Replace This with the Actual Title Using All Caps

MODELING NEW YORK STATE'S STOCKED STREAM TROUT FISHERY

A Thesis Presented to the Faculty of the Graduate School of Cornell University In Partial Fulfillment of the Requirements for the Degree of

Master of Science

Benjamin Marcy-Quay May 2015

ABSTRACT

A population dynamics model has been used in New York State since the

1980's to explore trout stocking scenarios, but the input information has become obsolete due to changes in predator communities and angler practices. We surveyed nine streams for a three year period and using a combination of methods. Creel surveys allowed us to estimate effort, harvest, and catch rates, while electrofishing surveys provided estimates of total mortality. Geostatistical methods were used to explore patterns of angler origin. The effect of angler specialization was assessed using regression to compare calculated travel distance to angler income, gear, and stream choice. All three factors were found to be significant (P<0.05).Using nonlinear parameter estimation, we obtained estimates for apparent natural mortality and catchability nearly an order of magnitude higher than historical estimates.

Predictions from the updated model exhibited much better fit than those from the model previously used to inform stocking decisions. BIOGRAPHICAL SKETCH

Born in Ann Arbor, Michigan in 1988, Ben spent the majority of his childhood exploring the woods, streams, and lakes around his home and abroad. While on family vacations that ranged from treks in Canada's Northwest Territories to houseboat trips down the Mississippi River, he cultivated a keen interest in the natural world around him. Ben's high school years were spent at a small boarding school on the shore of Lake Michigan where he took every science course offered, including volunteering as a teaching assistant for a marine biology course in the British Virgin Islands. In 2010, Ben earned his Bachelor's of Science in Environmental Science from Allegheny College. During summer breaks and after graduation, he spent time in Utah backpacking, climbing, and working as a counselor for a therapeutic wilderness program where he helped teens overcome problems such as substance abuse and depression. Ben came to New York in 2011 when he took a job as a research technician with the Department of Environmental Conservation. In fall 2012, after two enjoyable seasons of field work, Ben began graduate work at Cornell University in the Department of Natural Resources.

iii

ACKNOWLEDGMENTS

This work would not have been possible without the help of many people. I would first and foremost like to thank my committee, Patrick Sullivan and Clifford Kraft, for their advice and knowledge. In the course of this project they have greatly deepened my understanding of ecology and statistics. Funding support was provided by the New York State Department of Environmental Conservation and the USGS New York Cooperative Fish and

Wildlife Research Unit. The members of the DEC Trout Team were instrumental to the completion of the project, particularly Fred Henson and Shaun Keeler, who provided leadership and guidance. I owe heartfelt gratitude to all of the state biologists and managers involved in the process for their help with study design, data collection, and for providing "real world" interpretation of the results. The entire Cornell DNR community has helped and supported me along the way and I am grateful for their help and friendship. In particular I would like to thank Suzanne Beyeler for both encouraging me to apply to graduate school and being an excellent mentor during the last several years. Finally, I would like to thank my parents, Sherry Marcy and Nancy Quay, for their unfailing love and support.

TABLE OF CONTENTS

Biographical Sketch ...... iii Acknowledgements ...... iv Table of Contents ...... v List of Figures ...... vi List of Tables ...... vii Chapter 1 – A Population Dynamics Model for Stocked Brown Trout (Salmo trutta) Using Nonlinear Parameter Estimation ...... 1 Abstract ...... 1 Introduction ...... 2 Methods ...... 4 Results ...... 10 Discussion ...... 18 References ...... 23

Chapter 2 – Angler Specialization and Travel Distance: Applying Two Complementary Statistical Methods for Interpreting Spatially Referenced Creel Survey Data ...... 29 Abstract ...... 29 Introduction ...... 30 Methods ...... 33 Results ...... 37 Discussion ...... 43 References ...... 48

LIST OF FIGURES

Figure 1.1. Comparison of current and historical estimates of total annual angler effort. Historical estimates are taken from single year creel surveys conducted between 1985 and 1995...... 11

Figure 1.2. Mean creel rate by year for all study streams...... 12

Figure 1.3. Fit of model predictions (lines) to observed densities from surveys (points) for all streams and years...... 16

Figure 1.4. Fit of model predictions (lines) to observed average daily angler catch-rates (points) for all streams and years...... 17

Figure 1.5. Current model fit versus the historical model for an example stream and year...... 18

Figure 2.1. Estimates and Bonferroni-corrected confidence intervals for all levels of the variable Stream...... 38

Figure 2.2. Variogram model fits for each study stream...... 41

Figure 2.3. Interpolated angler origination for all study streams with heat map transparency increasing with variation...... 42

LIST OF TABLES

Table 1.1. Candidate models ranked by lowest Akaike Information Criterion (AIC) score...... 13

Table 1.2. Estimated parameter values by stream for the top model...... 14

Table 1.3. Estimated parameter values by DEC stream type compared to historical values used in the previous model...... 14

Table 2.1. Number of anglers surveyed per stream from 2011-2013...... 34

Table 2.2. ANCOVA analysis of angler travel distance against angler specialization variables...... 39

Table 2.3. Variogram models and parameter estimates for the study streams ...... 40

vii

CHAPTER 1

A Population Dynamics Model for Stocked Brown Trout (Salmo trutta) Using Nonlinear

Parameter Estimation

Abstract

Stocking programs used to improve the quality of recreational fisheries can have positive effects on angler catch-rate and satisfaction, but may also negatively affect the local habitat and native fish community. A population dynamics model has been used in

New York State since the 1980's to explore alternative stocking scenarios, but the input information for developing these forecasts has likely become obsolete due to changes in predator communities and angler practices. In this study we surveyed nine streams for a three year period and used a combination of methods to assess natural and angler-related mortality. Creel surveys allowed us to estimate angler effort, harvest, and catch rates, while depletion surveys of marked cohorts provided estimates of natural mortality. Angler effort was lower than historically recorded for nearly all streams and release rates were much greater (75-80%) than the 0% previously assumed. Using nonlinear parameter estimation, we obtained estimates for daily apparent natural mortality and catchability that were almost an order of magnitude higher than historical estimates (0.015 and .088 vs. 0.002 and 0.008). Catchability was not linearly related to density, as had previously been assumed, suggesting that density-dependent effects may exist. Twelve alternative models were explored and, of them, the top model included a natural mortality factor separated by stream while in the second best model mortality was partitioned using a stream type metric based on habitat characteristics.

This suggests that habitat plays a key role in the survival of stocked trout. Predictions

1 from the updated model exhibited much better fit than those from the model previously used to inform stocking decisions.

Introduction

A central tenet in the management of recreational fisheries is that angler satisfaction is positively linked to catch-rate. All anglers desire to experience some degree of fishing success, regardless of their orientation towards consumption or catch- and-release fishing (Fedler & Ditton 1986; Arlinghaus 2006). Declines in angler effort occur in fisheries in which angler expectations are not being met, while those with high catch-rates often see increasing demand as new anglers are attracted (Post et al.

2012). When faced with alternative fishing opportunities an angler may shift effort to nearby waters where conditions are considered better (Carpenter and Brock 2004).

When anglers do not have readily available alternative options, as might be the case under large-scale changes to regional management policies, they often choose other recreational activities (Martin and Pope 2011).

Stocking is often used to improve the quality of recreational fisheries (Cowx

1994). Hatchery reared fish are released in waters with the intent of either establishing a permanent population or providing a temporary boost to angler catch-rates, often referred to as “put-and-take”. Decisions about the goal of stocking efforts require knowledge of such factors as habitat suitability, seasonal temperature variations, the current fish community composition, and angler effort (Cowx 1999). In addition, the time of stocking and the size of stocked fish may vary depending on the management goals.

For example, fish intended for put-and-take fishing are stocked at a harvestable size for anglers (Miko et al. 1995).

