile Atlantic salmon may be greatly these differences can provide an understanding of phenotypic impeded by the presence of brown trout. Lake Ontario rainbow attributes that will strengthen efforts of native species reintro- trout appear to have less of an influence on Atlantic salmon, duction (Curio 1996). One important attribute for successful albeit high species richness did impede the performance of At- establishment and persistence is aggression (as reviewed by lantic salmon. As such, we suggest avoiding stocking juvenile Holway and Suarez 1999). We have shown differences among Atlantic salmon in Lake Ontario streams with high densities of Atlantic salmon strains in both aggressive and feeding behav- brown trout or in streams with multiple established salmonid iors. Indeed, the Lac Saint-Jean strain initiated the most aggres- species. sion and lost the least mass of the three strains, suggesting they are better competitors against brown trout and rainbow trout, two nonnative species prevalent in Lake Ontario tributaries. ACKNOWLEDGMENTS The DFA confirmed these strain differences by showing that the We gratefully acknowledge the support of World Wildlife presence of nonnative trout species influenced the LaHave and Fund–Canada and the Ontario Ministry of Natural Resources. Sebago Lake strains the most but had less of an impact on the The research was also supported by funding from the Natural 916 VANZWOLETAL.

Sciences and Engineering Research Council of Canada through David, H. A. 1988. The method of paired comparisons. Oxford University Press, a postgraduate fellowship to J. A. Van Zwol and a Discovery New York. Grant to B. D. Neff. The manuscript was improved by con- Dewald, L., and M. A. Wilzbach. 1992. Interactions between native brook trout and hatchery brown trout: effects on habitat use, feeding, and growth. structive comments by T. Hain, S. Blanchet, F. Utter, and two Transactions of the American Fisheries Society 121:287–296. anonymous reviewers. We also thank S. Garner, A.L. Houde, N. Dietrich, J. P., J. N. Bowlby, B. J. Morrison, and N. E. Jones. 2008. The impacts Lobo, and S. Colborne for methodological and statistical advice of Atlantic salmon stocking on rainbow trout in Barnum House Creek, Lake and A. Henkel, A. Wojcik, and B. Sutton-Quaid as well as T. Ontario. Journal of Great Lakes Research 34:495–505. MacDonald, S. Ferguson, and E. Scarf at the Codrington Fish- Dimond, P., and J. Smitka. 2005. Evaluation of selected strains of Atlantic salmon as potential candidates for the restoration of Lake Ontario. Trout eries Research Facility for assistance in the field or laboratory. Unlimited Canada, Technical Report ON-12, Guelph, Ontario. Dunteman, G. H. 1984. Introduction to multivariate analysis. Sage Publications, Thousand Oaks, California. Edge, K. A., C. R. Townsend, and T. A. Crowl. 1993. Investigating anti-predator REFERENCES behaviour in three genetically differentiated populations of non-migratory Abbott, J. C., R. L. Dunbrack, and C. D. Orr. 1985. The interaction of size galaxiid fishes in a New Zealand river. New Zealand Journal of Marine and and experience in dominance relationships of juvenile steelhead trout (Salmo Freshwater Research 27:357–363. gairdneri). Behaviour 92:241–253. Elliott, J. M. 1990. Mechanisms responsible for population regulation in young Armstrong, J. D., P. S. Kemp, G. J. A. Kennedy, M. Ladle, and N. J. Milner. migratory trout, Salmo trutta: III. the role of territorial behaviour. Journal of 2003. Habitat requirements of Atlantic salmon and brown trout in rivers and Animal Ecology 59:803–818. streams. Fisheries Research 62:143–170. Ens, B. J., and J. D. Goss-Custard. 1984. Interference among oystercatchers, Beaugrand, J. P., D. Payette, and C. Goulet. 1996. Conflict outcome in male Haematopus ostralegus, feeding on mussels, Mytilus edulis, on the Exe estu- green swordtail fish dyads (Xiphophorus helleri): interaction of body size, ary. Journal of Animal Ecology 53:217–231. prior dominance/subordination experience and prior residency. Behaviour Fausch, K. D. 1998. Interspecific competition and juvenile Atlantic salmon 133:303–319. (Salmo salar): on testing effects and evaluating the evidence across scales. Bengtsson, J. 1993. Interspecific competition and determinants of extinction in Canadian Journal of Fisheries and Aquatic Sciences 55(Supplement 1):218– experimental populations of three rockpool Daphnia species. Oikos 67:451– 231. 464. Fransen, B. R., P. A. Bisson, J. W. Ward, and R. E. Bilby. 1993. Physical Bernatchez, L. 2001. The evolutionary history of brown trout (Salmo trutta and biological constraints on summer rearing of juvenile coho salmon (On- L.) inferred from phylogeographic, nested clade, and mismatch analyses of corhynchus kisutch) in small western Washington streams. Pages 271–288 in mitochondrial DNA variation. Evolution 55:351–379. L. Berg and P.W. Delaney, editors. Proceedings of the coho salmon workshop. Bernier, N. J., N. Bedard, and R. E. Peter. 2004. Effects of cortisol on food intake, Canadian Department of Fisheries and Oceans, Vancouver. growth, and forebrain neuropeptide Y and corticotropin-releasing factor gene Gage, J. F. 1963. The landlocked salmon or ouananiche in Ontario. Ontario Fish expression in goldfish. General and Comparative Endocrinology 135:230– and Wildlife Review 2(2):2–7. 240. Gammell, M. P., H. De Vries, D. J. Jennings, C. M. Carlin, and T. J. Hayden. Blair, A. A. 1938. Scales of Lake Ontario salmon indicate a land-locked form. 2003. David’s score: a more appropriate dominance ranking method than Copeia 1938:206. Clutton-Brock et al.’s index. Animal Behaviour 66:601–605. Blanchet, S., L. Bernatchez, and J. J. Dodson. 2009. Does interspecific com- Gibson, R. J. 1973. Interactions of juvenile Atlantic salmon (Salmo salar)and petition influence relationships between heterozygosity and fitness-related brook trout (Salvelinus fontinalis). International Atlantic Salmon Foundation behaviors in juvenile Atlantic salmon (Salmo salar)? Behavioral Ecology Special Publication Series 4:181–202. and Sociobiology 63:605–615. Gibson, R. J. 1981. Behavioural interactions between coho salmon (On- Blanchet, S., G. Loot, L. Bernatchez, and J. J. Dodson. 2008. The effects of corhynchus kisutch), Atlantic salmon (Salmo salar), brook trout (Salveli- abiotic factors and intraspecific versus interspecific competition on the diel nus fontinalis), and steelhead trout (Salmo gairdneri), at the juvenile fluvi- activity patterns of Atlantic salmon (Salmo salar) fry. Canadian Journal of atile stages. Canadian Technical Report of Fisheries and Aquatic Sciences Fisheries and Aquatic Sciences 65:1545–1553. 1029. Blankenship, S. M., M. R. Campbell, J. E. Hess, M. A. Hess, T. W. Kassler, Good, S. P., J. J. Dodson, M. G. Meekan, and D. A. J. Ryan. 2001. Annual C. C. Kozfkay, A. P. Matala, S. R. Narum, M. M. Paquin, M. P. Small, variation in size-selective mortality of Atlantic salmon (Salmo salar) fry.

Downloaded by [Department Of Fisheries] at 20:29 25 September 2012 J. J. Stephenson, K. I. Warheit, and P. Moran. 2011. Major lineages and Canadian Journal of Fisheries and Aquatic Sciences 58:1187–1195. metapopulations in Columbia River Oncorhynchus mykiss are structured by Grieg, L., B. Ritchie, L. Carl, and C. A. Lewis. 2003. Potential and strategy for dynamic landscape features and environments. Transactions of the American restoration of Atlantic salmon in Lake Ontario: a workshop report. Ontario Fisheries Society 140:665–684. Ministry of Natural Resources, Lake Ontario Management Unit, Peterbor- Caro, T. 1999. The behaviour–conservation interface. Trends in Ecology and ough, Ontario. Evolution 14:366–369. Hamilton, J. G., C. Holzapfel, and B. E. Mahall. 1999. Coexistence and inter- Crawford, S. S. 2001. Salmonine introductions to the Laurentian Great Lakes: ference between a native perennial grass and non-native annual grasses in an historical review and evaluation of ecological effects. Canadian Special California. Oecologia 121:518–526. Publication of Fisheries and Aquatic Sciences 132. Hearn, W. E., and B. E. Kynard. 1986. Habitat utilization and behavioural Crawford, S. S., and A. M. Muir. 2008. Global introductions of salmon and interaction of juvenile Atlantic salmon (Salmo salar) and rainbow trout (On- trout in the genus Oncorhynchus: 1870–2007. Reviews in Fish Biology and corhynchus mykiss) in tributaries of the White River of Vermont. Canadian Fisheries 18:313–344. Journal of Fisheries and Aquatic Sciences 43:1988–1998. Curio, E. 1996. Conservation needs ethology. Trends in Ecology and Evolution Hojesj¨ o,¨ J., J. D. Armstrong, and S. W. Grifﬁths. 2005. Sneaky feeding by 11:260–263. salmon in sympatry with dominant brown trout. Animal Behaviour 69:1037– D’Antonio, C. M., and P. M. Vitousek. 1992. Biological invasions by exotic 1041. grasses, the grass/ﬁre cycle, and global change. Annual Review of Ecology Holway, D. A., and A. V. Suarez. 1999. Animal behaviour: an essential compo- and Systematics 23:63–87. nent of invasion biology. Trends in Ecology and Evolution 14:328–330. EFFECTS OF NON-NATIVE SALMONIDS ON ATLANTIC SALMON 917

Huntingford, F. A., J. B. Metcalfe, J. E. Thorpe, W. D. Graham, and C. E. Adams. Parsons, J. W. 1973. History of salmon in the Great Lakes, 1850–1970. U.S. 1990. Social dominance and body size in Atlantic salmon parr, Salmo salar Fish and Wildlife Service Bureau of Sport Fisheries and Wildlife Technical L. Journal of Fish Biology 36:877–881. Paper 68. Inouye, R. S., G. S. Byers, and J. H. Brown. 1980. Effects of predation and Rex, A., U. Sondern, J. P. Voigt, S. Franck, and H. Fink. 1996. Strain differences competition on survivorship, fecundity, and community structure of desert in fear-motivated behavior in rats. Pharmacology Biochemistry and Behavior animals. Ecology 61:1344–1351. 54:107–111. Jones, R. B. 1977. Sex and strain differences in the open-field responses of the Ricciardi, A., R. J. Neves, and J. B. Rasmussen. 1998. Impending extinctions of domestic chick. Applied Animal Ethology 3:255–261. North American freshwater mussels (Unionoida) following the zebra mus- Keenleyside, M. H. A., and F. T. Yamamoto. 1962. Territorial behaviour of sel (Dreissena polymorpha) invasion. Journal of Animal Ecology 67:613– juvenile Atlantic salmon (Salmo salar L.). Behaviour 19:139–169. 619. Kerr, S. J. 2006. An historical review of fish culture, stocking and fish transfers Robertson, D. R., and S. D. Gaines. 1986. Interference competition struc- in Ontario, 1865–2004. Ontario Ministry of Natural Resources, Fish and tures habitat use in a local assemblage of coral reef surgeonfishes. Ecology Wildlife Branch, Peterborough, Ontario. 67:1372–1383. Kitano, S. 2004. Ecological impacts of rainbow, brown and brook trout in Satterthwaite, F. E. 1946. An approximate distribution of estimates of variance Japanese inland waters. Global Environmental Research 8:41–50. components. Biometrics Bulletin 2:110–114. Langeland, A., J. H. L’Abee-Lund,´ B. Jonsson, and N. Jonsson. 1991. Resource Scott, R. J., D. L. G. Noakes, F. W. H. Beamish, and L. M. Carl. 2003. Chinook partitioning and niche shift in Arctic charr Salvelinus alpinus and brown trout salmon impede Atlantic salmon conservation in Lake Ontario. Ecology of Salmo trutta. Journal of Animal Ecology 60:895–912. Freshwater Fish 12:66–73. Lowe, S., M. Browne, S. Boudjelas, and M. De Poorter. 2000. 100 of the world’s Scott, R. J., M. S. Poos, D. L. G. Noakes, and F. W. H. Beamish. 2005. Ef- worst invasive alien species: a selection from the global invasive species fects of exotic salmonids on juvenile Atlantic salmon behaviour. Ecology of database. Invasive Species Specialist Group of the World Conservation Freshwater Fish 14:283–288. Union, Auckland, New Zealand. Available: interface.creative.auckland.ac.nz/ Stanfield, L., and M. L. Jones. 2003. Factors influencing rearing success of database/species/reference files/100English.pdf. (December 2010). Atlantic salmon stocked as fry and parr in Lake Ontario tributaries. North MacCrimmon, H. R. 1977. Animals, man and change: alien and extinct wildlife American Journal of Fisheries Management 23:1175–1183. of Ontario. McClelland and Stewart, Toronto. Stewart, T. J., and T. Schaner. 2002. Lake Ontario salmonid introductions 1970 Madsen, J. D., J. W. Sutherland, J. A. Bloomfield, L. W. Eichler, and C. W. to 1999: stocking, fishery and fish community influences. Pages 12.1–12.10 Boylen. 1991. The decline of native vegetation under dense Eurasian water- in Lake Ontario fish communities and fisheries: 2001 annual report of the milfoil canopies. Journal of Aquatic Plant Management 29:94–99. Lake Ontario management unit. Queen’s Printer for Ontario, Picton, Ontario. Metcalfe, N. B. 1986. Intraspecific variation in competitive ability and food in- Stradmeyer, L., J. Hojesj¨ o,¨ S. W. Griffiths, J. D. Gilvear, and J. D. Arm- take in salmonids: consequences for energy budgets and growth rates. Journal strong. 2008. Competition between brown trout and Atlantic salmon parr of Fish Biology 28:525–553. over pool refuges during rapid dewatering. Journal of Fish Biology 72:848– Mittelbach, G. G. 1988. Competition among refuging sunfishes and effects of 860. fish density on littoral zone invertebrates. Ecology 69:614–623. Volpe, J. P., B. R. Anholt, and B. W. Glickman. 2001. Competition among Moretz, J. A., E. P. Martins, and B. D. Robison. 2007. Behavioral syndromes juvenile Atlantic salmon (Salmo salar) and steelhead (Oncorhynchus mykiss): and the evolution of correlated behavior in zebrafish. Behavioral Ecology relevance to invasion potential in British Columbia. Canadian Journal of 18:556–562. Fisheries and Aquatic Sciences 58:197–207. Nakano, S., S. Kitano, K. Nakai, and K. D. Fausch. 1998. Competitive inter- Ward, H. B. 1932. The origin of the landlocked habit in salmon. Proceedings of actions for foraging microhabitat among introduced brook charr, Salvelinus the National Academy of Sciences of the USA 18:569–580. fontinalis, and native bull charr, S. confluentus, and westslope cutthroat trout, Watts, D. 1999. The Presumpscot gets a chance at life. Maine Sportsman (July). Oncorhynchus clarki lewisi, in a Montana stream. Environmental Biology of Available: friendsofsebago.org/mspresumpnew.html. (December 2010). Fishes 52:345–355. Werner, E. E., and D. J. Hall. 1977. Competition and habitat shift in two sunfishes Olden, J. D., N. L. R. Poff, M. R. Douglas, M. E. Douglas, and K. D. Fausch. (Centrarchidae). Ecology 58:869–876. 2004. Ecological and evolutionary consequences of biotic homogenization. Young, K. A. 2001. Habitat diversity and species diversity: testing the compe- Trends in Ecology and Evolution 19:18–24. tition hypothesis with juvenile salmonids. Oikos 95:87–93. Downloaded by [Department Of Fisheries] at 20:29 25 September 2012 918 VANZWOLETAL.

APPENDIX: SUPPLEMENTAL DATA FOR ATLANTIC SALMON

TABLE A.1. Agonistic interactions, growth, and physical measurements (means ± SDs) of three strains of juvenile Atlantic salmon in four treatments in seminatural stream behavioral trials. The four experimental treatments were Atlantic salmon alone (N = 72 for each strain), Atlantic salmon with rainbow trout (N = 36 for each strain), Atlantic salmon with brown trout (N = 36 for each strain), and Atlantic salmon with both brown trout and rainbow trout (N = 24 for each strain). Physical measurements of the trout species are included with sample sizes in parentheses under the treatment column.

Total Initiated Received Food Standard Initial length aggression/ aggression/ David’s consumption growth rate Treatment mass (g) (mm) h h score (items/h) (%/d) Alone LaHave 30.0 ± 11.0 146 ± 20 4.1 ± 4.7 4.9 ± 3.8 0.0 ± 15.1 11.6 ± 10.8 –0.1 ± 0.9 Lac Saint-Jean 35.3 ± 8.8 160 ± 14 4.5 ± 5.3 5.4 ± 4.1 0.0 ± 15.8 12.3 ± 7.7 0.1 ± 0.6 Sebago Lake 55.2 ± 17.9 183 ± 19 1.3 ± 3.6 1.6 ± 1.9 0.0 ± 6.0 9.1 ± 8.1 0.0 ± 0.7 + Rainbow trout LaHave 31.5 ± 13.9 152 ± 24 1.5 ± 2.9 3.8 ± 4.5 –1.4 ± 8.9 8.7 ± 9.6 0.0 ± 1.1 Lac Saint-Jean 37.1 ± 8.9 162 ± 13 4.2 ± 4.2 6.4 ± 4.6 0.6 ± 15.0 8.5 ± 5.4 0.1 ± 1.1 Sebago Lake 48.9 ± 10.9 176 ± 14 1.6 ± 2.0 3.6 ± 3.0 0.2 ± 8.5 8.6 ± 7.0 0.0 ± 0.4 Rainbow trout (108) 21.1 ± 10.2 127 ± 20 +Brown trout LaHave 31.3 ± 11.0 152 ± 23 3.0 ± 5.2 12.2 ± 8.2 –13.8 ± 11.1 12.6 ± 12.2 –0.1 ± 1.1 Lac Saint- Jean 35.9 ± 12.2 159 ± 20 2.5 ± 3.3 9.1 ± 8.2 –13.9 ± 10.9 4.7 ± 4.8 –0.5 ± 0.7 Sebago Lake 53.1 ± 17.5 181 ± 18 1.0 ± 2.3 3.8 ± 4.1 –4.6 ± 8.7 5.2 ± 6.1 –0.4 ± 0.6 Brown trout (108) 39.5 ± 13.2 151 ± 17 +Brown trout and rainbow trout LaHave 32.6 ± 11.7 156 ± 24 0.8 ± 1.4 9.0 ± 4.9 –10.4 ± 7.9 8.4 ± 7.4 –0.2 ± 0.8 Lac Saint-Jean 40.7 ± 8.6 167 ± 13 1.6 ± 2.2 7.0 ± 4.8 –5.9 ± 8.2 4.8 ± 4.1 –0.1 ± 1.0 Sebago Lake 54.2 ± 14.3 180 ± 17 1.2 ± 1.5 5.6 ± 3.0 –3.8 ± 7.7 3.7 ± 4.5 –0.5 ± 0.4 Brown trout (72) 39. 1 ± 14.3 150 ± 18 Rainbow trout (72) 20.0 ± 8.5 126 ± 19 Downloaded by [Department Of Fisheries] at 20:29 25 September 2012 This article was downloaded by: [Department Of Fisheries] On: 25 September 2012, At: 20:30 Publisher: Taylor & Francis Informa Ltd Registered in England and Wales Registered Number: 1072954 Registered office: Mortimer House, 37-41 Mortimer Street, London W1T 3JH, UK

Transactions of the American Fisheries Society Publication details, including instructions for authors and subscription information: http://www.tandfonline.com/loi/utaf20 An Age-Structured Integrated Assessment of Chinook Salmon Population Dynamics in Lake Huron's Main Basin since 1968 Travis O. Brenden a , James R. Bence a & Emily B. Szalai a a Quantitative Fisheries Center, Department of Fisheries and Wildlife, Michigan State University, 153 Giltner Hall, East Lansing, Michigan, 48824, USA Version of record first published: 22 Jun 2012.

To cite this article: Travis O. Brenden, James R. Bence & Emily B. Szalai (2012): An Age-Structured Integrated Assessment of Chinook Salmon Population Dynamics in Lake Huron's Main Basin since 1968, Transactions of the American Fisheries Society, 141:4, 919-933 To link to this article: http://dx.doi.org/10.1080/00028487.2012.675910

PLEASE SCROLL DOWN FOR ARTICLE

Full terms and conditions of use: http://www.tandfonline.com/page/terms-and-conditions 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. The publisher does not give any warranty express or implied or make any representation that the contents will be complete or accurate or up to date. The accuracy of any instructions, formulae, and drug doses should be independently verified with primary sources. The publisher shall not be liable for any loss, actions, claims, proceedings, demand, or costs or damages whatsoever or howsoever caused arising directly or indirectly in connection with or arising out of the use of this material. Transactions of the American Fisheries Society 141:919–933, 2012 C American Fisheries Society 2012 ISSN: 0002-8487 print / 1548-8659 online DOI: 10.1080/00028487.2012.675910

ARTICLE

An Age-Structured Integrated Assessment of Chinook Salmon Population Dynamics in Lake Huron’s Main Basin since 1968

Travis O. Brenden,* James R. Bence, and Emily B. Szalai Quantitative Fisheries Center, Department of Fisheries and Wildlife, Michigan State University, 153 Giltner Hall, East Lansing, Michigan 48824, USA

Abstract We conducted a statistical catch-at-age (SCAA) assessment, also known as an integrated assessment, of the Chinook salmon Oncorhynchus tshawytscha population in Lake Huron’s main basin to describe the dynamics and assess population abundance since the species began being stocked in 1968. The purpose of this assessment was to determine whether recent declines in Chinook salmon recreational harvest were indicative of an overall decrease in population abundance, and to estimate the magnitude of that decline. Data sources integrated into the assessment model included recreational harvest and effort, returns to the Swan River weir by year-class, age composition of recreational harvest (overall and of mature fish), and proportion mature by age. Parameters estimated included time-varying age-0 natural mortality rates and recreational fishing catchabilities, age-specific recreational fishing selectivities, and the parameters of a logistic function relating maturation probability to fish age and weight. The fitted model did a reasonable job of predicting recreational harvest, Swan River weir returns, age composition of recreational harvest, and proportion mature by age. The model predicted that abundance of age-1 and older Chinook salmon has declined by approximately 98% since the mid-1980s, with an estimated abundance of age-1 and older fish of approximately 65,000 in 2009. The primary cause for this decline has been a substantial increase in age-0 natural mortality rates, which peaked at an instantaneous rate of around 6.3 in the late 2000s, although fishing mortality rates have also been quite high. Initially, declines in survival were associated with increases in overall predator abundance in the lake. The most recent and large declines were associated with a decrease in the abundance of alewives Alosa pseudoharengus. We suspect that declines in age-0 survival reflect both direct effects of food limitation and increased predation mortality by larger fish, and are consequences of decreased abundance of other prey fishes.

Chinook salmon Oncorhynchus tshawytscha began being from major lakewide declines in abundance of native preda- Downloaded by [Department Of Fisheries] at 20:30 25 September 2012 stocked into Lake Huron in 1968 as a means to reduce densi- tors, which resulted in few recreational angling opportunities ties of exotic pelagic prey fishes, primarily alewife Alosa pseu- and very little predation pressure on pelagic prey fishes. Other doharengus and rainbow smelt Osmerus mordax, and increase Pacific salmonid species, including kokanee O. nerka and coho recreational fishing opportunities in the lake (Berst and Spangler salmon O. kisutch, also began being stocked into Lake Huron 1972; Tanner and Tody 2002; Dobiesz et al. 2005). Although in the late 1960s and early 1970s (Berst and Spangler 1972); stocking of Chinook salmon in the Great Lakes occurred as early however, Chinook salmon quickly became the preferred species as the late 1800s, survival from these stockings was low and for stocking by both fishery managers and anglers because of its as a result plantings were eventually discontinued (Kocik and fast growth, high angling quality, low production cost, and high Jones 1999). Renewed interest in establishing Chinook salmon rate of alewife predation (Hansen and Holey 2002; Mills et al. in the Great Lakes in the late 1960s and early 1970s resulted 2003). By the late 1970s, more than 1 million Chinook salmon

*Corresponding author: [email protected] Received July 19, 2011; accepted February 13, 2012 Published online June 22, 2012 919 920 BRENDEN ET AL.

were being stocked annually into the main basin of Lake Huron. (T. Kolb, MDNR, unpublished data). Size at age and body con- By the mid-1980s, the number stocked annually had risen to dition of Chinook salmon have also declined considerably, even more than 3 million. to levels below those observed in Lake Michigan when massive As in the other Great Lakes, the establishment of Chinook die-offs of Chinook salmon occurred (Holey et al. 1998; Bence salmon in Lake Huron is widely considered as having been suc- et al. 2008). Although there is no evidence that similar die-offs cessful in achieving the dual goals of establishing a recreational have occurred in Lake Huron, Chinook salmon stocking rates fishery and reducing pelagic prey densities (Goddard 2002). have nevertheless been reduced in recent years out of concern In the 1980s and 1990s, Chinook salmon composed most of of possible predator–prey imbalances within the lake. The pri- the recreational harvest from Lake Huron’s main basin for both mary purpose of our research was to conduct an age-structured, Michigan and Ontario anglers (Bence and Smith 1999). Alewife integrated assessment of the population dynamics of Chinook and rainbow smelt densities also declined considerably due at salmon in Lake Huron’s main basin to provide age-specific es- least in part to increased predation (O’Gorman and Stewart timates of abundance based on a coherent set of assumptions 1999; Dobiesz 2003; Riley et al. 2008). It is important to note, and use of available data. The analysis was intended to place however, that not everyone views the introduction or continued the understanding of what Chinook salmon dynamics have been stocking of Chinook salmon as beneficial to Great Lakes fish on a firmer foundation. By more clearly identifying the extent communities because of the possibility of disease transmission, of declines over time on an age-specific basis, we also hoped agonistic interactions with native fish species, and alterations to to shed some light on what factors might be responsible for the the environment and fish communities (Crawford 2001). declines. Although Chinook salmon continue to be widely stocked into Lake Huron, recent research suggests that the population METHODS has become predominantly supported through wild reproduction (Johnson et al. 2010). Wild reproduction in Lake Huron tribu- Study Area taries was first detected in the mid-1980s, and by the mid-1990s Lake Huron is the second largest of the Laurentian Great 15% of recruits were estimated to be from wild reproduction Lakes in surface area and the fifth largest lake in the world. (Bence et al. 2008). Johnson et al. (2010) recently conducted Although it is hydrologically connected to Lake Michigan, the an evaluation of wild reproduction by marking all fish stocked two lakes are typically regarded as separate systems for man- into Lake Huron with oxytetracyclene and fin clips and then agement purposes. Lake Huron is considered to consist of three checking for presence of marks in fish collected from fishing relatively discrete water masses: the main basin (including Sag- tournaments, as bycatch in commercial fishery operations, and inaw Bay), North Channel, and Georgian Bay (Figure 1). The as part of spawning tributary returns. Those investigators esti- lake receives outflow from Lake Superior into the North Chan- mated a lakewide wild reproduction rate of approximately 80%, nel via the St. Mary’s River and discharges into Lake St. Clair with wild reproduction rates in some areas of the lake exceed- through the St. Clair River. With the exception of Saginaw Bay ing 95%. The factors that have contributed to this increased and several nearshore areas, Lake Huron is generally regarded wild reproduction of Chinook salmon, or alternatively decreased as oligotrophic (Dobiesz et al. 2005). Lake Huron is bordered survival of stocked fish, are not presently known. Possible and jointly managed by the state of Michigan and the Province contributory factors include declining abundance of alewives, of Ontario. Several Native American and First Nation tribes also which is believed to have led to resurgence in wild reproduc- participate in the management of Lake Huron fisheries (Brown tion of species such as emerald shiner Notropis atherinoides et al. 1999). Our assessment of the Chinook salmon population (Schaeffer et al. 2008), walleye Sander vitreus (Fielder et al. only included the main basin of Lake Huron as there was insuf-

Downloaded by [Department Of Fisheries] at 20:30 25 September 2012 2007), and lake trout Salvelinus namaycush (Riley et al. 2007), ficient information to include North Channel and Georgian Bay improved conditions for spawning in Lake Huron tributaries, in the modeling exercise. and higher rates of predation on stocked Chinook salmon (including cannibalism and predation by other species such as lake Data Collection trout and walleye) as a result of overall declines in prey abun- Recruitment.—Information on recruitment came from both dance (J. Johnson, Michigan Department of Natural Resources stocking data and estimates of wild reproduction. Stocking [MDNR], personal communication). numbers for the time period from 1968 to 2009, inclusive, Despite the apparent shift of the Lake Huron Chinook salmon were obtained from MDNR and Ontario Ministry of Natural population to one primarily supported through wild reproduc- Resources (OMNR) databases (A. Cottrill, Ontario Ministry tion, there is wide concern on the part of fishery managers of Natural Resources, unpublished data; J. He, MDNR, and recreational fishers about the overall status of the popu- unpublished data). Information on wild reproduction rates lation (Bence et al. 2008). Recreational fishery harvest of Chi- came from several sources. Estimates of wild reproduction nook salmon has declined considerably in recent years, although rates for Ontario and Michigan waters in the early 1990s were recreational fishing effort has also declined so it is not clear to provided by MDNR and OMNR biologists through the Lake what extent the harvest declines reflect declines in abundance Huron Technical Committee. Estimates for the 2000 to 2003 CHINOOK SALMON ASSESSMENT IN LAKE HURON 921

reproduction of Chinook salmon in this or other nearby streams (Fenske and Shouder 1992; Johnson et al. 2010). As a result, returns to the Swan River weir per number of fish stocked by year-class provided a fishery-independent means to evaluate survival of the population. Age composition, proportion mature, and weight at age.— Observations on age composition of the recreational harvest, weights at age during the month of August, and proportion mature by age during the month of July and August for the time period from 1983 to 2009 were obtained from an MDNR database of Chinook salmon biological data collected as part of the annual creel survey program (T. Kolb, MDNR, unpublished data). Age compositions for the Michigan recreational harvest were derived from scale readings, while maturation status was judged based on internal examination of harvested individuals. Calculated age compositions included all fish collected during the course of a year, as well as age composition of only mature Chinook salmon collected during the months of July and Au- gust. Weights at age of Chinook salmon for the month of August for the time period from 1973 to 1982 were predicted from a time-varying growth model (He and Bence 2007) that included measurements on fish weight from individuals collected from the creel survey program and individuals collected from spawn- FIGURE 1. Lake Huron depicting the main basin, Georgian Bay, and North ing weir returns (J. He, MDNR, unpublished data). Weights at Channel, along with the location of the Swan River. age of Chinook salmon for the time period 1968–1972 were set equal to 1973 predicted values. Lake Huron Chinook salmon year-classes were obtained from Johnson et al. (2010). Statistical Catch-at-Age Modeling Recreational fishery harvest and effort.—Recreational har- We constructed a statistical catch-at-age (SCAA) model to vest and effort data for Michigan jurisdictional waters from predict age-specific abundances and mortality rates of Chinook annual creel surveys were obtained from MDNR databases (T. salmon in the main basin of Lake Huron for the time period Kolb, MDNR, unpublished data). This data series covered the from 1968 to 2009. Statistical catch-at-age models consist of time period from 1986 to 2009 and were expanded to include a population submodel that projects abundances at age of the estimates from all major Chinook salmon harvest ports for Lake modeled population and an observation submodel that predicts Huron. Complete surveys of recreational harvest within Ontario observable quantities (e.g., recreational harvest, survey catch jurisdictional waters of Lake Huron are not regularly conducted. per effort) from these projected abundances. Definitions of pa- Based on the available creel surveys, recreational harvest and rameters and variables used as part of the SCAA model and effort during the 1989 to 1999 period has been estimated for On- given in equations that follow are presented in Table 1. tario waters of the main basin (L. Mohr, OMNR, unpublished Population submodel.—Predicted Chinook salmon abun-

Downloaded by [Department Of Fisheries] at 20:30 25 September 2012 data) and has averaged 9% (range: 4–17%) and 4% (range: 1– dances at age for age-0 to age-5 ﬁsh were computed using 8%) of harvest and effort in Michigan waters, respectively. We the exponential population model in monthly time steps, with a estimated overall harvest and effort for years before 1989 and pulse of mortality of ﬁsh that matured occurring between month after 1999 by increasing the Michigan recreational harvest and 9 and month 10. The monthly time step and pulse of mortality effort values by these percentages. was required, in contrast with the usual exponential model us- Swan River weir returns.—Estimates of the number of re- ing annual time steps, because Chinook salmon are semelparous. turning salmon to the Swan River weir were obtained from Thus, except for month 10, Chinook salmon abundances at age Fenske and Shouder (1992) and MDNR databases (J. Clevenger, at the start of each month were represented as MDNR, unpublished data). The Swan River is located in the northern lower peninsula of Michigan (Figure 1) and histori- Ny,m+1,a = Ny,m,a exp(−Zy,m,a), cally has been a major MDNR Chinook salmon stocking site. On average, more than 750,000 Chinook salmon have been and initial monthly abundances at age for all but the youngest stocked annually at the Swan River weir since the early 1980s. age-class were equal to Because the Swan River receives discharge from several nearby limestone quarries, there has been very little evidence of wild Ny+1,1,a+1 = Ny,12,a exp(−Zy,12,a). 922 BRENDEN ET AL.

TABLE 1. Description of equation symbols used in the model and text for age-structured assessment of Chinook salmon population dynamics in Lake Huron.

Symbol Description Indicator variables a Age-class (0–5) y Year (1968–2009) m Month (1–12) Estimated parameters M1982,0 Instantaneous natural mortality for age-0 Chinook salmon in 1982 γ Age-0 instantaneous natural mortality random walk deviations q1986 Recreational fishing catchability in 1986 Recreational fishing catchability random walk deviations s Recreational fishing selectivities θ Parameters of maturation mortality function qSR Catchability for Swan River weir returns σ 2 Standard dispersion parameter for objective function data components Calculated and assumed quantities N Abundance Nˆ Model estimated abundance N˜ Abundance at end of the ninth month after fishing and natural mortality have occurred N˜ˆ Model estimated abundance at end of the ninth month after fishing and natural mortality have occurred Z Total instantaneous mortality M Instantaneous natural mortality F Instantaneous recreational fishing mortality PwildEarly Wild reproduction rates for the 1987 to 1993 year-classes PwildLate Wild reproduction rates for the 2000 to 2003 year-classes Wild Wild reproduction of Chinook salmon Hˆ Model estimated recreational fishery harvest Hˆ Model estimated recreational fishery harvest of mature fish Pˆ Model estimated recreational fishery harvest age composition Pˆ Model estimated recreational fishery harvest age composition of mature fish Rˆ Model estimated abundance of fish returning to tributaries to spawn ˆ Model estimated proportion mature Iˆ Model estimated returns to Swan River weir per number stocked at the weir λ Weighting value for objective function data sources NE Effective sample size for age composition and proportion mature by age Data H Observed recreational fishery harvest Downloaded by [Department Of Fisheries] at 20:30 25 September 2012 E Observed recreational fishery effort W Weight Stock Number of stocked Chinook salmon P Observed recreational fishery harvest age composition P Observed recreational fishery harvest age composition of mature fish Observed proportion mature I Observed returns to Swan River weir per number of fish stocked at the weir Distributional parameters 2 σ Standard dispersion for fishery catchability random walk deviations 2 σγ Standard dispersion for age-0 natural mortality random walk deviations CHINOOK SALMON ASSESSMENT IN LAKE HURON 923

Maturation mortality, corresponding to that fraction of the Chi- tality rates as follows: nook salmon population that mature and return to streams to ⎧ spawn and die, was assumed to occur at the end of the ninth ⎨ qy ;fory = 1986 month of each year. Thus, projected abundances at age of Chi- qy = ⎩ qy− exp y− ; for 1987 ≤ y ≤ 2009 nook salmon at the end of the ninth month, subsequent to harvest 1 1 and natural mortality were represented as 2 y−1 ∼ N 0,σ .

N˜ y,9,a = Ny,9,a exp(−Zy,9,a), Catchabilities for the time period 1968–1985 were set equal to the 1986 value. Because measures of recreational fishing effort with abundance at the start of the 10th month set equal to were not available before 1986, we assumed effort increased linearly from a level of 0 in 1968 to the level measured in 1986. N = N˜ − . y,10,a y,9,a 1 y,a Recreational fishing selectivities were estimated as freely varying model parameters for age-1 to age-3 fish and were set Total mortality (excluding maturation mortality) was partitioned equal to 0 for age-0 fish and 1.0 for age-4 and older fish. The into two components: natural mortality and recreational fishing relative proportion of fishing that occurs in each month of the mortality; i.e., year was calculated from MDNR creel survey monthly fishing effort estimates for major Lake Huron salmonid fishing ports using the month of July as the standard. Zy,m,a = My,m,a + Fy,m,a. The probability of maturation was modeled as a logistic function of weight, age, and the interaction between weight and age: Instantaneous natural mortality rates for age-1 and older fish were assumed known and set equal to 0.3 for age-1 fish and 1 y,a = . 0.1 for age-2 and older fish, which are values that have been 1 + exp(−θ1 − θ2a − θ3Wy,a − θ4aWy,a) used in other Great Lake Chinook salmon modeling efforts (e.g., Stewart et al. 1981; Stewart and Ibarra 1991; Bence and Dobiesz We chose this function for modeling maturation probability as it 2000; Benjamin and Bence 2003). Monthly natural mortality previously has been found to be one of the best performing and rates were obtained by dividing annual rates by 12. For age- most parsimonious models for describing Lake Huron Chinook 0 Chinook salmon, instantaneous natural mortality rates were salmon maturation (J. R. Bence, Michigan State University, time-varying. For the time period from 1982 to 2008, age-0 unpublished data). natural mortality were estimated via a random walk process as The total number of age-0 fish at the start of each year com- follows: prised both stocked and wild-reproduced fish. This total number ⎧ was determined as follows. First, the total number of fingerling ⎨ My,0;fory = 1982 Chinook salmon stocked in Lake Huron’s main basin was cal- My,0 = culated by summing the numbers stocked in Ontario and Michi- ⎩ My− , exp(γy− ); for 1983 ≤ y ≤ 2008 1 0 1 gan jurisdictional waters. The total number of Chinook salmon 2 stocked was then multiplied by an assumed survival rate of γy− ∼ N 0,σ . 1 γ 75% to account for poststocking mortality, which gave the total number of recruits from hatchery sources. The number of wild Downloaded by [Department Of Fisheries] at 20:30 25 September 2012 Natural mortality rates for 1980 and 1981 were set equal to the recruits was determined from the estimated wild reproduction estimated 1982 rate, while the 2009 rate was set equal to that in rates available from the early 1990s and 2000s (see previous 2008. For the time period of 1968 through 1979, age-0 natural Data Collection section). The estimated wild reproduction rates mortality rates were assumed to change linearly from a starting from the early 1990s were assumed to apply to the 1987–1993 level of 0.70 to the estimated rate in 1982. year-classes, with the number of wild recruits produced in these Monthly recreational fishing mortalities were assumed to be years calculated as: products of annual fishing effort, age-specific selectivities, year- specific catchabilities, and the relative proportion of fishing that 1993 PwildEarly y=1987 Stocky occurs in each month of the year as follows: Wildy = ; 1 − PwildEarly (1993 − 1987) + 1 for 1987 ≤ y ≤ 1993. Fy,m,a = qy saEy fm.,

For Michigan waters of Lake Huron, PwildEarly was set at 15%, Year-speciﬁc catchabilities for the time period of 1986 to 2009 while for Ontario waters it was set at 45% based on input pro- were estimated via random walk similar to age-0 natural mor- vided by MDNR and OMNR biologists through the Lake Huron 924 BRENDEN ET AL.

Technical Committee. A similar approach was used to calculate Because Swan River weir return data were only available the number of wild recruits produced from 2000 to 2003 through the 2007 Chinook salmon year-class, the number of ages that could be included in the calculation of returns per 2003 Stock number stocked varied for the 2004 to 2007 year-classes, PwildLate y=2000 y Wildy = ; with the maximum age-class that could be included equal to 1 − Pwild (2003 − 2000) + 1 Late 2008 − y where y was the year-class for which returns were for 2000 ≤ y ≤ 2003, calculated. Model fitting.—The Lake Huron Chinook salmon SCAA where PwildLate was set equal to the mean wild reproduction model was implemented in AD model Builder (Fournier et al. rate estimated for these year-classes estimated in Johnson et al. 2012). We adopted a Bayesian-based approach, whereby we (2010). The wild recruitment level estimated for the 2000 to used the set of parameters with the maximum posterior density 2003 year-classes was assumed to apply to the 2004 to 2009 (highest posterior density estimates) as point estimates (Schnute year-classes as well. For the year-classes prior to 1987, we 1994). Operationally this was accomplished by defining an assumed that there was no wild recruitment of Chinook salmon objective function equal to the negative log posterior density in 1968 and 1969, but that wild recruitment increased linearly (ignoring some constants) and numerically searching for to 1987 levels for years 1970 and later. We also assumed that the parameters that minimized the negative log posterior. wild recruitment of Chinook salmon increased linearly between This was done using a quasi-Newton optimization algorithm 1993 and 2000 levels. with convergence on the minimum of the objective function Observation submodel.—Harvest at age (in numbers) for the based on AD model Builder default termination criteria. The Chinook salmon recreational fishery by year was calculated with objective function for the SCAA model consisted of the sum the Baranov catch equation: of seven negative log-likelihood or negative log prior (penalty) components (Table 2). Lognormal distributions were assumed Fˆ y,m,a for the negative log likelihoods for recreational fishery harvest Hˆ y,a = Nˆ y,m,a m Fˆ y,m,a + My,a/12 (equation T.2.1 in Table 2) and returns to the Swan River weir per number of fish stocked at the weir (equation T.2.2). × (1 − exp(−Fˆ y,m,a − My,a/12)). Similarly, the negative log prior components for the random walk perturbations for recreational fishery catchability (equa- Recreational harvest of mature fish was calculated by summing tion T.2.3) and age-0 natural mortalities (equation T.2.4) were July and August harvests and multiplying summed harvest by assumed to be distributed lognormally. We assumed that the predicted probability of maturation as follows: negative log likelihoods for the age compositions of recreational harvest (equation T.2.5) and recreational harvest of mature Hˆ = Hˆ + Hˆ ˆ y,a. y,a y,7,a y,8,a fish (equation T.2.6.6) followed multinomial distributions, and that the probability of maturation for each Chinook salmon Recreational harvest age composition was calculated by di- age-class followed a binomial distribution (equation T.2.7). viding harvests at age by total recreational harvest. Similarly, In the lognormal negative log-likelihood and negative recreational-harvest age composition of mature fish was calcu- log-prior densities and components listed in Table 2, σ 2 is a lated by dividing harvests at age of mature fish by total recre- “standard dispersion parameter” that is estimated as part of the ational harvest of mature fish. model-fitting process and the λi term is a weighting factor that The number at age of mature fish returning to tributaries to represents how far the dispersion parameter for a particular data

Downloaded by [Department Of Fisheries] at 20:30 25 September 2012 spawn was calculated as the product of the predicted probability source deviates from this “standard”. An iterative approach of maturation and the projected abundance of Chinook salmon for setting the λ-values was used whereby initial data source at the end of the ninth month, i.e.: weightings were designated based on perceived reliability of the data sources. The assumed dispersions were then compared to ˆ Rˆ y,a = N˜ y,9,aˆ y,a. the SCAA model mean-squared error for the different sources. The λ-values were then adjusted based on the differences The total number of returns to the Swan River weir per number between the assumed dispersions and fitted SCAA model stocked for each year-class was calculated by summing age- residual estimates, until the assumed dispersions and fitted specific returns divided by number stocked and multiplying by SCAA model residual estimates were approximately equal or a catchability value to account for factors such as straying and were reasonable given the data source (Table 3). Negative log source-specific survival, as follows: likelihood components for age composition of the recreational harvest (both overall harvest and harvest of mature fish) and 5 the proportion mature by age were weighted by the effective Iˆ = q Rˆ / . sample size, which was the number of fish for which ages or y ˆSR y+a,a StockSwan Rivery a=1 maturation were measured each year up to a maximum of 50. CHINOOK SALMON ASSESSMENT IN LAKE HURON 925

TABLE 2. Equations and descriptions of the negative log likelihood and negative log prior components for the Lake Huron main basin Chinook salmon statistical catch-at-age (SCAA) model; equation numbers correspond with those mentioned in the text. The objective function of the SCAA model, which consisted ofthe sum of these components, was minimized during the model-ﬁtting process.

Equation Equation Description number H 2 L = n √σ + . λH y H H loge 0 5 2 loge Recreational fishery harvest T.2.1 λH σ Hˆ y y I 2 L = n √σ + . λI y I I loge λ 0 5 σ 2 loge Iˆ Returns to Swan River weir per number of fish stocked at the weir T.2.2 I y y L = n √σ + . λ 2 loge λ 0 5 σ 2 loge( y ) Recreational fishery catchability random walk deviations T.2.3 y σ λγ 2 Lγ = nγ √ + . γy loge λ 0 5 σ 2 loge( ) Age-0 natural mortality random walk deviations T.2.4 γ y L =− N P Pˆ P EP y,a loge y,a Recreational fishery harvest age compositions T.2.5 y a L =− N P Pˆ P EP y,a loge y,a Recreational fishery harvest age compositions of mature fish T.2.6 y a during months of July and August L =− N + − E y,a loge ˆ y,a 1 y,a Proportion mature by age of fish from recreational fishery harvest T.2.7 y a during months of July and August × loge 1 − ˆ y,a

Diffuse, uniform (on a loge scale) priors were implicitly as- to determine if the chain had converged on a stationary distribu- signed to each estimated parameter, except for those for which tion (Geweke 1992). All MCMC chain convergence diagnostics there was an explicit component in the objective function. To were conducted in R (R Development Core Team 2010) using construct probability intervals, posterior distributions for pa- the “coda” package (Plummer et al. 2010). rameters and derived variables were obtained by Markov Chain Monte Carlo (MCMC) simulations with a Metropolis Hastings Sensitivity and Retrospective Analyses algorithm (Fournier et al. 2012). The MCMC chain was run Sensitivity analyses were performed to determine how model for 1 million steps sampling every 100th step. The initial 3,000 estimates were affected by some of the key assumptions that saved steps were discarded as a burn-in period. Convergence of were made when configuring the SCAA model. We conducted the MCMC chain was evaluated with three methods. First, trace sensitivity analyses on weightings of individual data sources, plots of each estimated parameter were constructed to ensure maximum effective sample sizes for estimating age composition the chain was well-mixed and did not exhibit substantial “stick- of recreational fishery harvest and proportions mature by age, iness” over long portions of the chain. Second, we compared the and assumed natural mortality rates for age-1 and older Chinook effective sample size of the saved MCMC chain with the actual salmon. For each of these variables, sensitivity was explored by chain sample size to determine if there were large differences increasing and decreasing their values threefold and refitting between the values. Lastly, the means of the first 10% and last the SCAA model. Sensitivity was judged by comparing model 50% of the saved chain were tested using a standard Z-score test Downloaded by [Department Of Fisheries] at 20:30 25 September 2012 estimates of abundance of age-1 and older fish, instantaneous natural mortality rates of age-0 fish, and mean instantaneous fishing mortality rates of age-2 and older fish with those obtained TABLE 3. Weighting factors (λ) for negative log likelihood and negative log from the baseline weightings, effective maximum sample sizes, prior components that were assumed to be lognormally distributed. and natural mortality rates for age-1 and older fish. Description λ A retrospective analysis was conducted to determine if the Chinook salmon SCAA model exhibited systematic biases in Recreational fishery harvest 1.00 model parameters or estimates. This analysis involved deleting Return to Swan River weir per number of fish stocked 0.25 a year of data from the SCAA model and refitting it (Mohn at the weir 1999); this was repeated until 8 years of data had been removed Recreational fishery catchability random walk 0.05 sequentially from the model. We then examined retrospective deviations patterns for age-1 and older fish abundance, the instantaneous Age-0 instantaneous natural mortality random walk 0.50 natural mortality rate for age-0 fish, and mean instantaneous deviations fishing mortality rates for age-2 and older fish. 926 BRENDEN ET AL.

RESULTS age-1 and older Chinook salmon abundance between the mid- 1980s and 2009. Model Fits The predicted decline in abundance has occurred despite the The Chinook salmon SCAA model successfully converged large increase in wild reproduction of Chinook salmon. Based on a solution. The MCMC chain was also judged to have on the wild reproduction rates provided by the Lake Huron converged based on all three criteria used to evaluate conver- Technical Committee and Johnson et al. (2010), annual wild gence. There was considerable contrast in observed recreational reproduction of Chinook salmon has increased from around fishery harvest and returns to the Swan River weir by year-class 790,000 recruits in the early 1990s to more than 10 million re- per number of Chinook salmon stocked, but the fitted SCAA cruits in the early 2000s (Figure 4). This large increase in wild model appeared to do a reasonable job of fitting both data reproduction appears to have been offset, however, by a large in- sources (Figure 2). The mean absolute percent error between crease in age-0 Chinook salmon mortality over the last 30 years observed and predicted recreational harvest was approximately (Figure 5). In the mid-1980s, age-0 instantaneous natural mor- 10%, while for returns to the Swan River weir it was approx- tality was estimated to be approximately around 1.0 (finite rate, imately 50%. The better fit to the recreational harvest data ∼60%). By the early 1990s, model estimates of natural mortal- were probably due in part to the lower weighting assigned ity were as high as 2.0 (finite rate, ∼85%). For the late 2000s, to the Swan River weir return data relative to that assigned model estimates of natural mortality were around 6.3 (finite rate, to the recreational harvest data. Both observed and predicted ∼99%). recreational fishery harvest exhibited substantial fluctuations As previously indicated, the recreational fishery effort in with no clear directional trend between 1985 and 2003. Since Lake Huron’s main basin has declined over the last 25 years 2003, however, recreational harvest has declined considerably (Figure 6). Between the late 1980s and late 2000s, total fishing (Figure 2). Observed and predicted returns to the Swan River effort has declined by approximately 65%. According to our weir per number of fish stocked by year-class also fluctuated fitted SCAA model, however, recreational fishing catchability somewhat, but unlike recreational harvest there has been a has generally increased during this same time period, which generally consistent decline in returns since the mid-1980s. somewhat compensated for declining effort (Figure 6). There Observed mean age of the recreational harvest, both over- was a substantial increase in estimated catchability between all and for mature fish during the months of July and August, 2002 and 2004, which was when there was a total recruitment exhibited similar fluctuating patterns as recreational harvest, failure of alewife in Lake Huron (Bence et al. 2008) suggesting but the fitted SCAA model did an adequate job of reproducing that perhaps Chinook salmon were very vulnerable to harvest these fluctuations (Figure 2). The mean absolute percent error as a result of declining prey abundance. Model estimates of between observed and predicted mean recreational harvest age fishing selectivity were 0.022, 0.096, and 0.288 for age-1, age-2, composition was approximately 3%, while for recreational har- and age-3 Chinook salmon, respectively. Mean age-2 and older vest age composition for mature fish during the months of July instantaneous fishing mortality rates were below 0.40 (finite rate, and August it was approximately 6%. <32%) through the late 1980s, but increased steadily thereafter The mean observed proportion of mature Chinook salmon (Figure 5). Fishing mortality spiked between 2002 and 2004 by age for the time period from 1983 to 2009 ranged from 16% with fishing rates ranging from 2.8 to 3.7 (finite rates, 94–97%). for age-1 fish to nearly 98% for age-4 fish (Figure 2). The mean Since 2007, mean instantaneous fishing mortality rates of age-2 predicted proportion of mature Chinook salmon by age for the and older fish averaged approximately 1.3 (finite rate, ∼73%). same time period ranged from 14% for age-1 fish to 93% for age-4 fish (Figure 2).

Downloaded by [Department Of Fisheries] at 20:30 25 September 2012 Sensitivity and Retrospective Analyses The SCAA model estimates were relatively insensitive to Model Estimates changes in weighting of data components and effective max- According to the fitted SCAA model, abundance of age- imum sample sizes for estimating recreational fishery harvest 1 and older Chinook salmon in 2009 was at levels not seen age compositions and proportions mature by age. Threefold in- since the early 1970s (Figure 3). Abundance of age-1 and older creases or decreases in weighting of individual data components, fish peaked in the mid-1980s at approximately 3.7 million fish maximum effective sample sizes, and age-1 and older natural (Table 4), but steadily declined thereafter through the late 1980s. mortality rates still resulted in an estimated 97–99% decline in Abundance stabilized at approximately 1.0 million age-1 and abundance between the mid-1980s and 2009. Estimated abun- older Chinook salmon from the early to late 1990s, before it dances of age-1 and older fish for the last year of the model began its precipitous drop to current levels (Figure 3). In 2009, ranged from approximately 36,000 to 104,000 fish. Similarly, the estimated abundance of age-1 and older Chinook salmon age-0 instantaneous natural mortality rates showed a generally was approximately 65,000 fish (Table 4), with lower and upper increasing trend with peak rates ranging from approximately 95% Bayesian probability limits of 22,000 and 114,000 fish, 5.9 to 7.1 in recent years. Peak estimates of mean age-2 and respectively. This level equates to a roughly 98% decline in older instantaneous fishing mortality rates across the sensitivity CHINOOK SALMON ASSESSMENT IN LAKE HURON 927 Downloaded by [Department Of Fisheries] at 20:30 25 September 2012

FIGURE 2. Model fits to recreational fishery harvest (in 10,000s) by year (upper left panel) and Swan River weir returns by year-class per number of Chinook salmon stocked at the weir (upper right panel) for Lake Huron’s main basin during 1986 to 2009 (1983 to 2007 for Swan River returns). Also shown are the mean age of recreational harvest and the mean age of recreational harvest of mature fish during the months of August and September from 1986 to 2009, and the proportion of fish mature by age. Model estimates are highest posterior density estimates and are represented by solid lines; observed values are represented by dots. 928 BRENDEN ET AL.

TABLE 4. Statistical catch-at-age model estimated abundance at age (in 100,000s) of Chinook salmon in Lake Huron since 1968.