Quantitative models have been used to develop and evaluate fisheries management strategies since the 1950's, beginning with the Beverton-Holt (1957) and the Schaefer models (1954). The ability to predict and study changes in a fish population over time can help guide management decisions (Agnew 1982; Quinn and

Deriso 1999). These models generally require information about mortality, reproduction, and growth to evaluate the influence of various external factors on the population dynamics of a specific fishery. These factors, often characterized as rates, can be directly estimated from available data using a modeling framework. Harvest, catch-rate, and effort can all be assessed by surveying fisherman (Pollock et al. 1994). Data from population surveys can be used to obtain instantaneous estimates of fish abundance at specific sizes and ages (Nielsen and Johnson 1983). However, some model parameters, such as catchability, can be difficult to estimate without more sophisticated techniques such as mark-recapture methods and must often be set at some constant value based on the literature (Arreguin-Sanchez 1996). This can lead to substantial uncertainty because catchability, can vary even among different strains of the same species (Brauhn and Kincaid 1982). Natural mortality is similarly difficult to estimate directly and many assessment methods rely on assuming that fishing mortality is known, which again necessitates knowledge of catchability (Paloheimo 1980). One way to account for variables that are difficult to measure directly is to estimate them simultaneously from the data through nonlinear estimation methods. Models employing this approach are widely used in evaluations of commercial marine fisheries, but are only beginning to be applied to sport fisheries, particularly freshwater ones (Radomski and Goeman 1996; Arlinghaus et al. 2009). Where they have been employed they have

3 proven useful for estimating multiple population parameters in the absence of traditional time-series data (Irwin et al. 2008).

We updated and extended a historical model used to guide stream trout stocking in New York State. The new model accounts for changes to the fishery, such as increased catch-and-release fishing rates. Estimates of fishing mortality were directly estimated using data collected during a series of multi-year creel surveys. Natural mortality and catchability rates were estimated within the model in order to fit population survey data and observed angler catch-rates. We then employed model selection techniques to explore alternate parameterizations of the model to test hypotheses about the factors of greatest influence to the fishery.

Methods

Data collection

A total of nine streams were chosen for inclusion in this study in order to encompass the range of stream conditions found throughout New York State based on the extent to which they were representative of their region and feasibility for conducting both electrofishing and creel surveys. . All study streams were surveyed for three years

(2011-2013), during which they were stocked with hatchery brown trout (Salmo trutta,

Rome strain). Stocking densities, locations, and the timing of stocking was kept as consistent as possible throughout the study period. Each stocking cohort (i.e. fish separated either by time, age, or both) were marked using a unique combination of fin clips. Wild brown trout were also present in all study streams, likely descended from previously stocked individuals. To assess population density, three-pass depletion electrofishing population surveys were conducted in the spring and fall of each year.

The results of these surveys were analyzed using a Leslie-Delury Binomial Depletion model (Sullivan, in preparation) to obtain density estimates. Surveys were removed from the dataset if the computed capture probability was < 0.3, indicating poor or no observed depletion in the survey and therefore likely violating depletion model assumptions. A total of 15 population surveys, or 8% of all surveys, were removed for this reason.

Angler behavior was assessed using roving creel surveys (one stream, the

Carmans River conducted access-point surveys instead) conducted for the full length of the angling season. Car counts were employed to assess total effort, while individual interviews were used to calculate catch-rate using methods described by Pollock

(1997). Consistent with Pollock's recommendations we removed those anglers who had fished for less than 0.5 hours and used the mean of the ratios estimator to compute daily catch-rates (Alexiades et al. 2015). The mean of the ratios estimator has a larger sample variance, which is appropriate developing estimates from incomplete trip surveys. Removing those anglers who fished for a very short time adds negligible bias and tends to improve mean squared error. A total of 192 population surveys were included in the model, along with 630 daily average catch-rate estimates.

Model

The State of New York has used a simple, survivorship-based population dynamics model to guide salmonid stocking decisions since the early 1980's. Originally implemented in FORTRAN by State fisheries biologist Robert Engstrom-Heg, the model proved to be relatively accurate in setting appropriate stocking rates and was subsequently transferred to an Excel workbook that continues to be used by state

5 biologists (Engstrom-Heg and Engstrom-Heg 1984). This model is a key component of a program known as Catch Rate Oriented Trout Stocking (CROTS) that is used to establish trout stocking policies designed to maximize angler catch rates without adversely affecting the ecosystems of stocked streams (Sullivan and Kraft 2005).

Structure – The CROTS model reflects standard population dynamics theory. The number of fish present at time t is related to the number of fish present in the previous time period multiplied by survivorship, an exponential function of the total instantaneous mortality rate ZT (Ricker 1975). This can be expressed as:

( 1)

Total mortality (ZT) is the sum of the rates for natural mortality (ZN), harvested catch

(ZC), and the portion of catch-and-release fish that die due to handling (ZH). It should be noted that natural mortality may be compounded with unaccounted for forms of non- fishing loss including predation, disease, environmental stress, and emigration.

( 2 )

Harvested mortality is a function of angler effort (Et) multiplied by catchability (q).

Catchability can vary due to a number of factors including gear, temperature, and date

(Francis et al. 2001). Our data indicate that catchability in the study streams did not vary in relation to any of these factors. Therefore catchability was assumed to be constant and was estimated as a single parameter over the season. As anglers do not keep all fish captured, the portion kept or creel rate (pK) was also taken into account.

( 3 ) Total annual angler effort for a particular stream is estimated using a roving-roving creel survey and instantaneous vehicle counts following the methods outlined in Pollock et al.

(1994). Total party hours were derived from arrival and departure counts of parked vehicles and adjusted to account for the average number of anglers per vehicle. The resulting product represented an estimate of the number of angler hours per creel day.

From these creel-day effort estimates we were able to derive an estimate of total angler effort per month, the sum of which represents the estimate of total annual effort.

However, distribution of effort was not always uniform across the fishing season.

Potential effort distribution patterns other than uniform include those with the majority of the effort occurring earlier in the season. To reflect this in the model, total effort (E) is multiplied by the portion of effort that occurs within that period (Pt) and divided by the total number of time units within that period (T).

( 4 ) Fish that are caught and released experience stress and injuries that may affect their ability to survive. The release mortality rate is dependent on the catch rate, the portion of fish not creeled, and the handling survival rate (S).Studies have suggested that hooking mortality depends on a number of factors including gear, size, and species, with brown trout having lower mortality rates than other species of trout (Taylor and

White 1992). A meta-analysis of 53 studies found that gear and species significantly influences mortality but that fish size did not (Bartholomew and Bohnsack 2005). As distribution of gear types was relatively similar for all waters studied, S was assumed to 7 be constant for all waters included in the study. The original value of 0.9 estimated by

Engstrom-Heg and Hulbert (1982) was used as gear and general fishing techniques have not changed significantly since that time.

( 5 ) By contrast with a wild fishery, the quantity of fish entering a stocked fishery is known, as is the time when they enter the system. When fish are stocked at multiple times throughout the season this has the effect of dividing the population into cohorts, each of which experience different amounts of natural and angling-related mortality. We marked each stocking cohort so that all captured individuals could be matched to their original cohort. Modeling these cohorts separately allows for the analysis of effects such as stocking time and post-stocking loss. Additionally, different mortality values can be applied to each group when a range of ages are stocked in separate cohorts. Therefore, at any time t the total abundance of stocked fish is the sum of fish from each cohort c.

Abundance of an individual cohort is a function of the initial number stocked N0 and the total mortality that has occurred ZT. An indicator, I is used to represent whether the cohort has been stocked at time t.

( 6 )

This expression of overall abundance can be multiplied by the catchability parameter to provide the estimated angler catch-rate for the given time period.

Parameter values were fit using a maximum likelihood estimation approach.

Model predictions were fit to observed values of cohort population density and angler

8 catch-rate using a weighted residual sum of square RSS structure. This structure is a combination of the two measurements of fit, along with a weighting term λ that accounts for differences in units and sample size between population density surveys and angler interviews. Values for λ were chosen so that both residual sum of squares elements were of the same magnitude, ensuring that the model tried to fit both sources of observations equally.

( 7 ) The residual sum of squares for density is the sum of the squared differences between the observed density of each cohort and the predicted density of that cohort at time t.

( 8 ) The residual sum of squares for the catch-rate follows in a similar fashion with the differences between the observed catch-rate and the predicted catch-rate :

( 9 ) Unlike with density, catch-rate observations are not divided by cohort as angler reports of fish markings were considered to be unreliable. This weighted residual sum of squares is then used to calculate a concentrated form of the negative log-likelihood for use in the optimization.