Age-class (years) Year Age 0 Age 1 Age 2 Age 3 Age 4 Age 5 1968 2.055 0.000 0.000 0.000 0.000 0.000 1969 1.875 1.020 0.000 0.000 0.000 0.000 1970 5.262 0.961 0.643 0.000 0.000 0.000 1971 7.581 2.785 0.606 0.301 0.000 0.000 1972 4.948 4.143 1.755 0.283 0.046 0.000 1973 9.007 2.792 2.609 0.818 0.043 0.002 1974 8.012 5.248 1.758 1.215 0.123 0.002 1975 7.544 4.820 3.299 0.790 0.165 0.004 1976 9.295 4.685 3.036 1.538 0.112 0.005 1977 8.329 5.960 2.955 1.417 0.223 0.003 1978 13.823 5.514 3.765 1.397 0.206 0.007 1979 14.325 9.449 3.473 1.782 0.210 0.006 1980 18.900 10.110 5.967 1.625 0.250 0.006 1981 17.613 13.773 6.367 2.872 0.262 0.009 1982 20.699 12.835 8.635 2.904 0.419 0.008 1983 26.350 15.084 8.100 3.721 0.334 0.009 1984 30.141 19.638 9.553 4.405 0.543 0.014 1985 29.712 19.564 12.449 4.657 0.788 0.023 1986 32.818 13.517 12.418 6.203 0.770 0.033 1987 37.047 10.126 8.550 6.048 1.137 0.036 1988 40.061 6.197 6.317 4.407 1.089 0.052 1989 42.679 3.954 3.913 3.191 0.711 0.038 1990 40.894 4.851 2.482 2.032 0.441 0.021 1991 34.917 5.446 2.977 1.097 0.314 0.010 1992 34.405 5.668 3.404 1.404 0.140 0.007 1993 36.897 4.764 3.536 1.628 0.200 0.003 1994 52.419 4.594 3.001 1.698 0.218 0.004 1995 69.804 6.447 2.893 1.415 0.205 0.004 1996 81.508 5.590 3.976 1.302 0.153 0.002 1997 94.424 3.622 3.469 1.764 0.132 0.002 1998 109.264 3.534 2.241 1.634 0.225 0.001 1999 121.665 2.203 2.188 1.100 0.254 0.003 2000 127.249 2.709 1.338 0.960 0.133 0.003 2001 128.513 2.894 1.669 0.574 0.104 0.001 2002 126.825 2.405 1.766 0.712 0.059 0.001 Downloaded by [Department Of Fisheries] at 20:30 25 September 2012 2003 127.155 1.207 1.387 0.596 0.036 0.000 2004 126.921 0.561 0.685 0.456 0.026 0.000 2005 125.929 0.329 0.314 0.217 0.021 0.000 2006 114.909 0.358 0.200 0.142 0.031 0.000 2007 114.474 0.570 0.217 0.084 0.016 0.000 2008 115.941 0.473 0.342 0.090 0.009 0.000 2009 114.661 0.208 0.287 0.147 0.012 0.000

scenarios that were evaluated ranged from 3.1 to 4.8 in the early (Figure 7). For instantaneous natural mortality of age-0 ﬁsh, 2000s. there was some variability associated with the terminal year’s No retrospective patterns were evident in abundances of age- estimated mortality rate (Figure 7). However, this was to be 1 and older salmon, age-0 instantaneous natural mortality rates, expected given that there was little information available in the or mean age-2 and older instantaneous ﬁshing mortality rates existing data sources to inform the estimation procedure as to CHINOOK SALMON ASSESSMENT IN LAKE HURON 929

4.5

4.0

3.5

3.0

2.5

2.0

1.5

Age-1+ Abundance 1.0

0.5

0.0 1970 1980 1990 2000 2010 Year

FIGURE 3. Estimated abundance (solid line) of age-1 and older Chinook FIGURE 5. Model estimates of mean age-2 and older instantaneous fishing salmon (in millions) in the main basin of Lake Huron from 1968 to 2009. mortality (gray line) and age-0 natural mortality (black line) rates for Chinook Estimated abundances are the higher posterior density estimates; dashed lines salmon in the main basin of Lake Huron from 1968 to 2009. Estimated mortality indicate 95% Bayesian probability intervals for abundance. rates are highest posterior density estimates; dashed lines indicate 95% Bayesian probability intervals of mortality estimates. what age-0 natural mortality rates should be for the terminal year. species that have begun showing signs of recovery if some- thing were to happen to the Chinook salmon populations. This perceived importance of Chinook salmon to Great Lakes fish DISCUSSION communities and the surrounding region is evident from the Since Chinook salmon began being stocked in the Great fact that the fish community objectives for several of the lakes Lakes in the late 1960s, the species has become an important clearly state that maintenance of Chinook salmon populations component of the fish communities within most of the lakes is an important management goal (Goddard 2002). (Hansen and Holey 2002). Although it is itself an exotic species, We conducted this study to obtain quantitative estimates of many advocates for Great Lakes restoration believe that Chinook age-specific abundances and mortality rates for the Chinook salmon stocking has served a restorative function by reducing salmon population in Lake Huron’s main basin to better alewife densities, which has helped lead to the recovery of some understand what factors were contributing to observed declines native fishes (Eshenroder and Burnham-Curtis 1999; Kocik and in recreational harvest and how this related to overall abundance Jones 1999; Bence and Mohr 2008; Madenjian et al. 2008). of the population. Our results suggest that observed declines Likewise, there is concern that exotic pelagic prey fishes may in recreational harvest are indeed a consequence of a large once again become overabundant and negatively affect native 1e-6 4.5 12.0 4.0 8e-7 10.0 3.5 Downloaded by [Department Of Fisheries] at 20:30 25 September 2012 3.0 8.0 6e-7 2.5

6.0 2.0 4e-7 1.5 4.0 1.0 2e-7 2.0 0.5 Recruits (millions) Recreational Fishing Catchability 0.0 0 Recreational Fishing Effort (angler hrs) 0.0 1985 1990 1995 2000 2005 2010 1970 1980 1990 2000 2010 Year

Year FIGURE 6. Total recreational fishing effort in angler-hours (solid line) and estimated time-varying recreational fishery catchability (dashed line) for Lake FIGURE 4. Recruitment of stocked (solid line) and wild (dashed line) Chinook Huron’s main basin from 1986 to 2009. See text for a description of how total salmon (in millions) to the main basin of Lake Huron from 1968 to 2009. fishing effort was calculated. 930 BRENDEN ET AL.

be the cause. One theory is that predation on age-0 Chinook salmon by other predators has increased as alewife densities have declined (J. Johnson, MDNR, personal communication), either as a result of predators switching from eating alewives to age-0 Chinook salmon or the resurgence of predators that historically were depressed by high alewife densities. Recently conducted diet studies of Lake Huron predators indeed confirm predation on age-0 Chinook salmon by a variety of predators, including lake trout, walleye, and adult Chinook salmon (J. Schaeffer, U.S. Geological Survey, personal communication). Another possible explanation for high mortality rates of age- 0 Chinook salmon is the outbreak of some epizootic possibly related to high densities of age-0 fish. There has recently been a variety of new diseases and parasites detected in Great Lakes fish populations that have been attributed to ecosystem level changes in the lakes stemming from the invasion of dreissenid mussels (Faisal et al. 2010; Loch and Faisal 2010a, 2010b); thus, it seems plausible that a new disease or parasite could be affecting Chinook salmon populations. Another factor that has been theorized as possibly impeding recruitment of Chi- nook salmon in the Great Lakes is early mortality syndrome (EMS), which results from thiamine deficiencies (Brown et al. 2005). Wolgamood et al. (2005) found a high rate of EMS in offspring of Chinook salmon from both Lakes Huron and Michi- gan in the late 1990s and early 2000s. In particular, Wolgamood et al. (2005) found a high rate (>70%) of EMS in offspring of FIGURE 7. Retrospective patterns for estimates of age-1 and older abundance Chinook salmon in 2001. Symptoms of EMS include loss of Lake Huron Chinook salmon (in millions, top panel), age-0 instantaneous of equilibrium, erratic swimming behavior, lethargy, hyper- natural mortality rates (middle panel), and mean age-2 and older instantaneous fishing mortality rates (bottom panel). The terminal year of data included in the excitability, hemorrhage, and death (Marcquenski and Brown model refits for the retrospective analyses is indicated by different line shadings 1997; Honeyfield et al. 1998). Early mortality syndrome in (colors in the online version) and patterns. Only the results for 1990 and later salmonids is generally associated with low egg thiamine con- are shown to increase distinction among the estimates. [Figure available online tent (Wolgamood et al. 2005) stemming from adult consump- in color.] tion of fish containing high concentrations of thiaminase, such as alewives and rainbow smelt (Fitzsimons et al. 1999; Tillitt decline in abundance of age-1 and older Chinook salmon, et al. 2005). It has generally been thought that consumption and not simply a result of declining fishing pressure. The of alewives would result in a Chinook salmon recruitment bot- decline in Chinook salmon abundance is supported by the tleneck (Brown et al. 2005), but this cannot explain the high fairly consistent decline in year-class returns to the Swan River mortality of young Chinook salmon in Lake Huron in recent weir over the last 30 years. This overall decline in abundance years, because alewife abundance is presently quite low (Riley

Downloaded by [Department Of Fisheries] at 20:30 25 September 2012 has occurred despite a purported major increase in Chinook et al. 2008). However, high thiaminase levels have been detected salmon wild reproduction in Lake Huron and is a result of in a variety of Great Lakes forage fish (Tillitt et al. 2005); thus, a simultaneous increase in age-0 natural mortality rates. It the possibility of a Chinook salmon EMS-related recruitment is not presently known what might be causing high natural bottleneck cannot totally be ruled out. mortality rates of age-0 Chinook salmon. In Lake Michigan, In our opinion, the most likely causes for the decline in sur- a positive relationship exists between abundance of age-1 vival of age-0 Chinook salmon in Lake Huron are the direct Chinook salmon and density of age-0 alewives (Warner et al. effects of food limitation and increased predation mortality by 2008), suggesting that as densities of young alewives decline larger fish, both of which ultimately stem from decreases in mortality of young Chinook salmon can increase. The alewife abundance of other prey fishes. Even though predation on age- population in Lake Huron, including age-0 fish, is considered 0 Chinook salmon by a variety of predator species has been to have collapsed around 2003 and has so far shown little detected, the encounter rates of Chinook salmon in diets of signs of recovery (Riley et al. 2008; Roseman and Riley 2009). other Chinook salmon, lake trout, and walleye would need to be Thus, based on the observed relationship in Lake Michigan, extremely high to account exclusively for the estimated age-0 greater age-0 Chinook salmon natural mortality may have been natural mortality rates. With regards to limited food resources, expected in Lake Huron, although it still is not clear what might there have been relatively few studies that have been conducted CHINOOK SALMON ASSESSMENT IN LAKE HURON 931

on Chinook salmon diets in the Great Lakes. Those that have Lake Huron was not reported by Adlerstein et al. (2008). The been conducted have found that Chinook salmon feed predomi- extent of immigration or emigration of Chinook salmon from nantly on alewives once they have converted to piscivory, which Lake Huron’s main basin into the North Channel or Georgian generally occurs at age 0 (Stewart et al. 1981; Jude et al. 1987; Bay is also unclear, as is how its occurrence may be affecting Warner et al. 2008; R. Claramunt, MDRNE, unpublished data). the accuracy of our modeling results. Future research to better Given the reliance of Chinook salmon on alewives when prey understand movement of Chinook salmon within Lake Huron populations are abundant, the collapse of prey fish populations and between Lake Michigan ultimately would probably prove in Lake Huron have raised the potential for Chinook salmon star- beneficial for addressing how these stocks of fish should be vation to occur. Although there has been a resurgence of other managed. potential prey species in Lake Huron, such as emerald shiner, Perhaps the largest assumption that we made in conduct- since the alewife populations collapsed, it remains uncertain as ing this research was that stocked and wild-produced Chinook to whether Chinook salmon can use these alternative resources salmon had equal rates of survival. We made this assumption due to differences in habitat use or some other behavioral trait out of necessity as there were no data sources available as far as that results in segregation of the populations (Schaeffer et al. we were aware that would permit separate estimation of survival 2008). for stocked and wild-produced fish. It is entirely conceivable, Although high natural mortality rates of age-0 Chinook perhaps even probable, that stocked Chinook salmon may have salmon appear to be the primary cause for the overall decline in lower rates of survival than their wild counterparts, particularly Chinook salmon abundance since the 1980s, it is important to during the first year of life. Several studies conducted on salmon recognize that recreational fishing mortality also may be a sig- species have indicated wild-produced fish have greater rates of nificant source of mortality for older fish. Annual recreational survival early in life than stocked fish (Jonsson et al. 2003; fishing mortality for fully selected age-classes in recent years Saloniemi et al. 2004; Hyatt et al. 2005). If survival of stocked has exceeded 90%, suggesting that a substantial proportion of Chinook salmon is lower than that of wild fish in Lake Huron, the remaining adult Chinook salmon population was harvested than the estimates of wild reproduction of approximately 10 mil- by anglers. As previously indicated, this high rate of exploitation lion fish may be extremely biased, and, as a consequence, our coincided with an utter recruitment failure of alewife, which is estimates of age-0 natural mortality rates are probably too large believed to have resulted in increased vulnerability of Chinook as well and partly function to reduce abundances to more appro- salmon to angling (Mohr and Ebener 2005) and which our esti- priate levels. We did not automatically include lower survival of mates of recreational fishing catchabilities support. Irrespective stocked fish in the SCAA model because, as previously stated, of its exact cause, the high fishing mortality rates estimated for there was no information available to suggest how survival might the Lake Huron Chinook salmon suggest that angling does have differ between stocked and wild fish or over what age-classes the potential to influence the structure of the population and that survival rates may be different. Further, some research has found harvest limits for the population may need to be considered. stocked fish to have higher survival than wild fish (Hyatt et al. We made a number of assumptions in developing the Lake 2005), so we did not feel comfortable simply making an assump- Huron main-basin Chinook salmon SCAA model. Among these tion as to what survival rates were for stocked and wild fish. assumptions were: (1) a single, well-mixed population, (2) con- Recently, Johnson et al. (2010) suggested that the Lake Huron stant age-specific selectivities of the recreational fishery, (3) Chinook salmon population was a self-regulated population constant natural mortality rates for age-1 and older fish, (4) no based on the perceived decline of the Chinook salmon popu- immigration or emigration of fish into or from the main basin, lation since the 2004 collapse of alewives in the lake. For a and (5) equal rates of survival for stocked and wild-produced population to be self-regulating, it must possess some intrinsic

Downloaded by [Department Of Fisheries] at 20:30 25 September 2012 fish at all ages. With regards to the first assumption, Weeder characteristic that limits its overexpansion, such as a behavioral et al. (2005) conducted a lakewide genetic variability study of modification that limits reproduction or recruitment once food Lake Michigan Chinook salmon and concluded that the pop- resources become limited (Lomnicki 1988; Wolff 1997). Ex- ulation consisted of a single, effectively panmictic population trinsic regulation of abundance from factors such as weather, with substantial gene flow and low genetic divergence among competition, disease, and predation is generally not considered wild reproducing stocks. Given the findings of Weeder et al. self-regulation (Lomnicki 1988; Wolff 1997). Based on this def- (2005) for Lake Michigan, it seems reasonable to assume that inition of self-regulating, we believe it is premature to declare the Lake Huron population is similarly well mixed. Regard- the Lake Huron Chinook salmon population as self-regulating. ing our assumption of no immigration or emigration of Chi- The population could be self-regulating if adult fish were pre- nook salmon, this may be more difficult to justify. Movement venting the population from overexpanding by cannibalizing of Chinook salmon from Lake Huron into Lake Michigan has wild-reproduced fish or if territoriality of breeding females was been reported (Adlerstein et al. 2007), although it is not certain limiting wild reproduction, but presently there is little definitive whether fish that move into Lake Michigan continue residing evidence to indicate that such intrinsic behavior is occurring. there or whether they eventually return to Lake Huron. Move- The potential danger in identifying the Lake Huron Chinook ment of Chinook salmon originating from Lake Michigan into salmon population as self-regulating is that it may give the 932 BRENDEN ET AL.

impression that management intervention of the population is management: a binational perspective. Michigan State University Press, East unnecessary due to the population’s ability to regulate itself. Lansing. This in turn could lead to a hands-off management approach, Brown, S. B., J. D. Fitzsimons, D. C. Honeyfield, and D. E. Tillitt. 2005. Implications of thiamine deficiency in Great Lakes salmonines. Journal of which might lead to wildly fluctuating abundances. We agree Aquatic Animal Health 17:113–124. with the assertion of Johnson et al. (2010) that Chinook salmon Crawford, S. 2001. Salmonine introductions to the Laurentian Great Lakes: stocking in Lake Huron should be re-evaluated given the cur- an historical review and evaluation of ecological effects. Canadian Special rent status of the population and the recent changes to the Lake Publication of Fisheries and Aquatic Sciences 132. Huron ecosystem. Dobiesz, N. E. 2003. An evaluation of the role of top piscivores in the fish community of the main basin of Lake Huron. Doctoral dissertation. Michigan State University, East Lansing. Dobiesz, N. E., D. A. McLeish, R. L. Eshenroder, J. R. Bence, L. C. Mohr, ACKNOWLEDGMENTS B. A. Henderson, M. P. Ebener, T. Nalepa, A. P. Woldt, J. E. Johnson, R. The authors thank the Great Lakes Fishery Commission for L. Argyle, and J. C. Makarewicz. 2005. Ecology of the Lake Huron fish providing funding that allowed this research to be conducted. community 1970–1999. Canadian Journal of Fisheries and Aquatic Sciences 62:1432–1451. We also acknowledge support from the Michigan Department Eshenroder, R. L., and M. K. Burnham-Curtis. 1999. Species succession and of Natural Resources and the U.S. Fish and Wildlife Service sustainability of the Great Lakes fish community. Pages 145–184 in W. W. Sportfish Restoration Program (through Project F-80-R to the Taylor and P. Ferreri, editors. Great Lakes fisheries policy and management: State of Michigan). The authors also thank T. Kolb, D. Clapp, a binational perspective. Michigan State University Press, East Lansing. J. Clevenger, J. He, and J. Johnson from the Michigan Depart- Faisal, M., W. Fayed, T. O. Brenden, A. Noor, M. P. Ebener, G. M. Wright, and M. L. Jones. 2010. Widespread infection of lake whitefish (Coregonus clu- ment of Natural Resources and A. Cottrill and L. Mohr from peaformis) with the swimbladder nematode Cystidicola farionis in northern the Ontario Ministry of Natural Resources for their willingness Lakes Michigan and Huron. Journal of Great Lakes Research 36(Supplement to share data. This is manuscript 2012-03 of the Quantitative 1):18–28. Fisheries Center at Michigan State University. Fenske, J. L., and M. F. Shouder. 1992. Swan River harvest weir report. Michigan Department of Natural Resources, Fisheries Technical Report 92-6, Ann Arbor. Fielder, D. G., J. S. Schaeffer, and M. V. Thomas. 2007. Environmental and REFERENCES ecological conditions surrounding the production of large year classes of Adlerstein, S. A., E. S. Rutherford, D. Clapp, J. A. Clevenger, and J. E. Johnson. walleye (Sander vitreus) in Saginaw Bay, Lake Huron. Journal of Great 2007. Estimating seasonal movements of Chinook salmon in Lake Huron Lakes Research 33:118–132. from efficiency analysis of coded wire tag recoveries in recreational fisheries. Fitzsimons, J. D., S. B. Brown, D. C. Honeyfield, and J. G. Hnath. 1999. A review North American Journal of Fisheries Management 27:792–803. of early mortality syndrome in Great Lakes salmonids and its relationship with Adlerstein, S. A., E. S. Rutherford, R. M. Claramunt, D. F. Clapp, and J. A. thiamine. Ambio 28:9–15. Clevenger. 2008. Seasonal movements of Chinook salmon in Lake Michigan Fournier, D. A., H. J. Skaug, J. Ancheta, J. Ianelli, A. Magnusson, M. N. based on tag recoveries from recreational fisheries and catch rates in gill-net Maunder, A. Nielsen, and J. Sibert. 2012. AD model builder: using automatic assessments. Transactions of the American Fisheries Society 137:736–750. differentiation for statistical inference of highly parameterized complex non- Bence, J. R., and N. E. Dobiesz. 2000. Estimating forage fish consumption linear models. Optimization Methods and Software 27:233–249. by predators in Lake Huron. Great Lakes Fisheries Commission, Project Geweke, J. 1992. Evaluating the accuracy of sampling-based approaches to Completion Report, Ann Arbor, Michigan. the calculation of posterior moments. Pages 169–193 in J. M. Bernado, Bence, J. R., J. E. Johnson, J. He, J. S. Schaeffer, S. Riley, R. J. Young, M. J. O. Berger, A. P. Dawid, and A. F. M. Smith, editors. Bayesian statistics 4. Ebener, D. Reid, L. C. Mohr, D. Gonder, A. Cottrill, A. Woldt, T. J. Morse, Oxford University Press, Oxford, UK. G. C. Christie, and M. Ridgway. 2008. Offshore predators and their fish Goddard, C. I. 2002. The future of Pacific salmon in the Great Lakes. Pages 243– community. Pages 11–36 in J. R. Bence and L. C. Mohr, editors. The state of 258 in K. D. Lynch, M. L. Jones, and W. W. Taylor, editors. Sustaining North Lake Huron in 2004. Great Lakes Fishery Commission, Special Publication American salmon: perspectives across regions and disciplines. American 08-01, Ann Arbor, Michigan. Fisheries Society, Bethesda, Maryland. Downloaded by [Department Of Fisheries] at 20:30 25 September 2012 Bence, J. R., and L. C. Mohr. 2008. State of Lake Huron: current and future. Hansen, M. J., and M. E. Holey. 2002. Ecological factors affecting the sus- Pages 71–75 in J. R. Bence and L. C. Mohr, editors. The state of Lake Huron tainability of Chinook and coho salmon populations in the Great Lakes, in 2004. Great Lakes Fishery Commission, Special Publication 08-01, Ann especially Lake Michigan. Pages 155–180 in K. D. Lynch, M. L. Jones, Arbor, Michigan. and W. W. Taylor, editors. Sustaining North American salmon: perspec- Bence, J. R., and K. D. Smith. 1999. An overview of recreational fisheries of tives across regions and disciplines. American Fisheries Society, Bethesda, the Great Lakes. Pages 259–306 in W. W. Taylor and P. Ferreri, editors. Great Maryland. Lakes fisheries policy and management: a binational perspective. Michigan He, J. X., and J. R. Bence. 2007. Modeling annual growth variation using a State University Press, East Lansing. hierarchical Bayesian approach and the von Bertalanffy growth function, Benjamin, D. M., and J. R. Bence. 2003. Statistical catch-at-age analysis of with application to lake trout in southern Lake Huron. Transactions of the Chinook salmon in Lake Michigan, 1985–1996. Michigan Department of American Fisheries Society 136:318–330. Natural Resources, Fisheries Research Report 2066, Ann Arbor. Holey, M. E., R. F. Elliott, S. V. Marcquenski, J. G. Hnath, and K. D. Smith. Berst, A. H., and G. R. Spangler. 1972. Lake Huron: effects of exploitation, 1998. Chinook salmon eipizootics in Lake Michigan: possible contributing introductions, and eutrophication on the salmonid community. Journal of the factors and management implications. Journal of Aquatic Animal Health Fisheries Research Board of Canada 29:877–887. 10:202–210. Brown, R. W., M. Ebener, and T. Gorenflo. 1999. Great Lakes commercial Honeyfield, D. C., J. D. Fitzsimons, S. B. Brown, S. V. Marcquenski, and G. fisheries: historical overview and prognosis for the future. Pages 307–354 McDonald. 1998. Introduction and overview of early life stage mortality. in W. W. Taylor and P. Ferreri, editors. Great Lakes fisheries policy and Pages 1–7 in G. McDonald, J. D. Fitzsimons, and D. C. Honeyfield, editors. CHINOOK SALMON ASSESSMENT IN LAKE HURON 933

Early life stage mortality syndrome in fishes of the Great Lakes and Baltic policy and management: a binational perspective. Michigan State University Sea. American Fisheries Society, Symposium 21, Bethesda, Maryland. Press, East Lansing. Hyatt, K. D., K. L. Mathias, D. J. McQueen, B. Mercer, P. Milligan, and D. Plummer, M., N. Best, K. Cowles, and K. Vines. 2010. Output analysis and P. Rankin. 2005. Evaluation of hatchery versus wild sockeye salmon fry diagnostics for MCMC, R package version 0.14-2. Available: CRAN.R- growth and survival in two British Columbia lakes. North American Journal project.org/package = coda. (April 2010). of Fisheries Management 25:745–762. R Development Core Team. 2010. R: a language and environment for statistical Johnson, J. E., S. P. DeWitt, and D. J. A. Gonder. 2010. Mass marking reveals computing. R Foundation for Statistical Computing, Vienna, Austria. emerging self regulation of the Chinook salmon population in Lake Huron. Riley, S. C., J. X. He, J. E. Johnson, T. P. O’Brien, and J. S. Schaeffer. 2007. North American Journal of Fisheries Management 30:518–529. Evidence of widespread natural reproduction by lake trout Salvelinus namay- Jonsson, N., B. Jonsson, and L. P.Hansen. 2003. The marine survival and growth cush in the Michigan waters of Lake Huron. Journal of Great Lakes Research of wild and hatchery-reared Atlantic salmon. Journal of Applied Ecology 33:917–921. 40:900–911. Riley, S. C., E. F. Roseman, S. J. Nichols, T. P. O’Brien, C. S. Kiley, and J. Jude, D. J., F. J. Tesar, S. F. Deboe, and T. J. Miller. 1987. Diet and selection of S. Schaeffer. 2008. Deepwater demersal fish community collapse in Lake major prey species by Lake Michigan salmonines, 1973–1982. Transactions Huron. Transactions of the American Fisheries Society 137:1879–1890. of the American Fisheries Society 116:677–691. Roseman, E. F., and S. C. Riley. 2009. Biomass of deepwater demersal for- Kocik, J. F., and M. L. Jones. 1999. Pacific salmonines in the Great Lakes basin. age fishes in Lake Huron, 1994–2007: implications for offshore predators. Pages 455–488 in W. W. Taylor and P. Ferreri, editors. Great Lakes fisheries Aquatic Ecosystem Health and Management 12:29–36. policy and management: a binational perspective. Michigan State University Saloniemi, I., E. Jokikokko, I. Kallio-Nyberg, E. Jutila, and P. Pasanen. 2004. Press, East Lansing. Survival of reared and wild Atlantic salmon smolts: size matters more in bad Loch, T. P., and M. Faisal. 2010a. Infection of lake whitefish (Coregonus clu- years. ICES Journal of Marine Science 61:782–787. peaformis) with motile Aeromonas spp. in the Laurentian Great Lakes. Journal Schaeffer, J. S., D. M. Warner, and T. P. O’Brien. 2008. Resurgence of emerald of Great Lakes Research 36(Supplement 1):6–12. shiners Notropis atherinoides in Lake Huron’s main basin. Journal of Great Loch, T. P., and M. Faisal. 2010b. Isolation of Aeromonas salmonicida sub- Lakes Research 34:395–403. species salmonicida from lake whitefish (Coregonus clupeaformis) inhabiting Schnute, J. T. 1994. A general framework for developing sequential fisheries Lakes Michigan and Huron. Journal of Great Lakes Research 36(Supplement models. Canadian Journal of Fisheries and Aquatic Sciences 51:1676–1688. 1):13–17. Stewart, D. J., and M. Ibarra. 1991. Predation and production by salmonine fishes Lomnicki, A. 1988. Population ecology of individuals. Princeton University in Lake Michigan, 1978–1988. Canadian Journal of Fisheries and Aquatic Press, Princeton, New Jersey. Sciences 48:909–922. Madenjian, C. P., R. O’Gorman, D. B. Bunnell, R. L. Argyle, R. F. Roseman, D. Stewart, D. J., J. F. Kitchell, and L. B. Crowder. 1981. Forage fishes and their M. Warner, J. D. Stockwell, and M. A. Stapanian. 2008. Adverse effects of salmonid predators in Lake Michigan. Transactions of the American Fisheries alewife on Laurentian Great Lakes fish communities. North American Journal Society 110:751–765. of Fisheries Management 28:263–282. Tanner, H. A., and W. H. Tody. 2002. History of the Great Lakes salmon fishery: Marcquenski, S. V., and S. B. Brown. 1997. Early mortality syndrome in a Michigan perspective. Pages 139–154 in K. D. Lynch, M. L. Jones, and W. salmonid fishes from the Great Lakes. Pages 135–152 in R. M. Rolland, W. Taylor, editors. Sustaining North American salmon: perspectives across M. Gilbertson, and R. E. Peterson, editors. Chemically induced alterations in regions and disciplines. American Fisheries Society, Bethesda, Maryland. functional development and reproduction of fishes: proceedings from a ses- Tillitt, D. E., J. L. Zajicek, S. B. Brown, L. R. Brown, J. D. Fitzsimons, D. C. sion at the 1995 Wingspread conference. SETAC (Society of Environmental Honeyfield, M. E. Holey, and G. M. Wright. 2005. Thiamine and thiaminase Toxicology and Chemistry), Pensacola, Florida. status in forage fish of salmonines from Lake Michigan. Journal of Aquatic Mills, E. L., J. M. Casselman, R. Dermott, J. D. Fitzsimons, G. Gal., K. T. Animal Health 17:13–25. Holeck, J. A. Hoyle, O. E. Johannsson, B. F. Lantry, J. C. Makarewicz, Warner, D. M., C. S. Kiley, R. M. Claramunt, and D. F. Clapp. 2008. The E. S. Millard, I. F. Munawar, M. Munawar, R. O’Gorman, R. W. Owens, influence of alewife year-class strength on prey selection and abundance L. G. Rudstam, T. Schaner, and T. J. Stewart. 2003. Lake Ontario: food of age-1 Chinook salmon in Lake Michigan. Transactions of the American web dynamics in a changing ecosystem. Canadian Journal of Fisheries and Fisheries Society 137:1683–1700. Aquatic Sciences 60:471–490. Weeder, J. A., A. R. Marshall, and J. M. Epifanio. 2005. An assessment of Mohn, R. 1999. The retrospective problem in sequential population analysis: an population genetic variation in Chinook salmon from seven Michigan rivers investigation using cod fishery and simulated data. ICES Journal of Marine 30 years after introduction. North American Journal of Fisheries Management Science 56:473–488. 25:861–875.

Downloaded by [Department Of Fisheries] at 20:30 25 September 2012 Mohr, L. C., and M. P. Ebener. 2005. Description of the fisheries. Pages 19–26 Wolff, J. O. 1997. Population regulation in mammals: an evolutionary perspec- in M. P. Ebener, editor. The state of Lake Huron in 1999. Great Lakes Fishery tive. Journal of Animal Ecology 66:1–13. Commission, Special Publication 05-02, Ann Arbor, Michigan. Wolgamood, M., J. G. Hnath, S. B. Brown, K. Moore, S. V. Marcquenski, O’Gorman, R., and T. J. Stewart. 1999. Ascent, dominance, and decline of the D. C. Honeyfield, J. P. Hinterkopf, J. D. Fitzsimons, and D. E. Tillitt. 2005. alewife in the Great Lakes: food web interactions and management strategies. Temporal and spatial variation of early mortality syndrome in salmonids from Pages 489–514 in W. W. Taylor and P. Ferreri editors. Great Lakes fisheries Lakes Michigan and Huron. Journal of Aquatic Animal Health 17:65–76. This article was downloaded by: [Department Of Fisheries] On: 25 September 2012, At: 20:31 Publisher: Taylor & Francis Informa Ltd Registered in England and Wales Registered Number: 1072954 Registered office: Mortimer House, 37-41 Mortimer Street, London W1T 3JH, UK

Transactions of the American Fisheries Society Publication details, including instructions for authors and subscription information: http://www.tandfonline.com/loi/utaf20 Distribution of U.S. Atlantic Salmon Postsmolts in the Gulf of Maine Gilles L. Lacroix a , Derek Knox a , Timothy F. Sheehan b , Mark D. Renkawitz b & Meredith L. Bartron c a Department of Fisheries and Oceans, Biological Station, 531 Brandy Cove Road, St. Andrews, New Brunswick, E5B 2L9, Canada b National Oceanic and Atmospheric Administration Fisheries Service, Atlantic Salmon Research and Conservation Task, 166 Water Street, Woods Hole, Massachusetts, 02543, USA c U.S. Fish and Wildlife Service, Northeast Fishery Center, 227 Washington Avenue, Lamar, Pennsylvania, 16848, USA Version of record first published: 21 Jun 2012.

To cite this article: Gilles L. Lacroix, Derek Knox, Timothy F. Sheehan, Mark D. Renkawitz & Meredith L. Bartron (2012): Distribution of U.S. Atlantic Salmon Postsmolts in the Gulf of Maine, Transactions of the American Fisheries Society, 141:4, 934-942 To link to this article: http://dx.doi.org/10.1080/00028487.2012.675915

PLEASE SCROLL DOWN FOR ARTICLE

The publisher does not give any warranty express or implied or make any representation that the contents will be complete or accurate or up to date. The accuracy of any instructions, formulae, and drug doses should be independently verified with primary sources. The publisher shall not be liable for any loss, actions, claims, proceedings, demand, or costs or damages whatsoever or howsoever caused arising directly or indirectly in connection with or arising out of the use of this material. Transactions of the American Fisheries Society 141:934–942, 2012 American Fisheries Society 2012 ISSN: 0002-8487 print / 1548-8659 online DOI: 10.1080/00028487.2012.675915

ARTICLE

Distribution of U.S. Atlantic Salmon Postsmolts in the Gulf of Maine

Gilles L. Lacroix* and Derek Knox Department of Fisheries and Oceans, Biological Station, 531 Brandy Cove Road, St. Andrews, New Brunswick E5B 2L9, Canada Timothy F. Sheehan and Mark D. Renkawitz National Oceanic and Atmospheric Administration Fisheries Service, Atlantic Salmon Research and Conservation Task, 166 Water Street, Woods Hole, Massachusetts 02543, USA Meredith L. Bartron U.S. Fish and Wildlife Service, Northeast Fishery Center, 227 Washington Avenue, Lamar, Pennsylvania 16848, USA

Abstract Atlantic salmon Salmo salar of U.S. hatchery origin were captured as postsmolts by surface trawling during spatial surveys in late May to mid-June in the northeastern Gulf of Maine in 2002 and 2003. Most marked and unmarked postsmolts were from the Penobscot River, but some were also from the Dennys and other rivers of the Gulf of Maine. The capture rate of stocked smolts was very low (≤0.01%), and it was highest for marked smolts from known rivers and proportional to numbers stocked. Marked postsmolts were caught 16–43 d after their release as smolts and 89–240 km from their river of origin. The rate of migration from different rivers differed and varied annually, and it accounted for some of the observed differences in capture rate. The weight of postsmolts from known rivers was greater than at smolt stocking and river exit, indicating early marine growth. The distribution of U.S. postsmolts from different rivers was similar, and they were often caught together, as were those from the same river. As a result of this close association during migration, postsmolt catches were aggregated at a few adjacent locations in several areas of the Gulf of Maine. Postsmolts crossing the Gulf of Maine were found east of Jordan Basin and most were along a corridor near shore at the eastern edge of the area surveyed along the southwest coast of Nova Scotia, Canada. Postsmolts of U.S. origin were found in areas used by other salmon populations also leaving the Bay of Fundy during May and June. This seasonal occurrence in a speciﬁc coastal area provides an opportunity to manage activities within that corridor to mitigate potential losses of Atlantic salmon from endangered and threatened populations in North America. Downloaded by [Department Of Fisheries] at 20:31 25 September 2012

Remnant Atlantic salmon Salmo salar populations in rivers evaluated in different studies (Kocik et al. 2009; Sheehan et al. of Maine are protected under the U.S. Endangered Species 2011). Generally, these reported that salmon smolt passage Act (NOAA 2009). Extensive multiple life stage stocking of through the freshwater zone was rapid and mortality was low hatchery-reared juvenile salmon maintains these populations during migration from rivers. However, after smolts entered the and artiﬁcially conserves their genetic diversity (Fay et al. estuaries, their mortality increased markedly and, consequently, 2006). The effectiveness of these strategies and resulting few postsmolts entered the Gulf of Maine. Thereafter, little survival of smolts in rivers, estuaries and embayments were is known about these Maine salmon populations until they

*Corresponding author: [email protected] Received August 30, 2011; accepted January 16, 2012 Published online June 21, 2012

934 U.S. HATCHERY POSTSMOLT DISTRIBUTION 935

are captured in salmon fisheries on feeding grounds in Davis capture of fast-swimming pelagic fish. Using extra flotation and Strait, the Labrador Sea, and off West Greenland (Baum 1997). adjustable pelagic trawl doors, the trawl was towed an average Postsmolts generally disperse according to prevailing residual speed of 4 knots with the headrope in constant contact with the surface currents in coastal areas, and strong currents are thought surface. A total of 155 tows were completed between 26 May to act as transport vectors that facilitate migration in open and 15 June 2002, and 143 tows were made between 4 and waters (Jonsson et al. 1993; Holst et al. 2000; Lacroix et al. 18 June 2003. Salmon were captured in 34% of tows in 2002 2004, 2005; Lacroix and Knox 2005; Sheehan et al. 2011). and 16% in 2003. Tow variables and conditions described by Based on extensive tag returns from coastal fisheries of the Lacroix and Knox (2005) did not differ between years or be- Northwest Atlantic, Meister (1984) postulated on the migration tween tows with and without salmon. Tows were kept short (ap- routes of Maine salmon at different life stages. However, only proximately 30 min for a distance of 4 km) and trawling was lim- two tagged salmon postsmolts were ever recovered within the ited to daylight hours to reduce the number of nontarget pelagic Gulf of Maine, and inferences on migration route and timing fish captured and prevent postsmolt mortality in the holding could not be made. This could be a critical period and habitat for tank. postsmolt survival (Kocik et al. 2009). The leading hypothesis Postsmolt origin and distribution.—Postsmolts were sedated is generally that mortality of salmon at sea is highest during the and sampled as described by Lacroix and Knox (2005) and later early, postsmolt stage of migration (Hansen et al. 2003; ICES released. They determined postsmolt origin using a combination 2009). During this period, river of origin and migration timing of external tags and marks, fin clips, and scale examination for can influence distribution and survival (Ritter 1989; Lacroix age and circuli spacing (Shearer 1992). Smolts of U.S. hatchery 2008). origin released in Gulf of Maine rivers were age 1 and had either Our study examined the origin, distribution, and timing of a coded visual implant elastomer (VIE) mark and a clipped catches of Atlantic salmon of U.S. hatchery origin in the sur- adipose fin or no marks, tags, or clipped fins. The VIE-marked face trawl surveys conducted by Lacroix and Knox (2005) in smolts were released in 2002 and 2003 in five rivers of Maine the Bay of Fundy and northeastern Gulf of Maine during the with most going into the Penobscot and Dennys rivers (Table 1). initial period of postsmolt migration. Lacroix and Knox (2005) Smolts with no fins clipped and no other identifying marks were reported on the annual distribution and abundance and growth released annually in the same rivers as VIE-marked smolts, and feeding habits of postsmolts during this period, including with the majority going into the Penobscot River (USASAC some comparisons between those of wild and hatchery origin 2006). However, 10–12% of the unmarked smolts were released or between those from several Canadian rivers. Although they annually in the Merrimack River, Massachusetts, which flows identified the presence of salmon of U.S. origin, they did not into the Gulf of Maine, and in 2002, 12% were released in the look at their marine distribution and migration timing, or dif- Lamprey River which empties along the Maine–New Hampshire ferentiate them by river of origin. We reexamined the data from border. Parr from U.S. hatcheries with and without a pelvic fin Lacroix and Knox (2005) to focus on salmon populations from clip only (no other marks) also were stocked annually in the fall the rivers of Maine because of their endangered status and the in the Penobscot and Dennys rivers. These fall parr releases were need to identify both the marine habitat and time frame for excluded from Table 1 because of the low proportion surviving postsmolt migration, all of which are necessary to effectively to migrate as smolts the next spring (T. F. Sheehan, unpublished manage those populations and implement a recovery strategy. data). None of the postsmolts caught had pelvic fin clips that matched releases from U.S. hatcheries, further justifying the exclusion of parr releases from the analysis. METHODS The VIE codes on postsmolts captured 1–2 months after tag-

Downloaded by [Department Of Fisheries] at 20:31 25 September 2012 Trawling gear and method.—During spring of 2002 and ging were easily seen and not faded or obscured, as can occur 2003, surface-oriented trawl surveys were conducted in north- on returning adults after many months or years of growth at eastern Gulf of Maine east of the International Maritime Bound- sea. They indicated that postsmolts were from either the Penob- ary to capture migrating Atlantic salmon postsmolts. The trawl scot River or Dennys River, and from different groups released and fishing method were described by Lacroix and Knox (2005), in April and May. In 2002, postsmolts with VIE marks (n = which followed methods successfully used by Holm et al. (2000) 24) were genotyped by Leetown Science Center (U.S. Geolog- and Holst and McDonald (2000) in Norway to capture salmon. ical Survey) because of code duplication between these rivers. They found that surface contact by the trawl headrope was Eleven microsatellite loci were used, which was consistent with needed to catch salmon and that fishing deeper was not suc- putative parents from the Penobscot and Dennys broodstocks cessful. (Bartron et al. 2006). Using Cervus, parental assignment, based The trawl was approximately 125 m long with an opening on a minimum of 10 loci, was possible for 16 of the 24 postsmolts 10 m deep and 60 m wide at the mouth and it had a large fish- (Marshall et al. 1998). The remaining 8 postsmolts were as- holding tank at the cod end to protect and retain captured fish signed to a specific broodstock by maximum-likelihood analy- for live release postsampling. The trawl and rigging were de- sis using GeneClass (Cornuet et al. 1999). There was no VIE signed and used materials to increase towing speed and facilitate code overlap among rivers in 2003. 936 LACROIX ET AL.

TABLE 1. Number of VIE-marked and unmarked Atlantic salmon smolts of U.S. hatchery origin released in the Gulf of Maine watershed and their capture rate and dates in the northeastern Gulf of Maine in 2002 and 2003 (percent of total released is in parentheses).

Number released Percent captured Capture dates 2002 VIE-marked 231,220 (32.4) 0.0104 Penobscot River 172,519 0.0110 26 May–04 June Dennys River 50,501 0.0099 28 May–30 May Other Maine rivers 8,200 0 Unmarked 482,980 (67.6) 0.0070 26 May–13 June Total hatchery 714,200 (100) 0.0081 2003 VIE-marked 262,306 (39.7) 0.0023 Penobscot River 197,152 0.0030 04 June–13 June Dennys River 55,871 0 Other Maine rivers 9,283 0 Unmarked 398,974 (60.3) 0.0018 04 June–11 June Total hatchery 661,280 (100) 0.0020

All other postsmolts of smolt age 1 with no fins clipped For each postsmolt of known river origin, the time interval and no identifying marks were assigned to the unknown U.S. was determined between its release date (based on the VIE code) rivers group in the Gulf of Maine watershed. Lacroix and Knox as a smolt and its capture, and the minimum distance traveled (2005) identified any other postsmolts caught of smolt age 1 to was measured using a straight line between the mouth of the be of either Canadian hatchery or aquaculture origin. All smolts river of origin and the capture site. In addition, the weight of stocked from Canadian hatcheries were adipose fin-clipped or postsmolts of known hatchery origin was compared with (1) otherwise uniquely tagged, and a few fish deemed to be of the mean weight of smolts with VIE marks at the hatchery aquaculture origin had extensive erosion of the dorsal and other just before stocking, and (2) migrating smolts with VIE marks fins and a different circuli spacing pattern. captured in rotary screw traps in the lower portions of the rivers The distribution of postsmolts was plotted using coordinates (NOAA Fisheries Service, unpublished data). for the starting point of the trawl set. The goodness of fit of the Poisson, negative binomial, and normal distributions of the catch data were evaluated using a single classification G-test RESULTS for comparison of the observed number of postsmolts caught Atlantic salmon postsmolts of U.S. hatchery origin from per tow and per day to the number predicted by each distribu- rivers of the Gulf of Maine were captured in 13% of tows in tion. This test is a version of the likelihood ratio test, where 2002 and 8% of tows in 2003. They were in good condition G is chi-square-distributed and goodness of fit increases as G with little scale loss. The ectoparasitic salmon louse Lepeoph- approaches zero (i.e., the null hypothesis). Rejection of the null theirus salmonis, was not found on any of the postsmolts; the

Downloaded by [Department Of Fisheries] at 20:31 25 September 2012 hypothesis suggests that a distribution was aggregated. Dis- sea louse Caligus elongatus was found infrequently (i.e., four persal was measured using the variance-to-mean ratio, which fish of unknown river origin in 2002 each had one louse). The equals 1 if captures are randomly dispersed and is significantly capture rate of VIE-marked smolts was very low but higher than >1 if the dispersion pattern is not random. The coefficient of that of unmarked smolts (Table 1). Slightly less than half of variation was used as a measure of relative dispersion, and the the postsmolts were traced back to the river of stocking in each negative binomial parameter (k) was calculated as a measure year, and most of these were from the Penobscot River (Ta- of aggregation, where the smaller the value of k the greater the ble 2). In 2002, the catch rate was similar for fish of Penobscot aggregation. and Dennys rivers origin and representative of the number of The null hypothesis of independence between catches of smolts released. However, no postsmolts of Dennys River origin postsmolts classified to be of different river origin (i.e., Penob- were captured in 2003, when the catch rate for other groups was scot and Dennys rivers, known and unknown U.S. rivers) was the lowest. In 2002, 68% of Penobscot postsmolts and 60% of evaluated with the Fisher exact probability test of association us- the Dennys postsmolts caught were from the last smolt release ing 2 × 2 contingency tables for presence and absence in differ- groups (6–9 May), when 50% of Penobscot and 46% of Dennys ent tows. In addition, the strength of association between groups smolts were released; the rest were from earlier smolt releases was measured with Cramer’s coefficient of association (V). (5–19 April). In 2003, 100% of Penobscot origin postsmolts U.S. HATCHERY POSTSMOLT DISTRIBUTION 937

TABLE 2. Number of Atlantic salmon postsmolts of U.S. hatchery origin captured during surveys in the northeastern Gulf of Maine in 2002 and 2003, and their mean fork length, weight and condition factor. For each year, mean values in a column followed by different letters are signiﬁcantly different (P < 0.05, ANOVA and Tukey HSD test).

Number captured Fork length (cm) Weight (g) Condition factor total (percent) mean (SD) mean (SD) mean (SD) 2002 Penobscot River 19 (33) 20.3 yz (1.3) 83.5 yz (14.2) 0.988 (0.064) Dennys River 5 (9) 21.1 z (0.7) 94.5 z (11.6) 0.996 (0.039) Unknown rivers 34 (58) 19.2 y (2.0) 71.6 y (20.1) 0.990 (0.069) 2003 Penobscot River 6 (46) 21.0 z (1.6) 91.9 z (19.5) 0.984 (0.054) Unknown rivers 7 (54) 16.7 y (1.5) 48.4 y (9.9) 1.032 (0.072)

were from early releases (16–22 April) when 43% of smolts dence between groups in the catch was rejected for postsmolts were released, and none of any origin was caught from later from the Penobscot and Dennys rivers (P < 0.0001 for Fisher releases in May. No postsmolts of the early smolt release (28 exact probability test; Cramer’s V = 0.454) and for those from April) in the Dennys River in 2003 were caught; that release known and unknown rivers of origin (P < 0.0001 for Fisher was later than in 2002 and also later than that in the Penobscot exact probability test; Cramer’s V = 0.546). In addition, these River in 2003. postsmolts of different origins when not captured together were Postsmolts of different origins were caught at similar times found in close proximity to one another in 2002 (Figure 1A). in late May to early June (Table 1). They were captured starting In 2003, catch frequency was low, and no postsmolts of known on the first day of fishing each year, and most were caught early and unknown origins were caught together; the null hypothesis in the surveys when the area along the southwest coast of Nova of independence was not rejected (P = 0.805 for Fisher ex- Scotia was fished. Catches of salmon of U.S. origin decreased act probability test; Cramer’s V = 0.040). Nevertheless, where in the second half of the surveys as fishing moved north into postsmolts of Penobscot River and unknown river origins were the Bay of Fundy. Postsmolts were captured in similar locations captured in 2003 they were usually in close proximity to one regardless of river origin or survey year (Figure 1). They were another (Figure 1B). in open water along the eastern edge of Jordan Basin and along the southwest coast of Nova Scotia. The largest concentrations TABLE 3. Results of distribution analysis for frequency of Atlantic salmon of postsmolts were found along the eastern edge of the area postsmolts of U.S. hatchery origin captured at different locations and dates surveyed in a pattern rounding southwest Nova Scotia close to during trawling surveys in the northeastern Gulf of Maine in 2002 and 2003. shore. No postsmolts were found in areas surveyed to the north either at the mouth of or inside the Bay of Fundy. In contrast, Metric Catch per tow Catch per date some postsmolts were caught in tows at or near the southernmost 2002 edge of the survey. Mean 0.3742 2.7619 When postsmolts of U.S. hatchery origin were caught, two or Variance 1.833 38.29

Downloaded by [Department Of Fisheries] at 20:31 25 September 2012 more postsmolts of that origin occurred in a tow in 60% of cases Variance-to-mean ratio 4.90 13.86 in 2002 and 18% in 2003. The average catch in these tows taking Coefficient of variation 3.62 2.24 U.S. hatchery salmon was 2.9 postsmolts (range, 1–10) in 2002 Negative binomial k 0.10 0.21 and 1.2 postsmolts (range, 1–2) in 2003 (Figure 1). Measures Poisson G 145a (6) 127a (4) of distribution applied to all tows (i.e., with and without salmon Negative binomial G 10a (5) 20a (3) of U.S. origin) showed that catches per tow and per date were Normal G 347a (5) 74a (3) aggregated in a few locations on several days in 2002 (Table 3). Capture locations in 2003 were generally concentrated in the 2003 same areas as in 2002 (Figure 1A, B). However, distribution Mean 0.0909 0.8667 measures were not reliable in 2003 because of the small numbers Variance 0.111 0.838 of postsmolts caught at any location (maximum, two) or on any Variance-to-mean ratio 1.23 0.97 date (maximum, three). Coefficient of variation 3.67 1.06 There was a strong association between postsmolts of differ- Negative binomial k 0.40 100.00 ent origins at capture locations in 2002 (i.e., fish from different aNull hypothesis rejected (P < 0.05 in G-test for goodness of fit; degrees of freedom groups often occurred together). The null hypothesis of indepen- is in parentheses). 938 LACROIX ET AL. Downloaded by [Department Of Fisheries] at 20:31 25 September 2012

FIGURE 1. Distribution and abundance (scaled pie charts) of Atlantic salmon postsmolts of U.S. hatchery origin captured in the northeastern Gulf of Maine in (A) 2002, and (B) 2003. Groups represented in pie charts are: Penobscot River (dark ﬁll), Dennys River (gray ﬁll), and unknown rivers (clear). Trawl sites with no U.S. postsmolts ( × ), the 100-m and 200-m isobaths (differential shading), and the International Boundary (dashed line) are shown. U.S. HATCHERY POSTSMOLT DISTRIBUTION 939

TABLE 4. Mean time interval, distance traveled, and migration rate (SD and range in parentheses) between release of Atlantic salmon smolts of known U.S. origin and their capture as postsmolts during surveys in the northeastern Gulf of Maine in 2002 and 2003. Mean values in a column followed by different letters are signiﬁcantly different (P < 0.05 in ANOVA and Tukey HSD test). Sample sizes same as in Table 2.

Time interval (d) Distance traveled (km) Migration rate (km/d) 2002 Penobscot River 26.9 (10.8; 16–43) 200.3 z (23.6; 164–240) 8.6 z (3.29; 3.6–14.3) Dennys River 28.0 (11.0; 19–41) 130.6 y (31.7; 89–157) 5.3 y (2.34; 2.2–8.3) 2003 Penobscot River 38.0 (3.6; 34–43) 191.3 z (36.1; 139–240) 5.1 y (1.25; 3.9–6.7)

The time interval between smolt release and postsmolt cap- facilitate rapid movement around Nova Scotia into the open ture in the Gulf of Maine ranged from 16 to 43 d for groups of North Atlantic. The timing of catches in this corridor in late known origin, and the minimum or straight-line distance trav- May and early June is consistent with Meister’s (1984) descrip- eled by postsmolts between the river mouth and capture location tion of postsmolt migrating along the Atlantic coast of Nova ranged from 89 to 240 km (Table 4). The distance traveled by Scotia later in June and July and to the south and north coast of postsmolts of Dennys River origin was significantly shorter than Newfoundland in July and August. for those of Penobscot River origin, but the time interval was The capture rate of U.S. origin postsmolts in the northeast- almost the same in 2002. Also, the mean time interval was at ern portion of the Gulf of Maine was low, considering the high least 10 d greater in 2003 than 2002. As a result, the rate of stocking numbers. However, it was similar to that reported by migration, inclusive of time spent in the river, was slower for Lacroix and Knox (2005) for hatchery salmon stocked in Cana- smolts and postsmolts from the Dennys River than from the dian rivers of the Bay of Fundy at a similar time as U.S. smolts. Penobscot River, and it was slower in 2003 than 2002 for fish A low catch rate was not entirely unexpected given the difficulty from the Penobscot River. in catching salmon at sea and the complexity of matching fishing The condition factor for postsmolts from all groups indicated time and location to fish migration. Surface trawling is one of that they were not emaciated but in good condition (Table 2). the few methods successfully used to capture migrating salmon In addition, postsmolts from groups of known-river origin had a greater mean weight than that of smolts of the same origin, both at the time of stocking and when sampled at traps in the lower section of the Penobscot and Dennys rivers (Figure 2). The individual weight of 42% of VIE-marked postsmolts in 2002 and 67% in 2003 was greater than the mean weight plus SD of marked smolts at stocking. The mean length and weight of postsmolts of known river origin tended to be slightly greater than those of unknown origin, but the differences were not significant for the Penobscot River group in 2002 (Table 2).

Downloaded by [Department Of Fisheries] at 20:31 25 September 2012 DISCUSSION We report the ﬁrst capture of Atlantic salmon postsmolts of U.S. hatchery origin in open waters of the Gulf of Maine proper. The restricted distribution of postsmolts and timing of captures indicated the use of a temporary migration corridor across the Gulf of Maine from Maine estuaries (Kocik et al. 2009) and near shore (Sheehan et al. 2011) to the closest previous captures of postsmolts along the Atlantic coast of Nova Scotia (Meister 1984). The distribution of postsmolts we found supports Meis- ter’s (1984) postulation that Maine salmon stocks on their out- ward migration move easterly across the Gulf of Maine, thence northerly along the south shore and eastern shore of Nova Sco- FIGURE 2. Mean weight ( + SD) of Atlantic salmon of U.S. hatchery origin tia. Similarly, Sheehan et al. (2011) proposed that, upon exiting with VIE marks from the Penobscot River in 2002 (cross-hatched bars) and Penobscot Bay, postsmolts traveled northeast across the Gulf of 2003 (clear bars) and the Dennys River in 2002 (dark bars) at different sampling Maine toward the coast of Nova Scotia. This direct route would locations. 940 LACROIX ET AL.

in open waters, but it is under development and still inefficient which would tend to delay or slow down migration (Lacroix (Holm et al. 2000, 2003; Holst et al. 2000; Lacroix and Knox et al. 2004, 2005). A higher annual mortality of smolts from the 2005; Sheehan et al. 2011). Towing speed, distance and position Dennys River also could account for the low catch of Dennys of trawl behind the ship’s wake, large mesh sizes, surface con- River postsmolts in 2002 and their absence in 2003. Salmon tact by headrope, cod-end tank design, and method of fishing from the Dennys River must migrate through areas containing (single versus paired trawlers) can all influence the capture of salmon farming facilities, where there is potential for increased fast-swimming salmon. Limiting tows to 30 min and daylight but annually varied exposure to predation, parasites, and disease hours to prevent salmon mortality also would have restricted (Lacroix et al. 2004; Chang et al. 2005). the catch. In addition, the difficulty in having a spatial survey Underreporting of marked salmon was not considered to be overlap with rapidly migrating postsmolts would probably have important in the surveys or to have biased the results. The early resulted in low catches. Catches would have been significantly capture of postsmolts, 16–43 d after smolt-tagging, ensured that higher had fishing been concentrated only along the corridor VIE marks were clearly visible. Also, the possibility of total loss identified. Many locations were fished that were probably never or shedding of a VIE mark was considered to be very low. The used by migrating U.S. postsmolts. Also, synchronizing survey VIE-mark detections rates of 97–99% on smolts 2–3 months timing with annually variable smolt migration from rivers is dif- after tagging at U.S. hatcheries (Fitzgerald et al. 2004; Sheehan ficult, and catch rates should be interpreted with caution. Many et al. 2011). Postsmolts with lost VIE marks and with adipose of the salmon of U.S. hatchery origin released in April and early fin clips would have been placed in the Canadian hatchery group May could have left the Gulf of Maine by the time surveys (Lacroix and Knox 2005) and excluded rather than assigned to started in late May and early June. Sheehan et al. (2011) found the wrong U.S. hatchery group. that U.S. hatchery smolts migrated soon after stocking and dis- Postsmolts from the Penobscot and Dennys rivers had a persed rapidly as they left confined embayments and entered the greater mean weight than smolts of the same origin at the hatch- Gulf of Maine. This and other factors affecting overlap between ery and leaving the rivers, an indication of early marine growth. fishing and postsmolts would certainly influence catch rate. For Furthermore, postsmolts were not emaciated, and the condition example, Lacroix and Knox (2005) reported the highest catch factor was indicative of growth and not excessive use of stored rates (0.32–0.69%) for postsmolts originating from rivers fur- energy reserves during early migration (Smith et al. 2009). thest inside the Bay of Fundy; these inner Bay of Fundy stocks Lacroix and Knox (2005) found a high frequency of food items have a long smolt migration period extending to mid-June (Flan- in the stomachs of hatchery origin postsmolts captured in the nagan et al. 2006), a slow rate of migration, and a tendency to Bay of Fundy and Gulf of Maine, and the food items were con- remain in the Bay of Fundy and Gulf of Maine (Lacroix et al. sistently larger than those eaten by the smaller wild postsmolts. 2005; Lacroix 2008), all features that would increase probability They also found increases in spacing of new circuli on scales of of capture. both wild and hatchery postsmolts in the marine environment, The catch rate of smolts was higher for those of known U.S.-origin salmon having the most extensive and rapid changes. than unknown river origin. For postsmolts of Penobscot and Hatchery smolts may have a growth advantage as they enter ma- Dennys origin, the catch was proportional to the number of rine habitat if they can prey more readily on large food items smolts of each origin released, except for the absence of Dennys or take in a larger number of prey items than the smaller wild postsmolts in the catch in 2003. Assuming equal catchability, smolts. This could lead to size-related differences in survival of these differences in catch rate could be indicative of differ- salmon in different habitats (Ritter 1989; Lacroix 2008). ences in migration route, migration timing and speed, or sur- The spatial and temporal distribution of U.S. hatchery vival among groups. Smolts from several rivers south of Maine postsmolts was not random. Catches from different river groups

Downloaded by [Department Of Fisheries] at 20:31 25 September 2012 were present annually in the unmarked group, and their poten- were aggregated within several areas both nearshore and off- tial dispersal further south than the surveys would effectively shore, and catches from the same river group tended to be aggre- have reduced the catch rate of unmarked smolts. The absence of gated at the same or adjacent locations. This indicates a possible those of Dennys River origin in 2003 was possibly related to the synchrony in migration at the scale of within-river release groups lateness of the ﬁrst smolt release: 28 April 2003 versus 19 April and an association that persisted at sea. Factors that initiate and 2002. In addition, all postsmolts of Penobscot River origin cap- synchronize smolt migration (Zydlewski et al. 2005) could lead tured in 2003 were from groups released earlier (16–22 April) to school formation or kin-structured congregations in rivers than the ﬁrst release in the Dennys. This suggests that smolts and estuaries (Olsen et al. 2004). Lacroix and Knox (2005) were possibly late in migrating or delayed in 2003 and, as a re- found that individually marked wild smolts leaving a river on sult, had a lower probability of capture. The rate of migration of the same date tended to be caught as postsmolts together in smolts and postsmolts of known origin was considerably lower the same tow. These captures, hundreds of kilometers away in 2003 than 2002, and it was similarly low for those of Dennys and weeks later, strongly suggest that some association of river River origin in 2002. The Dennys River drains into a series of stocks does persist during early stages of marine migration. large embayments, and postsmolts must travel through these and However, trawling in coastal habitat by Lacroix and Knox narrow channels with extreme tides to reach the Gulf of Maine, (2005) and Sheehan et al. (2011) did not yield large catches U.S. HATCHERY POSTSMOLT DISTRIBUTION 941

that would indicate schooling postsmolts. This may be due to manage activities within that corridor to mitigate potential to the low catchability of postsmolts by the trawl (Holst and losses of Atlantic salmon from endangered and threatened pop- McDonald 2000; J. C. Holst, Institute of Marine Research, ulations in North America. This would involve managing site Bergen, Norway, personal communication). In contrast, location for salmon farming to minimize exposure of postsmolts stationary gill-net captures provided evidence of postsmolt to predators, diseases, and parasites (Holst et al. 2003; Lacroix schooling in coastal waters of the Gulf of St. Lawrence (Dutil et al. 2004; Chang et al. 2005) and managing commercial pelagic and Coutu 1988), and loose feeding aggregations of adult and weir fisheries to minimize bycatch of postsmolts (Holst et al. salmon have been documented off the coast of West Greenland 2000; Holm et al. 2003; Lacroix and Knox 2005). in late summer and fall (ICES 2009). Most of the postsmolts of unknown U.S. origin caught probably migrated from the Penobscot River where the bulk of both ACKNOWLEDGMENTS VIE-marked and unmarked U.S. hatchery fish were stocked We thank the crews of the CCGS Alfred Needler and the many (USASAC 2006). Other unmarked smolts were released in sev- support staff and volunteers who participated in the trawling eral Gulf of Maine rivers south of Maine, and likelihood was surveys. Staff at National Oceanic and Atmospheric Adminis- high that they dispersed across the Gulf of Maine south of the tration Fisheries Service and the U.S. Fish and Wildlife Service survey areas. Therefore, with only five postsmolts of Dennys reared, marked, and released smolts into rivers of Maine. Barb River origin captured, the observed distribution of postsmolts Lubinski and Tim King at the U.S. Geological Survey Leetown most likely reflected dispersal and migration of the Penobscot Science Center provided the genotypes used for the parentage River population. Although they probably originated from the analysis. same river, unmarked postsmolts were slightly smaller than those of Penobscot River origin. This difference may have been REFERENCES the result of grading at the hatchery (Sheehan et al. 2011), but it Bartron, M. L., D. Buckley, T. King, M. Kinnison, G. Mackey, and T. Sheehan. could also be related to differences in time at sea before capture. 2006. Captive broodstock management plan for Atlantic salmon at Craig Grading involves the fall release of the smallest parr to maximize Brook national fish hatchery. Maine Atlantic Salmon Technical Advisory hatchery smolt production, which creates a size difference that Committee, Report, East Orland. persists at smolt migration (Renkawitz and Sheehan 2011). The Baum, E. T. 1997. Maine Atlantic salmon: a national treasure. Atlantic Salmon Unlimited, Hermon, Maine. presence of some postsmolts from parr stocking in the group of Chang, B. D., F. H. Page, and B. W. H. Hill. 2005. Preliminary analysis of coastal unknown U.S. origin could account for the bias towards smaller marine resource use and development of open ocean aquaculture in the Bay of size in this group compared with those of known river origin. Fundy. Canadian Technical Report of Fisheries and Aquatic Sciences 2585. The similar spatial distribution of postsmolts from the Penob- Cornuet, J. M., S. Piry, G. Luikart, A. Estoup, and M. Solignac. 1999. New scot River population to that of postsmolts of Canadian origin methods employing multilocus genotypes to select or exclude populations as origins of individuals. Genetics 153:1989–2000. (Lacroix and Knox 2005), coupled with annual aggregations Dutil, J. D., and J. M. Coutu. 1988. Early marine life of Atlantic salmon, Salmo along the coastal contour of southwest Nova Scotia close to salar, postsmolts in the northern Gulf of St. Lawrence. U.S. National Marine shore, support hypotheses about the temporary use of a com- Fisheries Service Fishery Bulletin 86:197–212. mon migratory corridor for different stocks of salmon leaving Fay, C., M. Bartron, S. Craig, A. Hecht, J. Pruden, R. Saunders, T. Sheehan, through the Gulf of Maine. This would imply that postsmolt and J. Trial. 2006. Status review for anadromous Atlantic salmon (Salmo salar) in the United States. Report to the National Marine Fisheries Ser- migration is partially the result of common oceanic conditions vice and U.S. Fish and Wildlife Service, National Oceanic and Atmospheric (e.g., surface current vectors and seasonal gyres, water temper- Administration, Office of Protected Resources, Silver Spring, Maryland. ature and salinity plumes, bathymetry and land constraints) as Fitzgerald, J. L., T. F. Sheehan, and J. F. Kocik. 2004. Visibility of visual implant

Downloaded by [Department Of Fisheries] at 20:31 25 September 2012 suggested by Meister (1984), Lacroix and Knox (2005), and elastomer tags in Atlantic salmon reared for two years in marine net-pens. Sheehan et al. (2011). Trawling (Lacroix and Knox 2005; Shee- North American Journal of Fisheries Management 24:222–227. Flannagan, J. J., R. A. Jones, and P. O’Reilly. 2006. A summary and evaluation han et al. 2011) and telemetry (Lacroix et al. 2004, 2005; Kocik of Atlantic salmon (Salmo salar) smolt monitoring and rotary screw fish trap et al. 2009) in these proposed migratory corridors have demon- activities in the Big Salmon River, 2001–2005. Canadian Technical Report strated that postsmolts tend to initially migrate close to shore of Fisheries and Aquatic Sciences 2646. and follow coastlines as they head to oceanic waters and that Hansen, L. P., M. Holm, J. C. Holst, and J. A. Jacobsen. 2003. The ecology of routes are consistent annually. post-smolts of Atlantic salmon. Pages 25–39 in D. Mills, editor. Salmon at the edge. Blackwell Scientific Publications, Oxford, UK. A common, temporary migratory corridor for postsmolts mi- Holm, M., J. C. Holst, and L. P. Hansen. 2000. Spatial and temporal distribution grating from U.S. and Canadian rivers to marine feeding grounds of post-smolts of Atlantic salmon (Salmo salar L.) in the Norwegian Sea and may result in their exposure to the same potential causes of mor- adjacent areas. ICES Journal of Marine Science 57:955–964. tality. Although the spatial surveys did not capture the full time Holm, M., J. C. Holst, L. P. Hansen, J. A. Jacobsen, N. O’Maoileidigh,´ and A. span of this migration, it can safely be assumed to occur mostly Moore. 2003. Migration and distribution of Atlantic salmon post-smolts in the North Sea and north-east Atlantic. Pages 7–23 in D. Mills, editor. Salmon during May and June, and may extend into July (Meister 1984; at the edge. Blackwell Scientific Publications, Oxford, UK. Ritter 1989; Lacroix 2008; Sheehan et al. 2011). This seasonal Holst, J. C., P. Jakobsen, F. Nilsen, M. Holm, L. Asplin, and J. Aure. 2003. passage through a specific coastal area provides an opportunity Mortality of seaward-migrating post-smolts of Atlantic salmon due to salmon 942 LACROIX ET AL.

lice infection in Norwegian salmon stocks. Pages 136–137 in D. Mills, editor. Marshall, T. C., J. Slate, L. E. B. Kruuk, and J. M. Pemberton. 1998. Statistical Salmon at the edge. Blackwell Scientific Publications, Oxford, UK. confidence for likelihood-based paternity inference in natural populations. Holst, J. C., and A. McDonald. 2000. FISH-LIFT: a device for sampling live Molecular Ecology 7:639–655. fish with trawls. Fisheries Research (Amsterdam) 48:87–91. Meister, A. L. 1984. The marine migration of tagged Atlantic salmon (Salmo Holst, J. C., R. Shelton, M. Holm, and L. P.Hansen. 2000. Distribution and possi- salar L.) of USA origin. ICES (International Council for the Exploration of ble migration routes of post-smolt Atlantic salmon in the north-east Atlantic. the Sea) CM 1984/M:27, Copenhagen. Pages 65–74 in D. Mills, editor. The ocean life of Atlantic salmon: envi- NOAA (National Oceanic and Atmospheric Administration) 2009. Endangered ronmental and biological factors influencing survival. Fishing News Books, and threatened species; determination of endangered status for the Gulf Oxford, UK. of Maine distinct population segment of Atlantic salmon. Federal Register ICES (International Council for the Exploration of the Sea). 2009. Report of 74:117(19 June 2009):29344–29387. the working group on North Atlantic salmon (WGNAS). ICES Advisory Olsen, K. H., E. Petersson, B. Ragnarsson, H. Lundqvist, and T. Jarvi. Committee, CM 2009/ACOM:06, Copenhagen. 2004. Downstream migration in Atlantic salmon (Salmo salar)smoltsib- Jonsson, N., L. P. Hansen, and B. Jonsson. 1993. Migratory behaviour and ling groups. Canadian Journal of Fisheries and Aquatic Sciences 61:328– growth of hatchery-reared post-smolt Atlantic salmon Salmo salar. Journal 331. of Fish Biology 42:435–443. Renkawitz, M. D., and T. F. Sheehan. 2011. Feeding ecology of early marine Kocik, J. F., J. P. Hawkes, T. F. Sheehan, P. A. Music and K. F. Beland. phase Atlantic salmon Salmo salar posts-molts. Journal of Fish Biology 2009. Assessing estuarine and coastal migration and survival of wild At- 79:356–373. lantic salmon smolts from the Narraguagus River, Maine using ultrasonic Ritter, J. A. 1989. Marine migration and natural mortality of North American telemetry. Pages 293–310 in A.Haro,K.L.Smith,R.A.Rulifson,C.M. Atlantic salmon (Salmo salar L.). Canadian Manuscript Report of Fisheries Moffitt, R. J. Klauda, M. J. Dadswell, R. A. Cunjak, J. E. Cooper, K. L. and Aquatic Sciences 2041. Beal, and T. S. Avery, editors. Challenges for diadromous fishes in a dynamic Shearer, W. M., editor. 1992. Atlantic salmon scale reading guidelines. ICES global environment. American Fisheries Society, Symposium 69, Bethesda, (International Council for the Exploration of the Sea) Cooperative Research Maryland. Report 188. Lacroix, G. L. 2008. Influence of origin on migration and survival of Atlantic Sheehan, T. F., M. D. Renkawitz, and R. W. Brown. 2011. Surface trawl survey salmon (Salmo salar) in the Bay of Fundy, Canada. Canadian Journal of for U.S. origin Atlantic salmon Salmo salar. Journal of Fish Biology 79:374– Fisheries and Aquatic Sciences 65:2063–2079. 398. Lacroix, G. L., and D. Knox. 2005. Distribution of Atlantic salmon (Salmo Smith, I. P., D. J. Booker, and N. C. Wells. 2009. Bioenergetic modeling of salar) postsmolts of different origins in the Bay of Fundy and Gulf of Maine the marine phase of Atlantic salmon (Salmo salar L.). Marine Environmental and evaluation of factors affecting migration, growth, and survival. Canadian Research 67:246–258. Journal of Fisheries and Aquatic Sciences 62:1363–1376. USASAC (U.S. Atlantic Salmon Assessment Committee). 2006. Annual re- Lacroix, G. L., D. Knox, and M. J. W. Stokesbury. 2005. Survival and behaviour port of the U.S. Atlantic salmon assessment committee, report no. 18: 2005 of post-smolt Atlantic salmon in coastal habitat with extreme tides. Journal activities. USASAC, Gloucester, Massachusetts. of Fish Biology 66:485–498. Zydlewski, G. B., A. Haro, and S. D. McCormick. 2005. Evidence for cu- Lacroix, G. L., P. McCurdy, and D. Knox. 2004. Migration of Atlantic salmon mulative temperature as an initiating and terminating factor in downstream postsmolts in relation to habitat use in a coastal system. Transactions of the migratory behavior of Atlantic salmon (Salmo salar) smolts. Canadian Jour- American Fisheries Society 133:1455–1471. nal of Fisheries and Aquatic Sciences 62:68–78. Downloaded by [Department Of Fisheries] at 20:31 25 September 2012 This article was downloaded by: [Department Of Fisheries] On: 25 September 2012, At: 20:31 Publisher: Taylor & Francis Informa Ltd Registered in England and Wales Registered Number: 1072954 Registered office: Mortimer House, 37-41 Mortimer Street, London W1T 3JH, UK

Transactions of the American Fisheries Society Publication details, including instructions for authors and subscription information: http://www.tandfonline.com/loi/utaf20 A Field-based Bioenergetics Model for Estimating Time- Varying Food Consumption and Growth Brett T. van Poorten a , Carl J. Walters a & Nathan G. Taylor b a Fisheries Centre, University of British Columbia, 2202 Main Mall, Vancouver, British Columbia, V6T 1Z4, Canada b Fisheries and Oceans Canada, 3190 Hammond Bay Road, Nanaimo, British Columbia, V9T 6N7, Canada

Version of record first published: 20 Jun 2012.