( 10 ) The use of the likelihood formulation allows for computation of an estimate of the standard error for the parameter estimates using a finite difference approximation to the 9 variance-covariance matrix. Parameter estimation was carried out using the AD Model

Builder (ADMB) environment. AD Model Builder uses a technique known as automatic differentiation that makes it possible to efficiently estimate many parameters simultaneously in complex nonlinear models (Fournier et al. 2012).

Twelve separate candidate models were developed by subdividing the natural mortality into multiple levels based on spatial, temporal, and biological characteristics expected to have an influence on mortality. Catchability was estimated as a global constant in all models, due to the limited data available from fishermen about observed fin clips, size, and age of caught fish. Models were then run using the same base data and initial parameter estimates and differing only in parameterization. Akaike

Information Criterion (AIC) was calculated using the final negative log-likelihood values from each model optimization to compare the relative fitness of each potential parameterization.

Results

Angler Behavior

Angler effort varied widely among streams, with values ranging from less than

100 hours per acre to over 1000 in the Carmans River, a popular Long Island fly fishing destination. With the exception of Carmans River all rivers had a lower annual effort than was found during previous surveys in the 1980s and '90s (Figure 1.1). The pattern of effort distribution throughout the season matched that from previous surveys, and all streams fit one of two patterns: 1) high effort in the spring with a gradual decline throughout the summer, or 2) extremely high effort in the first month with little to no effort during the rest of the season. The proportion of fish kept by anglers was quite

10 different from historical estimates. Previously, it was assumed that all legal size fish were harvested. Data from angler interviews in this study, however, showed that the overwhelming majority of fish were released, with only 10-20% being kept (Figure 1.2).

This is a substantial departure from the original CROTS model assumptions and may have great influence on the projected dynamics of the fishery.

Figure 1.1: Comparison of current and historical estimates of total annual angler effort. Historical estimates are taken from single year creel surveys conducted between 1985 and 1995.

Figure 1.2: Mean creel rate by year for all study streams.

Model Selection

We found that the model which included a separately estimated natural mortality parameter for each stream performed best, marginally outperforming the model that separated natural mortalities assigned according to the previous CROTS A-B stream classification system (ΔAIC=5.62). Other highly ranked models included terms for age of fish and year (Table 1.1).

Table 1.1: Candidate models ranked by lowest Akaike Information Criterion (AIC) score. Akaike Model # Model K AIC ΔAIC weight 6 q, ZN (Stream) 10 7133.00 0.00 0.93 2 q, ZN (Stream Type (A/B)) 3 7138.62 5.62 0.06 4 q, ZN (Age + Stream Type) 5 7142.44 9.44 0.01 10 q, ZN (Age + Stream) 16 7143.70 10.70 0.00 9 q, ZN (Stream + Year) 24 7144.60 11.60 0.00 5 q, ZN (Year) 4 7184.96 51.96 0.00 1 q, ZN 2 7189.00 56.00 0.00 8 q, ZN (Age + Year) 7 7190.90 57.90 0.00 7 q, ZN (Effort Pattern) 3 7190.96 57.96 0.00 3 q, ZN (Age) 3 7190.98 57.98 0.00 11 q, ZN (Age+Stream+Year) 39 7201.80 68.80 0.00 12 q, ZN (Stream Type + Effort 4 7396.20 263.20 0.00 Pattern)

Estimated Parameters

Natural mortality and catchability estimates for the top model were much larger than those used in previous versions of the model, which traditionally included a daily instantaneous natural mortality rate of .002 for type "A" trout streams. Estimates of natural mortality from the individual stream model ranged between .0075 and .1723, although the latter represented a stream where an entire cohort of stocked fish appeared to vanish just after stocking and was never detected by either population surveys or anglers (Kinderhook Creek). The estimated catchability rate for the top model was .00845 compared to the rate of .00152 used previously (Table 1.2).

Table 1.2: Estimated parameter values by stream for the top model.

Stream Type Estimated ZN Estimated q Big Creek A 0.01270 0.00845 Carmans River A 0.01205 0.00845 East Koy Creek A 0.01202 0.00845 Esopus Creek A 0.05033 0.00845 Kayaderosseras A 0.00750 0.00845 Creek Kinderhook Creek A 0.17226 0.00845 Meads Creek B 0.09436 0.00845 Oriskany Creek A 0.02250 0.00845 Otselic River B 0.06772 0.00845

Estimates of natural mortality for the A-B model were also much greater than historical values. Type "A" streams had an estimated instantaneous daily natural mortality rate of

.002 compared to .008 for "B" streams. Estimated catchability rate for the second model was similar to that of the first (.00879 versus .00845) (Table 1.3).

Table 1.3: Estimated parameter values by DEC stream type compared to historical values used in the previous model.

Stream Type Estimated ZN Historical ZN Estimated q Historical q A 0.01467 0.00200 0.00879 0.00152 B 0.08797 0.00800 0.00879 0.00152

Parameter estimates for the individual stream model were relatively comparable to those for their stream type with a clear division between "A" and "B" streams. Type "A" streams show greater variability although this could be due to the fact that many more of them were included in the study than "B" streams.

Model Fit

Model fits to electrofishing population survey observations were relatively good for most streams and years, especially during the fall survey (Figure 1.3). Greater

14 variability was seen during the spring surveys, which took place directly before or after stocking occurred and often in higher flow conditions. Fit to average daily angler catch- rates was poorer and noisy, with the worst fit observed in the fall when population surveys showed few fish but observed catch rates were still high (Figure 1.4). This suggests that the relationship between catch-rate and population density may not be linear as assumed in the model. Overall the fit of the model to the study observations was much better than the previous model and parameter values (Figure 1.5)

Figure 1.3: Fit of model predictions (lines) to observed densities from surveys (points) for all streams and years. "Spikes" are due to stocking events during which a known quantity of fish are added to the study area.

Figure 1.4: Fit of model predictions (lines) to observed average daily angler catch-rates (points) for all streams and years. "Spikes" are due to stocking events during which a known quantity of fish are added to the study area. Predictions include a constant density of wild fish determined from population surveys as angler catch-rate data does not distinguish between wild and stocked fish.

Figure 1.5: Current model fit versus the historical model for an example stream and year.

Discussion

The model formulation identified as best allowed the natural mortality rate to vary by stream. This variation is likely due to site-specific environmental characteristics such as cover availability, predator abundance, or competition within local fish communities, since variation in human factors, such as angling pressure and release rate, are accounted for within the model. The results suggest that these factors were stable within the study period, as the addition of study year to the model parameterization did not significantly improve fit. Such stability indicates that estimated parameter values from this study can be applied in future analyses in the absence of any significant environmental changes.

The model that parameterized natural mortality by the stream type classification system currently used by New York State emerged as the second best model. Stream type is calculated from a number of habitat ratings including overhead and in-stream cover, existing wild trout populations, food availability, and summer temperature regimes. This type of classification system, known as a "habitat suitability index", is a commonly used management tool (Vadas and Orth 2001). Critics of these models, however, contend that they are overly simplistic and often applied inappropriately

(Roloff and Kernohan 1999). While the performance of this model was significantly worse than that of the by-stream model (ΔAIC=5.62), the inclusion of it in the top two models suggests that this approach had some utility for predicting trout population dynamics. From a management perspective, the stream type habitat suitability model has significant advantages over the best-fit by-stream model. Determining a natural mortality rate for each stream requires at least a year of study, ideally several, and requires significant effort in the form of electrofishing surveys and creel agents to conduct angler interviews. Conducting such surveys for every stream under consideration would be extremely costly for what could be characterized as a limited benefit. The habitat-based stream type criteria uses existing data, requires a much simpler, less labor-intensive survey, and allows new streams to be quickly evaluated for stocking. The variability between streams shown by the top model also suggests that improvements to the existing stream type criteria could be explored. Data and model outcomes from this study, combined with more detailed information about the environmental conditions in streams, may elicit grouping criteria that better fit available data.