To cite this article: Brett T. van Poorten, Carl J. Walters & Nathan G. Taylor (2012): A Field-based Bioenergetics Model for Estimating Time-Varying Food Consumption and Growth, Transactions of the American Fisheries Society, 141:4, 943-961 To link to this article: http://dx.doi.org/10.1080/00028487.2012.675919

PLEASE SCROLL DOWN FOR ARTICLE

Full terms and conditions of use: http://www.tandfonline.com/page/terms-and-conditions 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. The publisher does not give any warranty express or implied or make any representation that the contents will be complete or accurate or up to date. The accuracy of any instructions, formulae, and drug doses should be independently verified with primary sources. The publisher shall not be liable for any loss, actions, claims, proceedings, demand, or costs or damages whatsoever or howsoever caused arising directly or indirectly in connection with or arising out of the use of this material. Transactions of the American Fisheries Society 141:943–961, 2012 C American Fisheries Society 2012 ISSN: 0002-8487 print / 1548-8659 online DOI: 10.1080/00028487.2012.675919

ARTICLE

A Field-based Bioenergetics Model for Estimating Time-Varying Food Consumption and Growth

Brett T. van Poorten* and Carl J. Walters Fisheries Centre, University of British Columbia, 2202 Main Mall, Vancouver, British Columbia V6T 1Z4, Canada Nathan G. Taylor Fisheries and Oceans Canada, 3190 Hammond Bay Road, Nanaimo, British Columbia V9T 6N7, Canada

Abstract Bioenergetics models are often used to describe the implications of changes in growth and consumption of specific wild populations, and yet most parameters are derived from a variety of laboratory studies on other populations or species, leading to questions regarding the validity of predictions. A novel bioenergetics approach was recently developed where many parameters are estimated from the population being modeled, but growth and consumption are assumed invariant over time, which would not hold true when manipulations to the system are known or suspected. In the present paper, a bioenergetics model with many key parameters estimated from field data are presented where temporal deviations in growth rates were directly estimated. A series of rainbow trout Oncorhynchus mykiss and northern pikeminnow Ptychocheilus oregonensis populations, which have undergone various population manipulations, were used to evaluate the model. Further, the model was fit to a series of rainbow trout size-classes stocked into each of the study lakes to compare with their wild counterparts and evaluate intercohort differences in growth and consumption. We found the model with time-varying consumption was more parsimonious compared with models where growth and consumption were assumed to be constant over time. Our field data demonstrated how the model can detect different patterns in growth and consumption across populations and species. The model detected highly variable growth and consumption in rainbow trout over time and between populations but did not seem to be particularly influenced by past population manipulations. By contrast, northern pikeminnow demonstrated differences between lakes, but showed little temporal variation in growth and consumption. Stocked rainbow trout demonstrated similar growth rates to their wild counterparts, helping to validate growth estimates. Our bioenergetics model moves beyond existing ones by allowing measurement and process errors to be explicitly represented, while also permitting growth and consumption to vary over time. Downloaded by [Department Of Fisheries] at 20:31 25 September 2012

Growth in poikilothermic animals can be quite plastic, de- and quality of different prey items consumed. Changes in net pending both on biotic factors, such as the quality and quan- consumption rates of a predator may have far-reaching implica- tity of food organisms available, and abiotic factors, such as tions for other organisms living in the system, through predation temperatures experienced, which influence production of food or competitive interactions. organisms as well as the consumption and metabolic rates of the The use of bioenergetics models for estimating growth and consumer (Jobling 1994). While changes in growth are partially consumption in fish has grown rapidly over the last several limited by the metabolic rate of an organism, it is net consump- decades (Hartman and Kitchell 2008). Bioenergetics models are tion rate that is likely to be the most variable over time and essentially mass-balance models that partition consumed energy depends on the assimilation efficiency and the relative quantity into growth and metabolism, or alternatively, attribute growth

*Corresponding author: b.vanpoorten@ﬁsheries.ubc.ca Received October 24, 2011; accepted February 22, 2012 Published online June 20, 2012 943 944 VAN POORTEN ET AL.

to the difference between consumed energy and metabolic costs suggests that it predicts growth and consumption at least as (Brandt and Hartman 1993; Ney 1993; Walters and Essington well as the Wisconsin model (van Poorten and Walters 2010). 2010). These models typically include environmental covari- One drawback of the general bioenergetics model presented ates, most commonly water temperature, to predict seasonal by Walters and Essington (2010) is that growth is based on a sin- variation in growth and consumption patterns. Direct estimates gle average lifetime trajectory with individual variation around of consumption and metabolic rates from bioenergetics mod- this trajectory, which assumes the parameters being estimated els can be incorporated into trophic interaction models where are stationary over time. In situations where data are collected consumption by one group is a direct component of mortal- over several months or years and known manipulations to the ity for its prey group (Irwin et al. 2003; Rose et al. 2008; system have taken place, this is unlikely (Matuszek et al. 1990; Tuomikoski et al. 2008; Myers et al. 2009). Similarly, con- Jobling and Baardvik 1994). Indeed, one of the key uses of sumption by predators can be predicted by abundance and avail- bioenergetics models is estimating how growth or consumption ability of its prey (Hayes et al. 2000; Beauchamp et al. 2004; varies over time or space through changes in consumption. In the Johnson et al. 2006). The obvious utility of bioenergetics mod- Wisconsin model, this is achieved by estimating a separate P for els has led to their use in studies for predicting or explaining each size and time interval of interest (Hanson et al. 1997). While changes in growth or consumption following various natural or the general bioenergetics model accounts for seasonal changes anthropogenic disturbances (Hansen et al. 1993; Jobling 1994; in consumption, metabolism, and prey availability through their Hanson et al. 1997; Chipps and Wahl 2008; Hartman and dependence with water temperatures, this probably underesti- Kitchell 2008). mates variation owing to other sources, such as changes in inter- By far the most commonly used bioenergetics model for and intraspecific competition through changes in the predator or fish is the Fish Bioenergetics software (Hanson et al. 1997), prey communities (Fraser and Gilliam 1992). To take advantage more commonly known as the “Wisconsin” model (Ney 1990; of the benefits of a field-based bioenergetics model, it would Hanson et al. 1997). This model uses laboratory or in situ es- be beneficial to be able to detect differences in growth over timates of various metabolic and physiological rates and func- time. tions to parameterize anabolism and catabolism. Based on the As an example, eight lakes in south-central British Columbia parameter values provided and physiological functions chosen, (Figure 1) were manipulated to reduce abundance of either rain- the model calculates the maximum consumption an individ- bow trout Oncorhynchus mykiss (hereafter referred to simply ual is capable of obtaining at a given temperature. The model as trout) or northern pikeminnow Ptychocheilus oregonensis estimates consumption rates at a given time interval by calcu- (hereafter referred to as pikeminnow) with the express purpose lating the proportion of maximum consumption (P) necessary of investigating interspecific competition and the influence of to account for observed growth between two points in time competition on recruitment. Depletion netting near the start (Hanson et al. 1997). There is considerable flexibility in this of the study reduced adult densities of one species or the model, particularly because the growth trajectory can be broken other in some lakes by between 60% and 90% (Taylor 2006; up into multiple stanzas, each with an independent P, whose O’Brien 2009), which would be expected to result in changes endpoints are defined as the age at which size or consumption in competition within and among populations for food, leading estimates are provided. to changes in consumption and growth across lakes and over An alternate bioenergetics model was recently developed time after removals. In a scenario like this, the assumption by Walters and Essington (2010) and is referred to as a general of constant consumption over time would almost certainly be bioenergetics model. This general bioenergetics model uses violated. Therefore, a new method of estimating time-varying length-increment and length-at-age data commonly collected consumption and growth using the general bioenergetics model

Downloaded by [Department Of Fisheries] at 20:31 25 September 2012 in field studies to directly estimate many of the parameters is certainly warranted. necessary to predict growth and consumption. By directly To account for changes in consumption that may occur ow- estimating parameters of the population of interest, and by ing to changes in community structure over time, we introduce summarizing the bioenergetics rates in terms of aggregate a variation on the Walters and Essington (2010) general bioen- parameters, at least some of the parameters to be estimated ergetics model where population- and year-specific changes in do not need to be borrowed from other populations or similar consumption and growth are directly estimated. If known per- species, as can be common with the Wisconsin model (Ney turbations occur over time or space, the model should be able to 1990, 1993; Petersen et al. 2008). Additionally, by estimating estimate their effects on growth and consumption. We evaluate parameters using likelihood or Bayesian estimation techniques, the model using data from the populations in Bonaparte Plateau uncertainty in parameter estimates and predicted growth and lakes (referred to here as the Bonaparte lakes). These lakes consumption are also estimated (van Poorten and Walters 2010). experience similar environmental conditions, but have varying This represents an improvement over the Wisconsin model, productivity, prey, and competitor densities and have experi- which does not admit uncertainty either in the parameters used enced changes to the fish community over the course of data in the model, or in the predicted growth and consumption. collection (Taylor 2006; O’Brien 2009). We additionally fit the A preliminary evaluation of the general bioenergetics model model to a series of differentiallysized cohorts of trout stocked USING A BIOENERGETICS MODEL FOR FOOD CONSUMPTION 945 Downloaded by [Department Of Fisheries] at 20:31 25 September 2012

FIGURE 1. Study area in the Bonaparte Plateau (bottom panel) and the study area position within British Columbia (top). Lakes used in the current study are ﬁlled and labeled as A: Meghan Lake; B: Cath Lake; C: Moose Pasture Lake; D: Wilderness Lake; E: Cheryl Lake; F: Dads Lake; G: Moms Lake; H: Nestor Lake. [Figure available online in color.]

into all of the study lakes over several years to validate findings growth can be modeled as from the wild population and examine intercohort differences in growth and consumption. dW = HWd f T − mW nf T , dt c( ) m( ) (1) MODEL DEVELOPMENT Below, we briefly describe the general bioenergetics model where the first and second terms are anabolism and catabolism, of Walters and Essington (2010). For further details of model respectively; H is a mass normalized net (of assimilation and structure, refer to the original article. The general bioenergetics specific dynamic action losses) rate of mass acquisition through model as introduced by Walters and Essington (2010) assumed feeding, W is whole-body mass, d is a scalar relating anabolism 946 VAN POORTEN ET AL.

to mass, m is a mass-normalized rate of mass loss through Size at age in 1 year built upon the size at age defined by param- catabolic processes, and n is a scalar relating catabolism to mass. eters of the previous years by updating γ l,y as the integration The two terms fc and fm were functions relating anabolism and proceeded over the years of the study. catabolism to temperature (T). They take the forms The model was fit to length-increment data of wild fish by finding the estimated age of each recaptured individual that min- e−g(T −Tm) T −T/¯ 10 imized the following likelihood (Walters and Essington 2010): fc(T ) = Qc 1 + e−g(T −Tm) LLRi(a1i ) and L − L¯ a Rˆ a 2 + L − L¯ a + Rˆ a 2 =−( 1i ( 1i ) i ( 1i )) ( 2i ( 1i ti) i ( 1i )) 2 2σm f T = QT −T/¯ 10, c( ) c Rˆ a − 2 − i ( 1i ) 1) , 2 (3) 2σR where fc(T) depicts consumption increasing exponentially at ◦ arateofQc for every 10 C increase above the mean annual temperature, T¯ , to a maximum and then declines to half the where a1i is the age of individual i at capture, L1i and L2i are L¯ a L¯ a +t maximum at a rate of g when temperature reaches T .Theterm observed length at capture and recapture, ( 1i ) and ( 1i i ) m σ 2 σ 2 f (T) simply results in metabolism increasing at a rate Q for are estimated ages at capture and recapture and m and R are m m Rˆ a every 10◦C. measurement and process error, respectively. The term i ( 1i ) Walters and Essington (2010) introduced equation (1) to de- is the growth deviation of individual i and is given by

scribe a seasonal model for growth that accounted for declines σ 2 L¯ a L + L¯ a + t L + m ( 1i ) 1i ( 1i i ) 2i 2 in growth, metabolism, and consumption with cooler water tem- σR Rˆ i a i = . peratures, as usually occurs in temperate climates. Walters and ( 1 ) σ 2 (4) L¯ a 2 + L¯ a + t 2 + m ( 1i ) ( 1i i ) σ 2 Essington (2010) termed this model the Continuous Allocation R model, with the assumption that there is a constant length– weight relationship over the life of a fish (i.e., W = aLb), which The model was fit to length-at-age data of wild fish by minimiz- is unlikely. To describe a more realistic interpretation of life- ing the following negative log-likelihood: time growth, they further developed the model to describe what L − L¯ a they term a Seasonal Reproduction, Skeletal Allocation (SRSA) ai ( 1i ) LLai =−wa , (5) 2 model, which accounted for complex length–weight dynamics σm + CVaL¯ (a1i ) owing to annual loss in mass to gonad development and variable allocation of consumed mass to skeletal and metabolizable where CVa in this equation is the coefficient of variation mass in times of metabolic stress (e.g., overwinter or after re- (SD/mean) in length of aged fish and wa is a weighting factor production). The SRSA model used the information on variable for downweighting the importance of aged fish in the likelihood. length–weight dynamics to predict the sequence of mean length The total likelihood of all data given model structure and pa- at age over a lifetime. Details of this model can be found in rameters was found by calculating the sum of equations (3) and Walters and Essington (2010). (5) combined. We evaluated a new parameterization of the general bioenergetics model, which included the possibility of consumption METHODS varying over time and across populations. We constructed the We analyzed data from eight study lakes located on the Bona- Downloaded by [Department Of Fisheries] at 20:31 25 September 2012 model to be able to estimate parameters from multiple, similar parte Plateau in south-central British Columbia (Figure 1). The populations simultaneously, but this is not necessary. In our region is a low-relief, mid-elevation (∼1,500 m) area where parameterization, we assumed that net consumption rate varied low productivity lakes experience a short open-water season both across populations and over time within each population, (Taylor et al. 2005). The lakes are covered with ice from October while all other parameters were fixed across populations and through May. Summer water temperatures rarely exceed 20◦C. over time. Each population was assumed to have been at equi- All eight lakes are within 5 km of one another and experience librium before and including the first year of data, which was similar weather and depth–temperature patterns, although lakes modeled with a lake-specific H, Hlake, and all other parameters farther downstream are slightly warmer (D. S. O’Brien, Fish- shared between lakes. From the second year on, the growth eries and Oceans Canada, unpublished data; B. T. van Poorten, in each lake and year was modeled as above, but with Hlake unpublished data). multiplied by a year-specific multiplier, γ l,y. Essentially, growth Fish populations in each study lake were expected to exhibit withinalake(lake) and lake-year (l,y) was defined by the varying growth rates owing to among-lake variation in produc- expression tivity, water temperatures, and fish community structure. Lakes farther downstream in each of the two watersheds typically dW d n had higher productivity and higher mean annual temperatures = H γl,yW fc(T ) − mW fm(T ). (2) dt lake (Taylor 2006). An important additional source of variation USING A BIOENERGETICS MODEL FOR FOOD CONSUMPTION 947

was that brought on by size-structured competition for and so were effectively removed from the population. These re- resources. Trout in the study lakes typically resided in the movals would not affect our growth estimates, although in cases same lake for their whole lives (∼6% straying rate, B.T.v.P., where large numbers died (i.e., owing to intentional depletions) unpublished data), resulting in similar size-structure across they would probably influence competition and predation in the lakes. Conversely, pikeminnow exhibited a complex spatial systems, and therefore result in changes in consumption and ontogeny, which led to variation in size-structure between growth rates from year to year. Fish with obvious misidenti- lakes in the same watershed (Taylor 2006). Adult pikeminnow fication errors because of either misread tags or transcription were present in all study lakes, but adults living in the most errors (identified as growth rate greater than 200 mm/year or upstream lakes typically spawned in outflow streams. Hatched length decrease greater than 5 mm) were removed from the age-0 pikeminnow drifted to lakes below those occupied by data set, as were fish moving between lakes (which account for their parents and reared for 2 to 3 years until they were large 6.6% of records) because the date of migration is unknown. In and strong enough to migrate between lakes. The result was total, 915 trout and 2,881 pikeminnow tagged and later recap- upstream (“headwater”) lakes that were devoid of the youngest tured were included for analysis. Subsamples of fish collected in age-classes and lakes farther downstream (“nursery lakes”) most years were sampled for aging structures. Trout were aged that had high densities of young pikeminnow as well as some using scales collected between the dorsal fin and lateral line, adult pikeminnow. This may have caused complex intercohort and pikeminnow were aged using lapilli otoliths. All ages were competition among pikeminnow that may have affected growth estimated by two independent readers and any differences were in these lakes. Further, trout in these nursery lakes might resolved by a third reader. Additionally, scales of known-age have supplemented their diet with young pikeminnow, thereby trout were randomly included among samples to verify reader resulting in increased trout growth in these lakes. Headwater accuracy. In total, 6,247 trout and 7,602 pikeminnow ages were lakes in this system included Moms, Nestor, Cheryl, and obtained across all lakes. Readers agreed on 49% of ages from Wilderness lakes; nursery lakes included Dads, Moose Pasture, trout scales and 89% of ages from pikeminnow otoliths. Valida- Cath, and Meghan lakes. Cath and Meghan lakes appeared to tion from known-age trout showed 41% accuracy. have extremely high densities of young pikeminnow owing to Water temperatures were derived from observed tempera- their relative position within the watershed (Figure 1). tures measured in each lake at various occasions over the course Among-year differences in growth might have occurred be- of the study. Observations were used to parameterize an annual cause of shifts in relative density of the two fish species over sinusoidal water temperature model with a minimum winter time owing to differential removals of the two species. Early temperature (similar to Taylor and Walters 2010). in the study, two lakes (Wilderness: 2001; Nestor: 2002) were Known numbers of hatchery-reared trout were stocked into partially depleted of adult trout and three lakes (Cheryl: 2001; the study lakes in the last 3 years of the study to examine interco- Moms: 2001; Moose Pasture: 2002) were partially depleted of hort differences in growth and consumption within a lake-year adult pikeminnow. Dads, Cath, and Meghan lakes were not de- (Table 1). Fish were raised from gametes collected from a wild pleted of any one species and were intended to serve as “control” population in Pennask Lake, British Columbia, and reared to one systems for assessing the impacts of depletion removals. Fur- of four size-classes. Age-0 trout were raised until they were a ther, all lakes were partially depleted of both species in the last sufficient size to fin-clip (∼0.5 g) and released in approximately 3 years of the study using a combination of gill nets set in a stan- mid-August of 2007 and 2008. Age-1 trout were raised for ap- dardized configuration (Post et al. 1999; Askey et al. 2007) and proximately 11 months at three different temperatures to create small-mesh fyke nets. These changes to the fish community in three size-classes with little or no size overlap. Age-1 trout were each lake over time were expected to lead to variations in growth released in early June of each year (see Table 1 for stocking de-

Downloaded by [Department Of Fisheries] at 20:31 25 September 2012 among years and lakes owing to variation in intraspecific and tails). All size-classes were given a unique combination of fin possible interspecific competition for resources. clips to identify size- and year-class. Subsets of released fish Data collection.—Fish sampling occurred throughout the from each size-class were measured to obtain length at release open-water season from 2001 to 2008, although not all lakes and the corresponding coefficient of variation (CV) in length. were sampled in every year. Cheryl, Dads, Moms, and Moose Intensive netting in late September of each year was used to re- Pasture lakes were not sampled in 2004, Wilderness Lake was capture stocked fish to estimate growth rates of each size-cohort not sampled in 2004 or 2005, and Cath and Meghan lakes were in each year. only sampled between 2006 and 2008. Diet data were collected from both species on three separate Sampling was conducted throughout the study primarily us- monthly occasions from six lakes in 2006. Fish were captured ing fyke nets and gill nets. Individual fish sampled with fyke using randomly placed gill nets set overnight in the benthic and nets were measured, and fish larger than 120 mm were marked pelagic zones. Stomachs were sorted into 30-mm size bins to with an individually numbered spaghetti tag and released alive. a maximum of 10 stomachs per species per bin. Diet compo- Recaptured individuals were checked for spaghetti tags and sition of each age-class was assumed to be represented as the measured. Fish captured with gill nets were all measured and proportion of each diet organism (biomass calculated by mul- checked for spaghetti tags, but usually died owing to sampling tiplying numbers by mean dry weight of sampled zooplankton 948 VAN POORTEN ET AL.

TABLE 1. Mean size and CV, date of stocking, and numbers of stocked ﬁsh from each cohort of hatchery-reared rainbow trout in each lake-year. All estimates of mean and CV in length are based on subsamples of 100 individual ﬁsh from each cohort in each year.

Stocking density (ﬁsh/ha) Mean Stocking length CV Moose Size-class date (mm) (mm) Cheryl Cath Dads Meghan Moms Nestor Pasture Wilderness Small age 1 12 June 2006 96 0.12 12 12 12 12 12 12 12 12 Medium age 1 12 June 2006 118 0.13 12 12 12 12 12 12 12 12 Large age 1 12 June 2006 159 0.17 10 10 10 70 Age 0 12 June 2007 43 0.08 1,650 1,650 500 1,650 1,650 500 500 500 Small age 1 12 June 2007 89 0.11 17 20 8 20 17 8 8 8 Medium age 1 12 June 2007 110 0.11 17 20 8 20 17 8 8 8 Large age 1 12 June 2007 142 0.12 17 20 8 20 17 8 8 8 Age 0 12 June 2008 43 0.07 500 3,500 3,500 3,500 500 500 3,500 500 Small age 1 7 June 2008 89 0.10 8 25 17 25 8 8 17 8 Medium age 1 7 June 2008 116 0.09 8 25 17 25 8 8 17 8 Large age 1 7 June 2008 153 0.11 8 25 17 25 8 8 17 8

and benthic invertebrates) across all size-bins that the fish of this Johnston (1999). We chose to include an informative normally age-class grow through over each year. As size at age changes distributed prior distribution function for Qm with a mean of 2.0 across years, the same age-class may have incorporated differ- and SD of 0.02. Finally, the influence of aged fish on the likeli- ent size-classes, resulting in small changes in diet proportions hood was downweighted for each species by setting wa to 0.5. over time. Diet was assumed to be unchanged for fish larger than To determine whether estimating year-specific variation in 240 mm and 150 mm for trout and pikeminnow, respectively, so net consumption rate results in an improvement in fit of the all bins were combined beyond this size. model to the data, we compared the model fit of the Time- Parameter estimation and model evaluation.—The full Varying Consumption model with one similar to the original model was used to evaluate changes in growth and consumption general bioenergetics model, which we refer to as the Time- for both the trout and pikeminnow populations in the Bonaparte Invariant Consumption model. As above, we assumed that all lakes. All populations for each species were run simultaneously, parameters apart from net consumption rate are shared among with m, Qc and Qm shared among populations. We simplified lakes and years. We further assumed that the net consumption analysis by fixing the anabolic and catabolic scalars to values rate is unique to each lake but does not vary over years. This 2 assumed in von Bertalanffy growth (i.e., d = /3 and n = 1.0; was akin to the Time-Varying Consumption model above, with Essington et al. 2001). The Hlake term was estimated for each γ l,y set to 1 for each year after the first year. In both species, this lake, and one γ l,y was estimated for each lake-year for which resulted in 11 parameters being estimated: Hlake for eight lakes, data were available after the initial year of data collection on m, Qc, and Qm, which are shared among lakes. We referred to each lake. Therefore, across all lakes and years, the parameter this as the Time-Invariant Consumption model.

Downloaded by [Department Of Fisheries] at 20:31 25 September 2012 vector being estimated included a vector Hlake equal to the For each species, the most parsimonious model was selected number of lakes, a vector γ l,y equal to the number of lake-years using the Bayesian information criterion (BIC; Schwarz 1978). where data exist, minus the first year, m, Qc, and Qm.We The principal of parsimony states that one should balance the fit referred to this model as the Time-Varying Consumption model. of a model against the number of parameters used to generate The number of lake-years where data exist varies across the the fit. As the number of parameters increases, bias declines, but two species owing to the differential ability to recapture marked variance increases (Hilborn and Walters 1992; Burnham and individuals. Lake-years where no fish were recaptured had γ l,y Anderson 2002). The BIC can favor lower dimension models set equal to the last year where data exist. In total, 50 parameters more strongly than Akaike’s information criterion (AIC; were estimated for trout (47 lake-years of data) and 45 param- Burnham and Anderson 2002), especially when the number eters were estimated for pikeminnow (42 lake-years of data). of observations is large (Schwarz 1978), resulting in a more Prior distributions for all parameters except Qm were assumed conservative measure of model parsimony. to be uniform with bounds given in Table 2. The metabolic Q10 We assumed the same model and parameter structure for parameter (Qm) is commonly confounded with Qc and difficult size–year cohorts of stocked trout as was used for the wild pop- to estimate. Walters and Essington (2010) recommend setting ulations above (cohort-specific H, all other parameters shared). Qm at or close to 2.0 based on the metanalysis of Clark and Growth of stocked fish was only estimated over one to several USING A BIOENERGETICS MODEL FOR FOOD CONSUMPTION 949

TABLE 2. Fixed parameters and prior distributions for parameters allowed to vary that were used in the general bioenergetics model for predicting growth. Prior distributions are denoted as either U(l,u), indicating a uniform prior distribution with lower and upper bounds in parentheses, or N(µ,σ ), indicating a normal prior distribution with mean and SD in parentheses.

Parameter value / prior distribution Parameter Description Rainbow trout Northern pikeminnow Units a Intercept coefficient of length–weight relationship 1.08 × 10−5 1.16 × 10−5 g·mm−b b Power coefficient of length–weight relationship 3.01 3.00 – −d −d −1 Hlake Net food consumption rate per W U(0,100) U(0,100) g·g ·year γ l,y Year-specific net food consumption multiplier U(0,10) U(0,10) m Standard metabolic rate per W−n U(0,20) U(0,20) g·g−n·year−1 d Food consumption power parameter 0.67 0.67 n Metabolism power parameter 1.0 1.0 ◦ Qc Proportional increase in feeding rate per 10 C U(0,10) U(0,10) temperature increase ◦ Qm Proportional increase in metabolism per 10 C N(2,0.02) N(2,0.02) temperature increase θ Slope parameter for decreasing allocation to 0.2 0.02 ∗ structural tissue as Ws/W varies around f s g Steepness parameter for decrease in feeding at 1.76 3.0 ◦C−1 high temperatures ◦ Tm Water temperature at which feeding drops by half 23.5 24 C Wma Weight at maturity 131 27 g pgonad Proportion of body weight lost to spawning 0.15 0.15 2 CVL-L Coefficient of variation of individual maximum 0.4 0.4 mm body lengths 2 CVL-A Coefficient of variation of individual maximum 0.2 0.1 mm body lengths for aged fish wa Weighting factor for aged fish 0.5 0.5 2 2 σm Measurement variance for L1 and L2 1.3 2.45 mm ae Assimilation efficiency 0.8 0.8 SDA Specific dynamic action 0.172a 0.163b

aRand et al. (1993). bPetersen and Ward (1999).

months so it was not possible to estimate curvature in growth Qm, and cohort-specific Hcoh, hypermeans, (µcoh) and hyper- rate owing to either seasonality or allometry. It was therefore precisions (τ coh) of net consumption rates were also estimated, necessary to assume metabolism and temperature dependence which defined the shape of the prior distributions for each size- was similar between wild and stocked trout populations by using class. Hypermeans for each age-class were normally distributed Downloaded by [Department Of Fisheries] at 20:31 25 September 2012 the posterior distribution functions for m, Qc, and Qm from the with mean of 6.0 and CV of 100. Hyperprecisions for each wild population as prior distribution functions for the stocked age-class were gamma distributed with shape and scale of 0.01. cohorts. A separate net consumption rate (Hcoh) for each cohort Likelihoods for estimating parameters for the stocked cohorts of size-classes stocked into each lake in each year was esti- based on individuals that were captured or captured, tagged, mated. We assumed a hierarchical structure for net consumption and recaptured were similar to those from wild fish (equations 5 rates within size-classes across years and lakes. The hierarchical and 3, respectively) except ages of all fish were known owing Bayesian analysis allowed information from other similar sam- to cohort-specific fin clips. Therefore, any fish that were tagged pling units to be incorporated by assuming that other units were after stocking and later recaptured were evaluated at the known not independent, but were exchangeable units within a defined age for that cohort in equations (3) and (4), rather than searching population of similar units, resulting in improved individual over all ages. estimates. Exchangeability implies that lake- and year-specific To compare growth rate estimates between wild and stocked prior distributions for net consumption rates were independently trout, as well as to determine if growth varies within a year drawn from a common distribution defined by hyperparameters among differentially sized fish, we estimated instantaneous (Askey et al. 2007). Therefore, in addition to estimating m, Qc, growth of wild and stocked fish. Instantaneous growth was 950 VAN POORTEN ET AL.

TABLE 3. Energy densities and sources for all prey items included in calcu- the Hastings–Metropolis algorithm found in AD model Builder lating consumption. Sources are 1: Cummins and Wuycheck (1971); 2: Mills and (Fournier et al. 2012). We ran four chains with a thinning inter- Forney (1981) based on juvenile yellow perch Perca flavescens of comparable size. val of 100. Burn-in, the number of iterations needed to reach the joint posterior distribution of all parameters, was evaluated us- Energy Density ing the Gelman and Rubin convergence statistics (Gelman et al. Prey group (J/g dry weight) Source 1995) and by visual examination of trace plots. After burn-in removal, a chain of 10,000 samples was saved as an approximation Daphnia 21,051 1 of the posterior distribution. Bosmina 21,905 1 Daily consumption per gram body weight was estimated us- Copepoda 24,036 1 ing the equation Cyclopoidae 24,233 1 Amphipoda 16,756 1 d H γ Ws,t fc(Tt ) Hirudinea 22,789 1 C = lake l,y , t W ·e· (6) Diptera 17,903 1 t 365 Trichoptera 20,930 1 Ephemeroptera 22,898 1 where Ws,t is structural weight at time t and e is growth efficiency, Odonata 21,424 1 calculated as Coleoptera 22,487 1 I p E p i=1 i i d,i Northern pikeminnow 20,704 2 e = ae(1 − SDA) . Other 21,492 Average α + βWt invertebrate Here, ae is the assimilation efficiency, SDA is the specific dynamic action, pi is the proportion of diet organism i in the diet, Ei is the energy density of prey organism i, and p is the proportion estimated as loge(initial length − final length)/t, where t d,i is the time interval between date of stocking and date of final of diet organism i that is digestible. Given information on diet removals in the fall. Initial and final lengths for stocked fish rep- composition, it is then possible to estimate total consumption resented the mean length at stocking and estimated mean length of a specific diet group by multiplying total consumption by in fall sampling for each cohort in each lake. For wild fish, estimates of prey-specific proportions of energy in the diet Wi Ei I we estimated what the final length would have been given the (i.e., Ct · Wt · W E , where Wi is the estimated i=1 i i estimated parameters if a wild fish had the same initial length as diet mass of diet group i). Information on prey items are shown in each stocked cohort. This allowed direct comparison between Table 3. The proportion of diet organisms indigestible (pd,i) the mean growth rate estimated across body lengths for wild was set to 0.9 for invertebrates and 0.97 for fish (Stewart et al. fish and size-specific growth rates estimated for each stocked 1983). All unidentified prey were invertebrates, so the mean size-class. energy density across invertebrates was used for this diet group. For all models, we approximated the posterior distribution The denominator of the growth efficiency formula depicts the using Markov chain Monte Carlo (MCMC) simulation with energy density of the predator (trout or pikeminnow), which we

TABLE 4. Diet organisms found in stomachs of rainbow trout and northern pikeminnow grouped into 30-mm length bins (values < 0.01 refer to trace populations).

Downloaded by [Department Of Fisheries] at 20:31 25 September 2012 Length bin (mm) Rainbow trout (n = 177) Northern pikeminnow (n = 134) Prey taxon 60 90 120 150 180 210 240 270+ 30 60 90 120 150 180+ Daphnia <0.01 <0.01 0.02 0.04 0.01 0.01 0.01 0.01 0.02 0.01 0.00 0.00 0.00 0.00 Amphipoda 0.00 0.06 0.03 0.02 0.02 0.42 0.08 0.16 0.9 0.37 0.15 0.05 0.00 0.94 Hirudinea 0.00 0.00 0.00 0.00 0.00 0.00 0.05 0.00 0.00 0.00 0.32 0.2 0.00 0.00 Trichoptera 0.01 0.02 0.03 0.08 0.01 0.01 0.04 0.01 0.07 0.1 0.03 0.06 0.02 0.06 Odonata 0.00 0.00 0.00 0.01 0.01 0.01 0.00 <0.01 0.00 0.00 0.00 0.00 0.00 0.00 Coleoptera 0.91 0.67 0.13 0.19 0.22 0.10 0.37 0.06 0.00 0.00 0.06 0.14 0.37 0.00 Diptera 0.00 0.12 0.64 0.19 0.60 0.21 0.19 0.63 0.00 0.5 0.25 0.45 0.17 0.00 Northern pikeminnow 0.00 0.00 0.00 0.00 0.00 0.00 <0.01 <0.01 0.00 0.00 0.00 0.00 0.00 0.00 Other or unidentiﬁed 0.03 0.05 0.07 0.06 0.08 0.12 0.14 0.04 0.00 0.02 0.03 0.02 0.05 0.00 USING A BIOENERGETICS MODEL FOR FOOD CONSUMPTION 951

TABLE 5. Model selection statistics for both model types and species evaluated. Bayesian information criteria (BIC) are calculated as −2loge(L) + k·loge(n), where loge(L) is the log-likelihood, k is the number of free parameters, and n is the number of observations (see text). The BIC column is the difference between the BIC of each model and the minimum BIC for that species. Values in bold italics represent the minimum BIC and therefore the selected model.

Number of Maximum posterior Species Model parameters (k) probability BIC BIC Rainbow trout Time-Invariant Consumption 11 −16,542.00 33,181.64 −7,327.61 Time-Varying Consumption 50 −12,705.1 25,854.03 0.00 Northern pikeminnow Time-Invariant Consumption 11 −2,647.62 5,397.07 −553.16 Time-Varying Consumption 45 −2,213.66 4,843.91 0.00

Cheryl Moms 0 200 400

Cath Nestor 0 200 400

Dads Moose Pasture Length (mm) (mm) Length Downloaded by [Department Of Fisheries] at 20:31 25 September 2012 0200400

Meghan Wilderness

Age-6 Age-5 Age-4 Age-3 Age-2 Age-1 0 200 400

2001 2003 2005 2007 2001 2003 2005 2007 Year

FIGURE 2. Length at age for rainbow trout in each lake across all study years. 952 VAN POORTEN ET AL.

Cheryl Moms 0 100 200 300

Cath Nestor 0 100 200 300

Dads Moose Pasture Length (mm) (mm) Length 0 100 200 300

Meghan Wilderness

Age-7 Age-13 Age-5 Age-11 Age-3 Age-9 Age-1 0 100 200 300

2001 2003 2005 2007 2001 2003 2005 2007 Year

FIGURE 3. Length at age for northern pikeminnow in each lake across all study years. Note only every other age is shown. Downloaded by [Department Of Fisheries] at 20:31 25 September 2012

assumed to increase linearly with weight for trout (α = 0.99, both trout and pikeminnow (Table 5). In both species, the Time- β = 5764; Rand et al. 1993) and to be constant for pikeminnow Varying Consumption model was found to be the most parsimo- (α = 0, β = 6703; Petersen and Ward 1999). Parameter values nious. The Time-Varying Consumption model had a BIC value are shown in Table 2. Proportional abundances of diet items of 7,328, 553 units lower than the Time-Invariant Consump- found in different size-classes of each species are shown in tion model, essentially giving full support to the more complex Table 4. Diet information was only available for the open-water Time-Varying Consumption model. season, so consumption was only estimated from June to Length at age estimated for trout and pikeminnow demon- October each year. Additionally, age-0 individuals from each strated obvious spatial and temporal variation (Figures 2, 3). species were not assessed for diet, so consumption was only In several lakes, this resulted in an increase in the maximum estimated for age-1 and older fish. sizes for trout by the final year of the study (Figure 2). Trout in other lakes, such as Cath, Meghan, and Dads lakes, which RESULTS were pikeminnow nursery lakes and not depleted of one Both the Time-Invariant Consumption model and the Time- species alone, showed constant or slight declines in length at Varying Consumption model were fit to the growth data for age over the years studied, whereas all other lakes generally USING A BIOENERGETICS MODEL FOR FOOD CONSUMPTION 953

Nursery Lakes: H Headwater Lakes: H

Lakes Lakes Cath Cheryl Dad's Mom's Meghan Nestor's Moose Pasture Wilderness 024681012 024681012

2.5 3.0 3.5 4.0 4.5 5.0 5.5 2.5 3.0 3.5 4.0 4.5 5.0 5.5 -d −1 -d −1 Net Consumption Rate (g g yr ) Net Consumption Rate (g g yr )

m Q m Q c Probability Density Probability 0 102030405060 05101520 024681012

0.3 0.5 0.7 1.8 1.9 2.0 2.1 2.2 3.0 3.4 3.8 -n −1 Standard Metabolic Rate (g g yr ) Metabolic Q10 Consumption Q10

FIGURE 4. Posterior probability densities for mean (over years) ﬁrst-year net consumption rates (Hl,1) in all eight lakes, standard metabolic rate (m), consumption and metabolic Q10 (Qm, Qc) parameters for rainbow trout. Net consumption rates are separated into those from populations in nursery lakes and those from headwater

Downloaded by [Department Of Fisheries] at 20:31 25 September 2012 lakes. Prior and posterior probability distributions for Qm are shown as dashed and solid lines, respectively.

showed slight increases in length at age in the same years. in headwater lakes than in pikeminnow nursery lakes as might These results indicate that the model findings reflect our be expected if the two species compete for some shared food biological understanding of the system. In contrast pikeminnow items. There was little information in the data to update the growth rates were much more consistent over time resulting prior probability distribution of the Q10 for metabolism (Qm); in similar lengths at age for different year-classes (Figure 3). hence, the posterior distribution was similar to the prior distribu- However, the model detected varied growth rates across lakes tion. The Q10 for consumption (Qc) was much higher than Qm, for pikeminnow, with populations in some lakes exhibiting indicating that consumption increases faster than metabolism markedly slower growth (e.g., Cath, Meghan, and Wilderness as temperature increases resulting in increased growth rates at lakes) than others (e.g., Nestor Lake). higher temperatures. Net consumption rates (Hlake) for trout at the start of the Net consumption rates (Hlake) for pikeminnow do not appear study were quite variable, ranging from 2.9 to 4.8 g·g−d·year−1 to be influenced by the size-structure of the population in each (Figure 4). Trout net consumption rates were typically lower lake, as evidenced by similar ranges in nursery and headwater 954 VAN POORTEN ET AL.

Nursery Lakes: H Headwater Lakes: H

Lakes Lakes Cath Cheryl Dad's Mom's Meghan Nestor's Moose Pasture Wilderness 024681012 024681012

1.0 1.5 2.0 2.5 3.0 3.5 4.0 4.5 1.01.52.02.53.03.54.04.5 -d −1 -d −1 Net Consumption Rate (g g yr ) Net Consumption Rate (g g yr )

m Q m Q c Probability Density Probability 0 5 10 15 20 012345 02468

0.6 1.0 1.4 1.8 1.9 2.0 2.1 2.2 3.0 3.4 3.8 -n −1 Standard Metabolic Rate (g g yr ) Metabolic Q10 Consumption Q10

FIGURE 5. Posterior probability densities for mean (over years) ﬁrst-year net consumption rates (Hl,1) in all eight lakes, standard metabolic rate (m), consumption and metabolic Q10 (Qm, Qc) parameters for northern pikeminnow. Net consumption rates are separated into those from populations in nursery lakes and those from

Downloaded by [Department Of Fisheries] at 20:31 25 September 2012 headwater lakes. Prior and posterior probability distributions for Qm are shown as dashed and solid lines, respectively.

lakes (Figure 5). The exception is pikeminnow in Nestor Lake, net consumption rates are relatively insensitive to pikeminnow which had a net consumption twice that in most other lakes. As removal experiments (Figure 6, top panels) since all three lakes with trout, there was little information in the data to update the follow the same annual pattern despite different treatment years. prior probability distribution of the metabolic Q10 (Qm), but net While this may seem like the model detected a general trend for consumption does increase with temperature at a much faster trout across lakes, there is a different pattern in the control lake rate than metabolism as shown by the difference between Qm (Dads; Figure 6, bottom-left panel). Similarly, no clear pattern and Qc. in trout net consumption rate appeared in lakes where adult The model was able to detect differences in net consumption trout were removed in 2001 (Figure 6, center-left panel). Over- rate both within populations over time and among populations all, the model demonstrated that trout net consumption rate, and (Figure 6), as reﬂected in differences in growth indicated in Fig- therefore growth, showed considerable variation across popu- ures 2 and 3. The model shows that both trout and pikeminnow lations and years that appeared to be independent of density USING A BIOENERGETICS MODEL FOR FOOD CONSUMPTION 955

Rainbow Trout Northern Pikeminnow Northern Pikeminnow Removal

Cheryl Lake Mom's Lake Moose Pasture Lake 2468 ) ) 1 − 0246 yr -d Rainbow Trout Removal

Nestor's Lake Wilderness Lake 2468 0246

Control

Net Consumption Rate (g g (g Rate Consumption Net Dad's Lake Cath Lake Meghan Lake 2468 0246

2001 2002 2003 2004 2005 2006 2007 2008 2001 2002 2003 2004 2005 2006 2007 2008 Year

FIGURE 6. Net consumption rates for rainbow trout and northern pikeminnow across all 8 years of the study. Lakes are grouped by treatment type: northern pikeminnow removal in 2001 (Cheryl and Moms lakes) and 2002 (Moose Pasture Lake); rainbow trout removal in 2001 (Nestors and Wilderness lakes) and

Downloaded by [Department Of Fisheries] at 20:31 25 September 2012 control lakes (Dads, Cath, and Meghan lakes). Years where lakes are not represented are years in which lakes were not studied or sample size was fewer than 10 individuals. Vertical bars represent 95% credible limits; some credible limits are obscured by the size of the marker.

manipulations of either species. The model showed that no consistent directions in differences across years, lakes, and pikeminnow demonstrated less variation than trout in net con- size-classes. sumption rate and growth over time (Figure 6, right panels). The model was also used to estimate mean daily consump- Changes in net consumption rates and growth were determined tion, which varied across years and lakes for both species to be more consistent across populations than in trout as well. (Figures 8, 9). Lake–year interactions in growth were common, The bioenergetics model estimated similar interannual trends but did not necessarily reflect changes in density of either species in instantaneous growth between stocked and wild trout owing to partial depletions. For example, growth increased in (Figure 7). This finding supports the bioenergetics model esti- trout across most cohorts in 2002, despite wide-ranging reduc- mates of lake- and year-specific differences in net consumption tions in abundance of either species in several lakes, but not in rates for wild trout since the two data sets were estimated in- others (Figure 8). The model predicted definite changes in con- dependently. Although absolute instantaneous growth estimates sumption within a lake across cohorts for trout, but year–cohort sometimes differed between wild and hatchery trout, there were interactions in consumption rates were rare. 956 VAN POORTEN ET AL. Cheryl Moms ] 051015 2 − Cath Nestor ) [x 10 [x ) 1 − 0 5 10 15 Dads Moose Pasture 051015 Meghan Wilderness Instantaneous Growth(d 051015 Age-0 Small Medium Large Age-0 Small Medium Large Age-1 Age-1 Age-1 Age-1 Age-1 Age-1

FIGURE 7. Instantaneous growth [log10(initial length − final length)/t] for four size-classes of stocked fish over 3 years in each of the study lakes. Squares, circles and triangles represent size-classes stocked from 2006 to 2008, respectively. Filled symbols represent growth of stocked hatchery trout, while open circles Downloaded by [Department Of Fisheries] at 20:31 25 September 2012 represent the expected growth of wild fish with identical initial size and time at large, given estimated wild trout parameters. Age-0 rainbow trout were not stocked in 2006, nor were large age-1 trout stocked in all lakes in 2006. Vertical bars represent 95% credible limits; some credible limits are obscured by the size of the marker.

DISCUSSION ing additional parameters to explain variation in consumption The new parameterization of the general bioenergetics model over time and among populations will lead to an increase in was able to detect differences in growth and consumption parameter uncertainty (Hilborn and Walters 1992), in situations for trout and pikeminnow within and between populations in where finer-scale understanding of growth is needed, the addi- spatially distinct lakes that experienced similar environmental tional flexibility may be warranted. However, our comparison conditions (Taylor 2006). The direction and magnitude of the with the simpler Time-Invariant Consumption model demon- estimated differences in growth within and between popula- strated that the Time-Variable Consumption model was much tions was corroborated using hatchery trout stocked into the more parsimonious and the use of additional explanatory pa- same lakes in several years and at various sizes. While includ- rameters was warranted. Overall, we found the flexibility of the USING A BIOENERGETICS MODEL FOR FOOD CONSUMPTION 957

Cheryl Moms ] 3 − 00.511.5

Cath Nestor ) [x10 ) 1 − day 1 − g . 00.511.5

Dads Moose Pasture 00.511.5

Meghan Wilderness

Age-1 Age-4 Age-2 Age-5 Age-3 Age-6 Mean Daily Consumption (g (g Consumption Mean Daily 00.511.5

2001 2003 2005 2007 2001 2003 2005 2007 Year

FIGURE 8. Mean daily consumption rates for wild rainbow trout in each lake studied across all study years. Consumption for trout of ages 1 to 6 years are shown. Downloaded by [Department Of Fisheries] at 20:31 25 September 2012

general bioenergetics model to be significantly improved as a timates of metabolic rates using the same general bioenergetics result of including the year- and lake-specific multipliers on net model as used in the present study suggested that groups of white consumption rate. sturgeon Acipenser transmontanus had different metabolic rates We assume metabolism-related parameters are shared among despite the fish being found in the same contiguous span of the populations within a species. Many authors warn against the Columbia River (van Poorten and McAdam 2010). One of the concept of “parameter borrowing” (Ney 1993), where func- benefits of the Bayesian analysis used to estimate parameters in tions and parameters developed from laboratory studies on one the bioenergetics model is the estimation of parameter uncer- species or population are used for predicting the growth or tainty in parameters such as metabolism. Estimating metabolism consumption of another. Recent empirical work suggests that as a population-independent parameter means that the uncer- metabolic rates and processes experienced by different popu- tainty within and among populations is combined, thereby per- lation within species can vary significantly, and this will result mitting potential differences between populations to be exposed. in varied estimates of growth and consumption in bioenergetics This means that parameters are not borrowed as in the Wisconsin models (Munch and Conover 2002; Tyler and Bolduc 2008). Es- model (Ney 1993; Tyler and Bolduc 2008), thereby eliminating 958 VAN POORTEN ET AL.

Cheryl Moms ] 3 − 00.511.52

Cath Nestor ) [x10 ) 1 − day 1 − g . 0 0.5 1 1.5 2

Dads Moose Pa sture 00.511.52

Meghan Wilderness

Age-1 Age-4 Age-2 Age-5 Age-3 Age-6 Mean Daily Consumption (g (g Consumption Mean Daily 00.511.52

2001 2003 2005 2007 2001 2003 2005 2007 Year

FIGURE 9. Mean daily consumption rates for northern pikeminnow in each lake studied across all study years. Consumption for pikeminnow of ages 1 to 6 years are shown. Downloaded by [Department Of Fisheries] at 20:31 25 September 2012

the misgivings of parameter-borrowing common in analysis of gape size (Mittelbach and Persson 1998) and gill raker spac- most bioenergetics models. ing (Keeley and Grant 2001) increases, allowing for larger and The general bioenergetics model was also used to estimate potentially more energetically rewarding prey to be consumed growth rates of different size-classes of stocked trout to deter- (Madenjian et al. 1998; Post 2003). As fish switch to larger diet mine if growth rates of differently sized fish deviated from the organisms, their anaerobic activity may drop, probably owing population mean estimated for the wild fish. Overall growth to a decrease in burst swimming accompanying fewer foraging estimates between stocked and wild fish were similar, both attempts (Pazzia et al. 2002; Sherwood et al. 2002). The result validating the year-specific net consumption rate estimates and is more energy gained in fewer predation attempts, which may indicating that little ontogenetic variation in growth occurs result in little ontogenetic variation in growth, as appeared to within trout in these lakes. While we attempted to account for occur in the present study. Alternately, if fish largely maintain ontogenetic changes in diet through separation of diet analysis their feeding activity, ontogenetic diet changes could result in into discrete size-classes, this does not account for changes in net much higher energy input, resulting in relatively rapid increases consumption rates as fish grow. Specifically, as fish grow, their in growth in one or more periods throughout the lifetime of a USING A BIOENERGETICS MODEL FOR FOOD CONSUMPTION 959

fish (Osenberg et al. 1988; Madenjian et al. 1998). Accounting 2008) such as relative abundance of primary food organisms, for ontogenetic diet shifts in growth can conceivably be done stream flows, or relative competition or predation risk. While using the generalized bioenergetics model in the same way as an infinite number of possibilities exist for increasing flexibility is done when modeling a multiphased growth model (He and to the general bioenergetics model, it is always necessary to Stewart 2002): estimate parameters for two growth phases and balance the need for increased model precision and parameter estimate the rate and size at which the diet shift occurs. This uncertainty (Hilborn and Walters 1992). Further, it is inherently approach would require sufficient data for most or all sizes of difficult and time consuming to attain the high number of re- fish in the population to ensure that parameters are estimable. captures necessary to estimate growth deviations in the general Again, this would add extra parameters to the model and should bioenergetics model. If sufficient data do not exist, it may not only be attempted in situations where obvious ontogenetic shifts be possible to estimate growth using the general bioenergetics in the population are suspected and when the need to account model as we have done here. for this change in growth rate is warranted. Our stocked trout of various size-classes indicate that no consistent shift in diet ACKNOWLEDGMENTS occurred in the study populations. This project was initiated through conversations with Eric Shifts in consumption rates may have implications for other Parkinson of the British Columbia Ministry of Environment. organisms in the ecosystem. With this in mind, a useful applica- Data collection and sample analysis were performed by many tion for our modification of the general bioenergetics model will field and laboratory assistants, and their dedication and hard be to include growth and consumption estimates in an ecosys- work made this project possible. Data on rainbow trout be- tem model where ecosystem implications can be modeled and fore 2005 was collected and generously contributed by David estimated directly. Several bioenergetics models have been inte- O’Brien as part of his Ph.D. research. Support for B.T.v.P. grated into foraging or ecosystem models to examine the inter- was provided by a Post Graduate Scholarship from the Nat- action of several trophic levels (Nibbelink and Carpenter 1998; ural Science and Engineering Research Council (NSERC) of Hayes et al. 2000; Aydin et al. 2005; Rose et al. 2008; Kishi Canada. Field and laboratory work were primarily funded et al. 2011). These models often use the Wisconsin model as through NSERC Discovery Grants to C.J.W., and grants from the framework with which fish bioenergetics are estimated, pri- both the British Columbia Small Lakes Management Commit- marily owing to the flexibility of the model. The present study tee to B.T.v.P. and Eric Parkinson and the Habitat Conservation shows that substantial growth information can be estimated di- Trust Foundation to Eric Parkinson. This manuscript was im- rectly from field data rather than predicted based on empirical proved by thoughtful suggestions and comments by two anony- relationships derived using laboratory studies. This information mous reviewers and the associate editor. AD model Builder code would be of use for ecosystem models, thereby permitting a for the Time-Varying Consumption model is available from the direct evaluation of how changes in growth rate and species tar- senior author. geting will affect the community in general (Werner and Gilliam 1984). For example, with size- or age-structured abundance estimates for the trout and pikeminnow in the study lakes, it would REFERENCES be possible to estimate total consumption of both species, which Askey, P. J., J. R. Post, E. A. Parkinson, E. Rivot, A. J. Paul, and P. A. Biro. could be used to help estimate mortality for all other groups of 2007. Estimation of gillnet efficiency and selectivity across multiple sampling species in the lakes, in a way similar to that of the EcoPath with units: a hierarchical Bayesian analysis using mark-recapture data. Fisheries Research 83:162–174. EcoSim model framework (Walters et al. 2000). The merger Aydin, K. Y., G. A. McFarlane, J. R. King, B. A. Megrey, and K. W. Myers. of bioenergetics and ecosystem models is still in its infancy, 2005. Linking oceanic food webs to coastal production and growth rates of

Downloaded by [Department Of Fisheries] at 20:31 25 September 2012 largely owing to the large number of parameters and substantial Pacific salmon (Oncorhynchus spp.), using models on three scales. Deep-Sea parameter uncertainty inherent in both approaches individually. Research, Part II Topical Studies in Oceanography 52:757–780. We hope that the model described here will help to promote Beauchamp, D. A., C. J. Sergeant, M. M. Mazur, J. M. Scheuerell, D. E. Schindler, M. D. Scheuerell, K. L. Fresh, D. E. Seiler, and T. P. Quinn. further integration as this avenue of research develops. 2004. Spatial–temporal dynamics of early feeding demand and food supply Our modification to the general bioenergetics model infor sockeye salmon fry in Lake Washington. Transactions of the American creases the utility of a field-based bioenergetics model for pre- Fisheries Society 133:1014–1032. dicting changes in growth and consumption over time. Like the Brandt, S. B., and K. J. Hartman. 1993. Innovative approaches with bioenergetics Wisconsin model, our modification makes it possible to estimate models: future applications to fish ecology and management. Transactions of the American Fisheries Society 122:731–735. how growth might have changed over a period of system per- Burnham, K. P., and R. S. Anderson. 2002. Model selection and multimodel turbations. It is possible to apply additional parameters to the inference: a practical information-theoretic approach, 2nd edition. Springer- bioenergetics model to predict cohort-specific effects on life- Verlag, New York. time growth (He and Bence 2007) or variation in consumption Chipps, S. R., and D. H. Wahl. 2008. Bioenergetics modeling in the 21st century: or metabolism to account for ontogenetic shifts in these rates reviewing new insights and revisiting old constraints. Transactions of the American Fisheries Society 137:298–313. (Quince et al. 2008). Alternately, net consumption rate (or some Clark, A., and N. M. Johnston. 1999. Scaling of metabolic rate with body mass other) parameters can be estimated with covariates (Kimura and temperature in teleost fish. Journal of Animal Ecology 68:893–995. 960 VAN POORTEN ET AL.