The top models presented here fit the data much more closely than the historical model and parameters (Figure 1.5) despite evidence that the previous model fit well at the time of its creation (Engstrom-Heg and Engstrom-Heg 1984). The newly estimated mortality rates for the two stream types represent a nearly tenfold increase over historical values. Such an extreme departure raises obvious questions about the changes to the fishery since Engstrom Heg and Hulbert conducted their original surveys. It is important to note, however, that the modern estimates are consistent with other estimates of mortality measured under similar conditions. A 2004 study of stocked, hatchery-reared brown trout used radiotelemetry to track 50 fish released into a Danish stream. After a 32 day period only 13 of the original fish remained, an instantaneous daily mortality rate of .0420 (Aarestrup et al. 2005). Another study that assessed the survival of marked hatchery fish using multipass depletion electrofishing found average instantaneous daily mortalities between 0.113 and 0.1893, with the highest rates 8-20 days immediately following stocking (Pedersen et al. 2003).

High apparent natural mortality rates for stocked hatchery-reared trout are not a new discovery; a 1954 study found that only 65-34% of stocked cohorts of cutthroat trout survived to fall (Miller 1954). Assuming an average of 90 days between stocking and measurement (stocking occasions were in mid-June and "fall" was undefined in this study) this represents a daily mortality rate between .012 and .005, either of which is far higher than the previous CROTS value. These studies, as well as the study that estimated the original CROTS mortality rates, were all conducted in streams known to contain high quality trout habitat and ample surplus carrying capacity. This descriptor applies to the type "A" streams, as characterized by the DEC, but not type "B" streams

20 which are considered to be "infertile and habitat deficient..." with "...relatively little potential for wild or holdover contribution" (Engstrom-Heg 1990). It naturally follows that such streams would have higher apparent natural mortality rates than "A" streams.

Engstrom-Heg estimated that the daily natural mortality rate in "B" streams would be

.008 or four times higher than that in "A" streams. Given the much greater modern estimate, which parallels rates observed in similar waters, an estimated "B" stream mortality rate that is six times higher than that in "A" streams is not surprising.

One potential explanation for greater apparent natural mortality is an increased predator population. Observations collected for a statewide breeding birds survey suggest that hooded mergansers (Lophodytes cullatus) have become much more prevalent since the 1980's (McGowan and Kimberley 2008). This concurs with a study in Massachusetts which found increases in abundance, reproductive success, and distribution between 1979-1983 and 1994-1998 (Heusmann et al. 2000). Northern river otters (Lontra canadensis) are also known to be predators of stream trout (Greer 1955) and their populations have greatly increased since the 1980s when they were extirpated from large areas of the state (Raesly 2001). This, combined with anthropogenic habitat changes, suggests a plausible mechanism for greatly increased estimates of natural mortality.

The apparent lack of a linear relationship between catch-rate and fish density, indicating a constant and estimable catchability, presents a difficult problem for a model intended to guide stocking decisions in such a way as to achieve a minimum catch-rate.

Density dependent catchability, where catch-rate plateaus above a certain density due to gear saturation is known to occur in sport-fishing populations (Peterman and Steer

1981). This can cause catch-rates to be "hyperstable" in that they are relatively unaffected by small changes to density (Dunn et al. 2000). On the other hand, it can mean that stocking above a certain density with the aim of increasing catch rates is likely to not have much influence. Understanding how catchability functions in a stocked fishery is therefore a necessary component for efficient and effective management. This importance extends beyond stocked fisheries, as knowledge of catchability is necessary to understand angler driven mortality in any managed fishery (Serns and Kempinger

1981). Additional studies have continued to find systems where catchability is most strongly related to factors other than density, such as growth rate (VanDeValk et al.

2011). Constant catchability is therefore an easy assumption to make, but an assumption that may not apply to many fisheries, especially sport fisheries where the gear is not designed for mass-capture as in commercial fisheries (Peterman and Steer

1981).

These results highlight the utility of modern statistical techniques in the re- evaluation of historical models used to manage stream fisheries. Population dynamics models are underused in freshwater sport fisheries, but when utilized they can be effective tools for understanding and managing the system in question. Where previous models exist, managers and biologists should take advantage of the opportunities presented by advances in analytical approaches and computing. By employing these techniques, new information can be gleaned from previously existing models and datasets.

REFERENCES

Aarestrup, K., N. Jepsen, A. Koed, and S. Pedersen. 2005. Movement and

mortality of stocked brown trout in a stream. Journal of Fish Biology

66(3):721–728.

Agnew, T. T. 1982. Models for fisheries management. Computers, Environment

and Urban Systems 7(3):145–154.

Alexiades, A. V., B. Marcy-Quay, P. J. Sullivan, and C. E. Kraft. 2015.

Measurement Error in Angler Creel Surveys. North American Journal of

Fisheries Management 35(2):253–261.

Arlinghaus, R. 2006. On the Apparently Striking Disconnect between Motivation

and Satisfaction in Recreational Fishing: The Case of Catch Orientation of

German Anglers. North American Journal of Fisheries Management

26(3):592–605.

Arlinghaus, R., S. Matsumura, and U. Dieckmann. 2009. ORIGINAL ARTICLE:

Quantifying selection differentials caused by recreational fishing: development

of modeling framework and application to reproductive investment in pike

(Esox lucius). Evolutionary Applications 2(3):335–355.

Arreguín-Sánchez, F. 1996. Catchability: a key parameter for fish stock

assessment. Reviews in Fish Biology and Fisheries 6(2):221–242.

Bartholomew, A., and J. A. Bohnsack. 2005. A Review of Catch-and-Release

Angling Mortality with Implications for No-take Reserves. Reviews in Fish

Biology and Fisheries 15(1-2):129–154.

23 Beverton, R. J. H., and S. J. Holt. 1957. On the dynamics of exploited fish

populations. Fishery Investigations Series 2 : Sea Fisheries 19.

Brauhn, J. L., and H. Kincaid. 1982. Survival, Growth, and Catchability of Rainbow

Trout of Four Strains. North American Journal of Fisheries Management

2(1):1–10.

Carpenter, S. R., and W. A. Brock. 2004. Spatial complexity, resilience, and policy

diversity: fishing on lake-rich landscapes. Ecology and Society 9(1):8.

Cowx, I. g. 1994. Stocking strategies. Fisheries Management and Ecology 1(1):15–

30.

Cowx, I. G. 1999. An appraisal of stocking strategies in the light of developing

country constraints. Fisheries Management and Ecology 6(1):21–34.

Dunn, A., S. J. Harley, I. J. Doonan, and B. Bull. 2000. Calculation and

interpretation of catch-per-unit-effort (CPUE) indices. Ministry of Fisheries.

Engstrom-Heg, R. 1990. Guidelines for stocking trout streams in New York State.

Bureau of Fisheries, New York State Department of Environmental

Conservation.

Engstrom-Heg, R., and P. Hulbert. 1982. Evaluation of trout regulations in streams,

1977-1980. Bureau of Fisheries, New York State Department of Environmental

Conservation.

Engstrom-Heg, V., and R. Engstrom-Heg. 1984. A FORTRAN Program for

Predicting Yields of Stocked Trout from Stream Fisheries under Various

Management Alternatives. North American Journal of Fisheries Management

4(4A):440–454.

24 Fedler, A. J., and R. B. Ditton. 1986. A framework for understanding the

consumptive orientation of recreational fishermen. Environmental

Management 10(2):221–227.

Fournier, D. A., H. J. Skaug, J. Ancheta, J. Ianelli, A. Magnusson, M. N. Maunder,

A. Nielsen, and J. Sibert. 2012. AD Model Builder: using automatic

differentiation for statistical inference of highly parameterized complex

nonlinear models. Optimization Methods and Software 27(2):233–249.

Francis, R., R. J. Hurst, and J. A. Renwick. 2001. An evaluation of catchability

assumptions in New Zealand stock assessments. Ministry of Fisheries.

Greer, K. R. 1955. Yearly Food Habits of the River Otter in the Thompson Lakes

Region, Northwestern Montana, as Indicated by Scat Analyses. American

Midland Naturalist 54(2):299–313.

Heusmann, H. W., T. J. Early, and B. J. Nikula. 2000. Evidence of an Increasing

Hooded Merganser Population in Massachusetts. The Wilson Bulletin

112(3):413–415.

Irwin, B. J., T. J. Treska, L. G. Rudstam, P. J. Sullivan, J. R. Jackson, A. J.

VanDeValk, and J. L. Forney. 2008. Estimating walleye (Sander vitreus)

density, gear catchability, and mortality using three fishery-independent data

sets for Oneida Lake, New York. Canadian Journal of Fisheries and Aquatic

Sciences 65(7):1366–1378.

Martin, D. R., and K. L. Pope. 2011. Luring anglers to enhance fisheries. Journal of

Environmental Management 92(5):1409–1413.

25 McGowan, K. J., and K. Corwin. 2008. The second atlas of breeding birds in New

York State. Cornell University Press New York, NY, USA.

Miko, D. A., H. L. Schramm, S. D. Arey, J. A. Dennis, and N. E. Mathews. 1995.

Determination of Stocking Densities for Satisfactory Put-and-Take Rainbow

Trout Fisheries. North American Journal of Fisheries Management 15(4):823–

829.

Miller, R. B. 1954. Comparative survival of wild and hatchery-reared cutthroat trout

in a stream. Transactions of the American Fisheries Society 83(1):120–130.

Neilsen, L. A., and D. L. Johnson. 1983. Fisheries Techniques. Amer Fisheries

Society.

Paloheimo, J. E. 1980. Estimation of Mortality Rates in Fish Populations.

Transactions of the American Fisheries Society 109(4):378–386.

Pedersen, S. S., C. Dieperink, and P. Geertz-Hansen. 2003. Fate of stocked trout

Salmo trutta L. in Danish streams: Survival and exploitation of stocked and

wild trout by anglers. Ecohydrology & hydrobiology 3(1):39–50.

Peterman, R. M., and G. J. Steer. 1981. Relation Between Sport-Fishing

Catchability Coefficients and Salmon Abundance. Transactions of the

American Fisheries Society 110(5):585–593.

Pollock, K. H., J. M. Hoenig, C. M. Jones, D. S. Robson, and C. J. Greene. 1997.

Catch Rate Estimation for Roving and Access Point Surveys. North American

Journal of Fisheries Management 17(1):11–19.

Pollock, K. H., C. M. Jones, and T. L. Brown. 1994. Angler survey methods and

their applications in fisheries management. American Fisheries Society.

26 Post, J. R., E. A. Parkinson, and P. Magnan. 2012. Temporal and spatial patterns

of angler effort across lake districts and policy options to sustain recreational

fisheries. Canadian Journal of Fisheries and Aquatic Sciences 69(2):321–329.

Quinn, T. J. (n.d.). II and RB Deriso. 1999. Quantitative fish dynamics. New York:

Oxford University Press.

Radomski, P. J., and T. J. Goeman. 1996. Decision Making and Modeling in

Freshwater Sport-fisheries Management. Fisheries 21(12):14–21.

Raesly, E. J. 2001. Progress and Status of River Otter Reintroduction Projects in

the United States. Wildlife Society Bulletin 29(3):856–862.

Ricker, W. 1975. Computation and interpretation of biological statistics of fish

populations. Bull. Fish. Res. Board Can 191:382.

Roloff, G. J., and B. J. Kernohan. 1999. Evaluating Reliability of Habitat Suitability

Index Models. Wildlife Society Bulletin 27(4):973–985.

Serns, S. L., and J. J. Kempinger. 1981. Relationship of Angler Exploitation to the

Size, Age, and Sex of Walleyes in Escanaba Lake, Wisconsin. Transactions of

the American Fisheries Society 110(2):216–220.

Sullivan, P. J., and C. E. Kraft. (n.d.). A Review of the Philosophy, Data and

Modeling that Supports the NYSDEC Catch-Rate-Oriented-Trout-Stocking

(CROTS) Program.

Taylor, M. J., and K. R. White. 1992. A Meta-Analysis of Hooking Mortality of

Nonanadromous Trout. North American Journal of Fisheries Management

12(4):760–767.

27 Vadas, R. L., and D. J. Orth. 2001. Formulation of Habitat Suitability Models for

Stream Fish Guilds: Do the Standard Methods Work? Transactions of the

American Fisheries Society 130(2):217–235.

VanDeValk, A. J., J. L. Forney, J. R. Jackson, L. G. Rudstam, T. E. Brooking, and

S. D. Krueger. 2005. Angler Catch Rates and Catchability of Walleyes in

Oneida Lake, New York. North American Journal of Fisheries Management

25(4):1441–144

CHAPTER 2

Angler Specialization and Travel Distance: Applying Two Complementary Statistical

Methods for Interpreting Spatially Referenced Creel Survey Data

Abstract

Effective fisheries management requires a comprehensive understanding of angler usage and behavior. Research has shown that angler attitudes and preferences vary based on their experience and degree of specialization. An accurate understanding of the origin and preferences of anglers utilizing a fishery can therefore help predict the influence that ecological, economic, and management changes will have on a fishery.

Geostatistics provides a powerful tool for exploring these relationships, but has so far seen little use in fisheries management. We evaluated the utility of this method in analyzing creel survey data from nine streams across New York State to make predictions about stream specific origin and technique specialization of anglers. Angler home zip codes were used to create interpolated maps of angler origin densities for each of the streams. The effect of angler specialization was assessed using regression to compare calculated travel distance to three potential factors influencing specialization: income, gear, and stream choice. Regression analysis showed that income and gear choice had a significant effect on travel distance (P<0.05). Travel distance also differed significantly between streams (P <0.05). Spatial variogram models were successfully fit to all streams, providing spatial estimates of angler origin and an intuitive method for the visualizing angler travel dynamics.

Introduction

The use of local streams for recreational fishing can be greatly influenced by changes in angler behavior. Anglers may initially choose sites based on cost, quality of the fishing experience, and environmental aesthetics (Hunt 2005). Once a site has been chosen, the angler is likely to revisit multiple times over the lifetime of their recreational experience (Schuett and Pierskalla 2007). Overall effort within a fishery is therefore likely to be relatively stable from year to year, barring any substantial changes to the perceived quality of the fishery.

Many recreational management strategies, including stocking, are driven by estimates of angler effort with the objective of providing a desired catch rate (Cowx

1994). Of course, high angler effort can lead to overexploitation, with the excessive pressure often targeted at particular species and size classes (Lewin et al. 2006). Such selective exploitation can disrupt food webs and lead to ecosystem-wide disturbances

(Post et al. 2002). Accurate knowledge of the distribution of angler effort is therefore necessary to design effective management strategies and better understand the effects of resource utilization.

Creel surveys or angler censuses are commonly used to estimate annual angler effort. While these methods are effective, they require extensive long-term planning and commitment of manpower because the surveys must be conducted throughout the entire fishing season. To mitigate these costs, surveys typically are conducted when management plans are first developed and then only periodically thereafter, as funding allows, to assess changes in effort over time. Additionally, the ability of management

agencies to conduct reassessments is often limited due to the large number of waters in their purviews (Hartzler 1988).

One potential solution to determining the optimal means for allocating sampling effort is to develop a better understanding of the behavior of anglers and the factors that influence them, such as fishing experience, willingness to travel, and preference for environmental aesthetics. These factors have historically been thought of as interrelated through the concept of “angler specialization”. Anglers can be classified based on their enthusiasm and knowledge, and anglers in these categories tend to demonstrate very different attitudes and behaviors (Bryan 1976). Bryan divided anglers into four categories: Occasional Fishermen, Generalists, Technique Specialists, and Technique-

Setting Specialists. Generally, Bryan (1977) found that investment and willingness to travel in order to enjoy an optimal fishing experience increased as anglers became more specialized. More recent studies have echoed Bryan’s findings and their utility for interpreting angler behavior (Morgan and Soucy 2009; Dorow and Arlinghaus 2012).

Such partitioning has indeed been noted in a number of fishing locales in New York

State, including the Salmon River, where angler technique was found to be correlated to attitudes towards conservation regulations (Dawson et al. 1992). These findings suggest that different types of anglers will vary in their reactions to changing fishery conditions.