Cummins, K. W., and J. C. Wuycheck. 1971. Caloric equivalents for investi- Matuszek, J. E., B. J. Shuter, and J. M. Casselman. 1990. Changes in gations in ecological energetics. Internationale Vereinigung fur¨ theoretische lake trout growth and abundance after introduction of cisco into Lake und angewandte Limnologie, Mitteilungen 18. Opeongo, Ontario. Transactions of the American Fisheries Society 119: Essington, T. E., J. F. Kitchell, and C. J. Walters. 2001. The von Bertalanffy 718–729. growth function, bioenergetics, and the consumption rates of fish. Canadian Mills, E. L., and J. L. Forney. 1981. Energetics, food consumption, and growth of Journal of Fisheries and Aquatic Sciences 58:2129–2138. young yellow perch in Oneida Lake, New York. Transactions of the American Fournier, D. A., H. J. Skaug, J. Ancheta, J. Ianelli, A. Magnusson, M. N. Fisheries Society 110:479–488. Maunder, A. Nielsen, and J. Sibert. 2012. AD model builder: using auto- Mittelbach, G. G., and L. Persson. 1998. The ontogeny of piscivory and its matic differentiation for statistical inference of highly parameterized complex ecological consequences. Canadian Journal of Fisheries and Aquatic Sciences nonlinear models. Optimization Methods and Software 27:233–249. Avail- 55:1454–1465. able: admb-project.org. (March 2012). Munch, S. B., and D. O. Conover. 2002. Accounting for local physiological Fraser, D. F., and J. F. Gilliam. 1992. Nonlethal impacts of predator invasion: adaptation in bioenergetic models: testing hypotheses for growth rate evo- facultative suppression of growth and reproduction. Ecology 73:959–970. lution by virtual transplant experiments. Canadian Journal of Fisheries and Gelman, A., J. B. Carlin, H. S. Stern, and D. B. Rubin. 1995. Bayesian data Aquatic Sciences 59:393–403. analysis. Chapman and Hall, New York. Myers, J. T., M. L. Jones, J. D. Stockwell, and D. L. Yule. 2009. Reassess- Hansen, M. J., D. Boisclair, S. B. Brandt, S. W. Hewett, J. F. Kitchell, ment of the predatory effects of rainbow smelt on ciscoes in Lake Superior. M. C. Lucas, and J. J. Ney. 1993. Applications of bioenergetics models Transactions of the American Fisheries Society 138:1352–1368. to fish ecology and management: where do we go from here? Transactions of Ney, J. J. 1990. Trophic economics in fisheries: assessment of demand-supply the American Fisheries Society 122:1019–1030. relationships between predators and prey. Reviews in Aquatic Sciences 2:55– Hanson, P. C., T. B. Johnson, D. E. Schindler, and J. F. Kitchell. 1997. Fish 81. bioenergetics 3.0. University of Wisconsin, Sea Grant Institute, Technical Ney, J. J. 1993. Bioenergetics modeling today: growing pains on the cutting Report WIS-CU-T-97-001, Madison. edge. Transactions of the American Fisheries Society 122:736–748. Hartman, K. J., and J. F. Kitchell. 2008. Bioenergetics modeling: progress Nibbelink, N. P., and S. R. Carpenter. 1998. Interlake variation in growth and since the 1992 symposium. Transactions of the American Fisheries Society size structure of bluegill (Lepomis machrochirus): inverse analysis of an 137:216–223. individual-based model. Canadian Journal of Fisheries and Aquatic Sciences Hayes, J. W., J. D. Stark, and K. A. Shearer. 2000. Development and test of 55:387–396. a whole-lifetime foraging and bioenergetics growth model for drift-feeding O’Brien, D. S. 2009. The effect of harvesting on size structured predatory brown trout. Transactions of the American Fisheries Society 129:315–332. and competitive interactions between rainbow trout (Oncorhynchus mykiss) He, J. X., and J. R. Bence. 2007. Modeling annual growth variation using a and northern pikeminnow (Ptychocheilus oregonensis). Doctoral dissertation. hierarchical Bayesian approach and the von Bertalanffy growth function, McGill University, Montreal. with application to lake trout in southern Lake Huron. Transactions of the Osenberg, C. W., E. E. Werner, G. G. Mittelbach, and D. J. Hall. 1988. Growth American Fisheries Society 136:318–330. patterns in bluegill (Lepomis macrochirus) and pumpkinseed (L. gibbosus) He, J. X., and D. J. Stewart. 2002. A stage-explicit expression of the von sunfish: environmental variation and the importance of ontogenetic niche Bertalanffy growth model for understanding age at first reproduction of Great shifts. Canadian Journal of Fisheries and Aquatic Sciences 45:17–26. Lakes fishes. Canadian Journal of Fisheries and Aquatic Sciences 59:250– Pazzia, I., M. Trudel, M. S. Ridgway, and J. B. Rasmussen. 2002. Influence of 261. food web structure on the growth and bioenergetics of lake trout (Salvelinus Hilborn, R., and C. J. Walters. 1992. Quantitative fisheries stock assessment: namaycush). Canadian Journal of Fisheries and Aquatic Sciences 59:1593– choice, dynamics and uncertainty. Kluwer Academic Publishers, Norwell, 1605. Massachusetts. Petersen, J. H., D. L. DeAngelis, and C. P. Paukert. 2008. An overview of Irwin, B. J., D. R. DeVries, and R. A. Wright. 2003. Evaluating the potential for methods for developing bioenergetic and life history models for rare and predatory control of gizzard shad by largemouth bass in small impoundments: endangered species. Transactions of the American Fisheries Society 137:244– a bioenergetics approach. Transactions of the American Fisheries Society 253. 132:913–924. Petersen, J. H., and D. L. Ward. 1999. Development and corroboration of a bioen- Jobling, M. 1994. Fish bioenergetics. Chapman and Hall, London. ergetics model for northern pikeminnow feeding on juvenile salmonids in the Jobling, M., and B. M. Baardvik. 1994. The influence of environmental manipu- Columbia River. Transactions of the American Fisheries Society 128:784– lations on inter- and intra-individual variation in food acquisition and growth 801. performance of Arctic charr, Salvelinus alpinus. Journal of Fish Biology Post, D. M. 2003. Individual variation in the timing of ontogenetic niche shifts

Downloaded by [Department Of Fisheries] at 20:31 25 September 2012 44:1069–1087. in largemouth bass. Ecology 84:1298–1310. Johnson, R. L., S. C. Blumenshine, and S. M. Coghlan. 2006. A bioenergetic Post, J. R., E. A. Parkinson, and N. T. Johnston. 1999. Density-dependent analysis of factors limiting brown trout growth in an Ozark tailwater river. processes in structured fish populations: interaction strengths in whole-lake Environmental Biology of Fishes 77:121–132. experiments. Ecological Monographs 69:155–175. Keeley, E. R., and J. W. A. Grant. 2001. Prey size of salmonid fishes in Quince, C., B. J. Shuter, P. A. Abrams, and N. P. Lester. 2008. Biphasic streams, lakes and oceans. Canadian Journal of Fisheries and Aquatic Sci- growth in fish II: empirical assessment. Journal of Theoretical Biology 254: ences 58:1122–1132. 207–214. Kimura, D. K. 2008. Extending the von Bertalanffy growth model using ex- Rand, P. S., D. J. Stewart, P. W. Seelbach, M. L. Jones, and L. R. Wedge. 1993. planatory variables. Canadian Journal of Fisheries and Aquatic Sciences Modeling steelhead population energetics in Lakes Michigan and Ontario. 65:1879–1891. Transactions of the American Fisheries Society 122:977–1001. Kishi, M. J., S. Ito, B. A. Megrey, K. A. Rose, and F. E. Werner. 2011. A review Rose, K. A., B. A. Megrey, D. Hay, F. Werner, and J. Schweigert. 2008. of the NEMURO and NEMURO.FISH models and their application to marine Climate regime effects on Pacific herring growth using coupled nutrient– ecosystem investigations. Journal of Oceanography 67:3–16. phytoplankton–zooplankton and bioenergetics models. Transactions of the Madenjian, C. P., T. J. DeSorcie, and R. M. Stedman. 1998. Ontogenetic and American Fisheries Society 137:278–297. spatial patterns in diet and growth of lake trout in Lake Michigan. Transactions Schwarz, G. 1978. Estimating the dimension of a model. Annals of Statistics of the American Fisheries Society 127:236–252. 6:461–464. USING A BIOENERGETICS MODEL FOR FOOD CONSUMPTION 961

Sherwood, G. D., I. Pazzia, A. Moeser, A. Hontela, and J. B. Rasmussen. 2002. Tyler, J. A., and M. B. Bolduc. 2008. Individual variation in bioenergetic rates of Shifting gears: enzymatic evidence for the energetic advantage of switching young-of-year rainbow trout. Transactions of the American Fisheries Society diet in wild-living fish. Canadian Journal of Fisheries and Aquatic Sciences 137:314–323. 59:229–241. van Poorten, B. T., and S. O. McAdam. 2010. Estimating differences in growth Stewart, D. J., D. Weininger, D. V. Rottiers, and T. A. Edsall. 1983. An ener- and metabolism in two spatially segregated groups of Columbia River white getics model for lake trout, Salvelinus namaycush: application to the Lake sturgeon using a field-based bioenergetics model. Open Fish Science Journal Michigan population. Canadian Journal of Fisheries and Aquatic Sciences 40: 3:132–141. 681–698. van Poorten, B. T., and C. J. Walters. 2010. Estimation of bioenergetics param- Taylor, N. 2006. Growth and survival responses to experimental fishing: models, eters for rainbow trout (Oncorhynchus mykiss) using capture-recapture data experiments and lessons from the pygmy pikeminnow of south central British with comparisons to estimates from a laboratory-based model. Open Fish Columbia. Doctoral dissertation. University of British Columbia, Vancouver. Science Journal 3:69–79. Taylor, N., and C. Walters. 2010. Estimation of bioenergetics parameters for a Walters, C., and T. Essington. 2010. Recovery of bioenergetics parameters stunted northern pikeminnow population of south central British Columbia. from information on growth: overview of an approach based on statisti- Open Fish Science Journal 3:110–121. cal analysis of tagging and size-at-age data. Open Fish Science Journal Taylor, N. G., C. J. Walters, and S. J. D. Martell. 2005. A new likelihood for si- 3:52–68. multaneously estimating von Bertalanffy growth parameters, gear selectivity, Walters, C., D. Pauly, V.Christensen, and J. F. Kitchell. 2000. Representing den- and natural and fishing mortality. Canadian Journal of Fisheries and Aquatic sity dependent consequences of life history strategies in aquatic ecosystems: Sciences 62:215–223. EcoSim II. Ecosystems 3:70–83. Tuomikoski, J. E., P. J. Rudershausen, J. A. Buckel, and J. E. Hightower. 2008. Werner, E. E., and J. F. Gilliam. 1984. The ontogenetic niche and species Effects of age-1 striped bass predation on juvenile fish in western Albemarle interactions in size-structure populations. Annual Review of Ecology and Sound. Transactions of the American Fisheries Society 137:324–339. Systematics 15:393–425. Downloaded by [Department Of Fisheries] at 20:31 25 September 2012 This article was downloaded by: [Department Of Fisheries] On: 25 September 2012, At: 20:33 Publisher: Taylor & Francis Informa Ltd Registered in England and Wales Registered Number: 1072954 Registered office: Mortimer House, 37-41 Mortimer Street, London W1T 3JH, UK

Transactions of the American Fisheries Society Publication details, including instructions for authors and subscription information: http://www.tandfonline.com/loi/utaf20 Geology as a Structuring Mechanism of Stream Fish Communities Margaret R. Neff a & Donald A. Jackson a b a Department of Ecology and Evolutionary Biology, University of Toronto, 25 Harbord Street, Toronto, Ontario, M5S 3G5, Canada b Centre for Environment, University of Toronto, 33 Willcocks Street, Toronto, Ontario, M5S 3B2, Canada

Version of record first published: 20 Jun 2012.

To cite this article: Margaret R. Neff & Donald A. Jackson (2012): Geology as a Structuring Mechanism of Stream Fish Communities, Transactions of the American Fisheries Society, 141:4, 962-974 To link to this article: http://dx.doi.org/10.1080/00028487.2012.676591

PLEASE SCROLL DOWN FOR ARTICLE

Full terms and conditions of use: http://www.tandfonline.com/page/terms-and-conditions 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. The publisher does not give any warranty express or implied or make any representation that the contents will be complete or accurate or up to date. The accuracy of any instructions, formulae, and drug doses should be independently verified with primary sources. The publisher shall not be liable for any loss, actions, claims, proceedings, demand, or costs or damages whatsoever or howsoever caused arising directly or indirectly in connection with or arising out of the use of this material. Transactions of the American Fisheries Society 141:962–974, 2012 C American Fisheries Society 2012 ISSN: 0002-8487 print / 1548-8659 online DOI: 10.1080/00028487.2012.676591

ARTICLE

Geology as a Structuring Mechanism of Stream Fish Communities

Margaret R. Neff* Department of Ecology and Evolutionary Biology, University of Toronto, 25 Harbord Street, Toronto, Ontario M5S 3G5, Canada Donald A. Jackson Department of Ecology and Evolutionary Biology, University of Toronto, 25 Harbord Street, Toronto, Ontario M5S 3G5, Canada, and Centre for Environment, University of Toronto, 33 Willcocks Street, Toronto, Ontario M5S 3B2, Canada

Abstract We examined the influence of the Precambrian Shield, a broad-scale geological feature of North America, on stream abiotic conditions and on the associated fish communities relative to an adjacent series of sites located on sedimentary geology (i.e., “off-Shield” streams) in south-central Ontario, Canada. Constrained and unconstrained multivariate analyses were used to quantify relationships in fish species composition, abiotic variables, and fish species’ functional traits. The results showed that for low-order lotic systems, streams located on the Precambrian Shield had fish communities that were distinctly different from those of off-Shield streams. Lotic fish communities on the Shield were associated with higher velocity, increased amounts of instream woody debris, higher dissolved oxygen, and shrub–marsh landscapes. In contrast, off-Shield communities were characterized by higher conductivity, alkalinity, water temperature, and turbidity and were frequently surrounded by meadow and cultivated or pastured land. In general, off-Shield sites were more species rich and were associated with species preferring a wide variety of habitats, whereas Shield sites were associated with a subset of the regional species pool that had more specific habitat associations. Overall, our results show that these sharp geological discontinuities result in distinct abiotic environments and fish communities between Shield and off-Shield low-order lotic systems within a small geographic range.

Community ecologists have long sought to determine the these attributes might also include water velocity, stream mor-

Downloaded by [Department Of Fisheries] at 20:33 25 September 2012 key factors that are important in influencing species richness, phology, and structural complexity of the environment. composition, abundance, and other metrics describing biolog- Determining which of these factors is most important in ical communities (e.g., McAuliffe 1984; Marsh-Matthews and structuring a particular community is no easy task, but in gen- Matthews 2000; Condit et al. 2002). In many cases, the impor- eral the literature suggests that species composition and richness tance of one factor versus another depends on the system and the are thought to be the product of a hierarchy of factors over a communities in question. For fish communities, a variety of fac- range of spatial scales. For example, regional patterns of fish tors concerning both the abiotic and biotic environment has been species distribution may be attributed to factors such as climate, identified as being particularly relevant. As reviewed by Jackson dispersal barriers, and historical biogeography (e.g., Ricklefs et al. (2001), these factors can include predation, competition 1987; Jackson and Harvey 1989; Minns 1989; Mandrak 1995), via resource partitioning, climate, dissolved oxygen (DO) lev- whereas local patterns may reflect habitat diversity, competition, els, acidity, and physical attributes of the habitat. For lotic fishes, and predation (e.g., Layman and Winemiller 2004; Buisson et al.

*Corresponding author: [email protected] Received September 29, 2011; accepted March 7, 2012 Published online June 20, 2012

962 GEOLOGY AS A STRUCTURING MECHANISM 963

2008). This hierarchical organization of processes is often de- (Chapman and Putnam 1984). This geological feature represents scribed as a series of filters through which the total species pool the surface geology for much of Canada, with extensions into the passes to reach a local assemblage (Smith and Powell 1971; northern United States (i.e., the Midwest, New York, and New Tonn 1990; Poff 1997; Jackson et al. 2001). Ricklefs (1987) England; NRC 2004). In south-central Ontario, the Shield geol- stressed the importance of examining multiple scales or consid- ogy stands in sharp contrast to the younger, primarily sedimen- ering both regional and local factors. tary bedrock of southern Ontario. The inherent differences in Numerous studies have demonstrated that regional factors weathering characteristics of Shield geology compared with the are integral in explaining various aspects of fish commu- limestone-dominated bedrock in “off-Shield” areas of Ontario nity structure at a local scale. For example, Angermeier and have a particularly strong influence on aquatic systems. Shield Winston (1998) found that regional fish diversity consistently streams and lakes are known to have extremely low conductivity predicted local diversity, whereas Tonn et al. (1990) found that and a propensity for acidification due to the poor buffering capa- for two systems with distinct regional and historical influences bilities of Shield bedrock (Jackson and Harvey 1989), and lotic but similar local characteristics, species richness was similar systems may further show different physical development, par- but assemblage patterns differed. More recently, studies have ticularly with a general lack of classical pool–riffle morphology incorporated aspects of landscape and watershed-level factors due to the erosion-resistant bedrock. By focusing on a smaller and analysis of fish functional traits to further enhance our abil- geographic area containing both Shield and off-Shield regions, ity describe aspects of fish assemblages. Wang et al. (2003) the influence of this strong environmental discontinuity on fish found that reach-scale characteristics are influenced by water- communities can be effectively examined in conjunction with shed characteristics, which in turn affect fish distribution, abun- the local environmental conditions. dance, and other community attributes. Hoeinghaus et al. (2007) Although the influence of the Canadian Shield on fish com- determined that taxonomic-based community patterns were best munity structure is relatively unstudied, there is also a lack explained by species’ geographic distributions, whereas func- of insight into the functioning of lotic ecosystems in this re- tional patterns in fish communities were explained by an equal gion. Historically, numerous lotic studies have been conducted influence of both local and regional factors. in the United States, but within the province of Ontario such Several studies have attempted to explain large-scale patterns studies have been primarily confined to off-Shield southern and in fish communities in the province of Ontario, Canada, with the southwestern areas, where streams and rivers are more likely to overarching conclusion that fish species distributions are largely conform to classical ideas concerning geomorphology and wa- shaped by historical and environmental processes. Jackson and ter chemistry (e.g., Bowlby and Roff 1986; Kilgour and Barton Harvey (1989) showed that regional patterns among lentic fish 1999). Many of the variables that are thought to be important assemblages reflect large-scale factors such as historical post- in fish community structure (as outlined by Jackson et al. 2001) glacial colonization routes, whereas variation within regions is may be directly influenced by Shield geology. For example, likely influenced by small-scale morphological and chemical Shield systems may be unable to support many acid-sensitive characteristics of individual lakes. Mandrak (1995) expanded species due to these systems’ inability to buffer against anthro- this examination further, finding broad-scale gradients in lentic pogenically generated acid precipitation inputs, which are of fish species richness and concluding that patterns in species particular concern in this area (e.g., Hall and Ide 1987; Kelso richness are directly related to postglacial colonization and cli- et al. 1990; Bowman et al. 2006). mate once spatial autocorrelation is removed from the analysis. Here, we address two main objectives. First, we identify the Hinch et al. (1991) highlighted certain species’ distributions that broad geographical patterns of fish community composition in are likely influenced by (1) the location and maximum extent low-order lotic systems, with particular attention to the differ-

Downloaded by [Department Of Fisheries] at 20:33 25 September 2012 of glacial Lake Algonquin, a water body composed of glacial ences attributed to the Canadian Shield. While previous work melt waters that historically covered the area of present-day has shown that historical colonization routes and climate are Lake Huron and surrounding lands; and (2) the Kirkfield Outlet, important factors determining fish species richness and assem- which connected Lake Algonquin to Lake Ontario and provided blage patterns in Ontario (Jackson and Harvey 1989; Hinch et al. an important dispersal route for coolwater and coldwater fishes 1991; Mandrak 1995), we propose that the conditions imposed into the new melt waters that were left behind by the retreating by the Shield act as a broad regional filter, and we predict that glaciers in the uplands of central Ontario. the abiotic environment created by Shield geology will directly Although these studies provide valuable insight into regional- influence fish community composition such that nearby Shield scale patterns in fish species distributions, most studies have and off-Shield lotic systems will have distinct fish communities. focused primarily on lentic systems and few have examined pat- By incorporating previous knowledge of historical colonization terns in relation to an important landscape feature at the same pathways, we will have a clearer understanding of how fish regional scale—the Canadian Precambrian Shield (hereafter, community composition is determined at the regional scale in “the Shield”). The Shield is a broad area characterized by Pre- Ontario. cambrian metamorphic and igneous bedrock that lies at or close The second objective is to identify the particular physical to the surface and is covered by a thin layer of nutrient-poor soil and chemical features that characterize Shield and off-Shield 964 NEFF AND JACKSON

systems in south-central Ontario and to relate these factors to of Lake Huron, climatic gradients exist in an east–west pat- characteristics of fish community composition. Aspects of water tern as well as in a north–south pattern). From the full set of chemistry (e.g., conductivity and acidity) and physical charac- sampled sites within the six districts, we restricted the data teristics that are representative of weather-resistant underlying set to small systems (first through third order, determined from bedrock should be important in distinguishing Shield systems, 1:50,000-scale topographic maps) to ensure that any differences and we expect that fish assemblages will primarily consist of in community composition were not due to the effects of stream species with greater acid tolerances and with preferences for order and that there was no systematic difference in stream size faster water flow and larger substrates. Overall, we expect to sampled in different districts (Neff and Jackson 2011). Within find lower fish diversity on the Shield due to (1) lower pro- the six districts, all sites that met these criteria were included in ductivity; (2) a chemical environment of lower-ionic-strength the analysis. waters and low pH, thereby imposing potential physiological Original records were converted from abundance to stress on fishes; and (3) potentially less-complex habitat (e.g., presence–absence so as to minimize any differences due to po- pools and riffles). Thus, the Shield systems may not provide the tential disparity in sampling effort among sites and districts. necessary conditions for supporting the broader group of fishes Although abundance data may provide additional insight into that are hypothesized to occur in adjacent off-Shield systems. patterns of fish community composition and structure, a lack of information on specific sampling methods and effort, particularly among OMNR districts, left us unable to describe or METHODS quantify any sampling biases that might impact estimates of This study capitalized on data available from a provincewide relative abundance. In addition, when the species of fish was sampling program conducted by the Ontario Ministry of Nat- not indicated (i.e., only a higher taxonomic level, such as fam- ural Resources (OMNR) in the 1970s and 1980s. The Aquatic ily or genus, was identified), such records were removed from Habitat Inventory (AHI) survey was implemented to provide the data set so that only species-level identifications were in- a record of fish and macroinvertebrate communities in both cluded. The final data set (data set A) included 104 sites and lotic and lentic systems across the province. Field crews from 50 fish species; 53 sites were identified as being on-Shield and OMNR district offices sampled streams and rivers via protocols 51 sites were identified as off-Shield based on a geographical outlined in the AHI manual (OMNR 1979). There is no specific information systems layer of bedrock geology (OGS 2003). To information available to describe how the sampling sites were examine these sites relative to historical colonization processes specifically chosen, but protocols indicate that site selection was that occurred in this region after deglaciation, the sites were also at the discretion of each district’s sampling crews, with the over- assessed for their location in relation to the maximum extent of all aim of obtaining a general survey of lotic fish communities. glacial Lake Algonquin and the Kirkfield Outlet (Hinch et al. The geographical extent of this historical data set allows for a 1991: their Figure 6), which connected Lake Algonquin to Lake large-scale analysis of fish communities in a large number of Ontario and provided an important dispersal route after the re- lotic systems. treat of the glaciers. Sixty-five sites were located to the north The sampling protocols for the AHI program were geared of the Kirkfield Outlet, 19 sites were located to the south of towards a qualitative assessment of all fish species present at the outlet, and 20 sites were in areas covered by the maximum a site. Various sampling gears were recommended for use, and extent of Lake Algonquin. backpack electrofishing was the predominant method of capture A subset of sites in data set A also included complete data (Gareth Goodchild, formerly of OMNR, personal communica- for a variety of abiotic variables (hereafter, data set B: 17 Shield tion). Although the AHI did not rigidly standardize the sampling sites and 36 off-Shield sites). These variables included water

Downloaded by [Department Of Fisheries] at 20:33 25 September 2012 procedure, there was an emphasis on exhaustive sampling such temperature, DO, pH, turbidity, alkalinity, conductivity (Cond), that all sites included in this survey were sampled extensively but amount of instream cover represented as organic debris (in- without providing standardized catch-per-unit-effort measures. cludes accumulations of leaves, twigs, and algae), large woody Fish species were identified in the field, and voucher specimens debris, instream rock cover, undercut banks, width, depth, ve- were collected for verification by experts at the Royal Ontario locity, and discharge. In addition, the original data also included Museum, Toronto. a number of variables recorded either as percent composition For this analysis, we chose a subset of all survey records or in categories; these variables were substrate composition, to examine patterns of fish community composition in Cana- canopy cover, gradient, and surrounding landscape type. As dian Shield and off-Shield systems. As surveys were conducted percent composition variables are not independent of each other through individual OMNR districts, data from six districts (Al- and most standard statistical approaches are unsuitable for ana- gonquin, Bancroft, Bracebridge, Minden, Huron, and Lindsay) lyzing such data (Jackson 1997), correspondence analysis (CA) were included to cover lotic systems from a wide area of south- was used to analyze the variation within each group of vari- central Ontario (Figure 1); these districts were selected to re- ables into one or more axes summarizing the dominant trends duce differences in climatic patterns among sites, particularly (Table 1). This method is well suited to analyzing such pro- between Shield and off-Shield sites (i.e., due to the influence portional or compositional data and summarizes the patterns GEOLOGY AS A STRUCTURING MECHANISM 965

FIGURE 1. Ontario Ministry of Natural Resources Aquatic Habitat Inventory sites overlying a map of surﬁcial geology in south-central Ontario. Circles denote sites with complete abiotic data (data set B), and triangles denote additional sites with ﬁsh community data (data set A). Dashed line indicates the geological

Downloaded by [Department Of Fisheries] at 20:33 25 September 2012 boundary between Precambrian Shield and off-Shield sites.

across the measured variables for each observation (i.e., site the case when considering the Canadian Shield, as the physical score). The corresponding site score was used as a new variable nature of this geological feature leaves Shield areas much for summarizing the position of the site along these dominant less suitable for agricultural development in comparison with environmental gradients. The summary variables included two off-Shield areas. Because high-quality land use data for the substrate variables (S1, S2), two landscape variables (L1, L2), entire catchments were not available for the geographical and one canopy cover variable, and one stream gradient (i.e., slope) temporal extent of the study area, we were unable to determine variable. whether differences in land use between Shield and off-Shield Other studies examining the effects of regional factors on bi- sites (rather than location relative to the Shield) could account ological communities often highlight the importance of land use for any potentially derived differences in structuring ﬁsh in inﬂuencing the ecological conditions of lotic systems (e.g., community composition in these systems. In an attempt to Allan 2004; Dow et al. 2006; Kratzer et al. 2006). Geology examine this issue, we performed additional analyses wherein and land use are often closely associated, and this is certainly the original landscape data collected at the time of the sampling 966 NEFF AND JACKSON

TABLE 1. Results of correspondence analysis (CA) on each of the percent TABLE 2. Five functional attributes and their respective categories used in composition variables from the raw data: substrate composition, landscape com- this analysis. Individual fish species were classified into a single category for position, canopy cover, and stream gradient. For each group of compositional each functional attribute. Numbers in parentheses denote the number of species variables, the CA scores for each site were used to create new variables (S1, S2, in the data set that fell within each specified category. L1, L2, Cov, and G) that could be incorporated in further analyses. Functional attribute Categories Variable Abbreviation CA score Description Temperature Cold (9), cool (26), or warm (14) Substrate S1 Positive Marl and muck preferences composition Negative Sand, clay, silt, Trophic status Planktivore–detritivore (1), and rocks planktivore–invertivore (5), S2 Positive Rocks, rubble, detritivore (1), detritivore–invertivore and boulders (2), herbivore–invertivore (4), Negative Clay, muck, and invertivore (14), invertivore–carnivore sand (17), variable–omnivore (5) Landscape L1 Positive Shrub marsh Geomorphology Riffle (3), riffle–run (3), riffle–pool (3), composition Negative Meadow, preferences run–pool (7), pool (15), all slow cultivated habitats (pools, backwaters, bogs) L2 Positive Meadow, (16), variable (2) cultivated, Substrate Small (3), vegetation–small (13), pasture preferences small–medium (10), medium (11), Negative Open marsh, vegetation–medium (1), upland conifer, medium–large (2), variable (5), swamp conifer vegetation–variable (4) Canopy cover Cov Positive Dense cover Reproductive Brood hider–nonguarder (13), open Negative Open cover methods or substrate–nonguarder (19), complex ◦ Gradient G Positive Medium (10–60 ) behaviors nester (17) Negative Low (<10◦)

represent characteristics of Ontario populations of these species. (i.e., variables L1 and L2) were replaced with 2001–2003 This exercise resulted in five frequency matrices, which were land use data from the Aquatic Landscape Information System then transformed to relative frequency prior to statistical (OMNR 2008), which provides percentages of different land analysis. use types for the entire catchment of each site. As much of Statistical analysis.—To analyze patterns in fish community the land use in the region has not changed to any great extent composition across south-central Ontario, we used CA on the between the two time periods, these recent data provided an fish community data in data sets A and B. We then used canon- opportunity to examine this issue. ical correspondence analysis (CCA) to test whether Shield and In addition to assessing differences in fish community off-Shield sites differed statistically based on their community structure based on taxonomic composition, we also considered composition. The CCA constrained the CA by using a single potential differences in the functional traits of fish communities. predictor variable to indicate a site’s location (on- or off-Shield)

Downloaded by [Department Of Fisheries] at 20:33 25 September 2012 For example, Shield systems may comprise predominantly and then performed a permutation test (n = 1,000) to determine riffle- or run-preferring species as opposed to those preferring whether the resulting axis was significant—in other words, this slower habitats. For this analysis, a combination of resources approach tested whether the communities from the two regions was used to classify the fish species in data sets A and B were significantly different from each other or could be viewed according to five categories of fish functional traits or attributes: as a homogeneous set. The same analysis was carried out on temperature preferences, trophic status, stream geomorphology each of the five frequency matrices representing fish functional preferences, substrate preferences, and reproduction behaviors traits (temperature preference, trophic position, geomorphology (Table 2). Information on species functional traits as described preference, substrate preference, and reproductive behavior), al- by Goldstein and Meador (2004) was used to categorize lowing the examination of patterns in fish functional traits across species; we cross-checked these classifications with those in the Shield and off-Shield sites. Through this approach, we tested Ontario Online Fish Database (Eakins 2010), which primarily whether Shield sites could be statistically distinguished from synthesizes information from Coker et al. (2001). When data in off-Shield sites based on taxonomic composition, functional the Ontario Online Fish Database differed from traits described trait composition, or both. by Goldstein and Meador (2004), we selected traits as described Fish community composition patterns were then assessed in in the database because that information was more likely to the context of the glacial history of the region. Using the CA GEOLOGY AS A STRUCTURING MECHANISM 967

result, sites were recoded according to their location north of the 12.0 a) Sites Shield Kirkfield Outlet or south of the Kirkfield Outlet and within the Off-Shield maximum extent of glacial Lake Algonquin. Correspondence 10.0 analysis was also applied to data summarizing fish thermal pref- 8.0 erences as described above, and CCA was used with a single predictor variable (location: north or south of the Kirkfield Out- 6.0 let) and permuted (n = 1,000) to test for significant associations between thermal preferences and site location in relation to the 4.0 glacial waters. To examine patterns in abiotic conditions between Shield 2.0 and off-Shield sites, a jackknifed linear discriminant analysis 0.0 (LDA) was used to summarize the covariation between the abiotic variables in data set B and to statistically test whether the -2.0 two types of sites differed in their environmental characteristics. -4.0 Using CCA, the community composition data were then con- -4.0 -2.0 0.0 2.0 4.0 6.0 8.0 10.0 12.0 strained by the abiotic variables to determine the variables that CA2 12.0 were most important in explaining fish community composition b) Species patterns across south-central Ontario. 10.0 All analyses were performed using the vegan, MASS, and labdsv libraries in R software (R Development Core Team 2010). 8.0

6.0 SLS RESULTS 4.0 RT Patterns in Fish Community Composition SPL BUR Off-Shield sites were significantly more species rich than 2.0 SILBT SS MUS MSCFTD RS = = < SELRDLD JD LS Shield sites (Kruskal–Wallis chi-square 11.87, df 1, P BKTRB SB FF MS BDWS BLM CHC CCBBPS TP WLE 0.001), with a median of 10 species for off-Shield sites and 7 0.0 BNSLP NRDCSFMLB CMPDBS YP = BG GS species for Shield sites (means 9.7 and 8.0, respectively). A FDHCBCS LCBM ID -2.0 NP CA plot of the first two axes for data set A shows that sites were LED grouped based on their Shield or off-Shield location but with -4.0 considerable overlap (Figure 2a; species codes are defined in -4.0 -2.0 0.0 2.0 4.0 6.0 8.0 10.0 12.0 Table 3). Although the ordination plot shows a classic V-shape CA1 common to many CA plots, the ability to readily recognize the endpoints of the summarized gradient indicates that the ordi- FIGURE 2. Correspondence analysis (CA) results for data set A: (a) pat- = nation provides an interpretable summary of the multivariate terns among sampling sites (black circles Precambrian Shield sites; open circles = off-Shield sites) and (b) associations between fish species. Species pattern. Burbot and muskellunge were two rare (i.e., present at labels (defined in Table 3) are positioned at the center of each point, with mi- <5% of sites) species found at a small number of off-Shield nor adjustments made for ease of reading. The first two axes for data set A sites, and they had highly positive values on CA axis 1 (Fig- summarize 16.4% of the variation. ure 2b). The rosyface shiner, longnose sucker, channel catfish,

Downloaded by [Department Of Fisheries] at 20:33 25 September 2012 and mimic shiner were other less-common species with posi- ilar pattern, and the constrained axis was also significant (n = tive values on CA axis 1 but were more associated with a small 1,000, P < 0.001; Figure A.1). number of Shield sites. The slimy sculpin, rainbow trout, and Patterns of fish species composition were analyzed accord- splake (lake trout × brook trout hybrid) appear on the far end ing to species’ preferences for water temperature, trophic group, of the second axis and were associated with a small number of geomorphology, substrate, and reproduction (Figure 3). There off-Shield sites. Both the rainbow trout and splake are stocked was no clear visual distinction between Shield and off-Shield species and represent introductions via human vectors. Removal sites for temperature preferences, and the constrained CCA axis of these species from the analysis did not change the results, and was not significant (P = 0.125; Figure 3a). Analysis of geomor- so they were retained to accurately portray the fish community phology preferences indicated that Shield sites tended to be as- during this time period. When constrained by a single predictor sociated with strictly riffle or strictly pool species (e.g., northern variable indicating on- or off-Shield location, the constrained redbelly dace and common shiner, both pool species), whereas axis was significant (n = 1,000, P < 0.001), indicating a sig- off-Shield sites were additionally associated with species that nificant difference in the fish community present at Shield sites use a variety of habitat types (e.g., central mudminnow and relative to off-Shield sites. A CA plot of the first two axes for bluntnose minnow; P < 0.001; Figure 3b). Shield sites were data set B (i.e., a subset of the sites in data set A) showed a sim- also associated with species that prefer smaller substrates, such 968 NEFF AND JACKSON

TABLE 3. List of the fish species included in data set A and definitions of the as the northern redbelly dace, creek chub, and common shiner, species abbreviations used in Figure 2. whereas off-Shield sites were associated with species that pre- < Species Species code fer a broader range of substrates (P 0.001; Figure 3c). For trophic group variables, there was not a clear distinction be- Silver lamprey Ichthyomyzon unicuspis SIL tween groups of sites on the first two CA axes, but some Sea lamprey Petromyzon marinus SEL Shield–off-Shield groupings were apparent on axes 3 and 4 Northern redbelly dace Phoxinus (= Chrosomus) NRD and the constrained CCA axis was significant (P < 0.001; Fig- eos ure 3d). Shield sites were associated with herbivore–invertivore = Finescale dace Phoxinus ( Chrosomus) neogaeus FD species and planktivore–invertivore species, whereas off-Shield Lake chub Couesius plumbeus LC sites were more associated with strict invertivores and detriti- Brassy minnow Hybognathus hankinsoni BM vores. Lastly, Shield sites tended to include species that do not Hornyhead chub Nocomis biguttatus HC guard their eggs or young (e.g., lake chub, blackchin shiner, and Golden shiner Notemigonus crysoleucas GS golden shiner), whereas off-Shield sites encompassed species Common shiner Luxilus cornutus CS with a broader variety of reproductive behaviors (P < 0.001; Blackchin shiner Notropis heterodon BCS Blacknose shiner Notropis heterolepis BNS Figure 3e). Rosyface shiner Notropis rubellus RS For both data set A (Figure A.2a) and data set B (Fig- Spotfin shiner Cyprinella spiloptera SS ure A.2b), the analysis of fish community composition incorpo- Mimic shiner Notropis volucellus MS rating the glacial history of the region showed patterns similar Bluntnose minnow Pimephales notatus BLM to those observed when examining community composition ac- Fathead minnow Pimephales promelas FM cording to on- or off-Shield location. Analysis of fish thermal Eastern blacknose dace Rhinichthys atratulus BD preferences showed a weak association of coolwater and cold- Longnose dace Rhinichthys cataractae LD water species with sites north of the Kirkfield Outlet, but the Creek chub Semotilus atromaculatus CC constrained axis was nonsignificant (P = 0.084). Fallfish Semotilus corporalis FF Pearl dace Margariscus margarita PD Patterns in Abiotic Conditions Longnose sucker Catostomus catostomus LS The LDA correctly classified 94.3% of sites into their respec- White sucker Catostomus commersonii WS tive Shield or off-Shield group by using the abiotic variables. By Brown bullhead Ameiurus nebulosus BB region, 94.1% of Shield sites (16 of 17) and 94.4% of off-Shield Channel catfish Ictalurus punctatus CHC sites (34 of 36) were correctly classified in a cross-validated, Northern pike Esox lucius NP jackknifed analysis, indicating that these streams can be accu- Muskellunge Esox masquinongy MUS rately classified into geological groups based on their physical Central mudminnow Umbra limi CM and chemical traits. Rainbow trout Oncorhynchus mykiss RT Brown trout Salmo trutta BT Patterns in Abiotic and Biotic Composition between Brook trout Salvelinus fontinalis BKT Shield and Off-Shield Systems Splake (lake trout Salvelinus namaycush × brook SPL Using CCA to assess the association between fish species trout) composition and abiotic conditions, we found that the first two Trout-perch Percopsis omiscomaycus TP Burbot Lota lota BUR CCA axes separated sites into their respective Shield and off- Brook stickleback Culaea inconstans BS Shield groups (Figure 4). High conductivity, alkalinity, temper- Downloaded by [Department Of Fisheries] at 20:33 25 September 2012 Mottled sculpin Cottus bairdii MSC ature, turbidity, meadow and cultivated landscapes (L2), and pH Slimy sculpin Cottus cognatus SLS were the dominant abiotic variables associated with off-Shield Rock bass Ambloplites rupestris RB sites, whereas higher velocity, more woody debris, shrub marsh Pumpkinseed Lepomis gibbosus PS (L1) and, to a lesser degree, higher DO were associated with Bluegill Lepomis macrochirus BG Shield sites. Notably, substrate variables were not important Smallmouth bass Micropterus dolomieu SB abiotic drivers in separating Shield and off-Shield sites. Largemouth bass Micropterus salmoides LB An additional CCA was performed wherein the original (i.e., Yellow perch Perca flavescens YP historical) landscape variables L1 and L2 were replaced with Walleye Sander vitreus WLE the present-day land use data. By partialling out the variation Rainbow darter Etheostoma caeruleum RD attributed to present-day land use, we found that the land use Iowa darter Etheostoma exile ID variables only accounted for a very small portion of the total Fantail darter Etheostoma flabellare FTD variation in fish communities (5.2%) compared with the other Least darter Etheostoma microperca LED abiotic variables (35.2%). Joint variation among land use and Johnny darter Etheostoma nigrum JD abiotic variables explained 1.21% of the total variation, and Logperch Percina caprodes LP 58.35% of the variation was unexplained. This result suggests GEOLOGY AS A STRUCTURING MECHANISM 969

a) Riffle Shield Warm b) Off-Shield Pool 2.0

Variable Riffle/run

Riffle/pool 0.5 1.5 Cold 0.0 All slow Cool -1.5 -1.0 -0.5 0.0 0.5 1.0

-0.5 1.0 Run/pool -2.5 -2.0 -1.5 -1.0-0.5 0.0 0.5 -2.0 -1.5 -1.0 -0.5 0.0 0.5 1.0 d) Planktivore-Invertivore c)

Invertivore-Carnivore Detritivore-Invertivore Planktivore-Detritivore Medium-Large Small-Medium Medium

Vegetation-Medium CA2 Invertivore Small Variable Vegetation-Variable Vegetation-Small Variable -2.0 -1.0 0.0 1.0

-2.0 -1.0 0.0 1.0 2.0 Detritivore Herbivore-Invertivore 0.0 1.0 2.0 3.04.0 5.0 -1.00.0 1.0 2.0 3.0 e) Open, nonguard Downloaded by [Department Of Fisheries] at 20:33 25 September 2012

Brood hider,non guard

-1.0 -0.5 0.0 0.5 1.0 Complex nester -1.0 -0.50.0 0.5 1.0 CA1

FIGURE 3. Correspondence analysis (CA) results for the ﬁve functional attribute analyses: (a) temperature, (b) geomorphology, (c) substrate, (d) trophic status, and (e) reproductive behavior (black circles = Precambrian Shield sites; open circles = off-Shield sites). 970 NEFF AND JACKSON

The creek chub is known as a broadly tolerant species that is Shield Off-Shield able to live in systems with lower DO, but it prefers small, clear, cold streams (Scott and Crossman 1998). The northern redbelly dace is also widespread throughout southern and central Ontario but is often found in dystrophic waters on the Shield (Holm et al. 3.0 2009).

1.0 We predicted that Shield sites would be dominated by larger substrates due to the surface proximity of bedrock on the Shield V CCA2 and that this condition would potentially inﬂuence ﬁsh species L1 DO 1.0 4.0 composition. However, neither of the two composite substrate WD Cov variables was an important abiotic driver in separating Shield

OD 0.0 Discharge G and off-Shield sites, perhaps reflecting (1) the comparable size S1 Turb UCBank Depth pH of the systems considered and (2) that substrate size is related Width S2 Rocks L2 generally to stream gradient and system size. Analysis of fish Alk functional traits indicated that Shield sites were more associ- Cond T ated with species preferring smaller substrates. Although the -2.0 -1.0 0.0 2.0 -1.0 -2.0 0.0 2.0 4.0 most important substrate variable associated with Shield sys- CCA1 tems (i.e., S1) represented marl and muck substrates, this variable was not an important characteristic of Shield systems, either FIGURE 4. Canonical correspondence analysis (CCA) results for data set B (black circles = Precambrian Shield sites; open circles = off-Shield sites). in the abiotic analysis or in the analysis of environment and fish Variables include conductivity (Cond), alkalinity, turbidity (Turb), pH, undercut communities. Overall, these results suggest that substrate com- banks (UCBank), gradient (G), canopy cover (Cov), water temperature (T), position was not one of the most important physical differences dissolved oxygen (DO), woody debris (WD), water velocity (V), discharge, between Shield and off-Shield streams and instead may be more width, depth, instream rock cover (Rocks), marl–muck–rubble substrate (S1), important for structuring community composition within Shield rock–rubble–boulder substrate (S2), meadow–cultivated landscape (L1), and shrub–marsh landscape (L2). or off-Shield sites. Although Shield and off-Shield sites can be distinguished that while differences in land use may be important in structuring based on their fish assemblages, patterns in species richness and these fish communities, other abiotic variables are of greater analysis of functional traits support the idea that Shield species importance. are a subset of the total regional species pool, whereas the off- Shield sites contain a broader set of the species pool. Fish functional attributes differed significantly between Shield sites and DISCUSSION off-Shield sites, but associations with those functional attributes Our study demonstrates that Shield and off-Shield sites have showed that Shield systems comprise species that represent only distinct physical and chemical conditions and that these differ- one or two functional traits, whereas off-Shield sites are home ences correspond with distinct fish community composition pat- to species representing a broader suite of traits. As species rich- terns. Systems located on the Canadian Shield are characterized ness was significantly lower in Shield sites, the lower degree of by higher current velocity, DO, woody debris, and discharge functional diversity in Shield systems may partly be a function and by lower conductivity, alkalinity, water temperature, and of lower species richness. However, if functional traits were not pH. Analysis of fish functional traits shows that Shield sites are associated with Shield versus off-Shield habitat, then we might

Downloaded by [Department Of Fisheries] at 20:33 25 September 2012 associated with fast-water species that prefer smaller substrates expect to observe a wide diversity of traits even within the small and that are less likely to be nest guarders. The brook trout, a number of species found on the Shield; this was not the case here. species that is intolerant of low DO levels and high tempera- In general, Shield sites were most associated with strictly riffle tures, was strongly associated with Shield systems, as was the or strictly pool species that prefer smaller substrates and that do common shiner, a species that is often found in pools within not exhibit nest-guarding behavior. In contrast, off-Shield com- clear, cool streams. Off-Shield sites are characterized by higher munities tended to exhibit a wider variety of preferences in every conductivity, alkalinity, temperature, and pH and meadow or category. Furthermore, in most cases, there were many species cultivated landscapes; these sites are associated with species that found off-Shield that were either absent or rare at Shield sites. are tolerant of warm temperatures (e.g., fathead minnow, central The lack of Shield-specific species provides further support to mudminnow, and brassy minnow) and species with wide pH tol- the idea that (1) Shield systems represent a lower diversity of erances (e.g., brook stickleback). Overall, the general patterns habitats than off-Shield systems and (2) only the species that are suggest that off-Shield sites are associated with broadly toler- best adapted to Shield-like environments are able to successfully ant species (e.g., fathead minnow, central mudminnow, creek inhabit those systems. chub, and northern redbelly dace), which have habitat prefer- The present results suggest that aspects of the abiotic or bi- ences that could describe either Shield or off-Shield streams. otic environment in Shield systems have “filtered” the regional GEOLOGY AS A STRUCTURING MECHANISM 971

species pool to a great degree such that only a specific subset of an important abiotic factor in structuring Shield and off-Shield fish species is found in Shield systems. One potential filter could fish communities in south-central Ontario, at least in support- be temperature, an abiotic factor that is known to be important in ing populations of particular species. The role of temperature structuring fish communities and that was an important variable in structuring fish community composition in lotic systems has for distinguishing Shield and off-Shield systems in this analy- been demonstrated in numerous studies, whether across broad sis. The postglacial dispersal of fishes in this region after the spatial scales (e.g., Lyons 1996; Marsh-Matthews and Matthews last glaciation is thought to be tightly linked to species’ thermal 2000), as a local-scale factor (e.g., Wang et al. 2003), or within tolerances. The location of glacial Lake Algonquin, as noted a system moving from headwaters to higher-order streams (e.g., by Hinch et al. (1991), is an important indicator of fish species Townsend et al. 1983). The role of temperature with regard to distributions based on thermal tolerances. During this time, the fish communities of the Shield remains unclear; to further ex- Kirkfield Outlet flowed southeast from glacial Lake Algonquin plore this factor, it will be necessary to perform an analysis that to Lake Ontario and connected waterways that drained northern is specifically designed to test this hypothesis. uplands in a north-to-south direction (Hinch et al. 1991: their In a previous analysis of Shield and off-Shield lotic systems, Figure 6). Recolonization of waters located to the north of the we (Neff and Jackson 2011) showed that many gastropod taxa outlet after the recession of the glaciers was only possible for a were rare in Shield systems and that lower conductivity and short period of time because isostatic rebound altered the con- alkalinity levels may have contributed to these differences. In nection of these waterways. As the glacial melt waters would the present study, the two strongest abiotic factors separating have been very cold, it is likely that recolonization would have Shield and off-Shield sites were the chemical variables of initially been limited to fishes that were adapted to cool or cold conductivity and alkalinity. This is not surprising given the waters. However, by the time the melt waters warmed enough geological differences between the two areas, especially as for colonization by warmwater species, the Kirkfield Outlet no many off-Shield areas in south-central Ontario are almost longer existed, effectively restricting colonization to northern entirely limestone. Lower conductivity levels may contribute areas. Therefore, we might expect more coolwater and coldwa- to increased physiological stress and energy expenditures to ter fishes to be found in waters north of the Kirkfield Outlet and maintain suitable body requirements. For example, one study warmwater species to be restricted to southern Ontario. There found that fathead minnow raised in low-conductivity water was a weak association between prevalence of coolwater and (15 ± 0.11 mg/L CaCO3 [mean ± SE]; conductivity = coldwater fish species and sites north of the outlet. However, 105 µS/cm) had lower survival but greater biomass than fish this association was not significant, suggesting either that this that were raised in high-conductivity water (65 ± 1.5 mg/L relationship is not very strong or that by the time these sys- CaCO3; conductivity = 502 µS/cm; Blanksma et al. 2009). tems were surveyed, a sufficient number of warmwater species The difference in conductivity levels in the Blanksma et al. had colonized or been introduced into this area. As present-day (2009) experiment resembles the difference between Shield drainage patterns of lotic systems in Ontario are tied to historical and off-Shield systems, although Shield systems generally had connections and dispersal pathways, we also investigated the re- lower conductivity than that in the experiment. It has also been lationship between watershed and fish community patterns. The demonstrated that acclimation to soft water can cause a signifi- effect of secondary watershed—which for this study, included cant reduction in the swimming ability of juvenile rainbow trout, central Ottawa, eastern Georgian Bay, Lake Ontario and the with a greater cardiac output necessary to achieve the same Niagara River, and northern Lake Erie—revealed no apparent amount of exercise as fish acclimated in hard water (Dussault influence on the observed patterns in fish communities, as there et al. 2008). Many factors can influence conductivity, including was fair representation of the eastern Georgian Bay watershed levels of cations such as calcium and magnesium. Calcium,

Downloaded by [Department Of Fisheries] at 20:33 25 September 2012 in both Shield and off-Shield sites. along with phosphorus, is important in the development and We did not find an association between certain tempera- maintenance of the skeletal system, and the majority of the ture preferences and Shield and off-Shield sites, indicating that required calcium is absorbed from the water via the gills (Lall warmwater, coolwater, and coldwater species inhabit both types and Lewis-McCrea 2007). The effects of calcium have also of site. It should be noted that the location of the Kirkfield Outlet been prominently studied in relation to the effect of low pH and more or less follows the Shield boundary, making it difficult to heavy metal toxicity to fishes; water calcium has been shown determine whether patterns in fish composition are due to a site’s to ameliorate the negative effects of low pH and aluminum on location relative to the glacial lake and outlet or are due to loca- the ionoregulatory mechanisms in fishes (e.g., Wood et al. 1990 tion relative to the Shield boundary. However, there was a num- and references therein). The effects of low pH on fishes is also ber of off-Shield, “above-outlet” sites among the “below-outlet” relatively well known and include reproductive failure, harm to and “Lake Algonquin” sites in the analysis, indicating that the gill mucus and membranes, loss of salts, and lowered capacity patterns in fish community composition are better explained of hemoglobin (Matthews 1998 and references therein). There- by on- or off-Shield location. There were some associations fore, low conductivity and low pH of streams on the Shield of typically coolwater and coldwater species (e.g., brook trout) may be key chemical drivers or filters, either excluding various with Shield sites, indicating that water temperature may yet be fish species due to low tolerance to these conditions or creating 972 NEFF AND JACKSON

more stressful environments that further exacerbate the effects fish communities, but it is clear that the Canadian Shield itself of competition, predation, or other unfavorable interactions. has a distinct effect on patterns in fish community composition We acknowledge that differences in land use patterns be- in south-central Ontario. tween Shield and off-Shield areas may influence patterns in fish community composition. The nature of the Shield, with its thin, nutrient-poor soils and large expanses of exposed bedrock, ACKNOWLEDGMENTS effectively limits the agricultural development in the Shield re- We thank the OMNR for assistance in accessing the AHI gion. It is possible that the differences in fish communities be- fish survey records, and we are grateful to Jesse Elders for as- tween Shield and off-Shield streams are due not to the Shield sistance in digitizing the original data. Nick Mandrak provided itself but rather to differences in the amount of agricultural, ur- valuable comments and criticism on an earlier version of this ban, or other land uses within each catchment. A lack of data work. Funding was provided by the Natural Sciences and En- describing catchment-scale land use patterns for the sampling gineering Research Council of Canada and the University of period prevented any in-depth analysis on this subject, but anal- Toronto. ysis using current land use data for these sampling sites showed that the variation attributed solely to land use was a very small portion of the overall variation in fish community composition. REFERENCES We suggest that differences in land use between Shield and off- Allan, J. D. 2004. Landscapes and riverscapes: the influence of land use on stream ecosystems. Annual Review of Ecology, Evolution, and Systematics Shield areas be viewed not as a confounding factor but instead as 35:257–284. another abiotic factor (i.e., like conductivity and pH) that is in- Angermeier, P. L., and M. R. Winston. 1998. Local vs. regional influences on fluenced by the geology of the Shield and that in turn ultimately local diversity in stream fish communities of Virginia. Ecology 79:911–927. affects fish community composition. Blanksma, C., B. Eguia, K. Lott, J. M. Lazorchak, M. E. Smith, M. Wratschko, Overall, the results indicate that abiotic and biotic conditions T. D. Dawson, C. Elonen, M. Kahl, and H. L. Schoenfuss. 2009. Effects of water hardness on skeletal development and growth in juvenile fathead in low-order lotic systems of the Shield and off-Shield areas are minnows. Aquaculture 286:226–232. distinct and that Shield systems exhibit environmental condi- Bowlby, J. N., and J. C. Roff. 1986. Trophic structure in southern Ontario tions of a more limited range, a different range, or both. These streams. Ecology 67:1670–1679. conditions may thereby restrict the fish species pool to a sub- Bowman, M. F., K. M. Somers, R. A. Reid, and L. D. Scott. 2006. Temporal set of that found off-Shield in south-central Ontario, suggesting response of stream benthic macroinvertebrate communities to the synergistic effects of anthropogenic acidification and natural drought events. Freshwater that the Canadian Shield may also be a regional environmental Biology 51:768–782. filter within the hierarchy of factors limiting local fish assem- Buisson, L., L. Blanc, and G. Grenouillet. 2008. Modelling stream fish species blages in south-central Ontario (Smith and Powell 1971; Tonn distribution in a river network: the relative effects of temperature versus 1990; Jackson et al. 2001). These findings differ from those of a physical factors. Ecology of Freshwater Fish 17:244–257. South American study (Esselman et al. 2006), which found that Chapman, L. J., and D. F. Putnam. 1984. The physiography of southern Ontario, special volume 2. Ontario Geological Survey, Sudbury. geology affected various physical and chemical parameters in Coker, G. A., C. B. Portt, and C. K. Minns. 2001. Morphological and ecological headwater streams but that fish communities did not differ. Al- characteristics of Canadian freshwater fishes. Canadian Manuscript Report though barriers (e.g., falls and rapids) may have limited histori- of Fisheries and Aquatic Sciences 2554. cal and contemporary dispersal, thereby influencing current dis- Condit, R., N. Pitman, E. G. Leigh Jr., J. Chave, J. Terborgh, R. B. Foster, P. ´ ˜ tributions of fish species, we (Neff and Jackson 2011) previously Nunez, S. Aguilar, R. Valencia, G. Villa, H. C. Muller-Landau, E. Losos, and S. P. Hubbell. 2002. Beta-diversity in tropical forest trees. Science 295:666– showed distinct differences in invertebrate communities encom- 669. passing Shield and off-Shield regions. As the invertebrates are Dow, C. L., D. B. Arscott, and J. D. Newbold. 2006. Relating major ions and nu-

Downloaded by [Department Of Fisheries] at 20:33 25 September 2012 not limited in their dispersal in the same manner as fishes (e.g., trients to watershed conditions across a mixed-use, water-supply watershed. many invertebrates have aerial adult forms), dispersal limita- Journal of the North American Benthological Society 25:887–911. tions cannot sufficiently explain the differences in invertebrate Dussault, E. B., R. C. Playle, D. G. Dixon, and R. S. McKinley. 2008. Effects of soft-water acclimation on the physiology, swimming performance, and car- communities between the two regions—thus, environmental dif- diac parameters of the rainbow trout, Oncorhynchus mykiss. Fish Physiology ferences must account for such differences. Similarly, although and Biochemistry 34:313–322. the fish species composition of some individual sites may have Eakins, R. J. 2010. Ontario freshwater fishes life history database, version 3.88. been influenced by dispersal limitations, it is likely that envi- Available: www.fishdb.ca. (October 2010). ronmental conditions account for these broad-scale differences Esselman, P. C., M. C. Freeman, and C. M. Pringle. 2006. Fish-assemblage variation between geologically defined regions and across a longitudinal in fish communities between Shield and off-Shield regions. Al- gradient in the Monkey River basin, Belize. Journal of the North American though some species may be excluded from Shield sites for Benthological Society 25:142–156. abiotic reasons, exclusions could also be due to biotic interac- Goldstein, R. M., and M. R. Meador. 2004. Comparisons of fish species traits tions, particularly those that become more pronounced in a more from small streams to large rivers. Transactions of the American Fisheries stressful abiotic environment (i.e., low conductivity, low pH, or Society 133:971–983. Hall, R. J., and F. P. Ide. 1987. Evidence of acidification effects on stream insect both). It is likely that a combination of abiotic and biotic factors communities in central Ontario between 1937 and 1985. Canadian Journal of ultimately defines the distinction between Shield and off-Shield Fisheries and Aquatic Sciences 44:1652–1657. GEOLOGY AS A STRUCTURING MECHANISM 973

Hinch, S. G., N. C. Collins, and H. H. Harvey. 1991. Relative abundance Neff, M. R., and D. A. Jackson. 2011. Effects of broad-scale geological changes of littoral zone fishes: biotic interactions, abiotic factors, and postglacial on patterns in macroinvertebrate assemblages. Journal of the North American colonization. Ecology 72:1314–1324. Benthological Society 30:459–473. Hoeinghaus, D. J., K. O. Winemiller, and J. S. Birnbaum. 2007. Local and NRC (Natural Resources Canada). 2004. The atlas of Canada: geological regional determinants of stream fish assemblage structure: inferences based provinces map. NRC, Ottawa. Available: atlas.nrcan.gc.ca/site/english/maps/ on taxonomic vs. functional groups. Journal of Biogeography 34:324–338. environment/geology/geologicalprovinces. (September 2011). Holm, E., N. Mandrak, and M. Burridge. 2009. The ROM field guide to fresh- OGS (Ontario Geological Survey). 2003. Surficial geology of southern water fishes of Ontario. Royal Ontario Museum, Toronto. Ontario—(including Toronto): miscellaneous release data 128. OGS, Sud- Jackson, D. A. 1997. Compositional data in community ecology: the paradigm bury. or peril of proportions? Ecology 78:929–940. OMNR (Ontario Ministry of Natural Resources). 1979. Aquatic habitat inven- Jackson, D. A., and H. H. Harvey. 1989. Biogeographic associations in fish tory surveys: manual of instructions. OMNR, Peterborough. assemblages: local vs. regional processes. Ecology 70:1472–1484. OMNR (Ontario Ministry of Natural Resources). 2008. Land information On- Jackson, D. A., P. R. Peres-Neto, and J. D. Olden. 2001. What controls who is tario: aquatic landscape information system (ALIS) dataset. OMNR, Peter- where in freshwater fish communities: the roles of biotic, abiotic, and spatial borough. factors. Canadian Journal of Fisheries and Aquatic Sciences 58:157–170. Poff, N. L. 1997. Landscape filters and species traits: towards mechanistic un- Kelso, J. R. M., M. A. Shaw, C. K. Minns, and K. H. Mills. 1990. An evaluation derstanding and prediction in stream ecology. Journal of the North American of the effects of atmospheric acidic deposition on fish and the fishery resource Benthological Society 16:391–409. of Canada. Canadian Journal of Fisheries and Aquatic Sciences 47:644–655. R Development Core Team. 2010. R 2.10.1: a language and environment for Kilgour, B. W., and D. R. Barton. 1999. Associations between stream fish statistical computing. R Foundation for Statistical Computing, Vienna. and benthos across environmental gradients in southern Ontario, Canada. Ricklefs, R. E. 1987. Community diversity: relative roles of local and regional Freshwater Biology 41:553–566. processes. Science 235:167–171. Kratzer, E. B., J. K. Jackson, D. B. Arscott, A. K. Aufdenkampe, C. L. Dow, Scott, W. B., and E. J. Crossman. 1998. Freshwater fishes of Canada. Galt House L. A. Kaplan, J. D. Newbold, and B. W. Sweeney. 2006. Macroinvertebrate Publications, Oakville, Ontario. distribution in relation to land use and water chemistry in New York City Smith, C. L., and C. R. Powell. 1971. The summer fish communities of drinking-water-supply watersheds. Journal of the North American Bentho- Brier Creek, Marshall County, Oklahoma. American Museum Novitiates logical Society 25:954–976. 2458. Lall, S. P., and L. M. Lewis-McCrea. 2007. Role of nutrients in skeletal Tonn, W. M. 1990. Climate change and fish communities: a conceptual frame- metabolism and pathology in fish: an overview. Aquaculture 267:3–19. work. Transactions of the American Fisheries Society 119:337–352. Layman, C. A., and K. O. Winemiller. 2004. Size-based responses of prey Tonn, W. M., J. J. Magnuson, M. Rask, and J. Toivonen. 1990. Intercontinental to piscivore exclusion in a species-rich neotropical river. Ecology 85:1311– comparison of small-lake fish assemblages: the balance between local and 1320. regional processes. American Naturalist 136:345–375. Lyons, J. 1996. Patterns in the species composition of fish assemblages among Townsend, C. R., A. G. Hildrew, and J. Francis. 1983. Community structure Wisconsin streams. Environmental Biology of Fishes 45:329–341. in some southern English streams: the influence of physiochemical factors. Mandrak, N. E. 1995. Biogeographic patterns of fish species richness in Ontario Freshwater Biology 13:521–544. lakes in relation to historical and environmental factors. Canadian Journal of Wang, L., J. Lyons, P. Rasmussen, P. Seelbach, T. Simon, M. Wiley, P. Kanehl, Fisheries and Aquatic Sciences 52:1462–1474. E. Baker, S. Niemela, and P. M. Stewart. 2003. Watershed, reach and riparian Marsh-Matthews, E., and W. J. Matthews. 2000. Geographic, terrestrial and influences on stream fish assemblages in the northern lakes and forest ecore- aquatic factors: which most influence the structure of stream fish assemblages gion, U.S.A. Canadian Journal of Fisheries and Aquatic Sciences 60:491– in the Midwestern United States? Ecology of Freshwater Fish 9:9–21. 505. Matthews, W. J. 1998. Patterns in freshwater fish ecology. Chapman and Hall, Wood, C. M., D. G. McDonald, C. G. Ingersoll, D. R. Mount, O. E. Johannsson, New York. S. Landsberger, and H. L. Bergman. 1990. Effects of water acidity, calcium McAuliffe, J. R. 1984. Competition for space, disturbance and the structure of and aluminum on whole body ions of brook trout (Salvelinus fontinalis) a benthic stream community. Ecology 65:894–908. continuously exposed from fertilization to swim-up: a study by instrumen- Minns, C. K. 1989. Factors affecting fish species richness in Ontario lakes. tal neutron activation analysis. Canadian Journal of Fisheries and Aquatic Transactions of the American Fisheries Society 118:533–545. Sciences 47:1593–1603. Downloaded by [Department Of Fisheries] at 20:33 25 September 2012 974 NEFF AND JACKSON

APPENDIX: SUPPLEMENTAL CORRESPONDENCE ANALYSIS RESULTS

10.0 Shield 12.0 a) Sites a) Data set A Above outlet Off-Shield Below outlet 8.0 10.0

6.0 8.0

6.0 4.0 4.0 2.0 2.0 0.0 0.0

-2.0 -2.0

-4.0 -4.0

CA 2 -10.0 -8.0 -6.0 -4.0 -2.0 0.0 2.0 4.0 -4.0 -2.0 0.0 2.0 4.0 6.0 8.0 10.0 12.0 CA2 16.0 MUS 10.0 b) Species b) Data set B 14.0 8.0 12.0

10.0 6.0 8.0

6.0 4.0

4.0 2.0 PS 2.0 RBLB BG SB BNS BSID RD WLE WS BM FFJDBLMCM HC LED MSC RB FMNP 0.0 0.0 BT NRDBCS BKT CC CS FD RT BD LS PD LD SLS GS SPL TP BB -2.0 SIL LC -2.0 -4.0 -16.0 -14.0 -12.0 -10.0 -8.0 -6.0 -4.0 -2.0 0.0 2.0 4.0 -4.0 CA1 -10.0 -8.0 -6.0 -4.0 -2.0 0.0 2.0 4.0 FIGURE A.1. Correspondence analysis (CA) results for data set B: (a) pat- CA1 terns among sampling sites (black circles = Precambrian Shield sites; open FIGURE A.2. Correspondence analysis (CA) results for (a) data set A and (b) circles = off-Shield sites) and (b) associations between fish species. Species data set B, with sites coded according to their location north of the Kirkfield labels (defined in Table 3) are positioned at the center of each point, with mi- Outlet (above outlet) or south of the Kirkfield Outlet and in areas covered by the nor adjustments made for ease of reading. The first two axes for data set B maximum extent of glacial Lake Algonquin (below outlet; according to Hinch

Downloaded by [Department Of Fisheries] at 20:33 25 September 2012 summarize 18.8% of the total variation. et al. 1991: their Figure 6). This article was downloaded by: [Department Of Fisheries] On: 25 September 2012, At: 20:33 Publisher: Taylor & Francis Informa Ltd Registered in England and Wales Registered Number: 1072954 Registered office: Mortimer House, 37-41 Mortimer Street, London W1T 3JH, UK

Transactions of the American Fisheries Society Publication details, including instructions for authors and subscription information: http://www.tandfonline.com/loi/utaf20 Systemwide Evaluation of Avian Predation on Juvenile Salmonids from the Columbia River Based on Recoveries of Passive Integrated Transponder Tags Allen F. Evans a , Nathan J. Hostetter a , Daniel D. Roby b , Ken Collis c , Donald E. Lyons d , Benjamin P. Sandford e , Richard D. Ledgerwood e & Scott Sebring f a Real Time Research, Inc., 52 Southwest Roosevelt Avenue, Bend, Oregon, 97702, USA b U.S. Geological Survey, Oregon Cooperative Fish and Wildlife Research Unit, Department of Fisheries and Wildlife, Oregon State University, 104 Nash Hall, Corvallis, Oregon, 97331, USA c Real Time Research, Inc., 52 Southwest Roosevelt Avenue, Bend, Oregon, 97702, USA d Department of Fisheries and Wildlife, Oregon State University, 104 Nash Hall, Corvallis, Oregon, 97331, USA e National Oceanic and Atmospheric Administration, Northwest Fisheries Science Center, Fish Ecology Division, 2725 Montlake Boulevard, Seattle, Washington, 98112-2097, USA f Pacific States Marine Fisheries Commission, 205 Southeast Spokane Street, Suite 100, Portland, Oregon, 97202, USA Version of record first published: 21 Jun 2012.