Anglers also vary in their interest and willingness to travel for their fishing experience. Due to high site fidelity of recreational anglers, spatially referenced creel survey data can be useful for estimating where angling effort for a particular waterbody originates and providing relatively accurate and temporally stabile measures of angler

effort. While the usage of spatial statistics in social science is growing (Goodchild and

Janelle 2010), historically this has been an underdeveloped technique in the study of angler behavior. While some believe the subjective nature of some social science datasets makes them unsuitable for spatial interpolation analyses (Alessa et al. 2008), there are many other instances where these techniques have been applied successfully. For example, geodemographic analysis, a form of cluster analysis that attempts to classify areas based on the demographic attributes of their residents, has seen increasing use as a tool for marketing and public service planning (Troy 2008).

The geostatistical method known as kriging has been used to spatially interpolate information between data collection points (Davis et al. 2002). Such analyses make use of the spatial correlation that often exists in social and environmental processes.

Originally employed for mining, geostatistics is now widely used in a number of fields including geochemistry, fisheries management, and remote sensing (Petigas 1993;

Wallace et al. 2000; Wang et al. 2001)

The information needed to determine angler specialization, such as only using a particular gear type, is often collected during creel surveys. . While questions that obtain explicit information about the spatial origin of anglers are rare, implicit indicators of origin are commonly collected (e.g., home zip codes). When this data does get used, it is often treated in a one-dimensional fashion, most often as a way to calculate relative travel distance and measured linearly centroid-to-centroid. However, more in depth analyses have been employed, such as using travel distance to estimate travel cost

(Kling and Thomson 1996), examining the potential spread of nuisance species (Gates et al. 2009), and categorizing anglers into broad classes of angler origin, such as

metropolitan and non-metropolitan (Page and Radomski 2006). Further, zip code data have been used identify the home demographics of urban anglers for the purpose of directed marketing (Taylor et al. 2008). While these analyses are relatively simple and limited in scope, researchers in other fields have used more complex statistical techniques, such as zip code based interpolation, to great effect. This has been especially true in epidemiology, where symptoms and causative agents are often tied to location and spatial patterns are analyzed for the purposes of disease monitoring and control (Moore and Carpenter 1999).

In this study we apply geostatistical methods to creel survey information, coupling that with the concept of angler specialization, to gain a better understanding of the composition and distribution of angler usage patterns in trout streams in New York

State. We explore the relationship between factors related to angler specialization and use creel survey data to construct visualizations of angler distribution throughout the

State. Finally, we use this information to make inferences about the behavior and resilience of each study stream fishery.

Methods

Anglers at eight streams in New York State were surveyed throughout the fishing season from 2011-2013. Streams were chosen by New York Department of

Environmental Conservation (NYSDEC) personnel to be representative of trout streams across the State. Roving surveys were conducted for five out of seven days each week, including one weekend day. All surveyed stream sections were managed under general statewide regulations for trout, with no special regulations as to season, creel limit, or gear. Effort was determined through a combination of car counts and direct angler

interviews using a stratified random sampling design (Pollock et al. 1997). Information about fishing success and angler demographics, including home zip code, was collected from each angler surveyed. Because anglers from outside of New York may have additional motivations for visiting within the State, such as proximity to visited relatives, we excluded data from surveys of non-New York State residents. Restricting the data in this way eliminates sources of potential bias about fishing preference associated with out of state travel. When exploring the effects of travel distance and the economic considerations of travel cost, we assumed that New York State resident anglers travelled to their fishing sites by car. In total, 6861 anglers were surveyed during the study period, 3842 of which provided a valid New York State zip code. The number of resident anglers surveyed per stream ranged from 233 (Kayaderosseras Creek) to 739

(East Koy Creek). Esopus Creek had the largest number of anglers surveyed overall

(1585) but only 31% of those indicated that their home was within New York (Table 2.1).

Table 2.1: Number of anglers surveyed per stream from 2011-2013.

NY State Resident Stream Interviews Residents Proportion Carmans River 817 527 0.65 East Koy Creek 1070 739 0.69 Esopus Creek 1585 499 0.31 Kayaderosseras Creek 519 233 0.45 Kinderhook Creek 413 381 0.92 Meads Creek 613 402 0.66 Oriskany Creek 877 491 0.56 Otselic River 912 549 0.6 Total 6861 3842 0.56

Angler counts for each zip code were summed and divided by the total population for that zip code area based on 2010 U.S. Census data. This angler-per- capita calculation standardized angler occurrence across zip codes with varying

population densities. Angler-per-capita data are a proportion and consequently often show a non-normal statistical distribution and were therefore transformed using the logit function. This transformation is appropriate for proportional data and may have advantages over the conventional arcsine transformation (Warton and Hui 2010).

Models

We applied a linear modeling approach to assess the relationship between angler travel distance and stream, median income, and gear choice. Centroid points for each New York zip code polygon were determined using ESRI ArcGIS 10.1 (ArcGis

2012) and angler per-capita values were assigned to each centroid .Travel distance was calculated as the straight line distance from the centroid of each angler's home zip code to the center point of the stream section where that respective angler was interviewed, using the NAD83 UTM Zone 18N projection. Travel distance and median income were log-transformed to meet the assumptions of normality. Median income for each zip code tabulation area was obtained from 2010 U.S. Census results and assigned to each angler by their home zip code. Technique specialization is most likely to occur in anglers who fly fish (Bryan 1977). That is, anglers start with bait and lure fishing and gradually progress in skill and investment towards solely fly fishing-based recreation.

Research also suggests that angler specialization is a discrete process, rather than continuous (Fisher 1997). Therefore, for the purposes of our regression analysis, we coded anglers into two categories of gear specialization: "generalist" (i.e., anglers who used bait or lures) and "specialist" (i.e., anglers that fished solely with fly equipment).

Analysis of covariance (ANCOVA) was used with angler travel distance represented as a function of income (USD), gear choice (bait and lure or fly) and

stream, which was treated as a factor with 7 levels (for the 8 streams included in the study) :

Y     Income   Gear   Stream ...  Stream   0 1 2 3 1 7 7

( 11 ) All statistical analyses were conducted using program R version 2.15.2 (R

2012).

A geostatistical procedure was applied to the angler counts collected from the creel surveys. This procedure, known as kriging, is based on the estimation and application of a model known as a variogram. These models provide curves representing the spatial continuity of the data and are defined by three parameters known as the sill, nugget, and range (Matheron 1963). The standard variogram defines the expected difference value E squared based on the distance factor h.

( 12 ) Multiple types of variograms exist in order to represent a variety of underlying processes. Variogram choice is often accomplished subjectively, but Cressie (1985) considers weighted least squares to be a better method. We used this approach, and the "sp" (Pebesma and Bivand 2005) and "gstat" packages (Pebesma 2004) for R, to fit a separate variogram for each stream, iterating through a set of standard models to determine the best-fit model. Using this approach allows the variograms to differ by fitting each stream separately, as would be expected if the processes driving the data

(i.e. angler behavior) vary between streams.

Interpolation was used in our analysis to estimate the angler density at each location based on the weighted linear combination of the sampled points. This weighting

was based on the distance of each sample point and the variogram model. Thus, closer sample points were more heavily weighted and the degree to which weighting decreases with distance was a function of the model type and the estimated sill, nugget, and range for that stream. The estimate for an unobserved location x0 was therefore a sum of the weights for that location wi(x0) multiplied by the observed values Z(xi).

( 13 ) Results

Using these methods we obtained significant fits for all models and were able to reliably estimate parameters. Both linear and spatial models showed significant between-stream differences in angler demographics and behavior. In all cases post-hoc analysis indicated that model assumptions were met.

In our analysis of angler travel distance relative to median income, gear choice, and study stream all predictors were found to be statistically significant (P<0.05). All individual variables were significantly different from the base level, Carmans River, with the exception of the coefficient for Oriskany Creek (Table 2.2). In addition, pairwise comparisons using the Bonferroni's method showed that streams differed from each other and could be split into three distinct groups (Figure 2.1).Coefficient values were positive with a wide range of values. Oriskany Creek had the lowest coefficient value, with an estimate of 0.08, compared to the highest of 1.28 for East Koy Creek. Adjusted r2 for the model was 0.25 indicating that the chosen factors are helpful for explaining the variability in angler travel distance. The particularly high significance values obtained in

nearly all cases may be due to the relatively large sample size used in the analysis

(n=3842).