To cite this article: Allen F. Evans, Nathan J. Hostetter, Daniel D. Roby, Ken Collis, Donald E. Lyons, Benjamin P. Sandford, Richard D. Ledgerwood & Scott Sebring (2012): Systemwide Evaluation of Avian Predation on Juvenile Salmonids from the Columbia River Based on Recoveries of Passive Integrated Transponder Tags, Transactions of the American Fisheries Society, 141:4, 975-989 To link to this article: http://dx.doi.org/10.1080/00028487.2012.676809

PLEASE SCROLL DOWN FOR ARTICLE

Full terms and conditions of use: http://www.tandfonline.com/page/terms-and-conditions 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. The publisher does not give any warranty express or implied or make any representation that the contents will be complete or accurate or up to date. The accuracy of any instructions, formulae, and drug doses should be independently verified with primary sources. The publisher shall not be liable for any loss, actions, claims, proceedings, demand, or costs or damages whatsoever or howsoever caused arising directly or indirectly in connection with or arising out of the use of this material. Transactions of the American Fisheries Society 141:975–989, 2012 C American Fisheries Society 2012 ISSN: 0002-8487 print / 1548-8659 online DOI: 10.1080/00028487.2012.676809

ARTICLE

Systemwide Evaluation of Avian Predation on Juvenile Salmonids from the Columbia River Based on Recoveries of Passive Integrated Transponder Tags

Allen F. Evans and Nathan J. Hostetter Real Time Research, Inc., 52 Southwest Roosevelt Avenue, Bend, Oregon 97702, USA Daniel D. Roby U.S. Geological Survey, Oregon Cooperative Fish and Wildlife Research Unit, Department of Fisheries and Wildlife, Oregon State University, 104 Nash Hall, Corvallis, Oregon 97331, USA Ken Collis* Real Time Research, Inc., 52 Southwest Roosevelt Avenue, Bend, Oregon 97702, USA Donald E. Lyons Department of Fisheries and Wildlife, Oregon State University, 104 Nash Hall, Corvallis, Oregon 97331, USA Benjamin P. Sandford and Richard D. Ledgerwood National Oceanic and Atmospheric Administration, Northwest Fisheries Science Center, Fish Ecology Division, 2725 Montlake Boulevard, Seattle, Washington 98112-2097, USA Scott Sebring Paciﬁc States Marine Fisheries Commission, 205 Southeast Spokane Street, Suite 100, Portland, Oregon 97202, USA

Abstract We recovered passive integrated transponder (PIT) tags from nine piscivorous waterbird colonies in the Columbia

Downloaded by [Department Of Fisheries] at 20:33 25 September 2012 River basin to evaluate avian predation on Endangered Species Act (ESA)-listed salmonid Oncorhynchus spp. populations during 2007–2010. Avian predation rates were calculated based on the percentage of PIT-tagged juvenile salmonids that were detected as passing hydroelectric dams and subsequently were consumed and deposited by birds on their nesting colonies. Caspian terns Hydroprogne caspia (hereafter, “terns”) and double-crested cormorants Pha- lacrocorax auritus (hereafter, “cormorants”) nesting on East Sand Island in the Columbia River estuary consumed the highest proportions of available PIT-tagged salmonids, with minimum predation rates ranging from 2.5% for Willamette River spring Chinook salmon O. tshawytscha to 16.0% for Snake River steelhead O. mykiss. Estimated predation rates by terns, cormorants, gulls of two species (California gull Larus californicus and ring-billed gull L. delawarensis), and American white pelicans Pelecanus erythrorhynchos nesting near the conﬂuence of the Snake and Columbia rivers were also substantial; minimum predation rates ranged from 1.4% for Snake River fall Chinook salmon to 13.2% for upper Columbia River steelhead. Predation on ESA-listed salmonids by gulls and American white pelicans were minor (<2.0% per ESA-listed salmonid population) relative to predation by terns and cormorants. Cu- mulative impacts were greater for Snake River and upper Columbia River salmonids than for salmonids originating

*Corresponding author: [email protected] Received November 30, 2011; accepted March 13, 2012 Published online June 21, 2012 975 976 EVANS ET AL.

closer to the estuary because upriver salmonids must migrate past more bird colonies to reach the ocean. Predation rates adjusted for colony size (per capita rates) were signiﬁcantly higher for terns and cormorants nesting at inland colonies (upstream of Bonneville Dam) than for those nesting in the estuary, suggesting that inland colonies have a greater reliance on salmonids as a food source. Management actions to increase salmonid survival by reducing avian predation in the estuary could be offset if birds that disperse from the estuary relocate to inland nesting sites on or near the Columbia River.

Predation on juvenile salmonids Oncorhynchus spp. during encountered by out-migrating salmonids, ranged from 9% to out-migration to the Pacific Ocean is considered a limiting factor 15% depending on the year (Collis et al. 2001; Ryan et al. in the recovery of Columbia River basin salmonid populations 2003). With the few exceptions noted above, similar trends in that are listed for protection under the U.S. Endangered Species salmonid susceptibility to and overall impacts of predation from Act (ESA; NOAA 2008). Studies of avian predation in the birds nesting at inland colonies have not yet been evaluated. Columbia River basin have focused on colonial waterbirds that Previous studies of avian predation impacts on the survival nest in the estuary (Collis et al. 2001; Roby et al. 2003; Ryan of salmonids from the Columbia River basin have focused on et al. 2003; Lyons et al. 2010), which currently hosts the largest individual nesting colonies (Collis et al. 2001; Roby et al. 2003; known colonies of Caspian terns Hydroprogne caspia (hereafter, Antolos et al. 2005; Maranto et al. 2010) as opposed to the “terns”) and double-crested cormorants Phalacrocorax auritus cumulative effects of numerous colonies located on or near the (hereafter, “cormorants”) in western North America (Lyons et al. Columbia and Snake rivers. Information on salmonid losses to 2010). Previous research has demonstrated that cormorants and avian predation at larger spatial and temporal scales, however, terns nesting on East Sand Island in the Columbia River estuary is paramount in order to fully understand and effectively man- consume millions of juvenile salmonids annually (Lyons et al. age avian predation and thereby maximize the potential benefits 2010), including salmonids from evolutionarily significant units to ESA-listed salmonid ESUs throughout the basin. Further- (ESUs) and distinct population segments (hereafter collectively more, the ephemeral nature of many of the colony sites and referred to as ESUs; NOAA 2011) that are listed under the ESA. the frequency of intercolony movements documented in these Breeding colonies of piscivorous colonial waterbirds, however, bird species (Conover et al. 1979; Cuthbert 1988; Quinn and are not limited to the Columbia River estuary but are distributed Sirdevan 1998; Wires et al. 2000) necessitate a systemwide as- throughout the Columbia River basin. Nearly 150,000 piscivo- sessment of avian predation management plans. For example, rous colonial waterbirds representing five species at 20 different these data are crucial in order to confirm that increases in smolt colonies were documented as nesting at inland sites (upstream survival associated with piscivorous waterbird management in of the estuary) during 2007–2010 (BRNW 2011). Published the estuary are not offset by increased ESU-specific avian pre- research on the impacts of predation by these inland bird dation rates along the mid-Columbia and lower Snake rivers. colonies on survival of juvenile salmonids has been limited to Resource management agencies and conservation groups the tern colonies on Crescent Island in the mid-Columbia River working in the Columbia River basin recognize the importance (Antolos et al. 2005) and on Goose Island in Potholes Reservoir, of addressing avian predation in efforts to restore ESA-listed Washington (Maranto et al. 2010; Figure 1). salmonid ESUs (USFWS 2005; NOAA 2008). Plans to recover Since 1987, passive integrated transponder (PIT) tags have ESA-listed ESUs have been developed by the United States gov- been placed in juvenile salmonids from the Columbia River ernment and specifically call for development of strategies to

Downloaded by [Department Of Fisheries] at 20:33 25 September 2012 basin to study their behavior and survival after release. Passive manage avian predation as a means to bolster juvenile salmonid integrated transponder tags can provide specific information on survival (NOAA 2008, 2010). With the exception of terns nest- individual fish, including species, run type, and migration timing ing on East Sand Island in the Columbia River estuary (USFWS (based on detections of live fish passing hydroelectric dams). Re- 2005), however, plans have not specified (1) which bird colonies coveries of PIT tags on bird colonies have previously been used pose the greatest risks to juvenile salmonid survival or (2) the to calculate minimum predation rates and to measure the rela- potential benefits of management initiatives to reduce avian pre- tive susceptibility of different salmonid ESUs to avian predation dation, particularly in terms of increased salmonid survival. (Collis et al. 2001; Ryan et al. 2003; Antolos et al. 2005; Maranto The main objectives of this study were to (1) determine et al. 2010). Passive integrated transponder tags that were re- colony-specific and cumulative predation rates on ESA-listed covered from large tern and cormorant colonies in the Columbia salmonids by avian predators located on or near the Columbia River estuary revealed that steelhead O. mykiss ESUs were con- and Snake rivers, (2) evaluate relative differences in avian preda- sumed disproportionately in comparison with other PIT-tagged tion rates among salmonid ESUs, (3) assess whether differences salmonid ESUs. Predation rates on PIT-tagged steelhead de- in predation rates are based on the location of the bird colony tected as passing Bonneville Dam, the last Columbia River dam (estuary versus inland), and (4) determine whether per capita AVIAN PREDATION ON JUVENILE SALMONIDS 977

FIGURE 1. Map of the Columbia River basin, showing the bird colonies that were scanned for passive integrated transponder tags from consumed juvenile salmonids, the river systems associated with salmonid populations, and the hydroelectric dams that were used to determine smolt availability during 2007–2010.

(per bird) predation rates differ among bird species and colony and near the confluence of the Snake and Columbia rivers (here- locations. Objectives 1 and 2 address the paucity of knowledge after, “the confluence”; Figure 1). Specific breeding colonies regarding which ESA-listed salmonid ESUs are most affected by that were scanned for PIT tags included tern colonies on East avian predation on a systemwide scale. Objectives 3 and 4 will Sand Island (rkm 8; estuary), the Blalock Islands (rkm 445; aid current and future management efforts by identifying which mid-Columbia River), Crescent Island (rkm 510; confluence), bird species and colonies pose the greatest risk to salmonid and Goose Island (an off-river colony in Potholes Reservoir, populations in the region and by specifying where reductions in Washington, near the confluence); cormorant colonies on East avian predation would most enhance juvenile salmonid survival. Sand Island and Foundation Island (rkm 518; confluence); gull Downloaded by [Department Of Fisheries] at 20:33 25 September 2012 colonies (i.e., California gulls Larus californicus and ring-billed gulls L. delawarensis; hereafter, “gulls”) on Miller Rocks Is- STUDY AREA lands (rkm 333; mid-Columbia River) and Crescent Island; and Our study area included breeding colonies of piscivorous a colony of American white pelicans Pelecanus erythrorhynchos waterbirds from the mouth of the Columbia River to the upper (hereafter, “pelicans”) on Badger Island (rkm 511; confluence; Columbia River, a distance of approximately 730 river kilome- Figure 1). ters (rkm; Figure 1). In total, nine individual bird colonies were The designation of ESUs for ESA-listed salmonids fol- surveyed for this study. These colonies were selected based on lowed those of the National Oceanic and Atmospheric Ad- previous surveys for PIT tags (Ryan et al. 2003; Antolos et al. ministration (NOAA; NOAA 2011) and included both wild 2005; Maranto et al. 2010), the large size of the colonies, the and hatchery-reared fish. Passive integrated transponder tagged close proximity of the colonies to out-migrating salmonids, or salmonids that originated from within the geographic bound- a combination of large size and close proximity. Colonies were ary of the NOAA-defined ESU were included in the study. All located in the Columbia River estuary, inland along the mid- ESA-listed salmonid ESUs that originated entirely upstream of Columbia River (between Bonneville Dam and McNary Dam), Bonneville Dam on the Columbia River were included in 978 EVANS ET AL.

the study (Figure 1). Upper Willamette River spring Chinook These periods were selected because they encompassed the time salmon O. tshawytscha were also included, as the majority periods when juvenile salmonids were out-migrating and there- of fish from this ESU originates above Sullivan Dam on the fore available as prey to nesting birds. The total number of Willamette River (Figure 1). However, upper Willamette River control PIT tags that were sown varied by colony and year, with steelhead were excluded from the study due to the small sample sample sizes ranging from 100 to 600 PIT tags/colony in any sizes of PIT-tagged individuals. Overall, eight different ESA- given year. The number of discrete time periods during which listed salmonid ESUs were evaluated by this study: Snake control tags were sown also varied but was no less than two River steelhead, Snake River sockeye salmon O. nerka, Snake (at the beginning and end of the nesting season) and no more River spring–summer Chinook salmon, Snake River fall Chi- than four. During each release, control tags were randomly sown nook salmon, upper Columbia River steelhead, upper Columbia throughout the entire area occupied by nesting birds during the River spring Chinook salmon, mid-Columbia River steelhead, breeding season. Priorities for sowing control tags were based on and upper Willamette River spring Chinook salmon. colony size (with larger colonies receiving the most control tags) and our a priori expectation of salmonid predation at that colony, with tern and cormorant colonies generally receiving more con- METHODS trol tags than gull or pelican colonies (Collis et al. 2002). Scanning of PIT tags was conducted after birds dispersed Not all PIT tags egested by birds are subsequently deposited from their breeding colonies following the nesting season on their nesting colony. An unknown number of tags are presum- (August–November) during 2007–2010 (hereafter, “the study ably damaged during digestion or are regurgitated or defecated period”). We used the methods described by Ryan et al. (2001), off-colony at loafing, staging, or other areas utilized by birds whereby flat-plate and pole-mounted PIT tag antennas were during the breeding season. The number of consumed PIT tags used to detect PIT tags in situ by systematically scanning the that were deposited off-colony during this study was unknown. area that was occupied by birds during the nesting season. The Therefore, predation rate estimates are minimum estimates of area occupied by birds on each colony was determined based on salmonid losses to colonial waterbirds. aerial photographs of the colony and visits to the colony during Availability of PIT-tagged salmonids.—We queried the the nesting season. The entire colony area occupied by nesting regional salmonid PIT Tag Information System (PTAGIS) birds was scanned for PIT tags (referred to as a “pass”). Numer- database (maintained by the Pacific States Marine Fisheries ous passes were then conducted until the number of previously Commission) to acquire data on interrogations of ESA-listed undetected PIT tags that were found during a pass was less than PIT-tagged salmonids that were released in the Columbia or equal to 5% of the total number of PIT tags that were found River basin during the study period. Availability of PIT-tagged during all previous passes. The effort required to achieve this salmonids to predation by birds nesting on different colonies criterion ranged from 2–6 passes/colony, which took from 1 to was determined by interrogations of PIT-tagged fish at the near- 5 d to complete each year depending on the size (surface area) est upstream hydroelectric dam with juvenile fish interrogation of the colony. capabilities. Therefore, fish availability to birds nesting at East Passive integrated transponder tag detection efficiency.— Sand Island in the estuary was based on detections of PIT-tagged Not all PIT tags deposited by birds on the nesting colony are salmonids at Bonneville Dam (rkm 225) on the lower Columbia subsequently found by researchers after the nesting season. For River or at Sullivan Dam (rkm 206) on the Willamette River example, tags can be blown off of the colony’s nesting area dur- (Figure 1). For bird colonies on Miller Rocks Islands and the ing wind storms; washed away during high tides, rain storms, Blalock Islands in the mid-Columbia River, salmonid availabil- or other flooding events; or otherwise damaged or lost during ity was determined based on detections of PIT-tagged fish at

Downloaded by [Department Of Fisheries] at 20:33 25 September 2012 the course of the nesting season. Furthermore, the detection McNary Dam (rkm 470; Figure 1). For bird colonies near the methods used to find PIT tags on bird colonies are not 100% confluence (Crescent, Badger, Foundation, and Goose islands), efficient, as some proportion of detectable tags is missed by availability was determined from detections of PIT-tagged fish researchers during the scanning process (Ryan et al. 2003). To at Lower Monumental Dam (rkm 589) on the Snake River and at address these factors, PIT tags with known tag codes were in- Rock Island Dam (rkm 730) on the upper Columbia River (Fig- tentionally sown on each bird colony (hereafter, “control tags”) ure 1). Data on impacts to mid-Columbia River steelhead were throughout the nesting season to quantity PIT tag detection ef- limited to predation impacts by birds in the Columbia River ficiency. Control tags had the same dimension and length as estuary because the majority of PIT-tagged fish from this pop- PIT tags used to mark juvenile salmonids from the Columbia ulation entered the migration corridor downstream of McNary River basin (12-mm, 134.2-kHz, full-duplex tags). The sowing Dam. of control tags was conducted during several discrete stages of The distance between the dam used to determine fish avail- the birds’ nesting season: (1) prior to the initiation of egg laying ability and the downstream bird colony surveyed varied from a (March–April), (2) during the egg incubation period (April– minimum of 25 rkm (McNary Dam to the Blalock Islands) to May), (3) during the chick rearing period (May–June), and a maximum of 220 rkm (Rock Island Dam to Crescent Island; (4) immediately after the fledging of young (July–August). Figure 1). For most colonies in this study, the distance between AVIAN PREDATION ON JUVENILE SALMONIDS 979

the dam and the colony was beyond the maximum published for- generated for each bird colony by using all available PIT-tagged aging radius for the bird species (Baird 1976; Gill 1976; Ryder salmonids for 2007–2010 to evaluate colony-specific impacts on 1993; Anderson et al. 2004; Scoppettone et al. 2006), suggesting smolt survival during the entire study period. For all instances that birds from downstream colonies rarely consumed juvenile in which a bird colony consumed less than 0.1% of a given salmonids upstream of these dams. ESA-listed ESU, predation rates are noted as being less than Avian predation rates.—Predation rates on PIT-tagged 0.1% and are presented without confidence intervals due to the salmonids were calculated using a multistep approach. First, proximity of the estimate to zero. for each ESA-listed ESU, the proportion of PIT-tagged fish that To control for imprecise results that might arise from small were consumed by avian predators on day j (qˆj ) was estimated sample sizes, estimates of predation rates were only calcu- by dividing the number of PIT-tagged fish detected at a dam lated for ESA-listed ESUs from which at least 500 PIT-tagged on day j that were subsequently recovered on a bird colony salmonids were interrogated while passing an upstream dam in (recoveredj) by the total number of salmonids detected as pass- a given year. Additionally, only PIT-tagged salmonids that were ing that dam on day j (availablej): detected at a dam during the bird nesting season (1 March–31 August for colonies in the estuary; 1 April–31 July for inland recoveredj colonies) were included in these analyses, as these fish were qˆj = . (1) availablej believed to be available to birds nesting at the colony. Analyses were conducted using R software, with statistical significance Second, we used logistic regression to estimate colony- α setat0.05. specific daily detection efficiencies, whereby a binary response Per capita predation rates.—Predation rates adjusted for dif- of detections (detected or not detected) was modeled as a func- ferences in colony size (number of nesting adults) were gener- tion of time since control tags were placed on the bird colony: ated for each bird colony and each year to address how potential changes in bird colony size might affect overall predation rates β +β t e( 0 1 j ) on ESA-listed ESUs. Colony-size-adjusted predation rates (per p = , ˆ j β +β t (2) 1 + e( 0 1 j ) capita rates) were calculated by dividing predation rate estimates by the number of adult birds present at each colony in where pˆ j is the probability of detecting a control tag that was each year. The numbers of adult birds nesting at each colony in deposited on day j, β0 is the regression intercept, β1 is the regres- each year were obtained from Bird Research Northwest (BRNW sion slope, and tj is the independent variable for deposition date. 2011). Per capita predation rates were based on detections of all To calculate colony-specific adjusted daily predation rates (rˆj ), ESA-listed PIT-tagged salmonids that were interrogated while the proportion of available PIT-tagged salmonids whose tags passing the nearest upstream dam(s). were recovered from a bird colony on day j (qˆj ) was corrected Model assumptions.—Results from our multistep modeling for colony-specific detection efficiency on day j (pˆ j ): procedure for estimating minimum avian predation rates on PIT- tagged salmonids were based on the following assumptions: qˆj (1) salmonid release and detection information obtained from rˆj = . (3) pˆ j PTAGIS was complete and accurate; (2) PIT-tagged salmonids that were detected while passing an upstream dam were available To calculate annual predation rates, daily estimates of the to avian predators nesting downstream of that dam; (3) the detec- total number of PIT-tagged salmonids consumed were summed tion probability for control PIT tags was equal to the detection and divided by the total number of salmonids that were avail- probability for PIT tags that were naturally deposited by birds

Downloaded by [Department Of Fisheries] at 20:33 25 September 2012 able within that same time period. Reach-specific (estuary, mid- on-colony; (4) off-colony PIT tag deposition rates (i.e., tags that Columbia River, and confluence) predation rates were calculated were regurgitated or defecated by birds somewhere other than by summing predation rates from bird colonies in the same on the nesting colony) did not differ among bird species, among reach per salmonid ESU. Confidence intervals for predation colonies, or among years; and (5) PIT tags from consumed fish rates were estimated by a bootstrapping simulation technique were deposited on a bird colony on the same day that the PIT- (Efron and Tibshirani 1986; Manly 1998). The bootstrapping tagged fish were detected as passing the upstream dam. analysis consisted of 2,000 iterations of the model calculation, To verify assumption 1, irregular entries were either vali- with each iteration representing a unique bootstrap resample dated by tagging coordinators or eliminated from the analy- (random sample with replacement) of the observed detection sis. Detections of PIT-tagged salmonids at dams upstream of efficiency and salmonid PIT tag data sets. The 2.5th and 97.5th bird colonies were deemed the most appropriate measure of quartiles were used to represent the limits of a bootstrapped fish availability given the downstream movement of juvenile 95% confidence interval. Predation rate estimates and 95% con- salmonids, the ability to standardize data across all sites, and the fidence intervals were calculated for each unique ESA-listed ability to define unique groups of salmonids based on a known ESU of PIT-tagged fish consumed by a bird colony in each year. location and passage date (assumption 2). Detection efficiency A study period estimate and 95% confidence interval were then estimates (assumption 3) were generally high at all colonies (see 980 EVANS ET AL.

Results); thus, possible violations of assumption 3 would have Columbia River steelhead (n = 1,965), upper Columbia River little effect on estimates of predation rates. Variation in the spring Chinook salmon (n = 399), Willamette River spring Chi- proportion of consumed PIT tags deposited off-colony among nook salmon (n = 200), and Snake River sockeye salmon (n = bird species and among colonies (assumption 4) could result in 183). By river reach and bird colony, the largest number of PIT differences in minimum predation rate estimates. At this time, tags was recovered from bird colonies in the Columbia River however, there are no data available to support or refute as- estuary (n = 20,733 PIT tags recovered on the East Sand Island sumption 4 other than to note that during the nesting season, tern and cormorant colonies), followed by colonies near the con- some PIT tags presumably are damaged during digestion, are fluence (n = 8,831 PIT tags recovered on the Goose Island tern deposited off-colony, or both. Assumption 5 relates to the use colony, Crescent Island tern and gull colonies, Foundation Is- of the last date of live detection as a proxy for the date of PIT land cormorant colony, and Badger Island pelican colony), and tag deposition on a bird colony; this assumption needed only colonies in the mid-Columbia River between McNary and The to be roughly true because detection efficiency did not change Dalles dams (n = 2,500 PIT tags recovered on the Miller Rocks dramatically on a daily basis (see Results). Islands gull colony and Blalock Islands tern colony). Interro- gations of PIT-tagged salmonids overlapped almost completely with the nesting seasons of the avian colonies studied here; over RESULTS 98% of all PIT-tagged salmonids were detected as passing dams In total, 1,058,808 PIT-tagged salmonids from the eight during the nesting seasons. salmonid ESUs were used to determine fish availability to avian predators (Table 1). From these fish, 32,064 PIT tags were sub- Passive Integrated Transponder Tag Detection Efficiency sequently recovered by researchers on avian colonies during Detection efficiency of control PIT tags that were intention- the study period. Snake River steelhead represented the ESU ally sown on bird colonies during the nesting season was unique with the highest number of on-colony recoveries (n = 17,353 to each bird colony and each year. In general, detection efficien- PIT tags), followed by Snake River spring–summer Chinook cies were high across colonies and years (Table 2). Detection salmon (n = 4,858), upper Columbia River steelhead (n = efficiency estimates ranged from a low of 46.5% at the Goose 4,378), Snake River fall Chinook salmon (n = 2,728), mid- Island tern colony in 2009 to a high of 93.0% at the Blalock

TABLE 1. Numbers of PIT-tagged salmonids that were interrogated while passing Bonneville Dam (BON) on the lower Columbia River, Sullivan Dam (SUL) on the Willamette River (WR), McNary Dam (MCJ) on the mid-Columbia River (MCR), Lower Monumental Dam (LMJ) on the Snake River (SR), and Rock Island Dam (RIS) on the upper Columbia River (UCR) during 2007–2010. Salmonids were from Endangered Species Act-listed evolutionarily signiﬁcant units (ESUs); dashes denote PIT-tagged ESUs with too few interrogations for analyses (<500 detections/year).

Migration year Dam ESU 2007 2008 2009 2010 Total BON SR spring–summer Chinook salmon 23,830 11,425 17,396 38,441 91,092 SR fall Chinook salmon 2,005 24,136 16,314 17,974 60,429 SR sockeye salmon — — 1,845 1,382 3,227 SR steelhead 6,391 19,571 23,310 40,023 89,295 MCR steelhead 2,277 2,435 3,570 9,112 17,394 UCR spring Chinook salmon 2,268 1,662 2,064 5,972 11,966

Downloaded by [Department Of Fisheries] at 20:33 25 September 2012 UCR steelhead 3,021 2,494 2,213 12,196 19,924 SUL WR spring Chinook salmon 1,505 2,509 5,573 510 10,097 MCJ SR spring–summer Chinook salmon 74,905 27,288 60,155 52,129 214,477 SR fall Chinook salmon 7,374 36,857 43,461 29,587 117,279 SR sockeye salmon — — 2,088 1,327 3,415 SR steelhead 7,680 15,447 29,877 17,805 70,809 UCR spring Chinook salmon 6,764 4,713 3,982 6,192 21,651 UCR steelhead 3,102 3,204 3,220 3,942 13,468 LMJ SR spring–summer Chinook salmon 22,730 30,142 20,753 8,562 82,187 SR fall Chinook salmon 2,147 22,968 27,198 38,709 91,022 SR sockeye salmon — 767 2,651 568 3,986 SR steelhead 17,120 28,652 52,220 10,950 108,942 RIS UCR spring Chinook salmon — — 738 929 1,667 UCR steelhead 3,781 7,742 7,226 7,732 26,481 AVIAN PREDATION ON JUVENILE SALMONIDS 981

TABLE 2. Average detection efﬁciency (proportion) of control PIT tags that were sown on colonies of Caspian terns (tern), double-crested cormorants (cormorant), American white pelicans (pelican), and California gulls and ring-billed gulls (gull) in the Columbia River basin during 2007–2010. Sample sizes are provided in parentheses. In-season variation in detection efﬁciency is denoted by footnotes.

Island Bird colony 2007 2008 2009 2010 East Sand Tern 0.89 (600) 0.92a (600) 0.90a (600) 0.84a (400) East Sand Cormorant 0.58a (200) 0.69a (600) 0.70 (600) 0.76a (400) Miller Rocks Gull 0.87 (200) 0.83 (200) 0.78a (200) 0.75a (200) Blalock Tern 0.88 (200) 0.93 (100) 0.84 (100) 0.88c (NA) Crescent Tern 0.70a (800) 0.62a (400) 0.71a (400) 0.75a (400) Crescent Gull 0.63a (200) 0.74a (200) 0.73a (200) 0.79a (200) Badger Pelican 0.65a (200) 0.68 (200) 0.85a (200) 0.75a (200) Foundation Cormorant 0.68 (400) 0.74 (400) 0.73b (400) 0.63 (400) Goose Tern 0.53a (100) 0.64a (400) 0.47a (400) 0.58a (400)

aDetection efficiency significantly (P < 0.05) increased with Julian date of tag deposition. bDetection efficiency significantly (P < 0.05) decreased with Julian date of tag deposition. cDetection efficiency was based on the average from previous years because no tags were sown on the colony in 2010.

Islands tern colony in 2008 (Table 2). Within-season temporal (Figure 2). Terns and cormorants in the estuary consumed be- differences in detection efficiency were also observed at some tween 0.9% and 2.4% of available PIT-tagged Chinook salmon colonies but varied by colony and year (Table 2). Logistic regres- ESUs (Figure 2), which suggests that Chinook salmon ESUs sion results indicated that estimated detection efficiency could exhibited similar susceptibility to predation by terns and pre- increase, decrease, or remain stable throughout the nesting sea- dation by cormorants. The combined minimum predation rate son (Table 2). The most common temporal trend was increasing on Snake River sockeye salmon by terns and cormorants in the detection efficiency through the nesting season, and this rela- estuary was estimated at 3.0% (Figure 2); the predation rate tionship was observed in all 4 years at the Crescent Island tern on sockeye salmon was higher for cormorants (2.1%) than for colony, the Crescent Island gull colony, and the Goose Island terns (0.9%; Appendix Table A.1). Annual variability in tern colony (Table 2). predation rates was observed during the 4-year study, and predation rates by cormorants on salmonids (all ESUs) was lowest Avian Predation Rates Results indicated that avian predation on ESA-listed salmonids varied by bird colony, colony location or river reach, and salmonid ESU. By bird colony and location, minimum predation rates were highest from terns and cormorants nesting on East Sand Island in the Columbia River estuary. Terns and cormorants nesting on East Sand Island consumed a minimum of 2.5–16.0% (depending on the ESU) of the available PIT-tagged salmonids that were last detected as passing Bonneville Dam or Sullivan Dam during the study period (Figure 2). Of the

Downloaded by [Department Of Fisheries] at 20:33 25 September 2012 eight ESA-listed ESUs evaluated, minimum predation rates were highest on Snake River steelhead, with an estimated 16% consumed by terns and cormorants nesting on East Sand Is- land (Figure 2). Although combined losses were greatest for Snake River steelhead, similar losses to terns and cormorants nesting in the estuary were also observed for mid-Columbia River steelhead (14.1%) and upper Columbia River steelhead FIGURE 2. Estimated minimum predation rates (with upper 95% confidence (13.8%; Figure 2). Among avian predators in the estuary, pre- limit) on PIT-tagged juvenile salmonids by Caspian terns (tern) and double- dation on steelhead ESUs was significantly higher from terns crested cormorants (cormorant) nesting on East Sand Island (ESI) in the (9.7–10.7%) than from cormorants (3.1–5.5%; Figure 2). Of Columbia River estuary during 2007–2010; prey availability was calculated the four ESA-listed Chinook salmon ESUs evaluated, mini- based on the number of PIT-tagged salmonids that were last interrogated while mum predation rates by terns and cormorants in the estuary passing Bonneville Dam (Columbia River) or Sullivan Dam (Willamette River [WR]). Salmonid evolutionarily significant units (ESUs) are provided (SR = were highest on Snake River spring–summer Chinook salmon Snake River; UCR = upper Columbia River; MCR = mid-Columbia River). at 4.6% (Figure 2). Conversely, minimum predation rates were Only ESUs with at least 500 PIT-tagged individuals interrogated in any given lowest on Willamette River spring Chinook salmon at 2.5% year (see Table 1) are presented. 982 EVANS ET AL.

FIGURE 3. Estimated minimum predation rates (with upper 95% confidence limit) on PIT-tagged juvenile salmonids by Caspian terns (tern) nesting on the FIGURE 4. Estimated minimum predation rates (with upper 95% confidence Blalock Islands (BI) and California gulls and ring-billed gulls (gull) nesting limit) on PIT-tagged juvenile salmonids by Caspian terns (tern), double-crested on Miller Rocks Islands (MRI) in the mid-Columbia River during 2007–2010; cormorants (cormorant), California gulls and ring-billed gulls (gull), and Amer- prey availability was calculated based on the number of PIT-tagged salmonids ican white pelicans (pelican) nesting on islands near the Snake River–Columbia = that were last interrogated while passing McNary Dam. Salmonid ESUs are River confluence during 2007–2010 (PTI Goose Island, Potholes Reservoir; = = = provided (SR = Snake River; UCR = upper Columbia River). Only ESUs with CSI Crescent Island; FDI Foundation Island; BDI Badger Island); prey at least 500 PIT-tagged individuals interrogated in any given year (see Table 1) availability was calculated based on the number of PIT-tagged salmonids that are presented. were last interrogated while passing Lower Monumental Dam (Snake River [SR]) or Rock Island Dam (upper Columbia River [UCR]). Salmonid ESUs are provided. Only ESUs with at least 500 PIT-tagged individuals interrogated in in 2007 (Table A.1). Annual predation rates on salmonids by any given year (see Table 1) are presented. East Sand Island terns were less variable than predation rates by cormorants, but significant differences in annual predation little variation in annual predation rate estimates was observed rates among years and among salmonid ESUs were also noted for these colonies during the study period (<1.0% difference (Table A.1). The trend in which terns had the highest predation in ESU-specific predation rates in all yearly comparisons; Ta- rates on steelhead ESUs, however, was evident during each of ble A.1). The lack of variability in annual predation rate esti- the four study years. mates is associated with the close proximity of these estimates Predation on ESA-listed PIT-tagged salmonids by the two to zero. avian colonies in the mid-Columbia River between McNary Of the three river reaches examined, predation rates asso- and The Dalles dams (gulls on Miller Rocks Islands; terns on ciated with bird colonies near the confluence were the most Blalock Islands) was relatively minor (<2.0% of available fish variable: minimum predation rates on ESA-listed ESUs ranged per salmonid ESU; Figure 3) during the study period in com- from 1.4% to 13.2% during the study period (Figure 4). Of the parison with predation rates by terns and cormorants nesting five avian colonies in the confluence reach, the terns nesting on East Sand Island. Similar to avian predation in the estu- on Goose Island in Potholes Reservoir demonstrated the high-

Downloaded by [Department Of Fisheries] at 20:33 25 September 2012 ary, however, minimum predation rates by terns and gulls in est single colony-specific predation rate of 10.6%, which was the mid-Columbia River were significantly greater on steelhead observed for upper Columbia River steelhead (Figure 4). Preda- ESUs (1.6–1.9%) than on salmon ESUs (0.3–1.1%; Figure 3), tion by Goose Island terns was almost exclusively on salmonid with Blalock Islands terns and Miller Rock Islands gulls both ESUs originating from the upper Columbia River (2.1–10.6%), consuming disproportionately more steelhead relative to their whereas their predation rates on Snake River ESUs were sig- availability downstream of McNary Dam. During the study pe- nificantly lower (<0.3%; Figure 4). The Crescent Island tern riod, minimum predation rates from Blalock Islands terns and colony and Foundation Island cormorant colony were associ- Miller Rocks Islands gulls were less than 0.5% for the three ated with the highest predation rates on ESUs originating from ESA-listed Chinook salmon ESUs evaluated (upper Columbia the Snake River (Figure 4), and they consumed disproportion- River spring, Snake River fall, and Snake River spring–summer ately more Snake River sockeye salmon, Snake River spring– ESUs). Predation on Snake River sockeye salmon (1.1%), es- summer Chinook salmon, Snake River fall Chinook salmon, and pecially by gulls on Miller Rocks Islands, was higher than that Snake River steelhead compared with upper Columbia River on Chinook salmon ESUs, although data were limited to two of ESUs. Predation by Crescent Island terns was highest on Snake the four study years due to inadequate numbers of PIT-tagged River steelhead, as a minimum of 3.3% of available PIT-tagged Snake River sockeye salmon in 2007 and 2008 (Table 1). Very individuals were consumed during the study period (Figure 4). AVIAN PREDATION ON JUVENILE SALMONIDS 983

With the exception of upper Columbia River spring Chinook by avian colonies near the confluence, 1.9% by mid-Columbia salmon, which were predominately consumed by Goose Island River colonies, and 16.0% by estuarine colonies (Figures 2–4). terns, predation by Foundation Island cormorants on salmon Minimum predation rates on upper Columbia River steelhead (Chinook salmon and sockeye salmon) ESUs was similar to or were 13.2, 1.6, and 13.8% by avian colonies near the conflu- greater than predation by terns (Crescent and Goose islands) on ence, in the mid-Columbia River, and in the estuary, respectively the same salmon ESUs (Figure 4). Significant annual variation (Figures 2–4). Cumulative avian predation impacts were greater in predation rates was observed at bird colonies in the conflu- for upper Columbia River steelhead than for Snake River steel- ence reach. Predation rates on upper Columbia River steelhead head due to the high predation rates on upper Columbia River by Goose Island terns were particularly variable, ranging from steelhead by terns nesting on Goose Island (Figure 4). Preda- a low of 7.5% in 2008 to a high of 15.7% in 2009 (Table A.1). tion rates by the nine avian colonies evaluated here were often Similar to results in the estuary, the trend of higher predation significantly less for ESA-listed Chinook salmon and sockeye rates on steelhead populations, particularly by terns, was evident salmon ESUs than for steelhead populations but were still in in all study years. excess of 2.0% for most ESUs in most river reaches. Overall predation impacts were often greater on ESUs originating upstream of the confluence than on ESUs originating Per Capita Predation Rates lower in the basin (mid-Columbia River or Willamette River), as After accounting for differences in the size of each avian upriver ESUs (Snake River and upper Columbia River) were sus- colony, per capita (per bird) predation rates were highest for ceptible to predation by birds from several inland colonies that Crescent Island terns, Foundation Island cormorants, Blalock did not prey upon fish from lower-river ESUs (Figures 2–4). Rel- Islands terns, and Goose Island terns—all inland colonies (Fig- ative to their availability, upper Columbia River steelhead and ure 5). Predation rates adjusted for colony size were an order Snake River steelhead suffered the greatest cumulative impacts of magnitude greater for Crescent Island and Blalock Islands from the nine avian colonies evaluated here. Reach-specific terns than for East Sand Island terns (Figure 5). A similar dif- minimum predation rates on Snake River steelhead were 6.1% ference was evident between Foundation Island cormorants and Downloaded by [Department Of Fisheries] at 20:33 25 September 2012

FIGURE 5. Box-and-whisker plot showing annual estimated minimum per capita predation rates on PIT-tagged juvenile salmonids (Chinook salmon, sockeye salmon, and steelhead combined) by Caspian terns (tern), double-crested cormorants (cormorant), California gulls and ring-billed gulls (gull), and American white pelicans (pelican) nesting on islands in the Columbia River estuary (ESI = East Sand Island), in the mid-Columbia River (MRI = Miller Rocks Islands; BLI = Blalock Islands), or near the Snake River–Columbia River conﬂuence (CSI = Crescent Island; BDI = Badger Island; FDI = Foundation Island; PTI = Goose Island, Potholes Reservoir) during 2007–2010. Predation rates were calculated based on the number of PIT-tagged salmonids that were detected as passing Bonneville and Sullivan dams (for estuarine colonies), McNary Dam (for mid-Columbia River colonies), or Lower Monumental and Rock Island dams (for colonies near the conﬂuence). The sizes of bird colonies (number of nesting adults) were from Bird Research Northwest (BRNW 2011). Lines within the box represent the median, ends of the box represent the interquartile, and the whiskers represent the minimum and maximum. 984 EVANS ET AL.

East Sand Island cormorants; per capita predation rates were Even after adjustments for on-colony detection efficiency are approximately 13 times higher for cormorants nesting at the made, an unknown number of PIT tags consumed by colonial inland location (Foundation Island) relative to those nesting at waterbirds are presumably damaged during digestion, deposited the estuarine location (East Sand Island; Figure 5). Per capita off-colony, or both. Biology and foraging behavior differ among salmonid predation rates by gull and pelican colonies, regard- the five avian species evaluated here, and theoretically these dif- less of river reach, remained extremely low in comparison with ferences could influence the degree to which predation rates are those by tern and cormorant colonies (Figure 5). underestimated due to off-colony deposition and tag damage. Thus, although adjustments for detection efficiency increase the accuracy of predation rate estimates, the predation rates based DISCUSSION on PIT tags recovered from bird nesting colonies still consti- This study is among the first to document predation rates on tute minimum estimates of predation. Further study is needed different ESA-listed fish populations by multiple bird species to quantify off-colony PIT tag deposition and tag damage for nesting throughout a large river system. Predation impacts on comparisons among avian species, among colonies, and among PIT-tagged salmonids were evaluated for five previously un- different environmental conditions. studied piscivorous waterbird colonies in the Columbia River basin (gulls nesting at colonies on Miller Rocks Islands and Avian Predation Rates Crescent Island, cormorants nesting on Foundation Island, terns Results demonstrated that minimum rates of avian predation nesting on the Blalock Islands, and pelicans nesting on Bad- on salmonids varied considerably by bird colony, location (estu- ger Island) and compared them with updated predation rates ary, mid-Columbia River, or near the confluence), and salmonid from four colonies evaluated in the published literature (tern ESU. In general, the highest avian predation rates were observed and cormorant colonies on East Sand Island, terns on Crescent for steelhead ESUs. It is well documented that steelhead have Island, and terns on Goose Island in Potholes Reservoir). Re- a higher susceptibility to avian predation than other salmonid sults demonstrate that avian predation by these nine colonies ESUs in the Columbia River basin (Collis et al. 2001; Ryan was a substantial source of mortality for ESA-listed salmonids et al. 2003; Antolos et al. 2005; Maranto et al. 2010). Possible during out-migration. reasons for the greater susceptibility of steelhead in comparison with salmon include differences in smolt behavior during Passive Integrated Transponder Tag Detection Efficiency out-migration and differences in the relative size (length) of Overall, the efficiency of PIT tag detection on avian colonies salmonids. Collis et al. (2001) hypothesized that steelhead were was high, as the majority of control tags sown on-colony during more susceptible to avian predation because of their larger size the nesting season were subsequently recovered by researchers (length) and their greater surface orientation in comparison with after the nesting season. Variation in on-colony detection ef- Chinook salmon and sockeye salmon. The positive association ficiency of PIT tags among colonies and among years was between average fish length and avian predation rates described likely due to the unique characteristics of each island, including by Ryan et al. (2003) supports this hypothesis, with steelhead substrate type (sand, rock, or soil-based nesting substrate) and being larger on average and preyed upon at a higher rate than weather conditions. The loss of PIT tags during the birds’ nesting salmon. In a study of salmonid migration depth in the Columbia season (i.e., losses to wind storms, rain storms, or other factors) River, Beeman and Maule (2006) observed that steelhead mi- and the missed detections of functional PIT tags on-colony have grated closer to the surface than Chinook salmon during daylight not been incorporated into previously published studies (Collis hours, when bird predation occurs. et al. 2001; Ryan et al. 2003; Antolos et al. 2005; Maranto et al. Of the eight ESA-listed salmonid ESUs evaluated here, Snake

Downloaded by [Department Of Fisheries] at 20:33 25 September 2012 2010). Although Evans et al. (2011) documented substantial loss River steelhead and upper Columbia River steelhead experi- of coded wire tags on a Caspian tern colony in San Francisco enced the highest proportional reach-specific and cumulative Bay, published studies of on-colony loss and detection probabil- losses to avian predation. Steelhead from the Snake River and ities of PIT tags were lacking. Our results demonstrate that when upper Columbia River must pass all nine bird colonies during detection efficiency is ignored, predation rates based on PIT tag out-migration to the Pacific Ocean, and fish from these two ESUs recoveries can underestimate smolt losses to avian predators. experienced avian predation rates greater than 14% in the estuary Predation rates by terns nesting on Crescent and Goose islands and greater than 6% near the confluence. Relative to other doc- were especially influenced by within-season differences in de- umented factors that influence mortality during out-migration, tection efficiency and the relatively low estimates of detection avian predation—particularly by tern and cormorant colonies— efficiency (<60%) in comparison with the other bird colonies was a substantial source of mortality for out-migrating steel- evaluated. Salmonid smolt out-migration occurs across several head. Muir et al. (2001) estimated a 2–5% mortality rate for months; therefore, data describing the variation in detection ef- juvenile steelhead as they passed dams on the Snake River. ficiency of PIT tags deposited on bird colonies during these Rieman et al. (1991) estimated that piscivorous fish (northern time periods are necessary to make more accurate comparisons pikeminnow Ptychocheilus oregonensis, walleye Sander vitreus, across ESUs and across years. and smallmouth bass Micropterus dolomieu) consumed 11% AVIAN PREDATION ON JUVENILE SALMONIDS 985

of available juvenile steelhead in John Day Reservoir on the The highest estimated predation rates on PIT-tagged Columbia River. salmonids by birds nesting at inland colonies were from terns All of the ESA-listed salmonid populations included in this nesting on Crescent and Goose islands and cormorants nesting study comprise a mixture of wild and hatchery-raised fish on Foundation Island. Of the six ESA-listed salmonid popula- (NOAA 2011). As such, an evaluation of bird predation on ESA- tions that were evaluated while passing inland avian colonies, listed ESUs required the inclusion of both rearing types. Other upper Columbia River steelhead received the highest observed studies in the Columbia River basin have noted that hatchery- predation rates during the study period (10.6%; range = 7.5– reared salmonids are occasionally more susceptible to avian 15.7%) from terns nesting on Goose Island in Potholes Reser- predation than their wild counterparts (Collis et al. 2001; Ryan voir. The predation rate estimate for upper Columbia River steel- et al. 2003; Kennedy et al. 2007). Differences in the relative head was surprising because of the relatively small size of the susceptibility of wild and hatchery-raised fish in these studies, tern colony (<500 breeding pairs; BRNW 2011) and the lo- however, generally were not statistically significant and were not cation of the colony (at least 35 km from the upper Columbia consistently observed across salmonid species, avian colonies, River). Our estimates of predation rates on salmonid populations or years. by terns nesting on Goose Island differ considerably from rates Of the nine piscivorous waterbird colonies investigated, the previously reported by Maranto et al. (2010); those authors esti- colonies of terns and cormorants nesting on East Sand Is- mated an average predation rate of 0.6% (range = 0.4–1.1%) on land consumed the highest proportions of available PIT-tagged upper Columbia River steelhead by terns nesting on an island in salmonids. Smolt losses to tern and cormorant predation in the Potholes Reservoir during 2003–2006. There are several expla- estuary as presented here were higher than those reported by nations for this apparent discrepancy. First, during 2003–2006, Ryan et al. (2003), who investigated predation by terns and cor- the location of the tern colony in Potholes Reservoir shifted from morants on East Sand Island during 1998–2000. The increases in Solstice Island in the northern portion of the reservoir to Goose predation rates between these studies are likely due to the grow- Island in the southern portion of the reservoir (6 km closer to ing number of cormorants nesting on East Sand Island (14,324 the upper Columbia River). This move corresponded with a adults in 1998; 27,192 adults in 2010; BRNW 2011) and the change in the birds’ diet composition, as salmonid prey types fact that PIT tag detection efficiency was not incorporated into were more commonly observed in the diets of terns nesting on the 1998–2000 estimates. Goose Island (∼24% of prey items) compared with terns nesting Predation rate estimates based on PIT tag recoveries have on Solstice Island (∼2% of prey items; Maranto et al. 2010). excluded ESA-listed ESUs originating from the lower Columbia Second, the size of the Goose Island tern colony increased from River (lower Columbia River chum salmon O. keta, coho salmon a maximum of 323 breeding pairs in 2006 (Maranto et al. 2010) O. kisutch, Chinook salmon, and steelhead; Collis et al. 2001; to a maximum of 487 breeding pairs in 2009 (BRNW 2011). Ryan et al. 2003). Lower Columbia River salmonid populations Third, measures of PIT tag detection efficiency were not avail- were not considered in this study due to a lack of adequate PIT able prior to 2007—a substantial factor given that detection tag interrogation sites downstream of Bonneville and Sullivan efficiency was less than 65% during 2007–2010. Finally, smolt dams. Predation rates may differ to an unknown degree for availability to terns nesting at Potholes Reservoir was calculated salmonids originating from the lower Columbia River ESUs differently in the two studies. Maranto et al. (2010) based their and those originating from ESUs upstream of Bonneville and predation rate estimates on all PIT-tagged salmonids released Sullivan dams. In a study of smolt consumption (numbers of into the upper Columbia River, regardless of the distance of the fish) in the estuary, Lyons et al. (2010) concluded that coho fish’s release point to the tern colony at Potholes Reservoir. We salmon and subyearling Chinook salmon, two abundant lower limited our analysis to PIT-tagged salmonids that were last de-

Downloaded by [Department Of Fisheries] at 20:33 25 September 2012 Columbia River salmonids, were the most numerous salmonid tected as passing Rock Island Dam, which is 70 km from Goose prey type in the diets of cormorants nesting on East Sand Island. Island and therefore is near the estimated upper foraging range Avian predation in the Columbia River estuary affects ju- of terns nesting in Potholes Reservoir (Maranto et al. 2010). venile salmonids that have survived freshwater migration and Avian predation rates on Snake River and upper Columbia presumably have a higher probability of surviving to return River salmonid ESUs only included salmonids that migrated as adults relative to those ﬁsh that have yet to complete out- past inland bird colonies. However, not all Snake River migration (Roby et al. 2003). Additionally, juvenile salmonids salmonids pass inland bird colonies; a portion of salmonids are belonging to every ESA-listed Columbia River basin ESU must collected at Snake River dams and are transported (via barges pass through the Columbia River estuary and are therefore sus- or trucks) to release locations downstream of Bonneville Dam ceptible to predation by birds nesting on East Sand Island. At (Buchanan et al. 2006). Estimates of the percentage of Snake current colony sizes, management efforts focused on terns and River salmonids that are transported past inland bird colonies cormorants in the Columbia River estuary will consequently vary by ESU and year, and average transportation estimates beneﬁt a greater number of salmonid ESUs than will manage- for Chinook salmon, sockeye salmon, and steelhead during the ment of inland bird colonies (Roby et al. 2003; USFWS 2005; study period ranged from approximately 60% in 2008 to 40% Lyons et al. 2010). in 2010 (FPC 2011). Consequently, the effects of predation on 986 EVANS ET AL.

Snake River salmonids by inland bird colonies apply only to the Rice Island in the upper estuary (rkm 34) and those nesting portion of the smolt population that is not transported around on East Sand Island (rkm 8) revealed that East Sand Island bird colonies (Antolos et al. 2005); during the study period, terns were more reliant on marine forage fishes (northern an- roughly half of all Snake River salmonids were not transported. chovy Engraulis mordax, Pacific herring Clupea pallasii, etc.), However, all (100%) of the salmonids originating from the up- whereas Rice Island terns had a greater reliance on freshwater per Columbia River must out-migrate in-river past inland bird fishes (Roby et al. 2002). Furthermore, diet composition varied colonies in the confluence reach. Similarly, because transported substantially between the tern colonies in the estuary and those Snake River salmonids are released just downstream of Bon- at inland sites (Collis et al. 2002). neville Dam, all salmonids must out-migrate past bird colonies Differences in colonywide predation rates and per capita pre- in the estuary. dation rates indicate that current management efforts to increase Predation rates associated with tern and cormorant colonies smolt survival by reducing the number of nesting birds on East were almost always significantly higher than predation rates as- Sand Island could be offset if those birds relocate to inland sites sociated with gull and pelican colonies, regardless of salmonid in large numbers. Increases in colony size at inland sites, where ESU or river reach. Previous research indicates that fish, and per capita predation rates are higher, could have a negative im- salmonids in particular, constitute a very small proportion of the pact on smolt survival, especially for Snake River and upper diets of California gulls and ring-billed gulls nesting on inland Columbia River steelhead ESUs. Movement of terns from estu- islands of the Columbia River (Collis et al. 2002). Gut content arine to inland nesting locations is plausible given the ephemeral analysis of gulls nesting at Miller Rocks Islands and Crescent nature of waterbird nesting habitats and the documented inter- Island (Collis et al. 2002) indicated that juvenile salmonids colony movements of waterbird species (Conover et al. 1979; comprised less than 4% (by mass) of food biomass. In contrast, Cuthbert 1988; Quinn and Sirdevan 1998; Wires et al. 2000). salmonids comprised 74% (by mass) of the diets of Caspian terns nesting on Crescent Island (Antolos et al. 2005). Preda- Concluding Remarks tion rates on PIT-tagged salmonids by the Badger Island pelican Predation rates based on PIT tag recoveries from bird colony—the only breeding colony of American white pelicans in colonies provide minimum estimates of the proportion of avail- Washington State (Ackerman 1997)—were the lowest observed able fish that are consumed by avian predators and provide among the nine bird colonies investigated during our study. specific information on when and where salmonid populations Pelican predation rates were 0.1% or less for five of the six are most susceptible to predation by colonial waterbirds. To ESA-listed salmonid populations evaluated, and the predation more precisely measure predation impacts, additional research is rate on Snake River steelhead was only slightly higher (0.2%). needed to evaluate off-colony deposition of PIT tags by colonial Low predation rates on out-migrating salmonids by pelicans waterbirds. Research is also needed to determine whether reduc- nesting on Badger Island may be due to several factors, includ- tions in smolt losses to avian predation translate into commensu- ing (1) a reliance on larger fish (Scoppettone et al. 2006) or on rate increases in smolt survival and, ultimately, adult salmonid fish that congregate in shallow-water habitats (Knopf and Evans recruitment. For example, if avian predators are disproportion- 2004), (2) differences in foraging behavior that reduce the habi- ately consuming dead, diseased, injured, or otherwise moribund tat overlap between Badger Island pelicans and out-migrating fish relative to healthy fish, efforts to reduce avian predation will salmonids from the upper Columbia and Snake rivers, or (3) a not result in commensurate increases in smolt survival (Schreck combination of these. et al. 2006). Similarly, reductions in smolt mortality by reducing predation at one bird colony could be countered by increases Per Capita Predation Rates in predation at other colonies or by other piscivorous predators. Research aimed at addressing these uncertainties will help to de- Downloaded by [Department Of Fisheries] at 20:33 25 September 2012 The per capita predation rates on juvenile salmonids (i.e., rates adjusted for differences in colony size) were substantially termine the efficacy of avian management initiatives to recover higher at inland tern and cormorant colonies relative to their ESA-listed salmonid ESUs. counterparts in the estuary. Per capita salmonid predation rates associated with gull and pelican colonies were much lower than ACKNOWLEDGMENTS those associated with tern and cormorant colonies, regardless This project was funded by the U.S. Army Corps of Engi- of river reach. The higher per capita predation rates for inland neers (USACE) Walla Walla District with additional support tern and cormorant colonies are due to the higher prevalence of from the Bonneville Power Administration (BPA), the USACE juvenile salmonids in the diets of terns and cormorants nesting Portland District, and the Bureau of Reclamation. We especially at inland colonies versus estuarine colonies (Collis et al. 2002; thank Scott Dunmire, Rebecca Kalamasz, David Trachtenbarg, Antolos et al. 2005; Lyons 2010). Differences in diet composi- and Paul Schmidt (USACE) and John Skidmore (BPA) for their tion have been attributed to colony location, as food availability assistance and support. We are grateful to Dave Marvin of the differs throughout the Columbia River basin (Collis et al. 2002; Pacific States Marine Fisheries Commission for providing in- Roby et al. 2003; Lyons 2010; Maranto et al. 2010). A com- formation on PIT tag releases in the region via PTAGIS; Brad parison of diet composition between Caspian terns nesting on Ryan, an important collaborator during the early phases of the AVIAN PREDATION ON JUVENILE SALMONIDS 987

project; Manuela Huso and Nick Som for providing consulta- Gill, R. E. 1976. Notes on the foraging of nesting Caspian terns Hydroprogne tion on the statistical methods used to estimate predation rates; caspia (Pallas). California Fish and Game 62:155. and our field crews and field crew coordinators, particularly Kennedy, B. M., W. L. Gale, and K. G. Ostrand. 2007. Relationship between smolt gill Na + ,K+ ATPase activity and migration timing to avian predation Brad Cramer, Pete Loschl, Jessica Adkins, James Tennyson, and risk of steelhead trout (Oncorhynchus mykiss) in a large estuary. Canadian Melissa Carper, for their valuable contributions to this study. The Journal of Fisheries Aquatic Sciences 64:1506–1516. mention of trade or product names does not constitute endorse- Knopf, F. L., and R. M. Evans. 2004. The birds of North America on- ment by the U.S. Government. The study was performed under line: American white pelican (Pelecanus erythrorhynchos). Cornell Labora- the auspices of the Institutional Animal Care and Use Commit- tory of Ornithology, Ithaca, New York. Available: bna.birds.cornell.edu/bna/ species/057. (August 2011). tee, Oregon State University (Animal Care and Use Protocol Lyons, D. E. 2010. Bioenergetics-based predator-prey relationships between Number 3718). piscivorous birds and juvenile salmonids in the Columbia River estuary. Doctoral dissertation. Oregon State University, Corvallis. Avail- able: www.birdresearchnw.org/Project-Info/publications-reports/graduate- REFERENCES student-theses. (February 2011). Ackerman, S. M. 1997. Update: American white pelican colony. Washing- Manly, B. F. J. 1998. Randomization, bootstrap and Monte Carlo methods in ton Ornithological Society News 51 (October/November):1, 6. Available: biology, 2nd edition. Chapman and Hall, London. wos.org/newsletterarchives.htm. (February 2011). Maranto, C. J., T. P. Good, F. K. Wiese, and J. K. Parrish. 2010. Impact of the Anderson, C. D., D. D. Roby, and K. Collis. 2004. Conservation implications of Potholes Reservoir Caspian tern breeding colony on out-migrating juvenile the large colony of double-crested cormorants on East Sand Island, Columbia salmonids in the mid-Columbia River. Transactions of the American Fisheries River estuary, Oregon, USA. Waterbirds 27:155–160. Society 139:362–381. Antolos, M., D. D. Roby, D. E. Lyons, K. Collis, A. F. Evans, M. Hawbecker, and Muir, W. D., S. G. Smith, J. G. Williams, and B. P. Sandford. 2001. Survival of B. A. Ryan. 2005. Caspian tern predation on juvenile salmonids in the mid- juvenile salmonids passing through bypass systems, turbines, and spillways Columbia River. Transactions of the American Fisheries Society 134:466– with and without flow deflectors at Snake River dams. North American Journal 480. of Fisheries Management 21:135–146. Baird, P.A. 1976. Comparative ecology of California and ring-billed gulls (Larus NOAA (National Oceanic and Atmospheric Administration). 2008. Biologi- californicus and Larus delawarensis). Doctoral dissertation. University of cal opinion, Endangered Species Act section 7(a)(2) consultation: consul- Montana, Missoula. tation on remand for operation of the federal Columbia River power sys- Beeman, J. W., and A. G. Maule. 2006. Migration depths of juvenile Chi- tem, 11 Bureau of Reclamation projects in the Columbia basin and ESA nook salmon and steelhead relative to total dissolved gas supersaturation in section 10(a)(1)(A) permit for juvenile fish transportation program. NOAA a Columbia River reservoir. Transactions of the American Fisheries Society National Marine Fisheries Service, Northwest Region, F/NWR/2005/05883, 135:584–594. Seattle, Washington. Available: https://pcts.nmfs.noaa.gov/pls/pcts-pub/pcts BRNW (Bird Research Northwest). 2011. Research, monitoring, and evalua- upload.summary list biop?p id=27149. (September 2010). tion of avian predation on salmonid smolts in the lower and mid-Columbia NOAA (National Oceanic and Atmospheric Administration). 2010. Endangered River: colony counts from the 2007, 2008, 2009, and 2010 annual re- Species Act section 7(a)(2) consultation supplemental biological opinion: ports. BRNW, Bend, Oregon. Available: www.birdresearchnw.org/Project- supplemental consultation on remand for operation of the federal Columbia Info/publications-reports/unpublished-reports/default.aspx. (August 2011). River power system, 11 Bureau of Reclamation projects in the Columbia Buchanan, R. A., J. R. Skalski, and S. G. Smith. 2006. Estimating the effects of basin and ESA section 10(a)(I)(A) permit for juvenile fish transportation pro- smolt transportation from different vantage points and management perspec- gram. NOAA National Marine Fisheries Service, Northwest Region, Seattle, tives. North American Journal of Fisheries Management 26:460–472. Washington. Collis, K., D. D. Roby, D. P. Craig, S. Adamany, J. Y. Adkins, and D. E. NOAA (National Oceanic and Atmospheric Administration). 2011. Updated Lyons. 2002. Colony size and diet composition of piscivorous waterbirds on status of federally listed ESUs of West Coast salmon and steelhead. the lower Columbia River: implications for losses of juvenile salmonids to NOAA National Marine Fisheries Service, Northwest Region, Seattle, avian predation. Transactions of the American Fisheries Society 131:537– Washington. Available: www.nwr.noaa.gov/Publications/Biological-Status- 550. Reviews/Salmon.cfm. (September 2011). Collis, K., D. D. Roby, D. P. Craig, B. A. Ryan, and R. D. Ledgerwood. 2001. Quinn, J. S., and J. Sirdevan. 1998. Experimental measurement of nesting sub- Colonial waterbird predation on juvenile salmonids tagged with passive inte- strate preference in Caspian terns, Sterna caspia, and the successful coloniza- Downloaded by [Department Of Fisheries] at 20:33 25 September 2012 grated transponders in the Columbia River estuary: vulnerability of different tion of human constructed islands. Biological Conservation 85:63–68. salmonid species, stocks, and rearing types. Transactions of the American Rieman, B. E., R. C. Beamesderfer, S. Vigg, and T. P. Poe. 1991. Estimated Fisheries Society 130:385–396. loss of juvenile salmonids to predation by northern squawfish, walleyes, and Conover, M. R., B. C. Thompson, R. E. Fitzner, and D. E. Miller. 1979. Increased smallmouth bass in John Day Reservoir, Columbia River. Transactions of the populations of ring-billed and California gulls in Washington State. Western American Fisheries Society 120:448–458. Birds 10:31–36. Roby, D. D., K. Collis, D. E. Lyons, D. P. Craig, J. Y. Adkins, A. M. Myers, and Cuthbert, F. J. 1988. Reproductive success and colony-site tenacity in Caspian R. M. Suryan. 2002. Effects of colony relocation on diet and productivity of terns. Auk 105:339–344. Caspian terns. Journal of Wildlife Management 66:662–673. Efron, B., and R. J. Tibshirani. 1986. Bootstrap methods for standard errors, Roby, D. D., D. E. Lyons, D. P. Craig, K. Collis, and G. H. Visser. 2003. confidence intervals, and other measures of statistical accuracy. Statistical Quantifying the effect of predators on endangered species using a bioener- Science 1:54–75. getics approach: Caspian terns and juvenile salmonids in the Columbia River Evans, A. F., D. D. Roby, K. Collis, B. M. Cramer, J. A. Sheggeby, L. J. estuary. Canadian Journal of Zoology 81:250–265. Adrean, D. S. Battaglia, and D. E. Lyons. 2011. Recovery of coded wire Ryan, B. A., J. W. Ferguson, R. D. Ledgerwood, and E. P. Nunnallee. 2001. tags at a Caspian tern colony in San Francisco Bay: a technique to evaluate Detection of passive integrated transponder tags from juvenile salmonids impacts of avian predation on juvenile salmonids. North American Journal on piscivorous bird colonies in the Columbia River basin. North American of Fisheries Management 31:79–87. Journal of Fisheries Management 21:149–153. FPC (Fish Passage Center). 2011. FPC 2010 draft annual report. FPC, Portland, Ryan, B. A., S. G. Smith, J. M. Butzerin, and J. W. Ferguson. 2003. Relative Oregon. Available: www.fpc.org. (June 2011). vulnerability to avian predation of juvenile salmonids tagged with passive 988 EVANS ET AL.

integrated transponders in the Columbia River estuary, 1998–2000. Transac- Scoppettone, G. G., P. H. Rissler, D. Withers, and M. C. Fabes. 2006. Fish tag tions of the American Fisheries Society 132:275–288. recovery from the American white pelican nesting colony on Anaho Island, Ryder, J. P. 1993. The birds of North America online: ring-billed gull (Larus Pyramid Lake, Nevada. Great Basin Birds 8:6–10. delawarensis). Cornell Laboratory of Ornithology, Ithaca, New York. Avail- USFWS (U.S. Fish and Wildlife Service). 2005. Caspian tern management able: bna.birds.cornell.edu/bna/species/033. (August 2011). to reduce predation of juvenile salmonids in the Columbia River estuary: Schreck, C. B., T. P. Stahl, L. E. Davis, D. D. Roby, and B. J. Clemens. 2006. ﬁnal environmental impact statement. USFWS, Migratory Birds and Habitat Mortality estimates of juvenile spring-summer Chinook salmon in the lower Programs, Portland, Oregon. Columbia River and estuary, 1992–1998: evidence for delayed mortality? Wires, L. R., and F. J. Cuthbert. 2000. Trends in Caspian tern numbers and Transactions of the American Fisheries Society 135:457–475. distribution in North America: a review. Waterbirds 23:388–404.