Figure 2.1: Estimates and Bonferroni-corrected confidence intervals for all levels of the variable Stream. The dummy variable, Carmans River, is represented by a dashed line on the x-intercept. Significantly distinct groupings are denoted by the lettering above each estimate

Table 2.2: ANCOVA analysis of angler travel distance against angler specialization variables. All levels of the categorical variable "stream " are contrasted against the Carmans River intercept term. The model was significant (P <0.05) with an adjusted r2 of 0.25.

Coefficient Estimate Std. Error t-value P-value Intercept -2.07 0.59 -3.50 0.00 Gear choice 0.19 0.03 5.59 0.00 Median Income 0.43 0.05 8.29 0.00 East Koy Creek 1.28 0.06 22.96 0.00 Esopus Creek 1.15 0.05 21.25 0.00 Kayaderosseras Creek 0.53 0.07 7.78 0.00 Kinderhook Creek 0.51 0.06 8.76 0.00 Meads Creek 0.65 0.06 10.35 0.00 Oriskany Creek 0.08 0.06 1.27 0.20 Otselic Creek 0.96 0.06 15.99 0.00

Variogram parameter estimates for all eight streams were relatively similar(Figure 2.2)., with the greatest variation occurring in the sill parameter (Table

2.3). Given the variogram fits we were able to employ kriging to interpolate expected angler density within the state (Figure 2.3). Ten levels of transparency were used to indicate the estimation variance of each prediction pixel. These levels ranged from 0

(opaque) to 1 (transparent) representing even gradations of variance from <0.1 to >1.0.

This follows established methods for the visualization of uncertainty in interpolated results (Brus et al. 2013) and has the benefit of obscuring areas for which the estimated error is relatively high

Table 2.3: Variogram models and parameter estimates for the study streams

Stream Model Nugget Sill Range Carmans River Gaussian 0.89 1.55 0.65 East Koy Spherical 0.59 1.86 1.97 Esopus Creek Circular 0.86 5.32 2.05 Kayaderosseras Creek Spherical 0.54 2.01 0.52 Kinderhook Creek Power 0.71 4.13 1.73 Meads Creek Matern 0.46 3.28 0.66 Oriskany Creek Spherical 0.94 4.00 1.39 Otselic River Matern 0.84 4.44 2.29

Figure 2.2: Variogram model fits for each study stream.

Figure 2.3: Interpolated angler origination for all study streams with heat map transparency increasing with variation. Extent of study stream is marked with a black line and dots indicate major cities

Discussion

The ANCOVA showed a statistically significant relationship between angler travel distance and various factors involved in angler specialization. Travel distance was positively correlated with median income, a finding that is consistent with other research on willingness-to-travel for recreational experiences. For example, in a study of river recreation area users in Colorado, respondents with higher income chose higher valued trips (Walsh et al. 1990). While studies of other angler datasets have shown that those with higher incomes take trips less frequently (Shrestha et al. 2002), research into recreation of all types suggests that those with higher annual incomes are willing to pay more for the trips they do take (Dardis et al. 1981). This corresponds with the positive effect of income on travel distance observed in our study. It should be noted, however, that income was assessed by linking U.S. Census median income values to anglers based on their reported zip code. This assumes the average income within a zip code area is representative of the average income of anglers within that area. Inclusion of an income component in future creel surveys would be useful for determining the validity of this assumption, although this is often a difficult question to get an accurate response to. The coefficient for angler gear type was also found to be significant and positive, suggesting that fly fishers travel farther for their recreation. These findings concur with the results of other modeling approaches used to investigate angler specialization and fishing preferences (Beardmore et al. 2012). Finally, all levels for the categorical variable Stream were significant with the exception of Oriskany Creek. This suggests that the spatial and ecological aspects of the individual streams play a large role in the origin of their angler populations. Such a conclusion can be seen as a logical result of

the setting specialization discussed by Bryan (1977). One way to restate these results is that higher income anglers, using specialized gear, will travel farther in order to fish at a particular subset of streams.

Spatial statistical analysis provides a valuable tool for developing an improved understanding of angler behavior within a fishery. We were able to easily fit variogram models using commonly collected zip code data, which allowed for more nuanced analyses than straight-line distance metrics. For example, we can draw conclusions about the regions served by a particular fishery. In addition, this type of analysis allows for the detection of "hot spot" areas from which unusually high concentrations of anglers originate.

The variation in our observed variogram parameter estimates results from the heterogeneity of the angler population and the presence of natural travel barriers. The variogram parameter with the most variation was the sill, with estimates ranging from

1.55 to 5.32 (Table 2.3). Because this parameter represents a measure of large-scale variation within the model it is logical that different streams would show different degrees of variation. Streams that draw anglers from a larger area are likely to higher levels of overall variation than streams with a more local audience. The sill was also the value with the poorest observed fit in variogram models, which can be attributed to the heterogeneity of the landscape. The existence of unique, localized features within the data range can also lead deviations from the sill. This can be seen in the variograms for

Esopus Creek, where the presence of New York City increases the semivariance, and

East Koy Creek where the proximity of Lake Erie and Lake Ontario leads to a decrease in the semivariance as distance increases (i.e a hole effect). Another example of this

phenomenon is the Carmans River, a high quality trout stream situated on New York's

Long Island. More than 80% of the Carmans Rivers anglers interviewed preferred fly fishing and the annual effort at that location was the greatest of all study streams. This suggests that the stream provides a high-quality angling experience. Spatial analysis, however, reveals that nearly all of the anglers interviewed originate from within Long

Island. We hypothesized that origin of anglers is not more widespread due to the potential difficulty of traveling through nearby New York City.

Beyond observation of simple usage patterns, spatial statistical methods provide a more in-depth understanding of angler behavior and usage. This can be useful for a number of reasons, including the identification of streams targeted by urban anglers. In an era in which increasing urbanization coincides with decreasing sales of angler licenses, the recruitment and support of urban anglers has been receiving increased attention (Arlinghaus and Mehner 2004; Hutt and Neal 2010). A key to recruiting urban anglers is identifying waterbodies that this class of anglers consider suitable, both in terms of proximity and fishing quality (Balsman and Shoup 2008). Fishing locations currently used by urban anglers, which can easily be identified using spatially-based analyses, likely meet the criteria desired by potential recruits and present ideal locations for workshops and other educational outreach efforts. For example, in our study we found that Kinderhook Creek draws a large portion of anglers from Albany, NY.

Integrating Kinderhook Creek into outreach programs within that city would likely increase their effectiveness in recruiting additional urban anglers. These nonlocal anglers can also have large, beneficial impacts on local economies due to their greater requirements for services such as food and lodging (Steinback, 1999).

The potential to discover underserved markets also exists. For instance, New

York City has very few local trout fishing options, causing its urban anglers to explore opportunities further afield. This can be seen in the angler origination patterns for

Esopus Creek and Kinderhook Creek which have "hot-spots" extending to the urbanized areas in and around New York City (Figure 2.3). Carmans River, however, does not show a similar pattern in angler origin despite its comparable proximity. Interestingly, those anglers who do frequent the Carmans River are highly specialized, suggesting that the fishing experience at that location is of high quality. Our results indicate that urban anglers in New York City might simply not be aware of that location, which therefore presents a potential novel growth opportunity for marketing and recruitment programs within the city.

These results demonstrate that coupling spatial statistics with commonly collected creel survey information can be provide a better understanding the angler dynamics of stream fisheries. Combining the concept of specialization with spatial measures of origination allowed us to classify streams in a way that improved our understanding of angler preferences. This approach is useful in designing and meeting management objectives. Such information can be used to understand angler motivations when developing management strategies, or as a tool for predicting shifts in behavior due to outside factors, such as changes in economic conditions. Coupling this with spatial statistical techniques, such as kriging, gives biologists and managers a way to visualize these data and intuitively make complex connections that help understand angler behavior. Together these approaches offer a simple and cost-effective method of

using historical and future creel survey data to better understand and accomplish fisheries management objectives.

REFERENCES

Alessa, L. (Naia), A. (Anaru) Kliskey, and G. Brown. 2008. Social–ecological

hotspots mapping: A spatial approach for identifying coupled social–ecological

space. Landscape and Urban Planning 85(1):27–39.

ArcGIS, E. 2012. 10.1. Redlands, California: Environmental Systems Research

Institute.

Arlinghaus, R., and T. Mehner. 2004. A Management-Orientated Comparative

Analysis of Urban and Rural Anglers Living in a Metropolis (Berlin, Germany).

Environmental Management 33(3):331–344.

Balsman, D. M., and D. E. Shoup. 2008. Opportunities for urban fishing:

developing urban fishing programs to recruit and retain urban anglers. Pages

31–40 American Fisheries Society Symposium.

Beardmore, B., W. Haider, L. M. Hunt, and R. Arlinghaus. 2013. Evaluating the

Ability of Specialization Indicators to Explain Fishing Preferences. Leisure

Sciences 35(3):273–292.

Beeco, J. A., and G. Brown. 2013. Integrating space, spatial tools, and spatial

analysis into the human dimensions of parks and outdoor recreation. Applied

Geography 38:76–85.

Brus, J., V. Voženílek, and S. Popelka. 2013. An Assessment of Quantitative

Uncertainty Visualization Methods for Interpolated Meteorological Data. Pages

166–178 in B. Murgante, S. Misra, M. Carlini, C. M. Torre, H.-Q. Nguyen, D.

Taniar, B. O. Apduhan, and O. Gervasi, editors. Computational Science and

Its Applications – ICCSA 2013. Springer Berlin Heidelberg.

Bryan, H. 1976. The sociology of fishing: a review and critique. Marine recreational

fisheries. Sport Fishing Institute, Washington:83–92.

Bryan, H. 1977. Leisure value systems and recreational specialization: The case of

trout fishermen. Journal of Leisure Research 9(3):174–187.

Cowx, I. g. 1994. Stocking strategies. Fisheries Management and Ecology 1(1):15–

30.

Cressie, N. 1985. Fitting variogram models by weighted least squares. Journal of

the International Association for Mathematical Geology 17(5):563–586.

Dardis, R., F. Derrick, A. Lehfeld, and K. E. Wolfe. 1981. Cross-section studies of

recreation expenditures in the United States. Journal of Leisure Research

13(3):181–194.

Davis, J. C., and R. J. Sampson. 2002. Statistics and data analysis in geology.

Dawson, C. P., T. L. Brown, and N. Connelly. 1992. The angler specialization

concept applied: New York’s Salmon River anglers. Pages 7–9.

Dorow, M., and R. Arlinghaus. 2012. The Relationship between Personal

Commitment to Angling and the Opinions and Attitudes of German Anglers

towards the Conservation and Management of the European Eel Anguilla

anguilla. North American Journal of Fisheries Management 32(3):466–479.

Fisher, M. R. 1997. Segmentation of the Angler Population by Catch Preference,

Participation, and Experience: A Management-Oriented Application of

Recreation Specialization. North American Journal of Fisheries Management

17(1):1–10.

Gates, K. K., C. S. Guy, A. V. Zale, and T. B. Horton. 2009. Angler awareness of

aquatic nuisance species and potential transport mechanisms. Fisheries

Management and Ecology 16(6):448–456.

Goodchild, M. F., and D. G. Janelle. 2010. Toward critical spatial thinking in the

social sciences and humanities. GeoJournal.

Hunt, L. M. 2005. Recreational Fishing Site Choice Models: Insights and Future

Opportunities. Human Dimensions of Wildlife 10(3):153–172.

Hutt, C. P., and J. W. Neal. 2010. Arkansas Urban Resident Fishing Site

Preferences, Catch Related Attitudes, and Satisfaction. Human Dimensions of

Wildlife 15(2):90–105.

Kling, C. L., and C. J. Thomson. 1996. The Implications of Model Specification for

Welfare Estimation in Nested Logit Models. American Journal of Agricultural

Economics 78(1):103–114.

Lewin, W.-C., R. Arlinghaus, and T. Mehner. 2006. Documented and Potential

Biological Impacts of Recreational Fishing: Insights for Management and

Conservation. Reviews in Fisheries Science 14(4):305–367.

Matheron, G. 1963. Principles of geostatistics. Economic Geology 58(8):1246–

1266.

Miller, J. L. L., and M. L. Erickson. 1974. On Dummy Variable Regression Analysis

A Description and Illustration of the Method. Sociological Methods & Research

2(4):409–430.

Moore, D. A., T. E. Carpenter, and others. 1999. Spatial analytical methods and

geographic information systems: use in health research and epidemiology.

Epidemiologic reviews 21(2):143–161.

Morgan, M., and J. Soucy. 2009. Use of Recreation Specialization to Understand

Resource Knowledge of Trout Anglers. Applied Environmental Education &

Communication 7(4):155–163.

Oh, C.-O., and R. B. Ditton. 2006. Using Recreation Specialization to Understand

Multi-Attribute Management Preferences. Leisure Sciences 28(4):369–384.

Oh, C.-O., R. B. DITTON, D. K. ANDERSON, D. SCOTT, and J. R. STOLL. 2005.

Understanding Differences in Nonmarket Valuation by Angler Specialization

Level. Leisure Sciences 27(3):263–277.

Page, K. S., and P. Radomski. 2006. Compliance with Sport Fishery Regulations in

Minnesota as Related to Regulation Awareness. Fisheries 31(4):166–178.

Pebesma, E. J. 2004. Multivariable geostatistics in S: the gstat package.

Computers & Geosciences 30(7):683–691.

Pebesma, E. J., and R. S. Bivand. 2005. Classes and methods for spatial data in

R. R news 5(2):9–13.

Pollock, K. H., J. M. Hoenig, C. M. Jones, D. S. Robson, and C. J. Greene. 1997.

Catch Rate Estimation for Roving and Access Point Surveys. North American

Journal of Fisheries Management 17(1):11–19.

Post, J. R., M. Sullivan, S. Cox, N. P. Lester, C. J. Walters, E. A. Parkinson, A. J.

Paul, L. Jackson, and B. J. Shuter. 2002. Canada’s Recreational Fisheries:

The Invisible Collapse? Fisheries 27(1):6–17.

Schuett, M. A., and C. D. Pierskalla. 2007. Managing for Desired Experiences and

Site Preferences: The Case of Fee-Fishing Anglers. Environmental

Management 39(2):164–77.

Shrestha, R. K., A. F. Seidl, and A. S. Moraes. 2002. Value of recreational fishing

in the Brazilian Pantanal: a travel cost analysis using count data models.

Ecological Economics 42(1–2):289–299.

Steinback, S. R. 1999. Regional Economic Impact Assessments of Recreational

Fisheries: An Application of the IMPLAN Modeling System to Marine Party and

Charter Boat Fishing in Maine. North American Journal of Fisheries

Management 19(3):724–736.

Taylor, J. B., M. K. Scott, J. A. Henegar, R. K. Betsill, F. W. Janssen, and C. M.

Scofield. 2008. Using On-Site Angler Surveys and a Geographic Information

System for More Effective Marketing of an Urban Fishing Program. American

Fisheries Society Symposium.

Team, R. C., and others. 2012. R: A language and environment for statistical

computing.

Troy, A. 2008. Geodemographic Segmentation. Pages 347–355 Encyclopedia of

GIS. Springer US.

Wallace, C. S. A., J. M. Watts, and S. R. Yool. 2000. Characterizing the spatial

structure of vegetation communities in the Mojave Desert using geostatistical

techniques. Computers & Geosciences 26(4):397–410.

Walsh, R. G., L. D. Sanders, and J. R. Mckean. 1990. The Consumptive Value Of

Travel Time On Recreation Trips. Journal of Travel Research 29(1):17–24.

Wang, Y., T. Ma, and Z. Luo. 2001. Geostatistical and geochemical analysis of

surface water leakage into groundwater on a regional scale: a case study in

the Liulin karst system, northwestern China. Journal of Hydrology 246(1–

4):223–234.

Warton, D. I., and F. K. C. Hui. 2010. The arcsine is asinine: the analysis of

proportions in ecology. Ecology 92(1):3–10.