APPENDIX: AVIAN PREDATION RATES ON JUVENILE SALMONIDS

TABLE A.1. Minimum annual predation rates (%; with 95% confidence intervals in parentheses) by bird colonies on PIT-tagged salmonids from En- dangered Species Act-listed evolutionarily significant units (ESUs; SR = Snake River; MCR = mid-Columbia River; UCR = upper Columbia River; WR = Willamette River) during each year and across all years (study period, 2007–2010). Smolt availability for each river reach (Columbia River estuary [estuary], MCR, and SR–Columbia River confluence [confluence]) was based on the number of PIT-tagged salmonids that were interrogated at Sullivan Dam or Bonneville Dam (for estuarine colonies), McNary Dam (for MCR colonies), or Lower Monumental and Rock Island dams (for colonies near the confluence). Dashes denote PIT-tagged populations that were excluded from analysis because fewer than 500 individuals were interrogated during that year.

Salmonid ESU 2007 2008 2009 2010 Study period Double-Crested Cormorants on East Sand Island (Estuary) SR spring–summer Chinook salmon 0.8 (0.6–1.0) 1.7 (1.4–2.1) 3.3 (3.0–3.7) 2.6 (2.4–2.9) 2.2 (2.0–2.3) SR fall Chinook salmon 0.7 (0.3–1.3) 1.3 (1.1–1.5) 2.2 (1.9–2.5) 1.9 (1.7–2.2) 1.7 (1.6–1.9) SR sockeye salmon — — 2.7 (1.9–3.7) 1.2 (0.6–1.9) 2.1 (1.5–2.7) SR steelhead 1.7 (1.3–2.3) 7.3 (6.6–8.1) 8.1 (7.4–8.8) 3.7 (3.4–4.1) 5.5 (5.2–5.8) MCR steelhead 1.3 (0.7–2.1) 6.6 (5.2–8.1) 6.4 (5.4–7.5) 3.9 (3.4–4.5) 4.4 (4.0–4.9) WR spring Chinook salmon 0.4 (0.1–0.9) 1.6 (1.0–2.3) 0.7 (0.4–0.9) 1.8 (0.6–3.4) 0.9 (0.7–1.2) UCR spring Chinook salmon 1.3 (0.7–2.0) 1.7 (1.0–2.6) 1.3 (0.7–1.9) 1.6 (1.2–2.0) 1.5 (1.3–1.8) UCR steelhead 1.7 (1.1–2.4) 3.0 (2.2–3.9) 3.5 (2.6–4.6) 3.4 (2.9–3.8) 3.1 (2.8–3.4) Caspian Terns on East Sand Island SR spring–summer Chinook salmon 2.2 (2.0–2.4) 1.8 (1.5–2.0) 3.3 (3.0–3.6) 2.4 (2.2–2.6) 2.4 (2.3–2.6) SR fall Chinook salmon 2.3 (1.6–3.0) 1.3 (1.2–1.5) 1.4 (1.2–1.6) 0.5 (0.4–0.6) 1.1 (1.0–1.2) SR sockeye salmon — — 0.8 (0.4–1.3) 1.1 (0.5–1.7) 0.9 (0.6–1.3) SR steelhead 16.0 (15.0–17.2) 10.1 (9.6–10.7) 10.4 (9.9–10.9) 9.9 (9.3–10.6) 10.5 (10.2–10.9) MCR steelhead 12.9 (11.5–14.4) 9.8 (8.6–11.0) 10.1 (9.0–11.2) 8.7 (8.0–9.6) 9.7 (9.2–10.3) WR spring Chinook salmon 1.0 (0.5–1.5) 3.1 (2.4–3.8) 1.2 (0.9–1.5) 1.1 (0.2–2.3) 1.6 (1.4–1.9) UCR spring Chinook salmon 1.3 (0.8–1.8) 1.2 (0.7–1.8) 2.6 (1.8–3.3) 2.1 (1.7–2.5) 1.9 (1.6–2.2) UCR steelhead 11.1 (9.9–12.4) 11.7 (10.4–13.1) 14.0 (12.4–15.6) 9.8 (9.1–10.6) 10.7 (10.2–11.2) Downloaded by [Department Of Fisheries] at 20:33 25 September 2012 California Gulls and Ring-Billed Gulls on Miller Rocks Islands (MCR) SR spring–summer Chinook salmon 0.2 (0.1–0.2) 0.3 (0.3–0.4) 0.3 (0.3–0.4) 0.3 (0.2–0.3) 0.3 (0.2–0.3) SR fall Chinook salmon 0.5 (0.3–0.7) 0.4 (0.3–0.5) 0.6 (0.5–0.6) 0.1 (0.1–0.2) 0.4 (0.4–0.4) SR sockeye salmon — — 1.3 (0.8–1.9) 0.6 (0.2–1.1) 1.0 (0.7–1.4) SR steelhead 1.5 (1.2–1.8) 1.4 (1.2–1.6) 1.5 (1.3–1.7) 1.4 (1.2–1.6) 1.4 (1.3–1.5) UCR spring Chinook salmon 0.3 (0.2–0.5) 0.2 (0.1–0.4) 0.4 (0.2–0.6) 0.2 (0.1–0.3) 0.3 (0.2–0.3) UCR steelhead 1.3 (0.9–1.8) 1.1 (0.7–1.5) 1.0 (0.6–1.4) 1.1 (0.7–1.5) 1.1 (0.9–1.3) Caspian Terns on Blalock Islands (MCR) SR spring–summer Chinook salmon 0.1 (0.0–0.1) 0.1 (0.1–0.1) 0.2 (0.2–0.2) <0.1 0.1 (0.1–0.1) SR fall Chinook salmon 0.1 (0.0–0.1) 0.1 (0.0–0.1) <0.1 <0.1 <0.1 SR sockeye salmon — — <0.1 0.1 (0.0–0.3) <0.1 SR steelhead 0.6 (0.5–0.8) 0.5 (0.4–0.6) 0.4 (0.3–0.5) 0.6 (0.5–0.8) 0.5 (0.5–0.6) UCR spring Chinook salmon <0.1 <0.1 0.1 (0.0–0.2) <0.1 0.1 (0.0–0.1) UCR steelhead 0.7 (0.4–1.0) 0.4 (0.2–0.7) 0.3 (0.1–0.5) 0.7 (0.4–0.9) 0.5 (0.4–0.6) AVIAN PREDATION ON JUVENILE SALMONIDS 989

TABLE A.1. Continued.

Salmonid ESU 2007 2008 2009 2010 Study period Caspian Terns on Crescent Island (Near the Confluence) SR spring–summer Chinook salmon 0.3 (0.2–0.4) 0.6 (0.5–0.8) 1.0 (0.8–1.2) 0.3 (0.2–0.4) 0.6 (0.5–0.7) SR fall Chinook salmon 0.6 (0.3–1.0) 1.1 (0.9–1.3) 0.7 (0.6–0.9) 0.7 (0.6–0.8) 0.8 (0.7–0.9) SR sockeye salmon — 1.0 (0.2–2.0) 0.6 (0.3–1.0) 0.9 (0.2–1.8) 0.7 (0.5–1.0) SR steelhead 2.8 (2.5–3.1) 4.1 (3.8–4.6) 3.2 (2.9–3.5) 2.8 (2.4–3.2) 3.3 (3.1–3.5) UCR spring Chinook salmon — — <0.1 0.4 (0.0–1.0) 0.3 (0.0–0.6) UCR steelhead 1.7 (1.2–2.2) 2.0 (1.6–2.5) 1.6 (1.2–1.9) 1.2 (1.0–1.6) 1.6 (1.5–1.8) California Gulls and Ring-Billed Gulls on Crescent Island SR fall Chinook salmon <0.1 0.1 (0.1–0.1) 0.1 (0.1–0.1) <0.1 0.1 (0.0–0.1) SR sockeye salmon — 0.2 (0.0–0.5) 0.4 (0.1–0.7) <0.1 0.3 (0.1–0.5) SR steelhead 0.6 (0.4–0.8) 0.6 (0.5–0.7) 0.7 (0.6–0.8) 0.6 (0.4–0.7) 0.6 (0.6–0.7) UCR spring Chinook salmon — — <0.1 <0.1 <0.1 UCR steelhead 0.9 (0.5–1.3) 0.4 (0.2–0.5) 1.0 (0.7–1.3) 1.1 (0.8–1.4) 0.8 (0.7–1.0) American White Pelicans on Badger Island (Near the Confluence) SR spring–summer Chinook salmon <0.1 0.1 (0.0–0.1) 0.2 (0.1–0.2) <0.1 0.1 (0.0–0.1) SR fall Chinook salmon <0.1 0.1 (0.0–0.1) 0.1 (0.0–0.1) 0.1 (0.0–0.1) 0.1 (0.0–0.1) SR sockeye salmon — <0.1 <0.1 <0.1 <0.1 SR steelhead 0.3 (0.2–0.4) 0.2 (0.1–0.2) 0.3 (0.3–0.4) 0.3 (0.2–0.5) 0.3 (0.2–0.3) UCR spring Chinook salmon — — <0.1 <0.1 <0.1 UCR steelhead 0.1 (0.0–0.2) 0.1 (0.0–0.2) 0.3 (0.1–0.4) 0.1 (0.0–0.2) 0.1 (0.1–0.2) Double-Crested Cormorants on Foundation Island (Near the Confluence) SR spring–summer Chinook salmon 0.7 (0.6–0.9) 1.0 (0.9–1.1) 0.9 (0.7–1.1) 0.8 (0.5–1.0) 0.8 (0.7–0.9) SR fall Chinook salmon 0.9 (0.4–1.4) 0.4 (0.3–0.5) 0.5 (0.4–0.6) 0.4 (0.3–0.5) 0.4 (0.4–0.5) SR sockeye salmon — 1.1 (0.3–2.0) 2.1 (1.4–2.7) 1.7 (0.5–3.1) 1.8 (1.3–2.3) SR steelhead 2.3 (2.0–2.7) 2.3 (2.1–2.6) 1.7 (1.6–1.9) 1.3 (1.0–1.6) 1.9 (1.8–2.1) UCR spring Chinook salmon — — <0.1 <0.1 <0.1 UCR steelhead <0.1 0.1 (0.1–0.3) 0.1 (0.0–0.2) 0.1 (0.0–0.2) 0.1 (0.1–0.1) Caspian Terns on Goose Island (Potholes Reservoir, Near the Confluence) SR spring–summer Chinook salmon <0.1 <0.1 <0.1 <0.1 <0.1 SR fall Chinook salmon 0.1 (0.0–0.6) <0.1 <0.1 <0.1 <0.1 SR sockeye salmon — 0.2 (0.0–0.6) <0.1 <0.1 <0.1 SR steelhead <0.1 <0.1 <0.1 <0.1 <0.1 UCR spring Chinook salmon — — 3.6 (1.6–6.1) 1.0 (0.2–2.0) 2.1 (1.1–3.4) UCR steelhead 9.1 (6.3–14.0) 7.5 (6.5–8.5) 15.7 (13.6–18.2) 9.6 (8.3–11.2) 10.6 (9.7–11.6) Downloaded by [Department Of Fisheries] at 20:33 25 September 2012 This article was downloaded by: [Department Of Fisheries] On: 25 September 2012, At: 20:34 Publisher: Taylor & Francis Informa Ltd Registered in England and Wales Registered Number: 1072954 Registered office: Mortimer House, 37-41 Mortimer Street, London W1T 3JH, UK

Transactions of the American Fisheries Society Publication details, including instructions for authors and subscription information: http://www.tandfonline.com/loi/utaf20 Coupling Fish Community Structure with Instream Flow and Habitat Connectivity between Two Hydrologically Extreme Years Charles E. Stanley a b , Jason M. Taylor a c & Ryan S. King a a Center for Reservoir and Aquatic Systems Research, Department of Biology, Baylor University, One Bear Place, 97388, Waco, Texas, 76798-7388, USA b Lassen Volcanic National Park, Post Office Box 100, Mineral, California, 96063, USA c Department of Natural Resources, New York Cooperative Fish and Wildlife Research Unit, Cornell University, Ithaca, New York, 14853, USA Version of record first published: 26 Jun 2012.

To cite this article: Charles E. Stanley, Jason M. Taylor & Ryan S. King (2012): Coupling Fish Community Structure with Instream Flow and Habitat Connectivity between Two Hydrologically Extreme Years, Transactions of the American Fisheries Society, 141:4, 1000-1015 To link to this article: http://dx.doi.org/10.1080/00028487.2012.675893

PLEASE SCROLL DOWN FOR ARTICLE

Full terms and conditions of use: http://www.tandfonline.com/page/terms-and-conditions 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. The publisher does not give any warranty express or implied or make any representation that the contents will be complete or accurate or up to date. The accuracy of any instructions, formulae, and drug doses should be independently verified with primary sources. The publisher shall not be liable for any loss, actions, claims, proceedings, demand, or costs or damages whatsoever or howsoever caused arising directly or indirectly in connection with or arising out of the use of this material. Transactions of the American Fisheries Society 141:1000–1015, 2012 C American Fisheries Society 2012 ISSN: 0002-8487 print / 1548-8659 online DOI: 10.1080/00028487.2012.675893

ARTICLE

Coupling Fish Community Structure with Instream Flow and Habitat Connectivity between Two Hydrologically Extreme Years

Charles E. Stanley,1 Jason M. Taylor,2 and Ryan S. King* Center for Reservoir and Aquatic Systems Research, Department of Biology, Baylor University, One Bear Place, 97388, Waco, Texas 76798-7388, USA

Abstract Hydrologic variability and instream habitat connectivity play fundamental roles in structuring fish communities in lotic ecosystems. We collected fish assemblage and physical habitat data from 28 central Texas streams during the summers of 2006 (a drought year with minimal summer precipitation and low stream flow) and 2007 (an exceptionally wet year with periodic flooding in spring and sustained high flows throughout summer). We evaluated the correspondence between the magnitude of physical habitat and fish community composition change in stream reaches sampled in these two contrasting years using ordination, successional vector analysis, and indicator species analysis. In 2006, streams characterized by disconnected pools had different fish community structure and habitat characteristics than streams that had habitats connected by flowing water. The amount of interannual change in both fish community structure and habitat characteristics was greatest between streams that had disconnected pools in 2006 and their paired samples in 2007. Indicator species analysis identified species that had affinities to disconnected habitats during 2006, which included opportunistic life history strategists typical of temporary waters (western mosquitofish Gambusia affinis and blackstripe topminnow Fundulus notatus) and equilibrium strategists that rely on stable pool habitats for nesting (longear sunfish Lepomis megalotis and largemouth bass Micropterus salmoides). Conversely, indicator species of connected riffle–pool habitat included fluvial specialists (central stoneroller Campostoma anomalum, spotted bass Micropterus punctulatus, and bullhead minnow Pimephales vigilax). In summer 2007, the numbers of most species of fish declined markedly compared with 2006. Community structure between previously disconnected and connected stream types was also highly variable in 2007. However, strong recruitment of juveniles following spring flooding and sustained high summer flow significantly increased the frequency and abundance of two periodical strategists, channel catfish Ictalurus punctatus and flathead catfish Pylodictus olivaris in both types of streams in 2007. These findings provide important insights into how individual species’ life history strategies influence the response of fish community structure to extreme hydrologic events, which are likely to increase in frequency in many parts of the world due to climate change. Downloaded by [Department Of Fisheries] at 20:34 25 September 2012

Variation in discharge inﬂuences nearly all aspects of river systems (Allan 1995). Strong associations between both the ecosystem functioning (Richter et al. 2003). Discharge affects functional and taxonomic composition of ﬁsh communities and aquatic life and its related habitat, nutrient cycling, sediment hydrologic regimes occur at regional scales (Poff and Allan transport, water temperature, riverbank stability, groundwater 1995). Variability in discharge can lead to rapid changes in recharge, and many other ecological factors in freshwater lotic environmental conditions, such as temperature, turbidity, and

*Corresponding author: ryan s [email protected] 1Present address: Lassen Volcanic National Park, Post Ofﬁce Box 100, Mineral, California 96063, USA. 2Present address: Department of Natural Resources, New York Cooperative Fish and Wildlife Research Unit, Cornell University, Ithaca, New York 14853, USA. Received February 9, 2011; accepted March 1, 2012 Published online June 26, 2012

1000 FISH COMMUNITY STRUCTURE 1001

salinity (Ostrand and Wilde 2002), and these instream environ- streams. These stream reaches differed greatly between sam- mental changes can limit the spatial and temporal distribution pling events in their discharge and physical habitat variables of biota in lotic ecosystems. due to interannual differences in precipitation. The tempera- Fish community structure composition in streams is regu- tures and precipitation amounts immediately before and during lated by abiotic (density-independent) factors, such as discharge our index sampling periods differed markedly between 2006 and habitat connectivity, and local biotic (density-dependent) and 2007; the summer of 2006 was one of the hottest summers factors, such as predation and competition (Taylor 1997). The on record and was classified as an extreme drought year for relative importance of these two factors in structuring fish as- the study area (Palmer hydrological drought index; http://www. semblages varies longitudinally in streams (Ostrand and Wilde ncdc.noaa.gov/sotc/drought/2006/13), whereas 2007 was the 2002). Smaller streams in the Great Plains are prone to dis- wettest summer on record for the state of Texas, with a spring turbance and intermittence, and fish assemblage structure is characterized by extreme flooding and summer base flows well thought to be controlled primarily by abiotic factors and the fre- above median flows throughout our study area (http://www. quency of disturbances such as flood and drought (Echelle et al. ncdc.noaa.gov/sotc/drought/2007/13). 1972; Matthews and Styron 1981; Meador and Matthews 1992; Historically, as flow recedes during summer months, streams Taylor et al. 1993; Taylor 1997; Herbert and Gelwick 2003). As in the middle Brazos River and upper Trinity River basins be- a result, the predominant fish species in these stream types have come intermittent and are characterized as a continuum of dis- evolved to be physiologically tolerant of variable environmental connected, often perennial, pools within the streambed (Echelle conditions, able to move in response to changing conditions, or et al. 1972; Wilde and Ostrand 1999). This pattern was observed able to rapidly recolonize areas of local extinction caused by during the 2006 collection period in 16 of the 28 stream reaches intermittence (Winemiller 1989; Fausch and Bramblett 1991). sampled but in none of the sites sampled in 2007. Stream reaches While many previous studies have examined the longitudinal were classified into a priori groups based upon this habitat con- change of abiotic and biotic variables along the river contin- nectivity criterion, as determined by discharge measurements uum, disproportionately fewer studies have examined the effect (Table A.1 in the appendix) and visual evaluation of pool con- of drying within streams that leads to differing assemblages in nectivity. This variation presented a unique opportunity to exam- intermittent perennial pools during drought disturbance events. ine the relationship between extremes of hydrologic variability As perennial pools are formed, differing life history strategies and fish assemblages in central Texas streams. We used this and variation in tolerance of increasingly stressful abiotic envi- unique data set that spans two sequential years of contrasting ronmental conditions may contribute to species persistence in hydrological conditions to test the following hypotheses: these habitats. For example, opportunistic (O) strategists include 1. Extreme hydrologic conditions between years and associated small-bodied fishes with early maturation, low fecundity per effects on stream connectivity would influence habitat and spawning event, and low juvenile survivorship (Winemiller and fish assemblage structure; Rose 1992). This strategy, combined with tolerance of high tem- 2. Changes in fish assemblage structure between drought and peratures and low dissolved oxygen, may confer an advantage flood conditions would vary with habitat connectivity during during extreme low-flow conditions. Lack of connectivity pre- drought conditions; and vents the immigration or emigration of fish, and the continuing 3. Based on differences in life history strategies and habitat reduction of pool volume intensifies biotic interactions, poten- requirements, we would observe species that were charac- tially leading to local extinctions (Taylor 1997). However, stable teristic of species assemblage sorting among the extreme low-flow conditions and higher biotic interactions may benefit hydrologic and connectivity conditions observed during our equilibrium (E) strategists, small to medium-bodied fishes with study period. Downloaded by [Department Of Fisheries] at 20:34 25 September 2012 moderate maturation age, low fecundity, and high parental in- vestment (Winemiller and Rose 1992). For example, at lower, more stable flows, centrarchids nest success is more associated METHODS with biotic interactions and positively related to guarding male Site selection.—The Trinity and Brazos River drainage basins size (Noltie and Keenleyside 1986; Jennings and Philipp 1994; cover a large portion of central Texas. These drainage basins Dauwalter and Fisher 2007). In contrast, years with unusually span several ecoregions, including the Cross Timbers ecore- high-flow events may displace or reset fish assemblage structure gion, which encompasses the entirety of our study area. This and favor periodic (P) strategists, large-bodied fishes with late ecoregion is a transitional zone between the once-prairie, now maturation, high fecundity per spawning event, and no parental winter wheat–growing regions to the west and the forested low care (Winemiller and Rose 1992). This strategy, combined with mountains or hills of eastern Texas. A mosaic of forest, wood- high dispersal ability, may allow certain species to take ad- land, savanna, and prairie characterizes this ecoregion (Griffith vantage of new feeding and spawning habitats after high-flow et al. 2004). events. Stream reaches were selected from wadeable streams and We studied the relative importance of temporal variation in river tributaries throughout this ecoregion to provide broad ge- the physical habitat and fish assemblages of 28 central Texas ographic coverage and a range of landscape features (Figure 1; 1002 STANLEY ET AL. Downloaded by [Department Of Fisheries] at 20:34 25 September 2012

FIGURE 1. Map of the study area showing the locations of the stream reaches and watersheds within the Brazos and Trinity River basins and the Cross Timbers Level III Ecoregion (Grifﬁth et al. 2004). An explanation of the site codes is provided in Table A.1. FISH COMMUNITY STRUCTURE 1003

Pease et al. 2011). Over 50% of the land area in the Cross Tim- sampling) and discharge less than 0.001 m3/s were classified bers ecoregion lies within our study catchments; thus our sites as having disconnected instream habitat (2006 D). All stream spanned a substantial range of conditions and were likely rep- reaches sampled in 2007 had connected habitat throughout the resentative of other streams not included in our study. Selected stream reaches and discharge greater than 0.001 m3/s. The reaches were sampled once in 2006 and once in 2007 during an stream reaches sampled in 2007 were classified by whether index sampling period (1 May–15 October) defined by the Texas habitat was connected (2007 PC) or disconnected (2007 PD) Commission on Environmental Quality (TCEQ 2003). Thirty- during the 2006 sampling period. eight stream reaches were sampled in July and August during We conducted principal components analysis (PCA) to iden- 2006; however, due to high stream discharges during the index tify correlations among habitat variables and to reduce the sampling period only 28 streams were sampled during the 2007 shared variance among variables to a smaller number of syn- field season, primarily in August and September. Because this thetic axes for evaluating changes in habitat among interannual study is concerned with interannual variation, it is limited to the groups. Prior to the analysis, we reduced the number of vari- 28 sites that were sampled during both years of data collection. ables from an initial list of 49 to 22 by removing variables Instream physical habitat and fish sampling.—Stream phys- that were redundant within groups of similar variables (e.g., ical habitat and fish data collection during the summer of 2006 multiple measures of large woody debris). We further screened followed Surface Water Quality Monitoring Procedures proto- variables to meet the statistical assumption of approximately lin- cols for habitat and fish assessments (TCEQ 2003; Pease et al. ear relationships; in cases in which variables did not meet this 2011). Stream physical habitat and fish assemblages were as- assumption, we applied transformations (square root or loga- sessed within a sampling reach (40 times the average stream rithmic) to those variables (only) to achieve skew and kurtosis width but not less than 150 m or greater than 500 m in total < |1|. Principal components analysis was conducted on a cor- length). At each site, habitat variables were measured at five to relation matrix of stream reaches (28 sites × 2 years) and the six equally spaced transects (depending on reach total length) reduced set of 22 habitat variables. We assessed the significance or across the entire reach to characterize instream channel char- of each principal component by comparing eigenvalues obtained acteristics, stream morphology, and riparian environments. At from 1,000 random permutations of the data with the observed sites experiencing low-flow conditions in which less than 50% of eigenvalue (McCune and Grace 2002); PCs with eigenvalues the reach is covered in perennial pools, transects were adjusted significantly different than random (P < 0.05) were retained for to best characterize the available habitat in pools and additional further analysis and interpretation. measures of maximum length, width, and depth were recorded We estimated gradients in fish community composition for each pool, with the overall objective being to characterize among streams between years using nonmetric multidimen- available pool habitat. sional scaling (MDS). This is an effective ordination for eco- All discrete habitat and cover types within the entire reach logical species composition because it avoids the assumption were electrofished with a backpack electroshocker (Smith-Root of linear relationships necessary for other statistical techniques, Co., Vancouver, Washington). Actual shocking time was at least such as PCA (Clarke 1993; McCune and Grace 2002). Prior to 900 s but varied based on reach length and habitat complexity. the analysis, species abundances (the total number of fish cap- Six straight seine hauls covering a cumulative distance of at least tured per species at each site) were log10(x + 1) transformed in 60 m were conducted across all habitats, especially areas such order to reduce the influence of highly abundant, small-bodied as deep pools or shallow riffles where electrofishing may not species on the ordination. The MDS was conducted on a Bray- be as effective. In shallow riffle areas the seine net was placed Curtis dissimilarity matrix of 56 sample units (28 streams × downstream and the riffle habitat was disturbed by aggressively 2 years) by 34 species abundances. We used stress, a measure

Downloaded by [Department Of Fisheries] at 20:34 25 September 2012 agitating the 10-m riffle area by kicking substrate within the fit between the ordination and the original data, to select the riffle to drive fish into the net downstream. Deep pools were appropriate number of dimensions for the final ordination fol- sampled with 9.14-m or 4.57-m × 1.83-m × 0.64-cm-mesh lowing McCune and Grace (2002). We excluded fish collected seines, whereas riffles, runs, and small pools were sampled with in <5% of the sampling units because these were too infrequent 2-m or 5-m-long × 4-mm-mesh seines. All collected fish were to contribute useful information to the analysis (McCune and identified (Thomas et al. 2007; Hubbs et al. 2008), separated Grace 2002). The PCA and MDS analyses were conducted in into juvenile and adult age-classes, and counted in the field. At PC-Ord version 5.20 (McCune and Medford 2006). least two voucher specimens for each species and all unknown We related the habitat variables used in the final PCA to the species were preserved in 10% buffered formalin and returned MDS ordination of fish community composition using rotational to the laboratory for identification. vector fitting (Faith and Norris 1989) as implemented by the Data analyses.—Streams that had connected habitat (allow- envfit function in the vegan package for R 2.12 (Oksanen et al. ing movement of fish throughout the reach during the 2006 2011). Vector fitting determined the direction and magnitude of sampling) and total stream discharge greater than 0.001 m3/s the maximum correlation between predictor variables and the were classified as having connected instream habitat (2006 configuration sample units in the ordination space and aided C). Streams that had disconnected perennial pools (prevent- in the interpretation of gradients in fish species composition. ing movement of fish throughout the reach during the 2006 The probability of obtaining an equal or larger correlation from 1004 STANLEY ET AL.

random data (P) is estimated using 1,000 random permutations. tailed Student’s t-test for unequal variances was used to deter- We deemed correlations with P < 0.05 to be different from mine whether the mean vector lengths for interannually paired random. connected and disconnected sites were significantly different, a We tested for differences in habitat structure and species direct measure of the magnitude of change between years. Also, composition among the four hydrological groups and the dis- coordinates for successional vector diagrams were used as the persion of each group within ordination space using multivari- basis for additional MRPP comparisons. This MRPP analysis ate response permutation procedure (MRPP) and PERMuta- addresses a different question than the previous one, in that it tional analysis of multivariate DISPersion (PERMDISP). We is concerned with the differences in magnitude and direction of used MRPP (Mielke 1984) as implemented in PC-Ord 5.20 to the interannual change in habitat condition and species compo- describe differences in habitat and community structure using sition. Euclidean distances among sites in PCA and MDS ordination Finally, we used indicator species analysis (ISA; Dufreneˆ and space, respectively. This technique is a multivariate analog to a Legendre 1997) to test for affinities of different species to our Student’s t-test of differences between groups and determines hydrology and habitat connectivity groupings. This analysis as- the significance of separations of groups in ordination space and signs an indicator value (IndVal)to each taxon by calculating the thus significant differences in either fish community structure or product of the relative frequency (the percentage occurrence of a habitat characteristics between groups of streams. The MRPP taxon among sample units in each group) and the relative abun- was performed on coordinates from PCA and MDS ordination dance (the percentage of the total abundance of a taxon in each plots. Groups were weighted using a natural weighting, as rec- group) of each species to a group. The probability of achieving ommended by Mielke (1984). Pairwise comparisons were made an equal or larger IndVal value among groups (P) was estimated between the locations of group centroids in both PCA and MDS using 999 random permutations of the original data (Dufreneˆ and ordination space to determine the significance of differences in Legendre 1997). Species with significantly (P ≤ 0.05) higher group habitat or community composition. IndVals for a given group are likely to found in other regional We used the program PERMDISP2 (Anderson 2006) to de- streams with similar environmental conditions and suggest a hy- termine whether the dispersions (degree of variability among drological effect on that species. Indicator species analysis was sites) of each group around their group centroid were signifi- used to determine group indicator values for all species included cantly different from one another. This program is useful in de- in analyses. termining whether the separations between groups indicated by MRPP are affected by differences in the dispersions of groups. RESULTS MRPP distance comparisons can be sensitive to between-group differences in dispersion, and this technique is a multivariate Physical Habitat Analyses analog to Levene’s test on the Euclidean distances of indi- Principal components 1 and 2 had eigenvalues significantly vidual observations and their group centroid (Anderson 2006). different from random and accounted for 54.9% of the cumula- Anderson et al. (2006) also suggest that the overall dispersion tive variance among the 22 habitat variables (Table 1; Figure 2). of a group measured as average distance (or dissimilarity) from Eigenvalues for PCs 3 and higher were not significantly different an individual sampling unit to the group centroid may be used from random (P > 0.05). Reaches with increasingly positive val- as a multivariate measure of community beta diversity. We hy- ues along PC1 were higher-discharge streams with greater water pothesized that disconnected sites would have greater dispersion depth, diversity of depths and current speeds, and riffles. Nega- (beta diversity) due to increased biotic interactions, particularly tive values along PC1 were indicative of higher abundances of predation, within disconnected perennial pools, causing a de- macrophytes and filamentous algae and low discharge. Principal

Downloaded by [Department Of Fisheries] at 20:34 25 September 2012 coupling of fish species composition from habitat variables. component 2 was primarily a substrate variable axis with higher We interpreted the patterns in change of habitat variables and values corresponding to greater cover of sand, silt, and mud subspecies composition between annual sampling events for both strates; more woody debris; and higher substrate embeddedness. PCA and MDS ordinations using successional vector diagrams. Lower values corresponded to habitats that had less surface wa- These diagrams show the direction and magnitude of change ter that was slower moving, with erosional banks resulting in in both fish community composition and habitat characteristics muddy substrates. between interannual pairs of sites. Each interannual pair of sites The 2006 D group centroid was significantly separated from were translated to origin, meaning that all 28 sites sampled in all other groups in PCA space according to MRPP pairwise 2006 are located at the point (0, 0) of the two-dimensional comparisons (Figure 2; Table 2), with the greatest separation successional vector diagram (McCune and Grace 2002). from 2007 PD. The location of this group in PCA space sug- The corresponding paired 2007 sites were graphed in two- gests that the habitat conditions among streams indicative of this dimensional space around the origin and connected with vector group are characterized by comparatively high macrophyte and lines so that the direction and magnitude of the interannual filamentous algae abundance, bedrock substrates, and (by def- change in ordination space were visualized. Euclidean distance inition) no surface flow between disconnected perennial pools. was used to calculate the vector lengths between interannual The 2006 C group was not significantly different from either pairs of sites from the successional vector diagram. A one- 2007 group. These remaining groups’ positions in PCA space FISH COMMUNITY STRUCTURE 1005

TABLE 1. Results of principal components analysis on 22 stream habitat variables measured in 2006 and 2007. Only the ﬁrst two principal components (PCs) were deemed signiﬁcant following 1,000 random permutations (P < 0.05). Variables with the strongest loadings (r > 0.5) with each PC are shown in bold italics.

Variable code Variable PC 1 PC 2 BEDROCK Bedrock (%) –0.328 –0.835 BOULDER Boulder (%) –0.039 –0.508 CANOPY Canopy (%) –0.029 0.379 COBBLE Cobble (%) –0.115 –0.42 DISCHARG Discharge (m3/s) 0.883 0.003 EMBEDDED Embeddedness (%) 0.141 0.691 FILA ALG Filamentous algae abundance, cover class –0.617 –0.2 GRAVEL Gravel (%) 0.057 0.166 LWD Large woody debris abundance, cover class 0.348 0.514 MCRPH AB Macrophyte abundance, cover class –0.704 –0.32 MEANDEPT Water depth, mean (m) 0.742 0.303 MUDSILT Mud + silt (%) –0.026 0.663 NO RIFF Number of rifﬂes, reach 0.612 –0.283 POOL DEP Pool depth, maximum (cm) 0.387 0.478 POOL WID Pool width, maximum (m) 0.084 –0.449 ROOTS Submerged root abundance, cover class 0.357 0.015 SAND Sand (%) 0.449 0.689 SWD Small woody debris, cover class 0.129 0.693 THAL DEP Thalweg depth, mean (cm) 0.758 0.356 UNDERCUT Undercut banks (% of reach) 0.39 0.088 VELDEPTH Velocity–depth diversity (0–20) 0.625 –0.204 WETWIDTH Wetted width (m) 0.638 –0.304 Cumulative variance explained: 34.90% 54.8%

suggest conditions of higher discharge, mean depth, and pool brook silverside Labidesthes sicculus, redear sunﬁsh Lepomis depths than 2006 D sites, along with smaller substrate particle microlophus, ribbon shiner Lythrurus fumeus, spotted sucker sizes (sand, mud, and silt substrates). Minytrema melanops, golden shiner Notemigonus crysoleucas, The PERMDISP analysis of dispersion showed no signif- tadpole madtom Noturus gyrinus, freckled madtom Noturus icant differences between any pairwise comparison of groups nocturnus, and Texas logperch Percina carbonaria were also around their centroid in PCA habitat space, suggesting that the collected, but these species were excluded from analyses be- differences in MRPP were due to differences in habitat structure cause they were collected from less than 5% of the 56 sampling among groups alone. The PERMDISP and MRPP analyses con- units (McCune and Grace 2002). ducted on the position of stream sites in the PCA successional There was a large decline in the total number of ﬁshes col- vector plot revealed no differences in dispersion between inter- lected from 2006 to 2007 associated with the reduction in wetted

Downloaded by [Department Of Fisheries] at 20:34 25 September 2012 annual paired hydrologic groupings but a significant separation stream habitat in late 2006 followed by the sharp increases in between 2007 PD and 2007 PC instream habitat. This separa- discharge in response to heavy precipitation in 2007 (Table 4). tion can be interpreted as disconnected sites having a greater This decline was most notable in many of the species collected amount of change in habitat condition from 2006 to 2007 than from the families Cyprinidae and Centrarchidae, including cen- connected sites. Additionally, the successional vectors of 2007 tral stoneroller and the two bass species, large-mouthed bass PD streams were significantly longer than those of 2007 PC and spotted bass. There was also a pronounced reduction in streams (Figure 3), indicating a greater amount of interannual the numbers of blackstripe topminnow Fundulus notatus and change among streams with previously disconnected habitats. western mosquitofish Gambusia affinis. The few species that increased in abundance from 2006 to 2007 include green sun- Fish Community Analyses fish Lepomis cyanellus, channel catfish Ictalurus punctatus, and The fish that were included in our analyses totaled 38,897 flathead catfish Pylodictis olivaris. individuals from 34 species, 23 genera, 12 families, and 7 orders The MDS ordination of fish communities revealed a strong for the 2006 and 2007 collections (Table 4, A.2). A small num- gradient in fish community structure along axis 1 that was pos- ber of individuals of black bullhead Ameiurus melas, Mexican itively related to stream discharge, the number of riffles, stream tetra Astyanax mexicanus, smallmouth buffalo Ictiobus bubalus, width and depth, and velocity–depth diversity and negatively 1006 STANLEY ET AL.

FIGURE 2. Conﬁguration of stream reaches in a two-dimensional PCA representation of the habitat variable correlation matrix. The arrows show the directions Downloaded by [Department Of Fisheries] at 20:34 25 September 2012 and magnitudes of the loadings (r > 0.5) of habitat variables with principal components 1 and 2. See Table 1 for an explanation of the variables. Abbreviations are as follows: 2006-D = disconnected instream habitat in 2006 sampling, 2006-C = connected instream habitat in 2006 sampling, 2007-PD = disconnected instream habitat the previous year in 2007 sampling, and PC = connected instream habitat the previous year in 2007 sampling.

related to filamentous algae and macrophytes (Figure 4A; MDS ordination (Figure 5) implied that, consistent with the rela- Table 3). Streams were also dispersed widely along axis 2, and tive differences in habitat between years, the fish communities in this gradient was related only to substrate variables (Figure 4A; 2006 D sites had significantly greater magnitude of change from Table 3). 2006 to 2007 than those in 2006 C sites (Figure 5). However, Streams with negative scores on MDS 1 were almost exclu- the comparison of overall change in direction and magnitude sively sampled in 2006, whereas those with positive scores were in community structure suggested no predictable effect on the sampled in 2007 (Figure 4A). The MRPP pairwise comparisons pattern of community succession between years (P = 0.399; indicated differences in species composition between 2006 and Table 2), suggesting that high variation in direction of change 2007 but not between habitat connectivity groupings for either among streams was common to both groups despite a greater year (Table 2). However, successional vectors derived from the magnitude of change in the 2007 PD group. PERMDISP also FISH COMMUNITY STRUCTURE 1007

TABLE 2. Multiresponse permutation procedure (MRPP) pairwise comparisons between groups and mean distances of observations to the group centroid resulting from principal components analysis (PCA [habitat]) and nonmetric multidimensional scaling (MDS [ﬁsh]) analyses. Abbreviations are as follows: C = connected, D = disconnected, PC = previously connected, PD = previously disconnected, and t = Student’s t-statistic.

Pairwise comparison t α P Mean distance to centroid PCA (α = 0.19, P ≤ 0.0001) 2006 D vs. 2007 PD –12.74 0.24 ≤0.0001 2006 D 5.91 2006 D vs. 2006 C –5.04 0.11 0.0002 2006 C 4.88 2006 D vs. 2007 PC –11.42 0.25 ≤0.0001 2007 PD 4.66 2007 PD vs. 2006 C –1.97 0.04 0.05 2007 PC 4.74 2007 PD vs. 2007 PC 0.83 –0.02 0.84 2006 C vs. 2007 PC –2.31 0.06 0.03 MDS (α = 0.08, P ≤ 0.0001) 2006 D vs. 2007 PD –8.25 0.12 ≤0.0001 2006 D 1.17 2006 D vs. 2006 C 0.78 –0.01 0.78 2006 C 1.21 2006 D vs. 2007 PC –5.47 0.09 ≤0.0001 2006 PD 1.22 2007 PD vs. 2006 C –4.47 0.07 ≤0.0001 2007 PC 1.29 2007 PD vs. 2007 PC 1.21 –0.02 0.98 2006 C vs. 2007 PC –2.29 0.04 0.03

showed no significant differences in the dispersion of hydrologic sunfish, largemouth bass, blackstripe topminnow, and Lepomis groups in ordination space. spp. juveniles were significant indicator species for 2006 D Indicator species analysis conducted on both habitat connec- (Table 5; Figure 4B). Bullnose minnow, central stoneroller, and tivity (2006 D, 2006 C, 2006 PD, and 2006 PC) and interannual- spotted bass were significant indicator species of the 2006 C only groups (2006 versus 2007, regardless of connectivity) group. Channel catfish and flathead catfish were stronger in- revealed interesting patterns in fish community composition dicators of interannual differences than connectivity groupings (Table 5). Western mosquitofish was a strong indicator species (Table 5). Juveniles of both species were significantly more for the group 2006 D, further illustrated by its negative position abundant in 2007 than 2006, a result also revealed by their pos- along MDS 1 (Figure 4B). The ISA also indicated that longear itive scores on MDS 1 (Figure 4B). Downloaded by [Department Of Fisheries] at 20:34 25 September 2012

FIGURE 3. Diagram of successional vectors translated to the origin (2006) describing the magnitude and direction of change in two-dimensional PCA space of sites between 2006 and 2007 (left panel) and comparison of mean PCA vector lengths between previously disconnected (PD) and previously connected (PC) sites (right panel). 1008 STANLEY ET AL. Downloaded by [Department Of Fisheries] at 20:34 25 September 2012

FIGURE 4. Panel (A) shows a configuration of stream reaches in a two-dimensional NMS representation of the fish species composition dissimilarity matrix. The arrows show the directions and magnitudes of significant correlations between habitat variables and the ordination of fish species composition estimated using rotational vector fitting. See Table 3 for details. Panel (B) shows the weighted-average centroids of the 34 fish species in the same ordination space as in (A). FISH COMMUNITY STRUCTURE 1009

FIGURE 5. Diagram of successional vectors translated to the origin (2006) describing magnitude and direction of change in two-dimensional NMS space of sites between 2006 and 2007 (left panel) and comparison of mean MDS vector lengths between previously disconnected (PD) and previously connected (PC) sites (right panel).

DISCUSSION in 2007 with low fish abundance (presumably due to displace- Differences in stream discharge between years and instream ment by floods) favoring rapidly colonizing species of fish that habitat connectivity within years had a significant effect on fish could readily utilize new habitats. The two-thirds reduction in species composition in our study streams. Extended lack of the total number of individuals captured between the 2 years

Downloaded by [Department Of Fisheries] at 20:34 25 September 2012 precipitation in 2006 resulted in the drying of streams across may have partially resulted from reduced sampling efficiency central Texas and the development of disconnected perennial in 2007. Nevertheless, the shifts in fish species composition pools at many of our study sites. In contrast, hydrologic condi- can be explained by ecological relationships between species tions in 2007 were characterized by periodic extreme flooding that drove the observed shifts in community structure, along and sustained high flows throughout the summer across all 28 with changes in hydrologic conditions and associated habitat sites. Extended periods of precipitation prior to our sampling re- changes between the 2 years. sulted in flood disturbance conditions. Flooding causes massive While there were overall differences in habitat and fish debris flows, rearrangement of large instream substrates, in- community structure related to hydrologic differences between stream algal scouring, loss of riparian vegetation, and alteration years, the degree of change in habitat structure and fish of stream channel morphology. These events probably resulted assemblage structure between drought and flood conditions in the displacement of resident fishes and macroinvertebrate varied with habitat connectivity during drought conditions. prey items. The observed differences in fish community struc- There were several significant indicator species that help explain ture were likely the result of two factors: (1) low-flow conditions the differences in fish species composition between connectivity and disconnectivity in 2006 that created abiotic conditions fa- groups in 2006 as well as the overall differences between years. voring opportunistic species and (2) postdisturbance habitats It is helpful in synthesizing these results to consider the three 1010 STANLEY ET AL.

TABLE 3. Habitat variables signiﬁcantly (P < 0.05) related to the nonmetric multidimensional scaling (MDS) ordination of ﬁsh community composition in 2006 and 2007 (Figure 4A, B). The magnitudes of the correlations (r) corre- sponds to the relative lengths of the arrows in the ordination diagrams, whereas the direction of each vector is given by its relative loading on each axis. Pr(>r) = the probability of attaining an equal or larger correlation by randomization of the data. See Table 1 for an explanation of the variable codes.

Variable code MDS 1 MDS 2 r Pr(>r) BEDROCK –0.20 –0.98 0.51 0.001 CANOPY 0.41 0.91 0.39 0.012 DISCHARG 1.00 0.01 0.78 0.001 EMBEDDED –0.10 0.99 0.48 0.005 FILA ALG –0.89 –0.46 0.44 0.001 LWD 0.38 0.92 0.35 0.039 MCRPH AB –0.78 –0.63 0.52 0.001 MEANDEPT 0.95 0.30 0.43 0.005 MUDSILT –0.13 0.99 0.51 0.001 NO RIFF 0.96 –0.28 0.42 0.003 FIGURE 6. Conceptual model relating the life history strategies employed by POOL DEP 0.74 0.67 0.39 0.014 significant indicator species in the current study to the flow regimes encountered POOL WID 0.08 –1.00 0.35 0.037 in central Texas streams, including periods of disconnectivity during droughts SAND 0.57 0.82 0.40 0.004 and extreme high-flow events associated with sustained precipitation during extremely wet years. SWD –0.01 1.00 0.42 0.007 THAL DEP 0.93 0.37 0.43 0.005 VELDEPTH 0.99 –0.11 0.40 0.011 ing periods, whereas biological factors, which E strategists cope WETWIDTH 0.89 –0.46 0.43 0.004 with through high parental care, are more common during low- flow years (Jennings and Philipp 1994; Dauwalter and Fisher 2007). Low-flow conditions may confer some advantage on E–P basic life history strategies that can be related to stream hydrol- strategists, as Dauwalter and Fisher (2007) observed the highest ogy at the continental scale (Winemiller and Rose 1992; Olden reported densities of smallmouth bass nests ever reported dur- and Kennard 2010): periodic (P), opportunistic (O), and equilib- ing an exceptionally low-water year. Periodic strategists are not rium (E). Using life history strategy tables for Texas fishes from well adapted to the variable hydrologic and habitat conditions Hoeinghaus et al. (2007), our significant indicator species fall associated with central Texas streams, and central stonerollers into five life history strategy groups, namely, P, O, E, and two were the only P species identified by our indicator species anal- intermediate groupings (O–P and E–P; Table 5). The response ysis (Table 5). Central stonerollers were an indicator species for of these five life history strategies can be conceptually applied the 2006 C group, suggesting that they do prefer less variable to the extreme flow conditions observed during our study. conditions. However, central stonerollers are more likely O–P The natural ranges of variation in the summer low-flow con- strategists, as Spranza and Stanley (2000) reported this species ditions experienced by central Texas fishes probably favor a mix to exhibit faster growth rates and shifts to more opportunistic of species that employ equilibrium, periodic, or equilibrium– reproductive strategies in stream segments with greater environ-

Downloaded by [Department Of Fisheries] at 20:34 25 September 2012 periodic life history strategies, with the dominance of one life mental variation than in larger downstream segments exhibiting history strategy or species within a life history strategy varying more stable flows. with flow conditions and the relative influence of abiotic versus Extreme hydrologic conditions such as the low stream dis- biotic control on species composition (Figure 6). For example, charge and associated disconnectivity experienced in 2006 within equilibrium strategists, stable low flows and even discon- create abiotic conditions that favor opportunistic life history nected pools likely favor longear sunfish due to the availability strategies such as those employed by western mosquitofish of spawning habitats in shallow pools, whereas stable moder- (Figure 6). Western mosquitofish were a significant indicator ate flows may favor green sunfish by watering the undercut of the 2006 DC group. This species is morphologically adapted banks and riffle habitat used by both adults and juveniles. Like- to foraging at the water’s surface, where contact with the atmo- wise, within E–P strategists, pool conditions in streams with sphere increases dissolved oxygen slightly and gulping air may predictable low flows (2006 DC) likely favor largemouth bass, supplement oxygen intake during times of low dissolved oxy- whereas streams that maintain higher summer base flows (2006 gen (Pyke 2005). This ability to tolerate low-DO conditions in C) support greater numbers of spotted bass. Sunfish and bass disconnected pools, combined with an opportunistic life history nest failures due to substrate movement or nest abandonment strategy (quick maturation time, low fecundity, and low parental are common when high-flow conditions coincide with spawn- care), make this species well adapted to take advantage of the FISH COMMUNITY STRUCTURE 1011

TABLE 4. Changes in the number of individuals and relative abundance for ﬁsh species included in data analyses (i.e., occurring at >5% of the sampling locations).

Number of individuals Relative abundance (%) Family Species 2006 2007 Change 2006 2007 Change Atherinidae Inland silverside Menidia beryllina 29 54 25 0.10 0.53 0.43 Catostomidae River carpsucker Carpiodes carpio 58 4 −54 0.20 0.04 −0.16 Gray redhorse Moxostoma congestum 63 38 −25 0.22 0.37 0.15 Centrarchidae Redbreast sunfish Lepomis auritus 30 0 −30 0.10 0 −0.1 Green sunfish Lepomis cyanellus 805 1,592 787 2.81 15.58 12.77 Warmouth Lepomis gulosus 44 72 28 0.15 0.70 0.55 Orangespotted sunfish Lepomis humilis 24 11 −13 0.08 0.11 0.03 Bluegill Lepomis macrochirus 553 715 162 1.93 7.00 5.07 Longear sunfish Lepomis megalotis 3,540 1,037 −2,503 12.34 10.15 −2.19 Lepomis spp. 1,013 169 −844 3.53 1.65 −1.88 Spotted bass Micropterus punctulatus 343 39 −304 1.20 0.38 −0.82 Largemouth bass Micropterus salmoides 481 239 −242 1.68 2.34 0.66 White crappie Pomoxis annularis 31−2 0.01 0.01 0 Black crappie Pomoxis nigromaculatus 0 88 88 0 0.86 0.86 Clupeidae Gizzard shad Dorosoma cepedianum 286 3 −283 1.00 0.03 −0.97 Threadfin shad Dorosoma petenense 6 131 125 0.02 1.28 1.26 Cyprinidae Central stoneroller Campostoma anomalum 2,082 473 −1,609 7.26 4.63 −2.63 Red shiner Cyprinella lutrensis 2,818 1,276 −1,542 9.83 12.49 2.66 Blacktail shiner Cyprinella venusta 6,486 2,175 −4,311 22.62 21.28 −1.34 Common carp Cyprinus carpio 80 47 −33 0.28 0.46 0.18 Redfin shiner Lythurus umbratilis 3 29 26 0.01 0.28 0.27 Mimic shiner Notropis volucellus 475 139 −336 1.66 1.36 −0.3 Bullhead minnow Pimephales vigilax 1,279 400 −879 4.46 3.91 −0.55 Fundulidae Blackstripe topminnow Fundulus notatus 439 36 −403 1.53 0.35 −1.18 Ictaluridae Yellow bullhead Ameiurus natalis 199 179 −20 0.69 1.75 1.06 Channel catfish Ictalurus punctatus 176 595 419 0.61 5.82 5.21 Flathead catfish Pylodictis olivaris 9 64 55 0.03 0.63 0.6 Lepisosteidae Longnose gar Lepisosteus osseus 50 8 −42 0.17 0.08 −0.09 Moronidae White bass Morone chrysops 17 13 −4 0.06 0.13 0.07 Percidae Orangethroat darter Etheostoma spectabile 228 167 −61 0.80 1.63 0.83 Bigscale logperch Percina macrolepida 2 14 12 0.01 0.14 0.13 Dusky darter Percina sciera 12 19 7 0.04 0.19 0.15 Poeciliidae Western mosquitofish Gambusia affinis 7,040 387 −6,653 24.55 3.79 −20.76 Sciaenidae Freshwater drum Aplodinotus grunniens 4 6 2 0.01 0.06 0.05 All 28,677 10,220 −18,457 Downloaded by [Department Of Fisheries] at 20:34 25 September 2012

abiotic conditions associated with the temporary disconnection tween equilibrium and periodic life history strategies, and we of stream pools. Blackstripe topminnow have similar body mor- observed evidence of enhanced recruitment associated with high phology and body size and exhibit an opportunistic life history flows in 2007. strategy as well. The physical and behavioral adaptations of Flathead catfish and channel catfish were both significant these species allow them to persist in perennial pools such as indicators for 2007 (Table 5). This suggests that the postflood the ones present in the 2006 DC group. In addition, the small hydrologic conditions of the streams in 2007 were the primary size of both these fishes allows them to stay in very shallow determining factor of fish community composition independent water, where they avoid predation by larger-bodied fishes that of instream connectivity classification. These fishes are primar- are unable to access these shallow areas. ily lentic or large-river species found in large lakes and reser- Stable streamflow conditions may favor the growth and de- voirs, although channel catfish are heavily stocked in farm ponds velopment of large-bodied catfish species, but extreme spring as well; some of the collected individuals may have washed out and summer high flows (2007) trigger high recruitment and into our streams from these upstream source areas in 2007. In dispersal events (Figure 6). This represents an intermediate be- lotic habitats, the channel catfish occupies deeper habitats in 1012 STANLEY ET AL.

TABLE 5. Signiﬁcant (P < 0.05) indicator species and indicator values (IndVal) between years and among connectivity groups. Abbreviations are as follows: C = connected, D = disconnected, PC = previously connected, PD = previously disconnected; O = opportunist, OP = opportunist–periodic, P = periodic, EP = equilibrium–periodic, and E = equilibrium.

Species Life history class Group IndVal P Years Western mosquitofish O 2006 91.4 0.0001 Longear sunfish E 2006 74.6 0.0002 Blackstripe topminnow O 2006 49.5 0.0013 Blacktail shiner OP 2006 66.9 0.0085 Central stoneroller P 2006 66.9 0.0146 Spotted bass EP 2006 41.7 0.0235 Lepomis spp. — 2006 55.1 0.0353 Gizzard shad O 2006 21.2 0.0495 Channel catfish EP 2007 74.4 0.0001 Flathead catfish EP 2007 56.4 0.0003 Green sunfish E 2007 66.4 0.0112 Connectivity groups Central stoneroller P 2006 C 64.8 0.0001 Spotted bass EP 2006 C 40.8 0.0127 Bullhead minnow OP 2006 C 51.9 0.0326 Western mosquitofish O 2006 D 70.6 0.0001 Lepomis spp. — 2006 D 55.5 0.0055 Blackstripe topminnow O 2006 D 40.5 0.0088 Largemouth bass EP 2006 D 46.3 0.0258 Longear sunfish E 2006 D 42.4 0.0372 Flathead catfish EP 2007 PD 37.0 0.0258

higher-order streams for the majority of its adult life and feeds sites in 2006, and the amount of change experienced between primarily on detritus and benthic macroinvertebrates found in 2006 and 2007 was greater at sites that were disconnected in muddy substrates (Jackson 2004; Shephard and Jackson 2006). 2006. This suggests that, if anything, we might have seen an Adult flathead catfish also prefer deeper habitats but are primar- even stronger effect of drought on communities assuming that ily piscivorous as adults (Haas et al. 2001). These two catfish a few sites became disconnected after we sampled in 2006. Our species have been described as fast colonizers that move from original 2006 data set included 38 sites, but we were unable more typical, higher-order stream home ranges into these lower- to sample 10 streams in 2007 due to high flows through the order streams immediately following flood disturbance condi- end of the index period. This is another limitation of our study, tions in search of newly accessible resources such as food and in that we may have underestimated the effects of high flows spawning habitat (Ross 2001). Flathead catfish were also signifi- on streams by excluding the streams that were most severely cant indicators for the 2007 PD group, but based on the dispersal impacted by flooding.

Downloaded by [Department Of Fisheries] at 20:34 25 September 2012 ability of this species during high-flow years it is likely that this In conclusion, habitat connectivity and stream discharge are finding is more related to dispersal from connected habitat rather major influences on instream habitat as well as fish species com- than to the previous year’s conditions contributing to flathead position. Hydrologic extremes can result in varied balance be- catfish abundance at previously disconnected sites. Most of the tween abiotic and biotic controls on fish communities. Low-flow individuals collected were juveniles, indicating strong recruit- conditions and disconnected pools entailed abiotic factors that ment of catfishes following flood events. favored opportunistic life history strategists as well as equilib- Although sites that were disconnected in 2006 showed rium species that prefer stable low-flow environments, whereas greater change in 2007 than sites that were connected, this clas- stream conditions following sustained high flows favored equi- sification was based on the flow conditions during sampling. librium species whose habitat was enhanced by higher flows Some sites that were classified as connected may have become and the recruitment of species with high dispersal ability dur- disconnected later in the summer or fall, but we did not revisit ing flood events. Knowledge of how fish community structure sites to determine this. Thus, some variance in community struc- is impacted by extreme hydrologic events and what life history ture among sites in 2007 may have been due to a change in flow strategies are employed during these events is useful for the man- status after we sampled them. However, there were differences agement of stream fishes in a world where hydrologic alteration in species composition between disconnected and connected is increasingly becoming a threat to freshwater biodiversity. FISH COMMUNITY STRUCTURE 1013

ACKNOWLEDGMENTS McCune, B., and M. J. Medford. 2006. PC-ORD: multivariate analysis of eco- This study was funded by a contract from the Texas Commis- logical data, version 5. MjM Software Designs, Gleneden Beach, Oregon. sion on Environmental Quality to K. Winemiller and R. King Meador, M. R., and W. J. Matthews. 1992. Spatial and temporal patterns in fish assemblage structure of an intermittent Texas stream. American Midland (contract 582-6-80304). We especially thank M. Fisher and G. Naturalist 127:106–114. Easley for providing logistical support. The Texas Parks and Mielke, P. W., Jr. 1984. Meteorological applications of permutation techniques Wildlife Department provided collecting permits, and Texas based on distance functions. Pages 813–830 in P. R. Krishnaiah and P. K. Sen Agrilife Research and Baylor University gave administrative editors. Handbook of statistics. Elsevier, Amsterdam. support. We thank J. Grimm, D. Lang, A. Flores, S. Sumpaon- NOAA (National Oceanic and Atmospheric Administration). 2008. Na- tional climatic data center archives. NOAA, Washington, D.C. Available: goen, and J. Back for field assistance and three anonymous www.ncdc.noaa.gov/oa/ncdc.html. (November 2008). reviewers for comments that helped improve an earlier draft. Noltie, D. B., and M. H. A. Keenleyside. 1986. Correlates of reproductive success in stream dwelling male rock bass, Ambloplites rupestris (Centrar- chidae). Environmental Biology of Fishes 17:61–70. Oksanen, J., F. G. Blanchet, R. Kindt, P. Legendre, R. B. O’Hara, G. L. Simp- REFERENCES son, P. Solymos, M. Henry, H. Stevens, and H. Wagner. 2011. Vegan: com- Allan, J. D. 1995. Stream ecology: structure and function of running waters. munity ecology package, R package version 1.17-9. Available: CRAN.R- Kluwer Academic Publishers, Dordrecht, The Netherlands. project.org/package = vegan. (February 2011). Anderson, M. J. 2006. Distance-based tests for homogeneity of multivariate Olden, J. D., and M. J. Kennard. 2010. Intercontinental comparison of fish life dispersions. Biometrics 62:245–253. history strategies along a gradient of hydrologic variability. Pages 83–107 Anderson, M. J., K. E. Ellingsen, and B. H. McArdle. 2006. Multivariate dis- in K. B. Gido and D. A. Jackson, editors. Community ecology of stream persion as a measure of beta diversity. Ecology Letters 9:683–693. fishes: concepts, approaches, and techniques. American Fisheries Society, Clarke, K. R. 1993. Non-parametric multivariate analyses of changes in com- Symposium 73, Bethesda, Maryland. munity structure. Australian Journal of Ecology 18:117–143. Ostrand, K. G., and G. R. Wilde. 2002. Seasonal and spatial variation in a prairie Dauwalter, D. C., and W. L. Fisher. 2007. Spawning chronology, nest site stream-fish assemblage. Ecology of Freshwater Fish 11:137–149. selection and nest success of smallmouth bass during benign streamflow Pease, A. A., J. M. Taylor, K. O. Winemiller, and R. S. King. 2011. Multi- conditions. American Midland Naturalist 158:60–78. scale environmental influences on fish assemblage structure in central Texas Dufrene,ˆ M., and P. Legendre. 1997. Species assemblages and indicator species: streams. Transactions of the American Fisheries Society 140:1409–1427. the need for a flexible asymmetrical approach. Ecological Monographs Poff, N. L., and J. D. Allan. 1995. Functional organization of stream fish assem- 67:345–366. blages in relation to hydrological variability. Ecology 76:606–627. Echelle, A. A., A. F. Echelle, and L. G. Hill. 1972. Interspecific interac- Pyke, G. H. 2005. A review of the biology of Gambusia affinis and G. holbrooki. tions and limiting factors of abundance and distribution in the Red River Reviews in Fish Biology and Fisheries 15:339–365. pupfish, Cyprinodon rubrofluviatilis. American Midland Naturalist 88:109– Richter, B. D., R. Mathews, D. L. Harrison, and R. Wigington. 2003. Ecolog- 130. ically sustainable water management: managing river flows for ecological Faith, D. P., and R. H. Norris. 1989. Correlation of environmental variables integrity. Ecological Applications 13:206–224. with patterns of distribution and abundance of common and rare fresh-water Ross, S. T. 2001. Inland fishes of Mississippi. Mississippi Department of macroinvertebrates. Biological Conservation 50:77–98. Wildlife, Fisheries and Parks, Jackson. Fausch, K. D., and R. G. Bramblett. 1991. Disturbance and fish communities in Shephard, S., and D. C. Jackson. 2006. Difference in channel catfish growth intermittent tributaries of a western Great Plains river. Copeia 1991:659–674. among Mississippi stream basins. Transactions of the American Fisheries Griffith, G. E., S. A. Bryce, J. M. Omernik, J. A. Comstock, A. C. Rogers, B. Society 135:1224–1229. Harrison, S. L. Hatch, and D. Bezanson. 2004. Ecoregions of Texas. U.S. Spranza, J. J., and E. H. Stanley. 2000. Condition, growth, and reproduc- Geological Survey, Reston, Virginia. tive styles of fishes exposed to different environmental regimes in a prairie Haas, J. J., L. A. Jahn, and R. Hilsabeck. 2001. Diet of introduced flathead drainage. Environmental Biology of Fishes 59:99–109. catfish (Pylodictus olivarus) in an Illinois reservoir. Journal of Freshwater Taylor, C. M. 1997. Fish species richness and incidence patterns in isolated Ecology 16:551–555. and connected stream pools: effects of pool volume and spatial position. Herbert, M. E., and F. P. Gelwick. 2003. Spatial variation of headwater fish Oecologia 110:560–566. assemblages explained by hydrologic variability and upstream effects of im- Taylor, C. M., M. R. Winston, and W. J. Matthews. 1993. Fish species-

Downloaded by [Department Of Fisheries] at 20:34 25 September 2012 poundment. Copeia 2003:273–284. environment and abundance relationships in a Great Plains river system. Hoeinghaus, D. J., K. O. Winemiller, and J. S. Birnbaum. 2007. Local and Ecography 16:16–23. regional determinants of stream fish assemblage structure: inferences based TCEQ (Texas Commission on Environmental Quality). 2003. Surface water on taxonomic vs. functional groups. Journal of Biogeography 34:324–338. quality monitoring procedures, volume 2: methods for collecting and ana- Hubbs, C., R. J. Edwards, and G. P. Garrett. 2008. An annotated checklist of the lyzing biological assemblage and habitat data. TCEQ, Surface Water Quality freshwater fishes of Texas, with keys to identification of species, 2nd edition. Monitoring Program, Austin. Texas Academy of Science, Plainville. Thomas, C., T. H. Bonner, and B. G. Whiteside. 2007. Freshwater fishes of Jackson, D. C. 2004. Natural history and fisheries: biology and culture of channel Texas: a field guide. Texas A&M University Press, College Station. catfish. Elsevier, Amsterdam. Wilde, G. R., and K. G. Ostrand. 1999. Changes in the fish assemblage of Jennings, M. J., and D. P. Philipp. 1994. Biotic and abiotic factors affecting an intermittent prairie stream upstream from a Texas impoundment. Texas survival of early life history intervals of a stream-dwelling sunfish. Environ- Journal of Science 51:203–210. mental Biology of Fishes 39:153–159. Winemiller, K. O. 1989. Patterns of variation in life history among South Amer- Matthews, W. J., and J. T. Styron Jr. 1981. Tolerance of headwater vs. main- ican fishes in seasonal environments. Oecologia 81:225–241. stream fishes for abrupt physiochemical changes. American Midland Natu- Winemiller, K. O., and K. A. Rose. 1992. Patterns of life-history diver- ralist 105:149–158. sification in North American fishes: implications for population regu- McCune, B., and J. B. Grace. 2002. Analysis of ecological communities. MjM lation. Canadian Journal of Fisheries and Aquatic Sciences 49:2196– Software Designs, Gleneden Beach, Oregon. 2218. 1014 STANLEY ET AL.

APPENDIX: DETAILS AS TO SITES AND FISH SPECIES SAMPLED

TABLE A.1. Site locations, watershed size, connectivity classiﬁcation, and discharge for 28 stream reaches sampled in 2006 and 2007.

Discharge (m3/s) Watershed Connectivity Stream Site code Basin Latitude Longitude area (km2) class 2006 2007 Bear Creek BEAR Trinity 32.59442 97.51018 163.93 D <0.001 0.377 Bluff Creek BLUF Brazos 31.55536 97.47570 67.63 D <0.001 0.048 Clear Fork Trinity CFTR Trinity 32.70082 97.62979 327.23 C 0.014 0.07 Clear Creek CLEA Trinity 33.35909 97.25029 636.15 D <0.001 0.312 Coryell Creek CORY Brazos 31.39070 97.59826 220.45 D <0.001 2.118 Cowhouse Creek COWH Brazos 31.28327 97.88241 1180.23 C 0.002 0.699 Duffau Creek DUFF Brazos 32.01341 97.96521 156.99 D <0.001 0.101 Elm Fork Trinity EFTR Trinity 33.58631 97.13076 472.21 C 0.041 0.365 Harris Creek HARR Brazos 31.45960 97.29253 76.93 D <0.001 0.102 Hog Creek HOG Brazos 31.52264 97.28924 235.50 D <0.001 0.206 Lampasas River 1 LAMP1 Brazos 31.37802 98.18063 719.90 C 0.006 0.481 Lampasas River 2 LAMP2 Brazos 31.11558 98.05432 1570.99 C 0.006 1.112 Middle Bosque River MBOS Brazos 31.50748 97.35624 478.17 D <0.001 0.275 Meridian Creek MERI Brazos 31.81095 97.60911 479.59 D <0.001 3.818 North Bosque River 1 NBOS1 Brazos 32.18898 98.18258 257.05 C 0.504 0.034 North Bosque River 2 NBOS2 Brazos 32.04166 98.11330 488.83 D <0.001 0.276 North Bosque River 3 NBOS3 Brazos 31.97692 98.03974 924.56 D <0.001 0.621 Neils Creek NEIL Brazos 31.69952 97.53088 356.65 D <0.001 2.995 Nolan Creek NOLC Brazos 31.08916 97.48826 275.23 C 0.735 0.821 Nolan River 1 NOLR1 Brazos 32.25064 97.40433 450.54 C 0.140 0.326 Nolan River 2 NOLR2 Brazos 32.14660 97.40600 720.92 C 0.139 0.639 Paluxy River PALU Brazos 32.24960 97.84615 932.79 C 0.043 1.517 Plum Creek PLUM Brazos 31.49967 97.85686 226.49 D <0.001 0.231 Rocky Creek ROCK Brazos 30.94494 97.99117 220.57 C 0.029 0.904 Salado Creek SALA Brazos 30.91275 97.60105 214.87 C 0.023 0.216 South Bosque River SBOS Brazos 31.47359 97.27638 219.69 D <0.001 0.057 South Fork Trinity SFTR Trinity 32.69943 97.63202 252.00 D <0.001 0.231 South Leon River SLEO Brazos 31.83891 98.37763 518.30 D <0.001 0.057 Downloaded by [Department Of Fisheries] at 20:34 25 September 2012 FISH COMMUNITY STRUCTURE 1015

TABLE A.2. List of the 34 ﬁsh species collected in >5% of the 28 sites in 2006 and 2007. Lepomis spp. consist of juveniles that were identiﬁable only to genus due to their small size and lack of distinguishing markings.

Order Family Species Species code Siluriformes Ictaluridae Yellow bullhead AMEINATA Perciformes Sciaenidae Freshwater drum APLOGRUN Cypriniformes Cyprinidae Central stoneroller CAMPANOM Catostomidae River carpsucker CARPCARP Cyprinidae Common carp CYPRCARP Cyprinidae Red shiner CYPRLUTR Cyprinidae Blacktail shiner CYPRVENU Clupeiformes Clupeidae Gizzard shad DOROCEPE Clupeidae Threadfin shad DOROPETE Perciformes Percidae Orangethroat darter ETHESPEC Cyprinodontiformes Fundulidae Blackstripe topminnow FUNDNOTA Poeciliidae Western mosquitofish GAMBAFFI Siluriformes Ictaluridae Channel catfish ICTAPUNC Semionotiformes Lepisosteidae Longnose gar LEPIOSSE Perciformes Centrarchidae Redbreast sunfish LEPOAURI Centrarchidae Green sunfish LEPOCYAN Centrarchidae Warmouth LEPOGULO Centrarchidae Orangespotted sunfish LEPOHUMI Centrarchidae Bluegill LEPOMACR Centrarchidae Longear sunfish LEPOMEGA Centrarchidae Lepomis spp. LEPOSPP Cypriniformes Cyprinidae Redfin shiner LYTHUMBR Atheriniformes Atherinidae Inland silverside MENIBERY Perciformes Centrarchidae Spotted bass MICRPUNC Centrarchidae Largemouth bass MICRSALM Moronidae White bass MOROCHRY Cypriniformes Catostomidae Gray redhorse MOXOCONG Cyprinidae Mimic shiner NOTRVOLU Perciformes Percidae Bigscale logperch PERCMACR Percidae Dusky darter PERCSCIE Cypriniformes Cyprinidae Bullhead minnow PIMEVIGI Perciformes Centrarchidae White crappie POMOANNU Centrarchidae Black crappie POMONIGR Siluriformes Ictaluridae Flathead catfish PYLOOLIV Downloaded by [Department Of Fisheries] at 20:34 25 September 2012 This article was downloaded by: [Department Of Fisheries] On: 25 September 2012, At: 20:35 Publisher: Taylor & Francis Informa Ltd Registered in England and Wales Registered Number: 1072954 Registered office: Mortimer House, 37-41 Mortimer Street, London W1T 3JH, UK

Transactions of the American Fisheries Society Publication details, including instructions for authors and subscription information: http://www.tandfonline.com/loi/utaf20 Migrations of Common Snook from a Tidal River with Implications for Skipped Spawning Alexis A. Trotter a , David A. Blewett b , Ronald G. Taylor a & Philip W. Stevens b a Florida Fish and Wildlife Conservation Commission, Fish and Wildlife Research Institute, 100 8th Avenue Southeast, St. Petersburg, Florida, 33701, USA b Florida Fish and Wildlife Conservation Commission, Fish and Wildlife Research Institute, Charlotte Harbor Field Laboratory, 585 Prineville Street, Port Charlotte, Florida, 33954, USA

Version of record first published: 26 Jun 2012.

To cite this article: Alexis A. Trotter, David A. Blewett, Ronald G. Taylor & Philip W. Stevens (2012): Migrations of Common Snook from a Tidal River with Implications for Skipped Spawning, Transactions of the American Fisheries Society, 141:4, 1016-1025 To link to this article: http://dx.doi.org/10.1080/00028487.2012.675903

PLEASE SCROLL DOWN FOR ARTICLE

Full terms and conditions of use: http://www.tandfonline.com/page/terms-and-conditions 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. The publisher does not give any warranty express or implied or make any representation that the contents will be complete or accurate or up to date. The accuracy of any instructions, formulae, and drug doses should be independently verified with primary sources. The publisher shall not be liable for any loss, actions, claims, proceedings, demand, or costs or damages whatsoever or howsoever caused arising directly or indirectly in connection with or arising out of the use of this material. Transactions of the American Fisheries Society 141:1016–1025, 2012 C American Fisheries Society 2012 ISSN: 0002-8487 print / 1548-8659 online DOI: 10.1080/00028487.2012.675903

ARTICLE

Migrations of Common Snook from a Tidal River with Implications for Skipped Spawning

Alexis A. Trotter* Florida Fish and Wildlife Conservation Commission, Fish and Wildlife Research Institute, 100 8th Avenue Southeast, St. Petersburg, Florida 33701, USA David A. Blewett Florida Fish and Wildlife Conservation Commission, Fish and Wildlife Research Institute, Charlotte Harbor Field Laboratory, 585 Prineville Street, Port Charlotte, Florida 33954, USA Ronald G. Taylor Florida Fish and Wildlife Conservation Commission, Fish and Wildlife Research Institute, 100 8th Avenue Southeast, St. Petersburg, Florida 33701, USA Philip W. Stevens Florida Fish and Wildlife Conservation Commission, Fish and Wildlife Research Institute, Charlotte Harbor Field Laboratory, 585 Prineville Street, Port Charlotte, Florida 33954, USA

Abstract Common snook Centropomus undecimalis support a recreational fishery that contributes greatly to Florida’s economy and have been the subject of many diverse studies in which aspects of their life history, reproductive biology, and fishery dynamics have been defined. The current stock assessment contains assumptions based on historical reproductive schedules measured at the population level during the early 1990s. Recent research, however, suggests our understanding of habitat residency and spawning schedules may be overly generalized. We used acoustic telemetry to study the movements and migrations of common snook from the tidal reaches of the Caloosahatchee River, a major tributary of Charlotte Harbor, for 3 years (2005–2007). During this period, 60% of the tagged snook left the study area during at least one spawning season, presumably to enter higher salinity waters where they spawn. The onset and duration of these annual migrations are similar to those reported in previous reproductive studies of snook, but individual dynamics varied. Unexpectedly, 40% of tagged snook remained within the monitored portion of the river for up to three spawning seasons, providing initial evidence for skipped spawning. These detailed data indicate

Downloaded by [Department Of Fisheries] at 20:35 25 September 2012 differential individual contributions to total spawning biomass. If further investigation conﬁrms skipped spawning in common snook and shows similar rates of skipped reproduction throughout the snook population, this information should be incorporated into future stock assessments.

Common snook Centropomus undecimalis is a popular game 2006; R. Southwick, Southwick Associates, personal commu- fish that supports a fishery valuable to the Florida economy. In nication). High levels of exploitation of this popular fishery 2004, marine anglers in south Florida made approximately 1.8 have compelled the Florida Fish and Wildlife Conservation million trips that targeted common snook providing a direct Commission’s Fish and Wildlife Research Institute (FWRI) to economic value of about US$620 million (Muller and Taylor regularly assess the condition of the two coastal stocks (Gulf

*Corresponding author: [email protected] Received June 13, 2011; accepted February 1, 2012 Published online June 26, 2012 1016 MIGRATIONS OF COMMON SNOOK 1017

of Mexico and Atlantic Ocean) and to develop regulations portion of the population that spawns annually, (2) the duration designed to maintain high levels of abundance. Stringent regu- of the spawning season, and, therefore, (3) total egg production. lations have been in place since 1994 with the goal of attaining These quantitative dynamics of spawning and availability should a target spawning potential ratio (SPR) of 40%, but this goal thus be reevaluated and integrated into future stock assessments has not been achieved and the stocks are considered overfished. to improve their accuracy and ensure proper management. The common snook is a euryhaline, subtropical species that In this paper, we report on the movement patterns of common spawns over a protracted summer season (April through Septem- snook, specifically in regard to the duration of their residency ber) in major ocean inlets, passes, and high-salinity areas of in a southwest Florida river. We address two research needs estuaries (Peters et al. 1998; Taylor et al. 1998). The require- for common snook: (1) to identify areas used by snook within ments for successful spawning have been extensively studied. coastal rivers, and (2) to determine movement patterns and ex- As obligate marine spawners, salinity must remain above 24‰ change rates of adult snook between the river and the adjacent to maintain critical buoyancy of the fertilized eggs; once buoy- estuary. This research provides insight into how common snook ancy is lost and the surface tension is broken, mortality of the use the connected freshwater and saltwater habitats in southwest egg and developing embryos is imminent (Ager et al. 1978; Florida. Chapman et al. 1978; Tucker 1987). Further, Ager et al. (1978) demonstrated that saltwater was required for the activation of the sperm cells and that salinity greater than 35‰ resulted in METHODS motility in about 80% of sperm. Study area.—The Caloosahatchee River is one of three large After the spawning season, it has been assumed snook move rivers that supply freshwater to Charlotte Harbor, a major es- gradually from inlets and open estuarine shorelines into rivers, tuary located along the southwest coast of Florida (Figure 1). creeks, and canals to overwinter (Volpe 1959; Blewett et al. The river has been extensively altered to create the cross-Florida 2009). The major impetus for this migration was thought to be Okeechobee Waterway that connects Lake Okeechobee at its avoidance of potentially fatal winter water temperatures of 10– western terminus to Pine Island Sound within Charlotte Harbor. 14◦C (Shafland and Foote 1983; Howells et al. 1990). This mi- Three locks regulate the river flows and lake levels and fa- gration may also have allowed them to locate sufficient prey and cilitate navigation from the lake to the Gulf of Mexico. The escape predators when low water temperatures induce lethargy. river courses approximately 121 river kilometers (1 rkm = Recent research in a southwest Florida estuary and its rivers 1,000 m along the midline of the main stem of the river raised questions about this paradigm by showing that adult [ftp://nhdftp.usgs.gov/SubRegions/High/]), drains 3,569 km2, snook were relatively abundant year-round both in rivers and and discharges an average of 40.8 m3/s annually (Hammett along open estuarine shorelines, and that they were most abun- 1990). This study examines annual and seasonal movements dant in the rivers during fall, not winter (Blewett et al. 2009). of common snook along the tidally influenced (i.e., the low- These data indicated that snook migration patterns and their ermost 50 rkm) portion of the Caloosahatchee River, from the use of freshwater and marine habitats were more complex than mouth of the river east to 2.5 rkm beyond the W. P.Franklin Lock previously considered. Thus, critical information needed to cor- and Dam (Franklin Lock) located at rkm 48, the westernmost of rectly model the reproductive parameters in stock assessments three locks between Charlotte Harbor and Lake Okeechobee. is lacking, especially detailed information about the movements Acoustic receiver array.—An array of 23 Vemco VR2 acous- into and between freshwater habitats, the magnitude and dura- tic receivers was deployed by Mote Marine Laboratory, Sara- tion of residency in freshwater, and the levels of availability to sota, Florida, in 2003 in the estuarine portion of the Caloosa- anglers in these areas. hatchee River to passively track the movement of sharks and

Downloaded by [Department Of Fisheries] at 20:35 25 September 2012 Currently, snook stock assessments are based on age, in- rays fitted with acoustic transmitters (Heupel and Simpfendorfer tegrating life history characteristics with information on ages, 2008; Simpfendorfer et al. 2008). A detailed examination of the sex ratios, harvest levels, and catch rates. The majority of data performance (e.g., code detection efficiency, code rejection co- have come from snook captured within the estuaries where they efficients, and noise quotients) of this array can be found in are abundant, available to anglers, and easily sampled (Peters Simpfendorfer et al. (2008). In the present study, an additional et al. 1998; Taylor et al. 1998, 2000; Muller and Taylor 2006). eight receivers were deployed farther upriver on May 17, 2005. The assessment assumes that each female migrates annually into This extended the array to the Franklin Lock, which is consid- high-salinity waters, where they spawn continuously throughout ered to be the uppermost limit of tidal waters (Figure 1). An a 180-d spawning season (Tucker and Campbell 1988; Taylor additional VR2 receiver was placed approximately 2.5 km be- et al. 1998; Muller and Taylor 2006). An assumption of the yond the Franklin Lock to document snook movement through model is that all movements and migrations are the same across the structure. Additional receivers (n = 3) used by FWRI re- the population and, therefore, there are no unique exchanges searchers for tracking other species were deployed in August or differences in residency patterns between subpopulations. If 2007 in the residential canal system on the north side of the these assumptions are incorrect, then estimates of the following lower river. Receiver data were downloaded and our receivers reproductive parameters may be imprecise: (1) the size or pro- cleaned monthly through October 27, 2007, when our portion of 1018 TROTTER ET AL.

FIGURE 1. Florida (inset), Charlotte Harbor, and the Caloosahatchee River, depicting the location of VR2 receivers (ﬁlled circles) used to acoustically monitor tagged common snook from May 2005 through October 2007.

the array was dismantled. The remaining receivers were main- 9 modulator. Stunned snook were dipnetted aboard and placed tained by Mote Marine Laboratory and other FWRI researchers, in a 200-L holding tank with circulating water until they could

Downloaded by [Department Of Fisheries] at 20:35 25 September 2012 and data were regularly exchanged between the groups. Each be transferred to a surgery boat. Upon transfer, each snook was receiver recorded the date, time, and identity of tagged fish that inspected for injury and condition and then sedated by immer- swam within the detection range of the unit. The detection range sion in a tank containing low levels of carbon dioxide provided was documented at 600 m for the types of acoustic transmitters by dissolving eight Alka-Seltzer antacid tablets per 19 L (5 gal) and water conditions used in this study (M. R. Heupel, unpub- of water. Sedated snook were then moved to a tagging cradle fit- lished data; R. G. Taylor, unpublished data), which often allowed ted with a low-pressure water pump that supplied a continuous fish to be detected by more than one receiver simultaneously. flow of river water through the mouth and over the gills. Acoustic tagging.—On May 24, 2005, 25 adult common Each snook was measured for maximum total length (TL, snook were captured between the Franklin Lock and the Semi- mm) and surgically implanted with a Vemco RCODE acous- nole Gulf Railway (SGR) trestle (rkm 34) by electrofishing, and tic transmitter that was coded with a unique pulse series that acoustic transmitters were surgically implanted into their ab- operated on 69.0 kHz at random intervals from 60 to 120 s dominal cavities. Two electrofishers mounted on small research and had a suggested battery life of 36 months. The timing vessels were used to stun fish for capture; each rig produced of random-signal transmission helped prevent overlap of sig- 480–600 V of pulsating DC at 6–12 A, which was supplied nals from different tags, which could have blocked detections by a 9-kW generator and controlled by a Smith–Root model by a receiver. Twenty snook were tagged with V16 acoustic MIGRATIONS OF COMMON SNOOK 1019

transmitters (16 × 65 mm), and five were tagged with V8 acous- river during the study, or as migrants, which completed migra- tic tags (8 × 20.5 mm). Tags were inserted through a small inci- tions out of the study area. Overall, 27% (4 of 15; tag iden- sion in the ventral midline of the body wall that was then closed tification numbers [IDs]: 571, 570, 1992, 1989) of the tagged with 2 to 4 Vicryl sutures attached to a #4 curved, chisel-point snook did not leave the river during the time they were tracked surgical needle and sealed with Super Glue. Each snook was (Figures 2, 3A, B). The mean TL of the residents (654 mm TL, then tagged dorsally with a Hallprint PDL dart tag printed with n = 4) was significantly less (t-test: P = 0.02) than the mean a unique number, a message instructing anglers to release the TL of the migrants (814 mm TL, n = 11). Nine snook left the fish and report recapture information, and the promise of a re- river during at least one spawning season (April–September). ward. Tagged snook were held in the water over the gunwale of Five snook, 33%, (IDs: 1981, 174, 1995, 1977, 1986) migrated the boat until they had recovered sufficiently to swim away. once; two snook, 13%, (IDs: 1982, 1979) migrated twice; and Data analysis.—The position of detected fish along the river two snook, 13%, (IDs: 1984, 1991) migrated during all three was estimated using a mean-position algorithm (Simpfendorfer spawning seasons (Figure 2). et al. 2008). The mean position for a particular time period (4 h Individual fish were only included in annual calculations if in this study) was they were recorded within the river during all or part of an interval. Therefore, 15 snook are included in the first interval, i wi xi nine are included in the second interval, and only five fish are X¯ km = i wi included in the third interval. On an annual basis, 40% (6 of 15), 33% (3 of 9), and 40% (2 of 5) of individuals remained within the study area during May 2005–April 2006, May 2006–April 2007, where xi is the distance from receiver i to the river mouth, and wi is the number of valid detections recorded at each receiver during and May 2007–October 2007, respectively. In each interval, the 4-h period. Although Simpfendorfer et al. (2008) included residents were smaller than migrants although not significantly = = the code detection efficiency of receivers in calculations of so (t-test: interval 1, P 0.05; interval 2, P 0.06; interval 3, P = 0.31). Two snook in the first interval (IDs: 1983, 1980) and wi, the inclusion of this variable did not significantly improve the results of the algorithm and we did not use it. The 4-h posi- one fish in the second interval (ID 1983), although classified tion estimates of individual common snook were plotted so that as migrants, left the river during the winter months rather than we could examine overall movement patterns for the duration during the known spawning season (Figure 2). If our annual of the study. calculations only examined movements during the spawning Rates of exchange of snook between the river and presumed season, then 53% (8 of 15) and 44% (4 of 9) of individuals spawning areas were calculated for each year. Because the study remained within the river during the spawning seasons of the spanned a total of 30 calendar months, from May 2005 through first two intervals. Spawning season residents were significantly = October 2007, annual intervals are 12-month increments that smaller than spawning season migrants (t-test: interval 1, P = begin in May and continue through the following April. This 0.02; interval 2, P 0.02). resulted in two complete annual intervals and a 6-month partial The snook that resided within the river during the spawning interval. Consequently, the reported rates of exchange for 2005– seasons stayed upriver of the SGR trestle (rkm 34). Nearly all of 2006 and 2006–2007 are complete and, hence, more robust than the detections for the four resident snook (IDs: 571, 570, 1992 the rates for the third, partial interval. [Figure 3A], and 1989 [Figure 3B]) during the spawning seasons were upriver of that location, as were the detections for snook ID 1981 (Figure 3C) during the first and third spawning seasons. RESULTS Snook ID 1979 spent 80% of the 2005 spawning season above

Downloaded by [Department Of Fisheries] at 20:35 25 September 2012 Acoustically tagged common snook (n = 25) had a mean the trestle and made a 9-d trip to the lower river (downriver from TL of 761 mm (range, 543–1,085 mm). All study snook were rkm 12) in June (Figure 3D). Snook could successfully spawn assumed to be sexually mature because male and female snook within the river if salinities were high enough in the areas they on the Gulf coast begin maturation at approximately 250 and resided. During the spawning seasons of 2005 and 2006, surface 500 mm TL, respectively (Taylor et al. 2000). Mean surgery salinity was above 24‰ for a single day in the lower river (Figure duration was 4 min 22 s (range, 2 min 30 s to 6 min 30 s) and 4). In 2007, salinity in the lower river was above 24‰ through the mean number of sutures was three (range, 2–4). Ten individ- mid-July (67% of the spawning season) and, in the upper river, uals were removed from the analysis because of presumed or salinities were above 24‰ for 14 d in late May and early June observed tag loss or failure or immediate mortality. The mean (Figure 4). TL of the snook removed from the analysis (746 mm TL) was The mean date of departure from the river for completed not signiﬁcantly different (t-test: P = 0.72) from the mean TL spawning-season migrations (n = 10) was June 1 (range, March of the snook that remained in the study (771 mm TL). 12–July 8). Of those 10 migrations, 6 (60%) were completed The 15 snook that were analyzed were each classiﬁed as ei- during the month of August, and the mean return date was ther residents, which remained within the tidal portion of the August 18 (range, May 24–October 9; Table 1). The average 1020 TROTTER ET AL.

2005 2006 2007 Fish # (TL) May Jun Jul Aug Sep Oct Nov Dec Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec Jan Feb Mar Apr May Jun Jul Aug Sep Oct 571 (666) H - Harvested on March 26, 2006 in the Caloosahatchee River 570 (543) H - Harvested in mid-September 2006 in the Caloosahatchee River 1992 (653) S - Suspected harvested on May 27, 2006 at the Franklin Lock, Caloosahatchee River 1989 (754) 1981 (809) U

1982 (832) . U 1979 (748) 1984 (1010) 1991 (909) 174 (592) U 1995 (790) H - Harvested on April 5, 2006 in Boca Grande Pass 1977 (945) U 1986 (955) U 1983 (727) S - Suspected harvested on October 17, 2006 in the Caloosahatchee River 1980 (635) H - Harvested on November 9, 2005 in Hickory Pass

Fish detected in receiver array Fish moved downriver Fish left array Fish assumed to be at sea Fish moved upriver Fish re-entered array

FIGURE 2. Large-scale movements of 15 acoustically tagged common snook that were monitored in the Caloosahatchee River from May 2005 through October 2007. Fish identiﬁcation (ID) numbers are followed by the TL (mm) of each snook at tagging. Gray areas indicate primary spawning months (April through September). Fate of individual ﬁsh: (H) = reported harvested, (S) = suspected harvested, and (U) = unknown fate.

duration outside the study area for fish that were presumed ID 1981 (809 mm TL) was recorded 91% of the time in a 1-km to be spawning was 78 d (range, 40–124 d; Table 1). Snook area and returned to that same location in August 2006 after ID 1979 left the river before the spawning season in 2007. making a complete migration from the river during that summer However, because this fish was absent from the river for part of (Figure 3C). Similar patterns were observed in other snook (IDs: the spawning season, it was included as a spawning fish. If this 1979, 1991, 1984; Figures 3 and 5). migration is removed from our calculations, the mean dates of departure from and return to the river are June 10 and August TABLE 1. Schedule and duration (in days) of spawning season migrations 27, respectively, but the average duration outside of the study (n = 10) made by acoustically tagged snook (n = 5) out of and back into area remains 78 d. the Caloosahatchee River during the study (May 2005 through October 2007). The annual migrations of snook ID 1991 and snook ID 1984 Common snook spawning season occurs from April through September. Snook occurred during periods of maximum temperatures during all ID 1979 migrated from the river before the spawning season in 2007 (denoted with an asterisk). That migration did span part of the spawning season and is, 3 years and when flow rates were at their highest during the therefore, included here. summer of the first year (Figures 3E, 5). A subsequent period of high flow during late October and early November of the Fish ID Departure date Return date Duration (d) Downloaded by [Department Of Fisheries] at 20:35 25 September 2012 first year did not elicit a similar migration. Snook ID 1984 1979 May 3, 2006 Sep 4, 2006 124 also migrated during the highest flow rates in the second year 1979* Mar 12, 2007 May 24, 2007 73 (Figure 5A), whereas snook ID 1991 migrated out of the river 1981 Jun 2, 2006 Aug 26, 2006 85 before flow rates increased and was back in the upper river 1982 Jul 8, 2005 Aug 31, 2005 54 during the period of highest flow rates. The region experienced a 1984 Jun 26, 2005 Aug 29, 2005 64 record drought during 2007, resulting in minimum flows during 1984 Jul 2, 2006 Oct 9, 2006 99 that year. Both snook made migrations out of the river that 1984 May 20, 2007 Sep 4, 2007 107 summer. 1991 Jun 22, 2005 Aug 1, 2005 40 Whether an individual snook stayed in the river or left and 1991 May 28, 2006 Aug 11, 2006 75 later returned, most detections occurred within a small area, usu- 1991 Jun 17, 2007 Aug 11, 2007 55 ally within the range of two or three adjacent receivers (Figures 3 Mean Jun 1 Aug 18 78 and 5). Snook ID 1989 (754 mm TL) was recorded 68% of the Minimum Mar 12 May 24 40 time near rkm 42 and other than 1 d in January 2007 was never Maximum Jul 8 Oct 9 124 recorded farther than 8 km from that location (Figure 3B). Snook MIGRATIONS OF COMMON SNOOK 1021 Downloaded by [Department Of Fisheries] at 20:35 25 September 2012

FIGURE 3. Representative tracking charts describing the movement patterns of acoustically tagged common snook within the Caloosahatchee River. Four-hour position estimates are provided along the y-axis with date along the x-axis. Fish identiﬁcation (ID) numbers are followed by the TL (mm) of the snook at tagging. Plates (A) and (B) are resident snook that did not leave the study area during the time they were tracked. Plates (C) and (D) are migrant snook that left the study area during at least one spawning season (April through September). Plate (E) is one of only two snook that migrated from the study river during all threespawning seasons.

DISCUSSION those that did not migrate, were large enough to be mature and This study presents the ﬁrst evidence of skipped spawning capable of spawning (Taylor et al. 2000) and were of sizes seen in within the snook family Centropomidae. Forty percent of our spawning aggregations near our study site (range, 316–851 mm study population did not leave the river during the time they were standard length [SL]; mean, 422 mm SL; Adams et al. 2009). tracked and only 13% (2 of 15) migrated from the river during Milton and Chenery (2005) integrated otolith microchemistry all three spawning seasons. All of our study ﬁsh, including with tag-recapture studies of barramundi Lates calcarifer (an 1022 TROTTER ET AL.

FIGURE 4. Surface salinity (‰) measured at the Cape Coral Bridge (rkm 12; solid black line) and at the Franklin Lock (rkm 48; solid gray line). Surface salinity is provided along the y-axis with date along the x-axis. Fertilized snook eggs require a minimum salinity of 24‰ (dashed black line) to maintain critical buoyancy and surface tension with the water.

allied species of the family Latidae, which had previously been Hoplosternum littorale, an exotic species from South America assigned to the family Centropomidae; Otero 2004) and reported that forms bubble nests on floodplains, was the most frequent that 41% of fish had not migrated out of rivers and into marine prey item for snook during the study period. Therefore, the rel- spawning sites. Thus, the estimate that 40% of our snook may atively young snook may have been able to optimize growth skip spawning seems plausible. by not migrating to the spawning sites for reproduction and Although skipped spawning has been reported for many remaining in the upper river during flood conditions. fish species, the reasons are not well understood (Jørgensen Another explanation for year-round residency in the rivers is et al. 2006; Secor 2008). Secor (2008) outlined several types of that snook are not skip-spawning, but rather spawning within the

Downloaded by [Department Of Fisheries] at 20:35 25 September 2012 skipped reproduction, of which the most plausible for snook in river when conditions allow. However, based on the current un- this study is delayed maturation in combination with younger derstanding of snook reproductive biology, salinity above 24‰ “skipping”; an individual has never attained a sufficient surplus is required to maintain critical buoyancy of the fertilized eggs of energy to allow for migration and spawning. In the case of (Ager et al. 1978; Chapman et al. 1978; Tucker 1987). During snook inhabiting the Caloosahatchee River, the lack of suffi- our study, and specifically during the spawning seasons, salini- cient energy implies poor feeding conditions. However, the first ties were rarely above this level. It is possible that snook could 2 years of our study occurred during a period of high freshwater spawn within the river during a drought year, as evidenced in inflow caused by consecutive hurricanes that passed over the 2007. In that year, salinity was high enough in the lower river to Caloosahatchee River watershed (Bell et al. 2004). Barramundi support fertilized eggs for the first half of the spawning season. that spent time in freshwater grew faster than those that lived in Although spawning in the rivers during typical climatic condi- the estuary (Gillanders and Kingsford 2002), and higher flood tions seems highly unlikely, to fully eliminate this possibility, flows enhanced fish growth (Robins et al. 2006). Snook diet future studies should integrate gonadal sampling of both male data collected in the rivers through 2006 showed a pronounced and female riverine snook with archival tags that can provide peak in abundance of prey species that take advantage of flooded not only the fish location but also environmental data including conditions (Stevens et al. 2010). For example, the brown hoplo salinity and temperature. MIGRATIONS OF COMMON SNOOK 1023 Downloaded by [Department Of Fisheries] at 20:35 25 September 2012

FIGURE 5. Tracking charts describing the movements of common snook ID 1984 (1,010 mm TL) plotted with (A) the river ﬂow rate (ft3/s) measured at the Franklin Lock (rkm 48), and (B) the surface water temperature (◦C) recorded at the Franklin Lock (rkm 48) from May 2005 to October 2007. Four-hour position estimates are provided along the primary y-axis with date along the x-axis. River ﬂow rate or surface temperature is provided along the secondary y-axis. 1024 TROTTER ET AL.

The mean dates of departure from and return to the study site Subsequent techniques used to implant V16 acoustic tags in 62 (June 1 and August 18) coincide with the expected dates of the snook in rivers on Florida’s Atlantic coast did not include anes- onset and cessation of spawning activity on the Gulf coast of thesia and resulted in only 5% mortality (J. A. Wittington and Florida (April and September, respectively; Taylor et al. 1998); J. Dutka-Gianelli, FWRI, personal communication). The mean however, this study brings into question the duration of the TL of our suspected tagging mortalities (763 mm TL) is sig- spawning season. The mean duration of the 10 migrations that nificantly smaller (t-test: P = 0.008) than the mean TL of the occurred during the spawning seasons was 78 d (range, 40– suspected tagging mortalities on the Atlantic coast (987 mm 124 d), a substantial departure from the predicted 180-d spawn- TL), suggesting that smaller fish may not have been able to ing season (Taylor et al. 1998). Models of total egg production overcome the effects of the anesthesia as readily as larger fish. for snook are based on the assumption that each female spawns Another study employing identical methodologies to ours, in- continuously throughout the spawning season. Although our cluding the same type of anesthesia, had an assumed tagging calculation may be biased by a small sample size, our data mortality of 5%; however, because those snook were tagged in indicate that the length of the spawning season may be less open bays and estuaries during cooler months, they would not than half of that used in the models. In addition, an acoustic have been subjected to many of the water quality problems as- telemetry study conducted in the late 1990s on the Atlantic sociated with rivers during the warmer months (Bennett 2006). coast of Florida documented individual movement into and out Although Bennett (2006) did not include an exact measure of of the aggregation (Lowerre-Barbieri et al. 2003). Although the TL, or the TL of his tagging mortalities, 91% of his study snook spawning aggregation was present for the duration of the spawn- were greater than 660 mm TL compared with only 64% of our ing season, the individuals making up the aggregation varied, study fish. indicating that spawning frequency may also be overestimated. The results of this research should be viewed with caution, Future stock assessments should perhaps be adjusted to reflect especially with respect to findings that might be applied at the a shorter individual spawning period and reduced individual population level, primarily because of the small number of snook spawning frequency. monitored during the study. Conversely, estimates of annual It is possible that the movement patterns of the snook we rates and seasonal metrics may be considered more robust be- tagged were different from those of the population as a whole. cause the study captured migrations of snook for three consec- The majority of the snook tagged in this study showed strong site utive spawning seasons. The authors understand the weakness fidelity; even those that left the array and returned had a pattern inherent in assuming that movements out of the river in sum- of leaving the river (presumably to spawn) and then returning to mer indicate a spawning migration and that active spawning did the same relatively small area of the river (i.e., snook ID 1991 occur in higher-salinity locations closer to the inlets; we have and snook ID 1984). However, the proportion of fish exhibiting accepted this assumption to necessitate a closer inspection of long-term residency and affinity for the river may have been historical reproductive schedules of the Gulf coast snook stock. biased by the tagging date (May 24, 2005) and the tagging loca- This pilot study demonstrated that acoustic telemetry can tion. During the study, snook abundance in the river was higher be a powerful tool for describing reproductive and ecological in the fall and was three times greater than during our tagging characteristics of fishes, especially catadromous species such as season (Blewett et al. 2009). If this fall peak in abundance is snook, and it provided insight about these characteristics as they a result of a temporary influx of migrant fish, then a spring pertain to snook populations on the Gulf coast of Florida. More tagging event would have missed this component of the popula- work is needed to validate these findings; if skipped spawning is tion. If tagging had occurred during the summer, reported annual confirmed, the proportion of males and females that skip-spawn and multiyear residency would have been even greater because annually, the energetic cost of migration and spawning, and the

Downloaded by [Department Of Fisheries] at 20:35 25 September 2012 spawning fish would have already left, leaving only long-term population-level effects of skipped spawning on egg production residents available for tagging. Conversely, had we tagged fish per recruit are of utmost importance. It is equally important to in the inlet or estuary proper, we may have tagged a larger pro- understand and adjust these parameters for the Atlantic stock. portion of migrant snook. Therefore, the results of this study Long-term studies, combining acoustic telemetry over a larger probably reflect the migration patterns of the snook that primar- geographical scale with biological sampling (gonads, liver and ily reside within rivers, a facet of snook behavior that must be mesenteric fat), have been initiated on the east coast of Florida to better understood for a more complete life history model. More establish and validate rates and schedules of exchange between work is needed to determine whether snook inhabiting estuarine saltwater and freshwater habitats and to confirm and estimate the and marine environments show evidence of skipped spawning proportion of skipped spawning among snook on the Atlantic or whether it is unique to snook inhabiting freshwater habitat. coast. The combined effects of anesthesia, tagging season, and the size of our fish had unexpected results with regard to tagging mortality. The immediate loss of 40% (10 of 25) of the tagged ACKNOWLEDGMENTS snook in our study suggests there was an acute weakness in our We are indebted to E. Nagid, W.Strong, C. Keller, J. Willitzer, tagging methodology and that our methods should be improved. and B. Pouder of FWRI Freshwater Fisheries Research for their MIGRATIONS OF COMMON SNOOK 1025

time and assistance in collecting snook, and to W. Pine, Univer- Howells, R. G., A. J. Sonski, P. L. Shafland, and B. D. Hilton. 1990. Lower sity of Florida, J. Bennett, and L. Marcinkiewicz for their help in temperature tolerance of snook, Centropomus undecimalis. Northeast Gulf conceptualizing this study and their expert skills in implanting Science 11:155–158. Jørgensen, C., B. Ernande, Ø. Fiksen, and U. Dieckmann. 2006. The logic of the acoustic tags into our study fish. We also thank M. Heupel, skipped spawning in fish. Canadian Journal of Fisheries and Aquatic Sciences C. Simpfendorfer, B. Yeiser, and T. Wiley for their vast exper- 63:200–211. tise in acoustic telemetry and for sharing data collected on Mote Lowerre-Barbieri, S. K., F. E. Vose, and J. A. Whittington. 2003. Catch-and- Marine Laboratory’s receivers. We are grateful to Florida Power release fishing on a spawning aggregation of common snook: does it af- and Light Company for permission to deploy a receiver inside fect reproductive output? Transactions of the American Fisheries Society 132:940–952. the intake canal at their Orange River facility, and we thank B. Milton, D. A., and S. R. Chenery. 2005. Movement patterns of barramundi Tibble and B. Casey of Florida Power and Light Company for Lates calcarifer, inferred from 87Sr/86Sr and Sr/Ca ratios in otoliths, indi- guiding and protecting us as we serviced our equipment inside cate non-participation in spawning. Marine Ecology Progress Series 301: the intake canal. Finally, we thank S. Lowerre-Barbieri and A. 279–291. Collins. Their constructive reviews greatly benefitted this paper. Muller, R. G., and R. G. Taylor. 2006. The 2005 stock assessment update of common snook, Centropomus undecimalis. Florida Fish and Wildlife Con- This project was supported by funding from the Department servation Commission, Fish and Wildlife Research Institute, In-House Report of the Interior, U.S. Fish and Wildlife Service, Federal Aid for IHR 2006-003, St. Petersburg. Sport Fish Restoration, Project F-59. The views and conclusions Otero, O. 2004. Anatomy, systematics and phylogeny of both recent and fos- presented herein are those of the authors and do not necessarily sil latid fishes (Teleostei, Perciformes, Latidae). Zoological Journal of the reflect the views and conclusions of the federal government. Linnean Society 141:81–133. Peters, K. M., R. E. Matheson Jr., and R. G. Taylor. 1998. Reproduction and early life history of common snook, Centropomus undecimalis (Bloch), in Florida. Bulletin of Marine Science 62:509–529. REFERENCES Robins, J., D. Mayer, J. Staunton-Smith, I. Halliday, B. Sawynok, and M. Sellin. Adams, A., R. K. Wolfe, N. Barkowski, and D. Overcash. 2009. Fidelity to 2006. Variable growth rates of the tropical fish barramundi Lates calcarifer spawning grounds by a catadromous fish, Centropomus undecimalis. Marine (Bloch) under different freshwater flow conditions. Journal of Fish Biology Ecology Progress Series 389:213–222. 69:379–391. Ager, L. A., D. E. Hammond, and F. Ware. 1978. Artificial spawning of snook. Secor, D. H. 2008. Influence of skipped spawning and misspecified reproductive Proceedings of the Annual Conference Southeastern Association of Fish and schedules on biological reference points in sustainable fisheries. Transactions Wildlife Agencies 30(1976):158–166. of the American Fisheries Society 137:782–789. Bell, G., S. Goldenberg, C. Landsea, E. Blake, M. Chelliah, R. Pasch, and Shafland, P. L., and K. J. Foote. 1983. A lower lethal temperature for K. Mo. 2004. The 2004 North Atlantic hurricane season: a climate perspec- fingerling snook, Centropomus undecimalis. Northeast Gulf Science 6: tive. National Oceanic and Atmospheric Administration, National Weather 175–177. Service, Climate Prediction Center, Camp Springs, Maryland. Simpfendorfer, C. A., M. R. Heupel, and A. B. Collins. 2008. Variation in Bennett, J. 2006. Using acoustic telemetry to estimate natural and fishing mor- the performance of acoustic receivers and its implication for positioning tality of common snook in Sarasota Bay, Florida. Master’s thesis. University algorithms in a riverine setting. Canadian Journal of Fisheries and Aquatic of Florida, Gainesville. Sciences 65:482–492. Blewett, D. A., P. W. Stevens, T. R. Champeau, and R. G. Taylor. 2009. Use of Stevens, P. W., D. A. Blewett, T. R. Champeau, and C. J. Stafford. 2010. Posthur- rivers by common snook Centropomus undecimalis in southwest Florida: a ricane recovery of riverine fauna reflected in the diet of an apex predator. first step in addressing the overwintering paradigm. Florida Scientist 72:310– Estuaries and Coasts 33:59–66. 324. Taylor, R. G., H. J. Grier, and J. A. Whittington. 1998. Spawning rhythms of Chapman, P., G. Horel, W. Fish, K. Jones, and J. Spicola. 1978. Artificial culture common snook in Florida. Journal of Fish Biology 53:502–520. of snook: Rookery Bay, 1977. Annual Report on Sportfish Introductions, Taylor, R. G., J. A. Whittington, H. J. Grier, and R. E. Crabtree. 2000. Age, Florida Game and Fresh Water Fish Commission, Tallahassee. growth, maturation, and protandric sex reversal in common snook, Centropo- Gillanders, B. M., and M. J. Kingsford. 2002. Impact of changes in flow of fresh- mus undecimalis, from the east and west coasts of south Florida. U.S. National water on estuarine and open coastal habitats and the associated organisms. Marine Fisheries Service Fishery Bulletin 98:612–624.

Downloaded by [Department Of Fisheries] at 20:35 25 September 2012 Oceanography and Marine Biology: An Annual Review 40:233–309. Tucker, J. W., Jr. 1987. Snook and tarpon snook culture and preliminary evalu- Hammett, K. M. 1990. Land use, water use, streamﬂow, and water-quality ation for commercial farming. Progressive Fish-Culturist 49:49–57. characteristics of the Charlotte Harbor inﬂow area, Florida. U.S. Geological Tucker, J. W., Jr., and S. W. Campbell. 1988. Spawning season of common Survey, Record US9101367, Washington, D.C. snook along the east central Florida coast. Florida Scientist 51:1–6. Heupel, M. R., and C. A. Simpfendorfer. 2008. Movement and distribution of Volpe, A. V. 1959. Aspects of the biology of the common snook, Centropomus young bull sharks Carcharhinus leucas in a variable estuarine environment. undecimalis (Bloch), of southwest Florida. Florida Board of Conservation Aquatic Biology 1:277–289. Marine Research Laboratory Technical Series 31. This article was downloaded by: [Department Of Fisheries] On: 25 September 2012, At: 20:36 Publisher: Taylor & Francis Informa Ltd Registered in England and Wales Registered Number: 1072954 Registered office: Mortimer House, 37-41 Mortimer Street, London W1T 3JH, UK

Transactions of the American Fisheries Society Publication details, including instructions for authors and subscription information: http://www.tandfonline.com/loi/utaf20 Spring and Summer Distribution and Habitat Use by Adult Threatened Spotted Gar in Rondeau Bay, Ontario, Using Radiotelemetry W. R. Glass a , L. D. Corkum a & N. E. Mandrak b a Department of Biological Sciences, University of Windsor, 401 Sunset Avenue, Windsor, Ontario, N9B 3P4, Canada b Great Lakes Laboratory for Fisheries and Aquatic Sciences, Fisheries and Oceans Canada, 867 Lakeshore Road, Burlington, Ontario, L7R 4A6, Canada Version of record first published: 26 Jun 2012.

To cite this article: W. R. Glass, L. D. Corkum & N. E. Mandrak (2012): Spring and Summer Distribution and Habitat Use by Adult Threatened Spotted Gar in Rondeau Bay, Ontario, Using Radiotelemetry, Transactions of the American Fisheries Society, 141:4, 1026-1035 To link to this article: http://dx.doi.org/10.1080/00028487.2012.675904

PLEASE SCROLL DOWN FOR ARTICLE

Full terms and conditions of use: http://www.tandfonline.com/page/terms-and-conditions 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. The publisher does not give any warranty express or implied or make any representation that the contents will be complete or accurate or up to date. The accuracy of any instructions, formulae, and drug doses should be independently verified with primary sources. The publisher shall not be liable for any loss, actions, claims, proceedings, demand, or costs or damages whatsoever or howsoever caused arising directly or indirectly in connection with or arising out of the use of this material. Transactions of the American Fisheries Society 141:1026–1035, 2012 C American Fisheries Society 2012 ISSN: 0002-8487 print / 1548-8659 online DOI: 10.1080/00028487.2012.675904

ARTICLE

Spring and Summer Distribution and Habitat Use by Adult Threatened Spotted Gar in Rondeau Bay, Ontario, Using Radiotelemetry

W. R. Glass* and L. D. Corkum Department of Biological Sciences, University of Windsor, 401 Sunset Avenue, Windsor, Ontario N9B 3P4, Canada N. E. Mandrak Great Lakes Laboratory for Fisheries and Aquatic Sciences, Fisheries and Oceans Canada, 867 Lakeshore Road, Burlington, Ontario L7R 4A6, Canada

Abstract The spotted gar Lepisosteus oculatus is designated as threatened in Canada under the federal Species at Risk Act. Identiﬁcation and protection of critical habitat is an important component of recovering species at risk. To understand the habitat utilization of the adult life stage of the spotted gar in Rondeau Bay, a shallow coastal wetland of Lake Erie, external radio transmitters were surgically attached to 37 specimens in May 2007. These individuals were tracked at 224 discrete locations throughout the spring and summer of 2007. Aquatic macrophytes were present at 201 (90%) of these sites. Habitat and water chemistry data were collected at all tracked locations occupied by spotted gars. On the basis of electivity indices, in spring spotted gars showed a strong preference for shallow (<0.5-m) and deep (>2.5-m) waters with pH values <8.5. In summer, strong preference was shown for areas with mixed macrophyte beds. Spotted gars were found to relate to speciﬁc depths and cover rather than to shoreline features in Rondeau Bay. The study results are being used by the spotted gar Recovery Team to identify critical habitat in Rondeau Bay. This critical habitat designation will be used to ensure the protection of habitat needed to preserve the species in Canada, in part by curtailing the removal of aquatic vegetation in Rondeau Bay.

Preservation of the habitat that is used by a species at risk is risk are rare, so that defining their critical habitat may be difficult paramount to the long-term survival of the species (Rosenfeld (Naumann and Crawford 2009). One method of determining and Hatfield 2006). The Canadian Species at Risk Act defines the habitat used by a specific life stage of a species is to monitor Downloaded by [Department Of Fisheries] at 20:36 25 September 2012 this critical habitat for aquatic species as “spawning grounds and the movements of individuals using radio telemetry. In this nursery, rearing, food supply, migration and any other areas on manner, the feeding, spawning, nursery, and other important which aquatic species depend directly or indirectly in order to habitats can be determined for a species. This method has carry out their life processes” (Species at Risk Act 2002, s. 2[1]). been used on a wide variety of species at risk, including lesser Rosenfeld and Hatfield (2006) outlined four key information horseshoe bats Rhinolophus hipposideros (Bontadina et al. needs to identify critical habitat, including basic organism life 2002), giant barred river frogs Mixophyes iterates (Lemckert history (and habitat associations), habitat availability, recovery and Brassil 2000), and lake sturgeon Acipenser fulvescens targets, and habitat–abundance relationships. (Auer 1999). Radio-tagging and tracking in this manner have Habitat associations may not be known for rare or at-risk no negative effect on the behavior and swimming performance species, and thus an effective means of determining which of fish (Cooke 2003; Thorstad et al. 2001). Once habitat use habitat is used by the species is needed. By definition, species at by the species is determined, comparisons with the availability

*Corresponding author: [email protected] Received June 22, 2011; accepted February 27, 2012 Published online June 26, 2012 1026 THREATENED SPOTTED GAR IN RONDEAU BAY 1027

of habitat types are made using an electivity index (Jacobs 1974) to show whether certain habitat intervals are preferred or avoided (Moyle and Baltz 1985; Luttrell et al. 2002). ThespottedgarLepisosteus oculatus is designated as threatened under the Canadian Species at Risk Act. This species is at the northern edge of its range in Canada, inhabiting three coastal wetlands of Lake Erie: Point Pelee; Long Point Bay; and Ron- deau Bay, the largest of the Canadian populations (COSEWIC 2005). Spotted gars range as far south as the Gulf of Mexico, from eastern Texas in the west to the Florida panhandle in the east and are generally common south of the Great Lakes region (COSEWIC 2005). The threatened designation in Canada is due to its limited distribution and the threats posed by pollution, turbidity, and habitat loss (COSEWIC 2005) and means that the spotted gar is likely to become endangered if steps are not taken to reverse the factors leading to its extirpation in Canada (COSEWIC 2010). Although the movement and habitat use of the spotted gar was reported by Snedden et al. (1999) for a southern population in the Atchafalaya River basin of Louisiana, there has yet to be any characterization of habitat use by the species in Canada. The objectives of our study were to perform a radio- tracking survey of the spotted gar in Rondeau Bay, to describe the spring and summer distribution and critical habitat of this species in Canada, and to compare that habitat use with that of the Atchafalaya River basin population studied by Snedden et al. (1999).

METHODS Study site.—Our study was conducted in Rondeau Bay, a shallow (maximum depth of 3 m) coastal wetland on the north shore of the central basin of Lake Erie (Figure 1). Rondeau Bay is characterized by abundant submerged macrophyte growth, and its area (approximately 37 km2) is nearly enclosed. The bay is bounded by Rondeau Provincial Park on the east and by the town of Erieau in the south, with the remainder of the area being bordered by agricultural land with some residential development (Figure 1). There is a navigational channel in the

Downloaded by [Department Of Fisheries] at 20:36 25 September 2012 southern portion of the bay at Erieau that provides connectivity to the central basin of Lake Erie. Specimen collection and tagging.—Individual spotted gar specimens were collected from May 17 to May 23, 2007. Thirty- seven specimens were captured using 1.2-m fine-mesh fyke nets (6.35-mm bar mesh) set for approximately 24 h and retrieved in FIGURE 1. Maps showing (a) the location of Rondeau Bay on the north shore the morning. Nets were set in shallow areas adjacent to shore, of Lake Erie and (b) the sites within Rondeau Bay where spotted gar were captured, tagged, and released. targeting spawning-related movements (Figure 1b). After specimens were weighed (kg) and measured for total length (mm), fish were anesthetized in a 0.015% clove oil solution (3 mL 745 mm and weighed from 0.53 to 1.94 kg. The radio tags, man- clove oil emulsified with 5 mL ethanol, in 20 L water). Radio ufactured by Holohil Systems Limited (model PD-2), measured tags with unique frequencies (Table 1) were attached externally 23 × 12 × 6 mm, with an antenna 24 cm long; battery life to the dorsal musculature immediately behind the posterior in- was approximately 4 months. Tag weight (3.8 g) was <1% of sertion of the dorsal fin, following the procedure of Snedden the body weight of the smallest specimen. Small tag size and at- et al. (1999). Tagged specimens ranged in length from 515 to tachment at the base of the dorsal fin ensured that the swimming 1028 GLASS ET AL.

TABLE 1. Capture date, time at liberty, and number of times located for from a boat using a Lotek tracking receiver set to cycle spotted gar specimens in Rondeau Bay in spring 2007. through the tag frequencies. Once a specimen’s signal was Radio tag Date Days at Days Total located, its position was homed in on and a handheld GPS frequency tagged liberty located locations unit was used to determine the coordinates. Water depth (m), surface temperature (◦C), pH, and conductivity (µS/cm) were 151.242 May 23 130 1 1 measured using a Hydrolab Surveyor 4a with Datasonde 5. 151.270 May 23 130 13 15 Additionally, aquatic macrophyte samples were taken when 151.299 May 23 130 9 9 present and brought back to the laboratory for identification 151.320 May 23 130 1 1 to the genus level. Once fish were located, their tag frequency 151.340 May 23 130 7 7 was removed from the cycle list in the receiver so that 151.360 May 23 130 2 2 specimens were located a maximum of once per tracking 151.380 May 24 129 10 11 bout. 151.400 May 24 129 8 9 Tracking of specimens was conducted from the end of May 151.420 May 24 129 2 2 through September 2007 on at least 3 d per week and up to 151.440 May 24 129 7 7 5 d per week. Multiple tracking bouts were conducted over 151.460 May 17 136 2 29 a 24-h period on July 11 and 25, 2007. Tracking effort was 151.481 May 17 136 9 9 concentrated within Rondeau Bay; however, several attempts 151.500 May 17 136 10 10 were made to locate fish outside the bay, without success. Once 151.521 May 17 136 2 2 tracking was completed, ARCMap GIS software was used to 151.541 May 17 136 7 7 map all location coordinates for each individual (Figure 2a). 151.560 May 17 136 0 0 All the locations where spotted gars were tracked in Rondeau 151.579 May 17 136 4 4 Bay were noted (Figure 2b). We employed a modification of 151.600 May 17 136 6 6 the technique used by McGrath and Austin (2009) to determine 151.620 May 17 136 4 4 whether the number of times a specimen was located was 151.637 May 17 136 0 0 sufficient to describe its distribution. A series of minimum 151.661 May 17 136 2 2 convex polygons that enclosed all these points was created 151.680 May 17 136 2 2 (confer Winter 1977). Minimum convex polygons were built 151.700 May 17 136 7 7 after each tracking point was sequentially added to the map 151.720 May 17 136 9 9 (instead of daily tallies, as in McGrath and Austin 2009). The 151.740 May 17 136 7 7 area of the polygons was calculated using ARCMap. Once all 151.762 May 17 136 3 3 points had been mapped and the area of each polygon measured, 151.780 May 17 136 2 2 we plotted the area of the cumulative minimum convex polygon 151.800 May 18 135 6 6 against the number of times a specimen was located. The 151.820 May 31 122 20 24 leveling out of the curve for an individual specimen indicates 151.840 May 23 130 5 5 that there are sufficient data points to describe its distribution. 151.860 May 23 130 12 14 Several individuals exhibited a distinct clustering of points 151.880 May 23 130 15 16 (four or more points in proximity) where they were located 151.900 May 23 130 4 4 several times in the summer. To determine whether specimens 151.921 May 23 130 3 3 were associated with nearshore or offshore habitats, the distance

Downloaded by [Department Of Fisheries] at 20:36 25 September 2012 151.942 May 23 130 2 2 from shore to the closest of these clustered points was measured 151.961 May 23 130 3 3 for each individual. Also, the farthest linear distance between 151.980 May 23 130 7 7 two tracking locations and the maximum distance from point of capture were measured for each specimen as a surrogate for home range. Regression analysis was used to determine the relationships between (1) fish size (total fish length, weight) ability of specimens would not be impeded. Handling, surgeries, and distance from shore to the clustered points; (2) fish size and and recovery were conducted immediately at the site of capture. the maximum distance from capture; and (3) fish size and the Specimens were held in a recovery bin after surgery until they maximum distance between points. were able to maintain equilibrium. They were then released back Habitat variables.—Tracking locations were divided into into the bay at the capture site. All animal handling and surgeries two groups based on season: spring (May and June, which in- were approved by the animal care committees of the University cludes the spawning period for this species) and summer (July of Windsor and the Canada Centre for Inland Waters. to September). ARCMap was used to interpolate habitat values Tracking of specimens and distribution mapping.—The for the entire area of Rondeau Bay by inverse distance weight- movement and subsequent location of specimens were tracked ing based on the values collected at tracking locations. Habitat THREATENED SPOTTED GAR IN RONDEAU BAY 1029

where r is the proportion of individuals using the interval and p is the proportion of the overall habitat that has this value (Luttrell et al. 2002). These electivity indices are interpreted according to Moyle and Baltz (1985), whereby a value from –1.00 to –0.50 indicates strong avoidance, a value from –0.49 to –0.26 indicates moderate avoidance, –0.25 to + 0.25 indicates neutral selection, 0.26 to 0.49 indicates moderate selection, and 0.50 to 1.00 indicates strong selection. Population size and area of suitable habitat.—In May 2009 a mark–recapture study was conducted in Lake Pond, a marsh at Point Pelee National Park. The Point Pelee marsh is a coastal wetland of Lake Erie with habitat similar to Rondeau Bay. The contiguous surface area of the marsh is approximately 220 ha, and the marsh has no connection to the main basin of Lake Erie. Spotted gars were captured using 1.2-m ﬁne-mesh fyke nets (6.35-mm mesh) set overnight. Captured specimens (n = 93) were marked using PIT tags and released immediately after handling. A total of 99 spotted gars were captured and released, of which 6 were recaptured during the sampling. Based on this sampling, the total population of spotted gars in the Point Pelee marsh was estimated to be 483 individuals, with a density of 2.2 individuals/ha. To estimate the population size of spotted gars in Rondeau Bay, we used the population density estimate for the Point Pelee marsh and, assuming similar habitat and population density at the locations, multiplied it by the area of Rondeau Bay.

RESULTS Tracking and Distribution Mapping Of the 37 radio-tagged individuals, 35 were located at least once (Table 1). One tag was presumed lost when the individual was tracked on consecutive days to the same location in very shallow water and no ﬁsh was evident. All subsequent locations for this tag were removed from the analysis. The fate of the second tag that was not located is unknown. Each individual was located a mean ± SD of 6.19 ± 4.96 occasions, for a total of 224 discrete locations. When the cumulative area of the minimum convex polygon

Downloaded by [Department Of Fisheries] at 20:36 25 September 2012 was plotted against the number of times a fish was located, the curve appeared to level off for 10 individuals, (Figure 4) indicating that the tracking effort was sufficient to describe the FIGURE 2. (a) Locations of a single spotted gar specimen (tag number overall distribution for these specimens. 151.270), as determined by radio-tracking in Rondeau Bay during spring and summer 2007, and (b) all tracking locations of radio-tagged spotted gar speci- There was no significant relationship between fish length mens in Rondeau Bay during spring and summer 2007. and the offshore distance of clustered points (P = 0.17). The mean ± SD offshore distance of these clusters was 1.77 ± 1.58 km. There was a significant negative relationship layers were created for each of the measured variables separately between the loge-transformed weight of specimens and the by season. These habitat layers were then compared with the ob- offshore distance of the clusters (loge[offshore distance] = served habitat variables at all spotted gar locations to calculate 2 –0.68 loge[weight] + 5.02; R = 0.36, P = 0.02). electivity indices (Jacobs 1974). The electivity index (D)for When all specimens were considered, the mean ± SD far- each interval of a variable’s distribution is calculated as follows: thest distance from capture and mean ± SD farthest distance between two points were 2.95 ± 1.76 km and 3.47 ± 2.25 km, D = [r − p]/[(r + p) − 2rp], respectively. Regression analysis revealed no significant 1030 GLASS ET AL. Downloaded by [Department Of Fisheries] at 20:36 25 September 2012

FIGURE 3. Interpolated raster of the number of aquatic macrophyte genera present in Rondeau Bay.

relationship between the loge-transformed length and the far- Habitat Variables thest distance from capture (P = 0.17) or between length and the Interpolated raster layers were created for each habitat farthest distance between points (P = 0.19). There was, however, variable (e.g., Figure 3). The electivity indices showed strong a marginally signiﬁcant relationship between loge-transformed positive selection by the spotted gar for several habitat intervals weight and the farthest distance from capture and between in spring (Table 2) and summer (Table 3). In spring, spotted weight and the farthest distance between locations. These gars exhibited a preference for both the shallowest (<0.5-m) relationships were loge(distance from capture) = 1.02·loge and the deepest (>2.5-m) waters, areas with no macrophyte 2 (weight) – 5.94 (R = 0.13, P = 0.033) and loge(distance be- growth, waters with conductivity levels >325 µS/cm or 2 tween points) = 1.08·loge(weight) – 6.18 (R = 0.13, P = 0.036). <225–249.9 µS/cm, pH values <8.5, and pH values ≥9.50. THREATENED SPOTTED GAR IN RONDEAU BAY 1031

FIGURE 4. Plot of cumulative maximum convex polygon area versus number of times located. The inset shows the data for the individual with tag number 151.541 (see Table 1). Downloaded by [Department Of Fisheries] at 20:36 25 September 2012

The habitat interval of moderate depths (1.00–1.99 m) was A spawning event was witnessed on June 12. This spawning strongly avoided. In summer, habitats strongly selected by the activity took place in a mixed bed of macrophytes that included spotted gar were those with the deepest depths (>2.5 m) and the Myriophyllum spp. and Ceratophyllum spp. located 391 m from shallowest depths (<0.5 m), areas with two or more macrophyte shore. The spawning event consisted of a single large female sur- genera present, and waters with pH values between 8.0 and 8.49. rounded by three smaller males thrashing around in the shallow Of the 224 locations to which spotted gars were tracked, vegetation. 201 (90%) had some form of aquatic vegetation. Seven sites had emergent vegetation only, nine sites had both emergent and Population Size and Area of Suitable Habitat submerged vegetation, and 185 sites had submerged vegetation Based on the population density estimate (2.2/ha) from only. A large proportion of the sites contained complex, or highly the Point Pelee marsh and the total area of Rondeau Bay branched, vegetation. It was common to have sites represented (3,215 ha), the population of spotted gars in Rondeau Bay is by several genera of plants (Table 4). approximately 8,121 individuals. The area of suitable habitat 1032 GLASS ET AL.

TABLE 2. Electivity indices and level of selection for habitat variable intervals for spotted gars in May and June.

Habitat variable Habitat interval Electivity indexa Selection level Macrophyte growth No macrophytes 0.78 Strong selection Single macrophyte –0.36 Moderate avoidance Mixed macrophytes –0.44 Moderate avoidance Depth (m) <0.50 0.90 Strong selection 0.50–0.99 0.29 Moderate selection 1.00–1.49 –0.67 Strong avoidance 1.50–1.99 –0.72 Strong avoidance 2.00–2.49 –0.40 Moderate avoidance ≥2.50 0.84 Strong selection Temperature (◦C) 17.00–19.99 NA NA 20.00–22.99 0.25 Neutral selection 23.00–25.99 –0.3 Moderate avoidance ≥26.00 0.24 Neutral selection Conductivity (µS/cm) <225.0 NA NA 225.0–249.9 0.86 Strong selection 250.0–274.9 –0.04 Neutral selection 275.0–299.9 –0.57 Strong avoidance 300.0–324.9 –0.13 Neutral selection 325.0–349.9 0.67 Strong selection ≥350.0 0.97 Strong selection pH <8.0 0.99 Strong selection 8.0–8.49 0.57 Strong selection 8.50–8.99 –0.50 Moderate avoidance 9.0–9.49 –0.33 Moderate avoidance ≥9.50 0.74 Strong selection

aValues from –1.00 to –0.50 indicate strong avoidance, those from –0.49 to –0.26 moderate avoidance, –0.25 to + 0.25 neutral selection, 0.26 to 0.49 moderate selection, and 0.50 to 1.00 strong selection (Moyle and Baltz 1985). NA indicates that no values were recorded in that range in ﬁeld observations and thus did not appear in the interpolated layer.

based on our raster interpolation of vegetation complexity fish were mainly associated with flooded timber (Snedden et al. (Figure 3), conservatively determined by the proportion of 1999). In Rondeau Bay, spotted gars were often found in mixed Rondeau Bay with two or more macrophyte genera, is 1,543 beds of complex macrophytes. Like the timber in the Snedden ha. A less conservative estimate, the total proportion of the bay et al. (1999) study, complex macrophyte beds created a three- with either two or more macrophyte genera or no macrophytes dimensional environment in which the spotted gars could hide present, is 1,884 ha. These areas were chosen as surrogates and forage. This habitat type (specifically vegetation density) for suitable habitat area because spotted gar feeding success has been shown to be important for the feeding success of spotted has been shown to depend on the macrophyte complexity of gars (Ostrand et al. 2004). The potential loss of habitat is one of Downloaded by [Department Of Fisheries] at 20:36 25 September 2012 the cover present (Ostrand et al. 2004). Additionally, most the limiting factors for the recovery of spotted gar populations in spotted gars were found in areas with two or more macrophyte Canada (COSEWIC 2005). Specifically, the removal of aquatic genera present (Table 4). Areas with no macrophytes present vegetation by both physical and chemical means represents a were strongly selected by spotted gars in the spring (Table 2). high-impact activity that disturbs spotted gars in Rondeau Bay Other habitat variables, such as pH, temperature, and conduc- (Bouvier and Mandrak 2010). Removal of aquatic vegetation tivity, were not used to identify habitat area because they varied should be curtailed given the finding that the spotted gars in with changing weather conditions. Rondeau Bay are dependent on aquatic macrophytes throughout the spring and summer periods. There was also strong selection for areas without vegeta- DISCUSSION tion in the spring. Interestingly, our findings showed that only The spotted gar specimens tracked in this study were most 11% of Rondeau Bay lacked vegetation in the spring. These often found associated with aquatic vegetation. This association unvegetated areas may be used for postspawn feeding since the with aquatic macrophytes as cover shows an adaptation to local spring-spawning minnows (e.g., spottail shiner Notropis hud- conditions in Rondeau Bay when compared with the spotted gar sonius) present in sandy-bottomed areas (Scott and Crossman population of the Lower Atchafalaya River, Louisiana, where 1998) provide ample prey for spotted gars. THREATENED SPOTTED GAR IN RONDEAU BAY 1033

TABLE 3. Electivity indices and level of selection for habitat variable intervals for spotted gars in July through September.

Habitat variable Habitat interval Electivity indexa Selection level Macrophyte growth No macrophytes –0.32 Moderate avoidance Single macrophyte –0.46 Moderate avoidance Mixed macrophytes 0.50 Strong selection Depth (m) <0.50 0.64 Strong selection 0.50–0.99 0.04 Neutral selection 1.00–1.49 –0.61 Strong avoidance 1.50–1.99 –0.08 Neutral selection 2.00–2.49 0.42 Moderate selection ≥2.50 0.87 Strong selection Temperature (◦C) 17.00–19.99 0.63 Strong selection 20.00–22.99 0.05 Neutral selection 23.00–25.99 –0.40 Moderate avoidance ≥26.00 0.51 Strong selection Conductivity (µS/cm) <225.0 0.65 Strong selection 225.0–249.9 –0.56 Strong avoidance 250.0–274.9 0.29 Moderate selection 275.0–299.9 0.91 Strong selection 300.0–324.9 NA NA 325.0–349.9 NA NA ≥350.0 NA NA pH <8.0 NA NA 8.0–8.49 0.94 Strong selection 8.50–8.99 0.34 Moderate selection 9.0–9.49 –0.25 Moderate avoidance ≥9.50 0.09 Neutral selection

Early in the season, spotted gars were often found near shore. In the Atchafalaya River basin, Snedden et al. (1999) found Movement into the shallows was likely due to the spawning that spotted gars tended to migrate into ﬂooded areas in the behavior of the species. The spotted gar is known to spawn in spring, followed by the establishment of home ranges for the spring in shallow water among aquatic vegetation (Redmond duration of the high-water stage. The average distance from 1964). In the summer, spotted gars tended to move offshore the shore to the site of repeated location for the spotted gar and several individuals were repeatedly tracked to the same specimens in Rondeau Bay was much farther (mean ± SD = location. Similarly, Snedden et al. (1999) found that spotted 1.77 ± 1.58 km) than that reported by Snedden et al. (1999), gars established deﬁned home ranges in the summer. where 48% of all spotted gar movements were within 10 m from Downloaded by [Department Of Fisheries] at 20:36 25 September 2012

TABLE 4. Composition of submerged macrophytes present at spotted gar tracking locations. Asterisks indicate a complex, or highly branched, macrophyte type.

Number Sites as Sites dominant species Sites secondary species Genus of sites present lone species in mixed bed in mixed bed Chara* 68 21 39 8 Potamageton* 86 5 34 47 Myriophyllum* 61 6 25 30 Ceratophyllum* 20 1 5 14 Elodea* 40 0 4 Valisneria 59 1 2 56 Lemna 10 0 1 None present 22 NA NA NA 1034 GLASS ET AL.

shore. This difference in behavior likely results from habitat similarity of habitats, the population density should be similar differences between the two areas. Rondeau Bay is shallow, in the two locations. with an extended littoral zone and macrophyte cover throughout, Our sampling failed to collect any specimens less than 3 years while the Atchafalaya River basin is narrower and has depths old, which is the presumed age of maturity for spotted gars ranging from 3 to 5 m (low water stage in the Snedden et al. (Glass et al. 2011). Thus, additional studies are required to 1999 study area). The Atchafalaya River basin, unlike Rondeau identify the critical habitat for the young-of-the-year, juvenile, Bay, generally lacks aquatic vegetation (Snedden et al. 1999). and subadult life stages. Nevertheless, our current findings will Evidently, spotted gars are relating to specific depths and cover be used by the Spotted Gar Recovery Team to define critical rather than the shoreline features in Rondeau Bay. habitat and recovery targets for the spotted gar recovery strategy, Our habitat layers were created based on a relatively small leading to the protection of areas with aquatic macrophytes and number of points compared with the size of Rondeau Bay. The other critical areas of Rondeau Bay. These actions will assist in limitations in our method are apparent in cases where there the conservation of the species. were no observed values in a particular range. In such cases, the interpolated habitat layer also lacks values in the range. The observations on which interpolations were based were well ACKNOWLEDGMENTS spread throughout the bay. Given the lack of available habitat We thank Aaron Simpson for his assistance with the field maps and associated data for our study, we were limited to work and Christopher Bunt for demonstrating the tagging and interpolating habitat values for the entire study area. tracking procedure. We also thank Alice Grgikak-Mannion and The moderate preference for spring surface temperatures Hongcheng Zeng for assistance with GIS mapping and inter- ◦ (20–23 C) is indicative of the preferred spawning temperature polation, along with anonymous reviewers whose comments of spotted gars in spring. Snedden et al. (1999) reported that helped to improve the manuscript. Funding for this research spawning-related movements began when temperatures reached was provided by Fisheries and Oceans Canada’s species at risk ◦ 15 C. Boudreaux (2005) reported spawning activity in a labo- program and the Government of Canada’s interdepartmental re- ◦ ratory at a mean temperature of 20.6 C. covery fund. The strong selection of the high surface temperature interval (>26◦C) in the summer for the specimens in Rondeau Bay likely reflects preferred feeding temperatures. This temperature REFERENCES ◦ was much higher than the preferred water temperature of 16 C Auer, N. A. 1999. Population characteristics and movements of lake sturgeon reported by Coker et al. (2001) for spotted gars in Canada. The in the Sturgeon River and Lake Superior. Journal of Great Lakes Research physostomous gas bladder, common to all gar species, allows 25:282–293. Bontadina, F., H. Schofield, and B. Naef-Daenzer. 2002. Radio-tracking reveals the spotted gar to obtain atmospheric oxygen and thus provides that lesser horseshoe bats (Rhinolophus hipposideros) forage in woodland. an advantage over many other predatory species in warmwaters Journal of Zoology 258:281–290. and the low oxygen concentrations that often result. Smatresk Boudreaux, P. J. 2005. Acute ammonia toxicity and chloride inhibition of nitrite and Cameron (1982) showed that spotted gars increase their uptake in non-teleost actinopterygiian fishes. Master’s thesis. Nicholls State rate of air breathing when temperatures are higher, and the University, Thibodaux, Louisiana. Bouvier, L. D., and N. E. Mandrak. 2010. Information in support of a recov- use of the physostomous gas bladder is significantly higher at ◦ ◦ ery potential assessment of spotted gar (Lepisosteus oculatus) in Canada. 30 C than at 20 C. Our study also showed a preference for low Fisheries and Oceans Canada, Research Document 2010/079, Burlington, ◦ temperatures (17–19.9 C) later in the sampling period. This Ontario. finding was influenced by individuals inhabiting offshore areas Coker, G. A., C. B. Portt, and C. K. Minns. 2001. Morphological and ecological characteristics of Canadian freshwater fishes. Fisheries and Oceans Downloaded by [Department Of Fisheries] at 20:36 25 September 2012 in the early fall. Canada, Canadian Manuscript Report of Fisheries and Aquatic Sciences 2554, Conservation of the spotted gar, a native top predator, in Burlington, Ontario. Canada will hinge on protection of its critical habitat for all life Cooke, S. J. 2003. Externally attached radio transmitters do not affect the stages. Our study indicates that spotted gars use emergent and parental care behaviour of rock bass. Journal of Fish Biology 62:965–970. submerged aquatic macrophyte beds in both the nearshore and COSEWIC (Committee on the Status of Endangered Wildlife in Canada). 2005. offshore areas of Rondeau Bay for feeding, cover, and spawning. Assessment and update status report on the spotted gar Lepisosteus oculatus in Canada. COSEWIC, Ottawa. Long-term survival of the species in Canada will require at COSEWIC (Committee on the Status of Endangered Wildlife in Canada). 2010. least 1,400 adult spotted gars (Young and Koops 2010) and Assessment process and criteria. COSEWIC, Ottawa. at least 360 ha of suitable habitat (DFO 2010). We show that DFO (Department of Fisheries and Oceans). 2010. Recovery potential assess- the population of spotted gars in Rondeau Bay is large enough ment of spotted gar (Lepisosteus oculatus) in Canada. DFO Canada, Science (8,121 individuals) and has sufficient suitable habitat (1,543– Advisory Report 2010/047, Ottawa. Glass, W. R., L. D. Corkum, and N. E. Mandrak. 2011. Pectoral fin ray aging: 1,884 ha) to be viable in the long term. Although this population an evaluation of a non-lethal method for aging gars and its application to a estimate is based on Point Pelee marsh data, Point Pelee and population of the threatened spotted gar. Environmental Biology of Fishes Rondeau Bay are similar, albeit different in size. Based on the 90:235–242. THREATENED SPOTTED GAR IN RONDEAU BAY 1035

Jacobs, J. 1974. Quantitative measurement of food selection. Oecologia 14:413– Rosenfeld, J. S., and T. Hatfield. 2006. Information needs for assessing critical 417. habitat of freshwater fish. Canadian Journal of Fisheries and Aquatic Sciences Lemckert, F., and T. Brassil. 2000. Movements and habitat use of the endangered 63:683–698. giant barred river frog (Mixophyes iterates) and the implications for its con- Scott, W. B., and E. J. Crossman. 1998. Freshwater fishes of Canada. Galt House servation in timber production forests. Biological Conservation 96:177–184. Publications, Oakville, Ontario. Luttrell, G. R., A. A. Echelle, and W. L. Fisher. 2002. Habitat correlates of Smatresk, N. J., and J. N. Cameron. 1982. Respiration and acid-base physiology the distribution of Macrhybopsis hyostoma (Teleostei: Cyprinidae) in west- of the spotted gar, a bimodal breather: II. responses to temperature change ern reaches of the Arkansas River. Transactions of the Kansas Academy of and hypercapnia. Journal of Experimental Biology 96:281–293. Science 105:153–161. Snedden, G. A., W. E. Kelso, and D. A. Rutherford. 1999. Diel and seasonal pat- McGrath, P., and H. A. Austin. 2009. Site fidelity, home range, and tidal move- terns of spotted gar movement and habitat use in the lower Atchafalaya River ments of white perch during the summer in two small tributaries of the York basin, Louisiana. Transactions of the American Fisheries Society 128:144– River, Virginia. Transactions of the American Fisheries Society 138:966–974. 154. Moyle, P. B., and D. M. Baltz. 1985. Microhabitat use by an assemblage of Species at Risk Act of 2002. S.C. 2002, c. 29, Canadian Department of Justice, California stream fishes: developing criteria for instream flow determinations. Ottawa. Transactions of the American Fisheries Society 114:695–704. Thorstad, E. B., F. Økland, and T. G. Heggberget. 2001. Are long term negative Naumann, B. T., and S. S. Crawford. 2009. Is it possible to identify habitat for a effects from external tags underestimated? fouling of an externally attached rare species? shortjaw cisco (Coregonus zenithicus) in Lake Huron as a case telemetry transmitter. Journal of Fish Biology 59:1092–1094. study. Environmental Biology of Fishes 86:341–348. Winter, J. D. 1977. Summer home range movements and habitat use by four Ostrand, K. G., B. J. Braeutigam, and D. H. Wahl. 2004. Consequences of largemouth bass in Mary Lake, Minnesota. Transactions of the American vegetation density and prey species on spotted gar foraging. Transactions of Fisheries Society 106:323–330. the American Fisheries Society 133:794–800. Young, J. A. M., and M. A. Koops. 2010. Recovery potential modelling of Redmond, L. C. 1964. Ecology of the spotted gar (Lepisosteus oculatus)in spotted gar (Lepisosteus oculatus) in Canada. Fisheries and Oceans Canada, southeastern Missouri. Master’s thesis. University of Missouri, Columbia. Research Document 2010/078, Burlington, Ontario. Downloaded by [Department Of Fisheries] at 20:36 25 September 2012 This article was downloaded by: [Department Of Fisheries] On: 25 September 2012, At: 20:36 Publisher: Taylor & Francis Informa Ltd Registered in England and Wales Registered Number: 1072954 Registered office: Mortimer House, 37-41 Mortimer Street, London W1T 3JH, UK

Transactions of the American Fisheries Society Publication details, including instructions for authors and subscription information: http://www.tandfonline.com/loi/utaf20 Spring and Summer Distribution and Habitat Use by Adult Threatened Spotted Gar in Rondeau Bay, Ontario, Using Radiotelemetry W. R. Glass a , L. D. Corkum a & N. E. Mandrak b a Department of Biological Sciences, University of Windsor, 401 Sunset Avenue, Windsor, Ontario, N9B 3P4, Canada b Great Lakes Laboratory for Fisheries and Aquatic Sciences, Fisheries and Oceans Canada, 867 Lakeshore Road, Burlington, Ontario, L7R 4A6, Canada Version of record first published: 26 Jun 2012.

PLEASE SCROLL DOWN FOR ARTICLE

Full terms and conditions of use: http://www.tandfonline.com/page/terms-and-conditions 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. The publisher does not give any warranty express or implied or make any representation that the contents will be complete or accurate or up to date. The accuracy of any instructions, formulae, and drug doses should be independently verified with primary sources. The publisher shall not be liable for any loss, actions, claims, proceedings, demand, or costs or damages whatsoever or howsoever caused arising directly or indirectly in connection with or arising out of the use of this material. Transactions of the American Fisheries Society 141:1026–1035, 2012 C American Fisheries Society 2012 ISSN: 0002-8487 print / 1548-8659 online DOI: 10.1080/00028487.2012.675904

ARTICLE

Spring and Summer Distribution and Habitat Use by Adult Threatened Spotted Gar in Rondeau Bay, Ontario, Using Radiotelemetry

*Corresponding author: [email protected] Received June 22, 2011; accepted February 27, 2012 Published online June 26, 2012 1026 THREATENED SPOTTED GAR IN RONDEAU BAY 1027

FIGURE 3. Interpolated raster of the number of aquatic macrophyte genera present in Rondeau Bay.

TABLE 2. Electivity indices and level of selection for habitat variable intervals for spotted gars in May and June.

TABLE 3. Electivity indices and level of selection for habitat variable intervals for spotted gars in July through September.

TABLE 4. Composition of submerged macrophytes present at spotted gar tracking locations. Asterisks indicate a complex, or highly branched, macrophyte type.

Jacobs, J. 1974. Quantitative measurement of food selection. Oecologia 14:413– Rosenfeld, J. S., and T. Hatfield. 2006. Information needs for assessing critical 417. habitat of freshwater fish. Canadian Journal of Fisheries and Aquatic Sciences Lemckert, F., and T. Brassil. 2000. Movements and habitat use of the endangered 63:683–698. giant barred river frog (Mixophyes iterates) and the implications for its con- Scott, W. B., and E. J. Crossman. 1998. Freshwater fishes of Canada. Galt House servation in timber production forests. Biological Conservation 96:177–184. Publications, Oakville, Ontario. Luttrell, G. R., A. A. Echelle, and W. L. Fisher. 2002. Habitat correlates of Smatresk, N. J., and J. N. Cameron. 1982. Respiration and acid-base physiology the distribution of Macrhybopsis hyostoma (Teleostei: Cyprinidae) in west- of the spotted gar, a bimodal breather: II. responses to temperature change ern reaches of the Arkansas River. Transactions of the Kansas Academy of and hypercapnia. Journal of Experimental Biology 96:281–293. Science 105:153–161. Snedden, G. A., W. E. Kelso, and D. A. Rutherford. 1999. Diel and seasonal pat- McGrath, P., and H. A. Austin. 2009. Site fidelity, home range, and tidal move- terns of spotted gar movement and habitat use in the lower Atchafalaya River ments of white perch during the summer in two small tributaries of the York basin, Louisiana. Transactions of the American Fisheries Society 128:144– River, Virginia. Transactions of the American Fisheries Society 138:966–974. 154. Moyle, P. B., and D. M. Baltz. 1985. Microhabitat use by an assemblage of Species at Risk Act of 2002. S.C. 2002, c. 29, Canadian Department of Justice, California stream fishes: developing criteria for instream flow determinations. Ottawa. Transactions of the American Fisheries Society 114:695–704. Thorstad, E. B., F. Økland, and T. G. Heggberget. 2001. Are long term negative Naumann, B. T., and S. S. Crawford. 2009. Is it possible to identify habitat for a effects from external tags underestimated? fouling of an externally attached rare species? shortjaw cisco (Coregonus zenithicus) in Lake Huron as a case telemetry transmitter. Journal of Fish Biology 59:1092–1094. study. Environmental Biology of Fishes 86:341–348. Winter, J. D. 1977. Summer home range movements and habitat use by four Ostrand, K. G., B. J. Braeutigam, and D. H. Wahl. 2004. Consequences of largemouth bass in Mary Lake, Minnesota. Transactions of the American vegetation density and prey species on spotted gar foraging. Transactions of Fisheries Society 106:323–330. the American Fisheries Society 133:794–800. Young, J. A. M., and M. A. Koops. 2010. Recovery potential modelling of Redmond, L. C. 1964. Ecology of the spotted gar (Lepisosteus oculatus)in spotted gar (Lepisosteus oculatus) in Canada. Fisheries and Oceans Canada, southeastern Missouri. Master’s thesis. University of Missouri, Columbia. Research Document 2010/078, Burlington, Ontario. Downloaded by [Department Of Fisheries] at 20:36 25 September 2012 This article was downloaded by: [Department Of Fisheries] On: 25 September 2012, At: 20:42 Publisher: Taylor & Francis Informa Ltd Registered in England and Wales Registered Number: 1072954 Registered office: Mortimer House, 37-41 Mortimer Street, London W1T 3JH, UK

Transactions of the American Fisheries Society Publication details, including instructions for authors and subscription information: http://www.tandfonline.com/loi/utaf20 A Standardized Method and Analytical Approach for Predicting Female Reproductive Stage in Teleosts by Using Ovary Color and Female Characteristics A. C. Peer a , G. M. Selckmann a b & T. J. Miller a a Chesapeake Biological Laboratory, University of Maryland Center for Environmental Science, Post Office Box 38, Solomons, Maryland, 20688, USA b Department of Biology, Hood College, 401 Rosemont Avenue, Frederick, Maryland, 21701, USA Version of record first published: 26 Jun 2012.

To cite this article: A. C. Peer, G. M. Selckmann & T. J. Miller (2012): A Standardized Method and Analytical Approach for Predicting Female Reproductive Stage in Teleosts by Using Ovary Color and Female Characteristics, Transactions of the American Fisheries Society, 141:4, 1036-1044 To link to this article: http://dx.doi.org/10.1080/00028487.2012.681103

PLEASE SCROLL DOWN FOR ARTICLE

Full terms and conditions of use: http://www.tandfonline.com/page/terms-and-conditions 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. The publisher does not give any warranty express or implied or make any representation that the contents will be complete or accurate or up to date. The accuracy of any instructions, formulae, and drug doses should be independently verified with primary sources. The publisher shall not be liable for any loss, actions, claims, proceedings, demand, or costs or damages whatsoever or howsoever caused arising directly or indirectly in connection with or arising out of the use of this material. Transactions of the American Fisheries Society 141:1036–1044, 2012 C American Fisheries Society 2012 ISSN: 0002-8487 print / 1548-8659 online DOI: 10.1080/00028487.2012.681103

NOTE

A Standardized Method and Analytical Approach for Predicting Female Reproductive Stage in Teleosts by Using Ovary Color and Female Characteristics

A. C. Peer,* G. M. Selckmann,1 and T. J. Miller Chesapeake Biological Laboratory, University of Maryland Center for Environmental Science, Post Ofﬁce Box 38, Solomons, Maryland 20688, USA

essential for assessing individual and population reproductive Abstract potential. Determining and understanding patterns in the reproductive Reproductive classification of fishes based on macroscopic status of fishes are essential for assessing individual and population external and internal appearance (i.e., ovary or oocyte color reproductive potential. The least biased method for determining reproductive stage is through the use of ovarian histology; however, and shape) of the gonads has been conducted and developed this method can be time consuming and expensive. To overcome for at least a century (e.g., Hjort 1914; Graham 1924; Hickling these restrictions, we developed models to predict the reproductive 1935; Vladykov 1956; Powles 1958; Kynard and Kieffer 2002). stage of female striped bass Morone saxatilis by using ovary color Although these methods are fast and simple, they have the po- and other female characteristics. We provide a standardized and tential to produce biased and subjective classifications (Kjesbu calibrated method for quantifying ovary color, and we outline an analytical approach that utilizes binary and ordinal logistic regres- 2009). This has led to the recent increase in reproductive clas- sion. Our results indicated that binary models had high accuracy sification by histological techniques, which provide the most (<6% error) in distinguishing between the regressing phase (i.e., detailed information about reproductive status and the least bi- recently spawned) and the nonregressing phase (i.e., all other re- ased classification (Hunter and Macewicz 1985). Specifically, productive phases); we found even better accuracy (4% error) for histological analyses provide detailed information about cellular ordinal models in discriminating among four nonregressing reproductive phases. All of the best-fitting models required the inclusion substructures and their prevalence during the entire reproduc- of an ovary color variable and ovary energy density; however, good tive cycle (Tomkiewicz et al. 2003). Unfortunately, histological predictive accuracy was also obtained when we replaced ovary en- methods are also time consuming and more expensive than sim- ergy density with ovary percent water to produce models requiring ple macroscopic observation, making it difficult to evaluate a minimal laboratory processing time and cost. Although tested on large number of samples. striped bass, our method could be used to develop similar models for other species. We considered whether histological and macroscopic methods could be combined to develop a model for use in predicting Downloaded by [Department Of Fisheries] at 20:42 25 September 2012 Standardized determination of the reproductive status of reproductive status based on macroscopic and other easily at- individual fish is vital for the accurate quantification of tainable characteristics of females. By combining the detailed reproductive and stock dynamics (Brown-Peterson et al. 2011). accuracy of histology with rapid, low-cost macroscopic meth- Reproductive characterization provides an understanding ods, one may be able to develop a predictive model that can of species- and population-specific reproductive cycles and accurately determine the entire scope of reproductive plasticity characteristics, such as spawning stock size, temporal and based on only macroscopic methods and other easily attain- spatial spawning patterns, size and age at sexual maturity, able female characteristics. This approach has been attempted and the distinction between reproductive and nonreproductive once before by Bryan et al. (2007), who used nonlethal en- (i.e., skipped spawning) individuals. Consequently, deter- doscopic methods to determine oocyte color, which was then mining and understanding patterns in reproductive status are used to predict histologically determined reproductive stages of

*Corresponding author: [email protected] 1Present address: Department of Biology, Hood College, 401 Rosemont Avenue, Frederick, Maryland 21701, USA. Received November 9, 2011; accepted March 24, 2012 Published online June 26, 2012 1036 NOTE 1037

shovelnose sturgeon Scaphirhynchus platorynchus. Although series of 70% (1×), 80% (2×), 95% (2×), and 100% (3×) their model proved successful at distinguishing between repro- ethanol for 20 min at each grade. This was followed by 20-min ductive and nonreproductive females, it was unable to decipher changes in xylene (2×), 30-min changes in low-temperature greater detail and it did not provide (or aim to provide) a stan- paraffin (3×), and embedding in a final change of paraffin. dardized, calibrated macroscopic method that could be used in Embedded samples were desiccated overnight, and thin sections future studies for other species. (3.5 µm) were cut with glass knives at 5◦. Sections were placed We developed models to predict the reproductive stage of onto drops of distilled water on glass slides and were heated female striped bass Morone saxatilis by using ovary color and at approximately 60◦C on a hotplate to adhere the sections to other female characteristics (i.e., gutted weight, age, ovary en- the slides. Sections were then stained with periodic acid–Schiff ergy density, liver energy density, ovary percent water, and liver reagent. percent water). Here, we describe a standardized, calibrated Histologically prepared ovary sections were viewed under method for quantifying ovary color and we outline an analyti- a light microscope and categorized into one of six reproduc- cal approach that utilizes binary and ordinal logistic regression; tive phases (Figure 1): (1) immature, (2) early developing, (3) these methods can be used to develop similar models for other developing, (4) spawning capable, (5) regressing, and (6) regen- species. erating. Detailed descriptions of these phases can be found in Table 2 of Brown-Peterson et al. (2011). Ovary color analysis.—Ovary color was quantified using im- METHODS age analysis. To determine red, green, and blue color values for Field sampling and ovary processing.—The female striped each ovary, the frozen ovary lobe was thawed to allow removal bass used in this study were collected in the main stem of of the outer membrane. A composite sample of oocytes (∼5g) the Chesapeake Bay near the mouth of the Patuxent River, was then removed and immediately homogenized with a mortar Maryland, in 2009 (n = 31) and 2010 (n = 109). Upon capture, and pestle while the sample was still cold. A 60-mm diameter, each collected fish was euthanized with tricaine methanesul- 15-mm-deep petri dish was filled to overflowing with the ho- fonate (MS-222) at 500 mg/L and was placed in an ice–water mogenized sample, ensuring that no air pockets were present. slurry to prevent atresia of oocytes. In the laboratory, gutted The petri dish was placed on flat-bed scanner (CanoScan 8800F; weight (g) was measured, and the otoliths, liver, and ovaries Canon, Lake Success, New York) with the lid closed. Prior to were removed. A small section (∼6 g) of one lobe of the ovary taking images, the scanner was calibrated with an IT8 calibra- (for use in histological analyses) was then immediately placed tion target and calibration software (MonacoEZcolor). Images in a chilled (4◦C) 4:1 solution of formaldehyde : glutaraldehyde of the filled petri dish (at 300 dots per inch) were then taken (4F:1G) and was refrigerated at 4◦C for 24 h. The remaining with all advanced image settings turned off, and the images portion of the lobe was frozen in water for later color analy- were saved in tagged image file format. sis. To determine the energy density and percent water of the Image files were analyzed using ImageJ version 1.45h ovaries, we measured the wet weight of the second lobe and (Rasband 2011). The color histogram plug-in was used to deter- placed it in a drying oven at 65◦C until a consistent mass was mine the mean red, green, and blue color values for the imaged obtained after two consecutive measurements separated by at sample. Based on the mean color values, we also determined least 2 d. The same drying procedure was used to determine the the proportion of each color and the red : green, red: blue, and energy density and percent water of the liver. blue : green ratios. Prior to estimating the energy density of ovaries and livers, Statistical analyses.—To predict the reproductive stage of dried tissues were ground to a fine powder using either mor- female striped bass, we used both binary and ordinal logistic

Downloaded by [Department Of Fisheries] at 20:42 25 September 2012 tar and pestle or a coffee grinder. Ground tissues were then regression methods. All models were fitted in R (R Develop- formed into at least two composite pellets (∼0.5 g), and the ment Core Team 2011) using the rms package (Harrell 2011). energy density was estimated using an oxygen bomb calorime- Our preliminary model development indicated that we would ter (Model 6200; Parr Instrument Company, Moline, Illinois) not be able to adequately fit an ordinal logistic regression model that was standardized with benzoic acid pellets (26.444 kJ/g). that included all six of the reproductive phases described above. Replicate pellets of each tissue from each female were ignited This preliminary evaluation indicated that the regressing, im- in the calorimeter; if the two estimates differed by more than mature, and regenerating phases would have to be pooled given 400 kJ/g (∼2%), a third pellet was ignited. Energy density of that these adjacent phases were not distinguished by use of each tissue was equal to the mean of two composite tissue pellets the predictors. However, given our desire to predict which fe- that differed by less than 2%. males were in the regressing phase, we decided to split our Histology.—Samples fixed in 4F:1G were rinsed in 0.1-M statistical analyses into two stages. In the first stage, we used Sorenson’s sodium phosphate buffer (pH 7.2–7.4) twice (10 min a binary logistic regression model to determine our ability to per rinse) and were placed in cassettes for histological analyses. correctly classify females as either regressing (i.e., recently Sequential dehydration was accomplished in a graded ethanol spawned) or nonregressing (i.e., all other phases) by using ovary 1038 PEER ET AL.

color, mean blue ovary color, red : green ovary color ratio, red : blue ovary color ratio, and green : blue ovary color ratio. We also included mean liver energy density (kJ/g) and liver percent water in the full model due to the important role of the liver in packag- ing and transporting vitellogenin (the plasma precursor of yolk) to oocytes (Lubzens et al. 2010). Subsequently, we used a backward step-down variable deletion approach based on Akaike’s information criterion to find the model that yielded the greatest reduction in deviance. We then performed an enhanced bootstrap using backward step-down variable deletion to validate the best- fitting model and to determine the predictive reliability of the model (Harrell 2001). A simple bootstrap estimates an accuracy index directly from averaging indexes computed on the original sample; however, the enhanced bootstrap uses a more indirect approach by estimating the bias due to overfitting (i.e., the “optimism” in the final model fit; Harrell 2001). Using resampling with replacement (set to 2,500 resamples), bootstrapping allowed the derivation of bias-corrected estimates of predictive accuracy without the need to withhold data, which can have severe drawbacks (Harrell et al. 1996; Harrell et al. 1998). After the best-fitting binary model was determined and validated, we evaluated the predictive ability of the model using (1) Nagelk- 2 2 erke’s R index (RN ), which assesses the ability of the model to produce unbiased estimates of outcome (i.e., goodness of fit or calibration); and (2) the discrimination index (D = [model likelihood ratio χ2 –1]/n), which assesses the model’s ability to separate the subjects’ outcomes (i.e., discrimination; Harrell 2001). In addition to determining the best model, we used the same selection strategy described above to develop a predictive binary logistic model that would require the least amount of laboratory processing time and could be conducted on a small budget. For this model, we excluded mean ovary energy and mean liver energy from the initial full model, and we termed the resultant best model our “least effort” model. In the second stage of our analyses, we determined the ability of ovary color and other female characteristics to predict the reproductive stages of nonregressing females only. For this analysis, we used ordinal logistic regression with a continua-

Downloaded by [Department Of Fisheries] at 20:42 25 September 2012 tion ratio (CR) model rather than a proportional odds model, FIGURE 1. Reproductive phases (described by Brown-Peterson et al. 2011), reproductive stages (i.e., used for predictive models), and corresponding his- as preliminary analyses indicated that the CR model provided tological images and colors for female striped bass. Histological images show a better fit to our data. The CR model is based on conditional postovulatory follicles (POFs) and atretic (A), primary growth (PG), cortical probabilities and is suitable when subjects have to pass through alveolar (CA), primary vitellogenic (Vtg1), secondary vitellogenic (Vtg2), and one category to reach the next (Harrell et al. 1998), which is the tertiary vitellogenic (Vtg3) oocytes. Our binary logistic models tended to in- case for reproductive stage. As was done with the binary mod- correctly classify regressing∗ females (i.e., stage 0 but with a high proportion of atretic oocytes) as stage 1 females. All histology images were captured at els, we used a backward step-down variable deletion approach the same magnification (reference scale is shown in the image for the spawning based on Akaike’s information criterion to find the model that capable phase). yielded the greatest reduction in deviance to determine the best ordinal model and least effort ordinal model. After the best- color and other female characteristics. To determine the overfitting CR models were determined, we quantified the predic- all best model, we began with a full model that included mean tive ability of the model using the same strategy described above ovary energy density (kJ/g), ovary percent water, gutted female for the binary models; however, bootstrap resampling for ordi- weight (g), female age, mean red ovary color, mean green ovary nal models was stratified by the response variable so that all NOTE 1039

TABLE 1. Means (SD in parentheses) for all variables included in the initial full binary logistic and continuation ratio models used to predict the reproductive stages of female striped bass.

Gutted Ovary Liver Ovary color or color ratio body energy energy Ovary Liver weight Age density density percent percent Red : Red : Green : Stage Phase n (kg) (years) (kJ/g) (kJ/g) water water Red Green Blue green blue blue

0 Regressing 68 9.14 12.3 24.54 24.42 80.6 77.8 170.75 93.26 79.31 1.97 2.23 1.16 (2.31) (2.64) (2.28) (1.27) (8.4) (3.3) (19.91) (28.84) (19.16) (0.51) (0.40) (0.17) 1 Immature, regenerating, 8 4.95 7.6 25.07 27.58 83.3 76.3 185.80 105.09 76.88 1.87 2.47 1.36 and early developing (1.17) (1.40) (2.55) (2.41) (3.8) (4.1) (13.81) (28.39) (14.17) (0.46) (0.37) (0.23) 2 Developing 19 5.95 8.3 30.85 26.62 71.4 72.9 198.02 124.61 73.38 1.61 2.69 1.68 (1.52) (1.17) (1.69) (2.72) (5.5) (5.7) (9.66) (16.99) (11.34) (0.17) (0.46) (0.27) 3 Spawning capable 13 7.34 8.7 34.76 27.43 59.9 72.3 194.10 127.99 67.27 1.55 2.98 1.94 (early) (2.78) (1.62) (0.69) (3.45) (3.7) (6.6) (10.94) (21.06) (14.56) (0.25) (0.52) (0.30) 4 Spawning capable 32 8.41 10.1 34.67 25.53 53.9 74.6 166.82 133.44 80.20 1.26 2.18 1.72 (2.17) (2.00) (0.58) (2.56) (5.1) (4.9) (10.51) (14.05) (16.92) (0.13) (0.49) (0.29)

ordinal classes were represented at least once in every bootstrap Binary Classiﬁcation into the Regressing Phase or Other sample. Reproductive Phases The overall best binary logistic model for predicting whether females were in the regressing or nonregressing reproduc- RESULTS tive phase included ovary energy density, gutted body weight, The means and SDs of all independent variables included and the proportion of ovary color that was green (Table 2). in the full binary and ordinal regression models are shown in This model showed that for a female of average weight, the Table 1. probability of being in the regressing phase increased with

TABLE 2. Binary logistic regression model results for predicting whether female striped bass are in the regressing or nonregressing phase, and ordinal continuation ratio model results for distinguishing among females in four nonregressing reproductive stages. The overall best models and the models that require the least amount of laboratory processing time and cost (least effort models) are shown (LR χ2 = likelihood ratio chi-square; AIC = Akaike’s information criterion).

Variable or statistic n df LR χ2 AIC Coefﬁcient SE Wald ZP Binary: Best Model Overall model 140 3 154.5 44.3 <0.0001 Intercept 24.5816 5.58 4.41 <0.0001 Ovary energy density 1 −0.7277 0.17 −4.2 <0.0001 Gutted body weight 1 0.0011 0.00 3.59 0.0003 Proportion of ovary color = green 1 −41.84 13.4 −3.13 0.0018 Binary: Least Effort Model Overall model 140 3 137.2 66.7 <0.0001 Downloaded by [Department Of Fisheries] at 20:42 25 September 2012 Intercept −3.6009 5.01 −0.72 0.4700 Ovary percent water 1 11.0692 3.67 3.01 0.0026 Gutted body weight 1 0.0006 0.00 3.46 0.0005 Ratio of green : blue ovary color 1 −6.677 1.89 −3.53 0.0004 Ordinal: Best Model Overall model 72 3 170.9 30.9 <0.0001 Ovary energy density 1 −5.6416 1.85 −3.05 0.0023 Mean blue ovary color 1 −0.2672 0.10 −2.64 0.0083 Mean red ovary color 1 0.3486 0.11 3.19 0.0014 Ordinal: Least Effort Model Overall model 72 2 135.1 65.1 <0.0001 Ovary percent water 1 50.0961 10.4 4.84 <0.0001 Proportion of ovary color = red 1 43.0311 12.4 3.46 0.0005 1040 PEER ET AL.

2 2 TABLE 3. Bootstrap evaluation of Nagelkerke’s R index (RN ), discrimination index (D), and parameter estimates obtained from the binary logistic models (discrimination of regressing versus nonregressing females) and ordinal continuation ratio models (discrimination among females in four nonregressing reproductive phases). The best-ﬁtting overall models and the best models that included predictor variables requiring the least laboratory processing time and cost (least effort models) are evaluated. Original values of statistics, optimism values (i.e., bias due to overﬁtting), and bootstrap-corrected statistics are shown.

Statistic Original value Optimism Corrected value Binary: Best Model 2 RN 0.8878 0.0150 0.8728 D 1.1263 0.0407 1.0856 Intercept 0.0000 0.0326 −0.0326 Slope 1.0000 0.0735 0.9265 Binary: Least Effort Model 2 RN 0.8248 0.0355 0.7893 D 0.9485 0.0967 0.8518 Intercept 0.0000 −0.0242 0.0242 Slope 1.0000 0.1018 0.8982 Ordinal: Best Model 2 RN 0.9527 0.0001 0.9511 D 0.9588 0.0180 0.9408 Intercept 0.0000 0.0706 −0.0706 Slope 1.0000 0.0700 0.9300 Ordinal: Least Effort Model 2 RN 0.8059 0.0250 0.7800 D 0.6905 0.0204 0.6701 Intercept 0.0000 0.0532 −0.0532 Slope 1.0000 0.1063 0.8937

FIGURE 2. Results showing observed reproductive phases of female striped bass and the predicted probability of females being in the regressing phase The best-performing least effort binary logistic model in- based on (a) the best overall binary logistic model and (b) the least effort binary cluded ovary percent water, gutted body weight, and the green : logistic model. blue ovary color ratio (Table 2). This model showed less deviance reduction than the overall best model; however, much like the overall best model, the least effort model incorrectly decreasing ovary energy density and decreasing proportion of categorized a small but nearly equal proportion of females into green ovary color. Differences between the observed reproduc- the regressing (4%) and nonregressing phases (3%; Figure 2).

Downloaded by [Department Of Fisheries] at 20:42 25 September 2012 tive stage and the predicted probability of being in the regressing Furthermore, the characteristics of the incorrectly classified fe- stage showed that the best model incorrectly assigned a small males were identical to those observed for the best overall model. and equal proportion (∼3%) of females into the regressing and With the exception of two additional females that were incor- nonregressing phases (Figure 2). Exploration of the characteris- rectly classified as regressing by the least effort model, the same tics of incorrectly classified females indicated a high proportion individuals were misclassified by both models. Although bias- (23–43%) of atretic oocytes in regressing females that were in- corrected indices of predictive accuracy did indicate that the 2 correctly predicted to be nonregressing (Figure 1). In addition, model’s abilities to calibrate (RN = 0.79) and discriminate the four nonregressing females that were incorrectly classified as (D = 0.85) were slightly lower than those of the overall best regressing were either regenerating (n = 2) or early developing model, optimism was similar, again indicating good predictive with primary vitellogenic oocytes only (n = 2). Bias-corrected performance of the least effort model with future data (Table 3). indices of predictive accuracy also indicated a strong calibra- 2 tion ability (RN = 0.87) and discrimination ability (D = 1.09; Ordinal Classification of Nonregenerating Reproductive Table 3). In addition, bias due to overfitting (i.e., optimism) Phases was quite low for all indices and model parameters (Table 3), Initial ordinal model development indicated that ordi- indicating good performance of the model for future data. nal predictions were unable to distinguish among immature, NOTE 1041

FIGURE 3. Predictions from the best overall ordinal model, indicating the FIGURE 4. Predictions from the least effort ordinal model, indicating the observed reproductive stage of female striped bass and the predicted probability observed reproductive stage of female striped bass and the predicted probability of females being in (a) stage 1, (b) stage 2, (c) stage 3, and (d) stage 4. of females being in (a) stage 1, (b) stage 2, (c) stage 3, and (d) stage 4.

regenerating, and some developing phases (i.e., those with two stage 3 females to be in stage 2 (Figure 4). In addition, mostly primary growth and cortical alveolar oocytes and some one stage 1 female was incorrectly predicted to be in stage 2, primary vitellogenic oocytes). Because females in each of these one stage 2 female was predicted to be in stage 4, and three stage phases are not capable of spawning in the current year, we 4 females were predicted to be in stage 3 (Figure 4). Further- decided to group these phases together into “stage 1” for our more, bias-corrected indices of predictive accuracy indicated ordinal models (Figure 1). Because females in the later devel- 2 that the calibration (RN = 0.69) and discrimination (D = 0.67) oping phase (i.e., when ovaries also included secondary vitel- abilities were weaker than those of the best overall CR model; logenic oocytes) could be distinguished from those in stage however, optimism for all indices and model parameters was 1, we grouped these females into a separate category, stage 2 low, indicating that the least effort model would have lower pre- (Figure 1). In addition, our early model development indicated dictive accuracy than the overall best model but would exhibit that we could separate the spawning capable phase into early and similar unbiased performance with future data (Table 3). late subphases, which were called stages 3 and 4, respectively, in our ordinal model. The overall best CR model included mean ovary energy, DISCUSSION mean blue ovary color, and mean red ovary color (Table 2). We were able to use logistic models to predict the repro- Plots of predicted probabilities indicated very good ﬁt to the ob- ductive stages of female striped bass with high accuracy and served reproductive stages (Figure 3). There were three incorrect without the need for histological analysis by using a standard- predictions (4% error): one stage 1 female was incorrectly clas- ized and repeatable measure of ovary color in combination with

Downloaded by [Department Of Fisheries] at 20:42 25 September 2012 sified as stage 2; one stage 3 female was incorrectly classified other easily obtainable female characteristics. Many researchers as stage 4; and one stage 4 female was incorrectly classified have used color and other macroscopic measures (Hjort 1914; as stage 3 (Figure 3). Bias-corrected indices of predictive accu- Graham 1924; Hickling 1935; Vladykov 1956; Powles 1958; 2 racy also indicated excellent calibration ability (RN = 0.95) and Burnett et al. 1989; Kynard and Kieffer 2002) and biochemi- discrimination ability (D = 0.94); optimism was low for all pa- cal measures (i.e., Heppell and Sullivan 1999) to estimate the rameters and indices, indicating good predictive performance of reproductive stage of female teleosts, and some have tried to the model for future data (Table 3). Finally, the best-performing use color to predict the reproductive stage of females (i.e., least effort CR model included ovary percent water and the pro- Bryan et al. 2007); however, no one has provided a calibrated, portion of ovary color that was red (Table 2). Plots of predicted standardized, repeatable procedure for predicting the reproduc- probabilities for the least effort CR model indicated poorer fit tive stage of female teleosts by using ovary color or other fe- to the observed reproductive stages than was demonstrated for male attributes in the absence of histology. Our results indicate the overall best CR model, and there was a total of 11 incor- that for striped bass, predictive accuracy is highest (i.e., over- rect classifications (15% error; Figure 4). The greatest error for all best models) for models that include ovary energy density this model appeared to be in the prediction of stage 3, with and one or two ovary color variables. Our overall best binary the model predicting four stage 3 females to be in stage 4 and model showed an ability to discriminate between regressing and 1042 PEER ET AL.

nonregressing females very well, with less than 6% of the fe- should consider that more robust sample sizes may be necessary males being incorrectly classified. Of the eight incorrectly clas- for initial model development. sified females, the nonregressing females that were predicted to Despite the accuracy of our predictive models and the po- be regressing (only 3%) would not have spawned; this suggests tential for similar models to be developed for other species, that we might overpredict the number of spawners in a given our analysis was not without shortcomings. Specifically, we year. However, the regressing females that were misclassified as were unable to distinguish among immature, regenerating, and nonregressing were predicted by our overall best ordinal model early developing reproductive phases. The inability of our or- to be in phases that would not have spawned (i.e., immature, re- dinal models to discriminate among these phases may be due generating, or developing). Thus, the existing error in the binary to the small sample size of females representing these phases. logistic model likely would lead to little error in predicting the Based on our catches, the females in these phases were smaller numbers of spawners (i.e., spawning capable and regressing) and younger and did not appear to migrate with the spawn- and nonspawners (i.e., immature, regenerating, and developing) ing stock while in the Chesapeake Bay. Consequently, in the in a given year. future, additional sampling locations may be needed to target Our overall best ordinal regression model showed an even smaller striped bass females. Such constraints may not pre- greater ability to discriminate among striped bass females clude development of more inclusive models for other species. belonging to the four nonregressing reproductive stages we It is also possible, however, that our inability to discriminate established. Although we were forced to group immature, among the immature, regenerating, and early developing phases regenerating, and early developing-phase females into one was due to these phases’ very low ovary energy density and reproductive stage (stage 1) to fit our data well, the overall best their similarity in ovary color. If this is true, our results may ordinal model accurately discriminated among stage 1, stage also be typical of other species. However, this should not deter 2 (developing), stage 3 (early spawning capable), and stage 4 others from using our methods to develop their own species- (spawning-capable) females, with only 4% total error. Thus, specific models that are capable of discriminating among these ovary color (specifically, mean blue and mean red) and ovary reproductive phases. Although discrimination among the three energy density were very strong predictors of whether striped phases is not always imperative, their identification would be bass females were in the developing, early spawning capable, or required for accurate quantification of skipped spawners in a spawning capable phase; however, discrimination among imma- population. Discrimination of early developing females would ture, regenerating, and early developing phases was not possible. also be necessary to determine the beginning of the reproduc- Although ovary energy density did produce the best pre- tive season and the age or size at maturity. In such cases and dictive models, our least effort models indicated that if bomb in situations where discrimination among immature, regenerat- calorimetry of ovaries is too expensive or time consuming, ovary ing, and early developing phases is not possible, our methods energy density can be replaced by ovary percent water with are still useful for identifying the regressing, developing, and a slight reduction in model fit and discrimination ability but spawning-capable females. However, females classified as im- with similarly unbiased performance for future data. For ex- mature, regenerating, or early developing (i.e., stage 1 in our ample, when we used binary logistic regression to discriminate analysis) would have to undergo further discrimination using between regressing and nonregressing females, model fit was histology and microscopic examination. Although not ideal, the slightly reduced for the least effort model, with 7% of females use of these predictive models in combination with histology being incorrectly classified. However, we observed that despite can still substantially reduce laboratory time and costs. this error, the least effort model—like the overall best binary The inability of our striped bass ordinal models to dis- model—would have little error in predicting the numbers of criminate among immature, regenerating, and early developing

Downloaded by [Department Of Fisheries] at 20:42 25 September 2012 spawners and nonspawners. phases is a problem that is also associated with macroscopic Although our least effort ordinal model demonstrated re- methods. However, macroscopic methods also suffer from ad- duced fit relative to the overall best ordinal regression model, ditional deficiencies that are not present our methods. For ex- only 5% of the total error would be problematic for discrimi- ample, macroscopic methods use descriptive terminology that nating between nonreproductive (i.e., stage 2) and reproductive can be highly subjective and ambiguous (e.g., subjective ovary (i.e., stages 3 and 4) females. This is because most of the total color, flaccidity, and texture and oocyte clarity) and thus can lead error (10%) was due to incorrect classification between stages to interreader error, whereas our method is less biased because 3 and 4, which would not be a problem for most studies since it is an objective, calibrated, and standardized procedure. Con- both stages are spawning-capable groups. Still, the greater error sequently, our method essentially eliminates interreader error as evident in the least effort binary and ordinal models was likely long as the instruments that are used to quantify ovary color, due to the poorer resolving power of ovary percent water rela- ovary energy, and female size characteristics are appropriately tive to ovary energy density. It is possible that some of this error calibrated. This may be especially beneficial for long-term stud- could have been resolved by increasing the sample size. Thus, ies in which several individuals could be involved in determin- others that attempt to use this predictive approach to determine ing the reproductive phases of females. Another advantage of reproductive stages without using energetic densities of tissues our method is that it is validated for accuracy with histological NOTE 1043

techniques. Macroscopic methods have rarely been validated the collection of striped bass; and two anonymous reviewers for other than for commercially important gadid species, and helpful comments on the manuscript. This research was funded when validation is conducted, macroscopic methods often by the Steven Berkeley Marine Conservation Fellowship, Mary- show a strong tendency for misclassification at many stages land Sea Grant (NA10OAR4170072 SA75281114-EE), and an (Tomkiewicz et al. 2003; Vitale et al. 2006; Ferreri et al. 2009). Explorer’s Club Exploration Research Grant. This is Contribu- After initial model development, our method also requires very tion Number 4643 of the University of Maryland Center for little training, since the predictor variables included in our mod- Environmental Science. eling approach are easy to obtain using relatively simple instruments. This is especially true if future models can be developed without the need for ovary energy density. In contrast, due to the subjectivity and ambiguity of macroscopic techniques, there REFERENCES can be a steep learning curve before individuals can efficiently Brown-Peterson, N. J., D. M. Wyanski, F. Saborido-Rey, B. J. Macewicz, and S. K. Lowerre-Barbieri. 2011. A standardized terminology for describing and confidently evaluate the reproductive phases. Finally, our reproductive development in fishes. Marine and Coastal Fisheries: Dynamics, approach has the potential to accurately discriminate among Management, and Ecosystem Science 3:52–70. all reproductive phases if models are provided with appropriate Bryan, J. L., M. L. Wildhaber, D. M. Papoulias, A. J. DeLonay, D. E. Tillitt, predictor variables and sample sizes at each phase. Evidence and M. L. Annis. 2007. Estimation of gonad volume, fecundity, and repro- from the few validation studies that do exist indicates that other ductive stage of shovelnose sturgeon using sonography and endoscopy with application to the endangered pallid sturgeon. Journal of Applied Ichthyology macroscopic methods do not hold that potential (i.e., Vitale et al. 23:411–419. 2006; Ferreri et al. 2009). Burnett, J., L. O’Brien, R. K. Mayo, J. A. Darde, and M. Bohan. 1989. Finfish Despite the shortcomings of the ordinal model developed maturity sampling and classification schemes used during Northeast Fish- for striped bass, minimum estimates of skipped spawning still eries Center bottom trawl surveys, 1963–89. NOAA Technical Memorandum may be possible for female striped bass that are collected dur- NMFS-F/NEC-76. Chadwick, H. D. 1965. Determination of sexual maturity in female striped bass ing the spawning season, given that developing-phase females (Morone saxatilis). California Fish and Game 51:202–206. could be identified by using our models. Although developing Ferreri, R., G. Basilone, M. D’Elia, A. Traina, F. Saborido-Rey, and S. Mazzola. females collected in late fall or winter would be considered 2009. Validation of macroscopic maturity stages according to microscopic mature and reproductive, those collected during the spawning histological examination for European anchovy. Marine Ecology 30(Supple- season (April–May) are probably unable to spawn in the cur- ment 1):181–187. Graham, M. 1924. The annual cycle in the life of the mature cod in the North rent season given that vitellogenesis in striped bass can take Sea. Fishery Investigations Series 2 Marine Fisheries Great Britain Ministry several months to complete (Sullivan et al. 1997). In addition, of Agriculture, Fisheries and Food 6:1–77. it has been suggested that some striped bass may require more Harrell, F. E., Jr. 2001. Regression modeling strategies. Springer Verlag, New than 1 year from initiation of secondary growth until spawning York. (Merriman 1941; Chadwick 1965; Specker et al. 1987). Fur- Harrell, F. E., Jr. 2011. rms: regression modeling strategies, R package version 3.3-0. Available: CRAN.R-project.org/package=rms. (November 2011). thermore, evaluation of the length frequency distributions of Harrell, F. E., Jr., K. L. Lee, and D. B. Mark. 1996. Multivariable prognostic females collected on the two major Chesapeake Bay spawning models: issues in developing models, evaluating assumptions and adequacy, grounds (i.e., upper Chesapeake Bay and Potomac River) in- and measuring and reducing errors. Statistics in Medicine 15:361–387. dicated that developing-phase females, which were all smaller Harrell, F. E., Jr., P. A. Margolis, S. Gove, K. E. Mason, E. K. Mulholland, D. than 900 mm, were underrepresented on the spawning grounds Lehmann, L. Muhe, S. Gatchalian, and H. F. Eichenwald. 1998. Development of a clinical prediction model for an ordinal outcome: the World Health (authors’ personal observation). Consequently, we feel confi- Organization multicentre study of clinical signs and etiological agents of dent that developing female striped bass can be used to obtain pneumonia, sepsis and meningitis in young infants. Statistics in Medicine

Downloaded by [Department Of Fisheries] at 20:42 25 September 2012 minimum estimates of skipped spawning via our protocol and 17:909–944. models. This might also be feasible for other species if it can Heppell, S. A., and C. V. Sullivan. 1999. Gag (Mycteroperca microlepis) vitel- be determined that developing-phase females will not spawn logenin: purification, characterization and use for enzyme-linked immunosor- bent assay (ELISA) of female maturity in three species of grouper. Fish in the current season. Such an approach may be sufficient for Physiology and Biochemistry 20:361–374. detecting trends in a time series or for determining potential Hickling, C. F. 1935. Seasonal changes in the ovary of the immature hake, factors influencing the proportion of skipped spawners. How- Merlucius merluccius L. Journal of the Marine Biological Association of the ever, for determining accurate measures of skipped spawning in United Kingdom 20:443–461. Hjort, J. 1914. Fluctuations in the great fisheries of northern Europe, viewed in any species, histology would likely be needed, either alone or the light of biological research. Journal du Conseil Permanent International in combination with a model such as ours. pour L’Exploration de la Mer 20:1–228. Hunter, J. R., and B. J. Macewicz. 1985. Measurement of spawning frequency in multiple spawning fishes. NOAA Technical Report NMFS 8544. ACKNOWLEDGMENTS Kjesbu, O. S. 2009. Applied fish reproductive biology: contribution of individual reproductive potential to recruitment and fisheries management. Pages 293– We thank D. Loewensteiner, C. Sperling, G. Cantor, 332 in T. Jakobsen, M. J. Fogarty, B. A. Megrey, and E. Moksness, editors. S. Michaels, V. Ochoa, and M. Studivan for assisting with field- Fish reproductive biology: implications for assessment and management. work and laboratory processing; Captain Pete Ide for support in Wiley-Blackwell, New York. 1044 PEER ET AL.

Kynard, B., and M. Kieffer. 2002. Use of a borescope to determine the sex and Pages 162–174 in M. J. Dadswell, R. J. Klauda, C. M. Moffitt, R. L. Saunders, egg maturity stage of sturgeons and the effect of borescope use on reproductive R. A. Rulifson, and J. E. Cooper, editors. Common strategies of anadromous structures. Journal of Applied Ichthyology 18:505–508. and catadromous fishes. American Fisheries Society, Symposium 1, Bethesda, Lubzens, E., G. Young, J. Bobe, and J. Cerda.` 2010. Oogenesis in teleosts: how Maryland. fish eggs are formed. General and Comparative Endocrinology 165:367–389. Sullivan, C. V., D. L. Berlinsky, and R. G. Hodson. 1997. Reproduction. Pages Merriman, D. 1941. Studies on the striped bass (Roccus saxatilis) of the Atlantic 11–30 in R. M. Harrell, editor. Striped bass and other Morone culture. Else- coast. U.S. Fish and Wildlife Service Fishery Bulletin 50:1–77. vier, Amsterdam. Powles, P. M. 1958. Studies of reproduction and feeding of Atlantic cod (Gadus Tomkiewicz, J., L. Tybjerg, and Å. Jespersen. 2003. Micro- and macroscopic callarias L.) in the southwestern Gulf of St. Lawrence. Journal of the Fisheries characteristics to stage gonadal maturation of female Baltic cod. Journal of Research Board of Canada 15:1383–1402. Fish Biology 62:253–275. R Development Core Team. 2011. R: a language and environment for statistical Vitale, F., H. Svedang,¨ and M. Cardinale. 2006. Histological analysis invali- computing. R Foundation for Statistical Computing, Vienna. dates macroscopically determined maturity ogives of the Kattegat cod (Gadus Rasband, W. S. 2011. ImageJ: an image processing and analysis package. morhua) and suggests new proxies for estimating maturity status of individual U.S. National Institutes of Health, Bethesda, Maryland. Available: im- fish. ICES Journal of Marine Science 63:485–492. agej.nih.gov/ij/. (November 2011). Vladykov, V. D. 1956. Fecundity of wild speckled trout (Salvelinus fontinalis) Specker, J. L., D. L. Berlinsky, H. D. Bibb, and J. F. O’Brien. 1987. Oocyte in Quebec lakes. Journal of the Fisheries Research Board of Canada 13: development in striped bass: factors influencing estimates of age at maturity. 799–841. Downloaded by [Department Of Fisheries] at 20:42 25 September 2012