Characterization of Western Painted Turtle Bycatch in Fyke Nets During Freshwater Fish Population Assessments

Research focusing on fisheries bycatch has increased substantially, although most research has occurred in marine systems. Raby et al. (2011) noted that 1,115 of 1,152 papers on bycatch were devoted to marine systems, and only 37 focused on freshwater systems. Much of the research concerning freshwater bycatch has focused on commercial fisheries (Barko et al. 2004; Lowry et al. 2005; Larocque et al. 2011; Midwood et al. in press). The few studies pertaining to bycatch during freshwater biological sampling have focused on turtle bycatch during sampling for channel catfish Ictaluris punctatus (Sullivan and Gale 1999).

Researchers have recommended fyke nets for completing freshwater fish population assessments in lentic littoral areas (Miranda and Boxrucker 2009). Fyke nets commonly capture turtles as bycatch. In a comparison of five passive fishing gears, Barko et al. (2004) found fyke nets capture the most turtles and caused the highest turtle mortality. Turtle mortality generally results from drowning, which occurs when captured turtles are submerged for prolonged periods of time (Bishop 1983; Sullivan and Gale 1999; Barko et al. 2004; Larocque et al. 2011). We have observed western painted turtle Chrysemys picta bellii mortality during fish population assessments in northeastern South Dakota; however, the observed mortality has not been quantified.

Due to the high numbers of western painted turtles observed as bycatch in fyke nets, it is important for fisheries biologists to increase knowledge of turtle bycatch during fish population sampling. Therefore, our objectives were to 1) characterize the bycatch of western painted turtles caught in modified fyke nets during fish population assessments in lakes and impoundments and 2) relate the observed catch rates to weather factors. Understanding western painted turtle bycatch during fish population assessments would provide insight into how bycatch may affect turtle populations and help to identify methods to potentially minimize bycatch.

We sampled a total of 39 natural lakes and nine impoundments located in northeastern South Dakota (Table S1, Supplemental Material). Impoundments consisted of a dam constructed in a stream drainage to hold water in a reservoir. Among the lakes sampled, 35 are located within the Prairie Coteau (Figure 1). The Prairie Coteau is a 330-km-long plateau running from northern Iowa to the North Dakota border and contains thousands of wetlands and lakes created during the Illinoian and Wisconsin glacial periods (Willis et al. 2007). The other four lakes are located in the James River drainage along with eight of the impoundments. One impoundment is in the Red River drainage. Lakes and impoundments in northeastern South Dakota are relatively shallow, with maximum depths typically less than 8 m and the deepest being 12.5 m (Table S1, Supplemental Material). Waters in northeastern South Dakota are productive and are predominately eutrophic to hypereutrophic.

We collected western painted turtle bycatch data from modified fyke nets during fish population assessments in northeastern South Dakota from June to September of 2007 to 2012 at 39 lakes and nine impoundments. Modified fyke nets differ from regular fyke nets in that they have rectangular frames to enhance their stability (Hubert 1996). Modified fyke nets were constructed of 1.9-cm mesh (bar measure), with a 15.2-m lead, two 0.9 × 1.5 m rectangular frames, and three 0.9-m diameter hoops (Figure 2). The entrance was 7.6 cm wide × 0.9 m high, and had a maximum stretch width of approximately 17.8 cm. The throat was 18 cm in diameter, exhibited minimal stretch, and included a crow's foot to reduce fish escapement.

Net locations used during fish population assessments are standardized to allow comparison of trend data and we sampled no net locations more than once annually. Our standard sampling protocol is to set nets in the morning, lift them the following day, and move them to a different standard location each day of the survey. We staked leads on shore and stretched the nets perpendicular to the shoreline. Net depth and habitat at each location varied; however, we did not record depth or habitat data. We did not alter nets to allow turtles access to air (e.g., float, chimney, etc.) and the cod-end of the fyke net rested on the lake bottom. We did not add bait; therefore, the nets are considered unbaited.

Our annual netting effort varied depending on the waters sampled. We sampled six lakes and two impoundments annually and all other waters on a rotation ranging from 2 to 5 y. We sampled no waters more than once annually (Tables S2 and S3, Supplemental Material) and we assessed individual waters at approximately the same date to enable comparison of trend data for fishery management.

We recorded the number of western painted turtles captured, the curved carapace length (CCL; mm), and sex of all western painted turtles captured for each net. We measured the CCL with a flexible metal measuring tape and determined the sex of each turtle by observation of secondary sexual characteristics including fore-claw length and precloacal tail length (Ernst 1971; Gibbons and Lovich 1990; Mitchell 1994; Rowe 1997). Sex of western painted turtles less than 10 cm CCL is difficult to differentiate using secondary sexual characteristics; therefore, we classified them as undetermined. We did not determine sexual maturity of captured turtles. We calculated an overall sex ratio (females : males) for each sampling year.

We calculated catch rates as the number of turtles captured per net night. We compared catch rates between impoundments and lakes using a two-sample t-test. We used analysis of variance (ANOVA) to determine if annual variation in surface area and mean depth occurred for lakes and impoundments. We log₁₀(x + 1)–transformed data for the t-test and ANOVA to better meet the assumption of normality, and variances were similar for the data sets.

Catch rate data for lakes and impoundments were not normally distributed, nor was the assumption met concerning homogeneity of variances. Because of the violation of these assumptions we ranked the data and used a 1-way ANOVA on the ranked data (Conover and Inman 1981) to test for differences across years for both lake and impoundments. When ANOVA was significant, we separated mean ranks using the Bonferroni method. Although we completed statistical tests using ranks, we have presented data means for discussion. We used linear regression to analyze the relationship of catch rates with water mean depth and surface area for lakes and impoundments. We analyzed variations in catch rate by day of year using linear regression. Fish population assessments generally occur over multiple days; therefore data collected during each fish population assessment were assigned to the first day of the population assessment to match the mean catch rate to the day of year. We tested normality using the Shapiro–Wilk test and none of the data used in linear regression analysis exhibited normal distributions. Lack of normality is considered a minor violation of the assumptions for linear regression analysis of data sets with large sample size (Kleinbaum et al. 1988; Sokal and Rohlf 1995). We completed all aforementioned statistical tests in SYSTAT 13; alpha was set at 0.05.

We used an information theoretic approach to evaluate relative support of weather variables to observed annual catch rates of western painted turtles in the six lakes and two impoundments that we sampled annually. We used Akaike's information criterion (AIC) to evaluate the support of weather models. Because of the small sample size we used second-order AIC (AIC_c) values to calculate the difference in AIC_c between each model and the lowest-scoring model (Δ_i), and the difference in model weights (w_i) between each model (Burnham and Anderson 2002). We calculated model-averaged parameter estimates to account for model selection uncertainty, including 95% confidence intervals for each parameter. We calculated the residual sum of squares for each candidate model in SYSTAT 13 and the AIC_c and model-averaged parameter estimates in Excel (2010).

We included summer precipitation, winter precipitation, mean summer air temperature, and winter severity as weather variables in the analysis. Due to the potential for substantial variation in annual water levels, we separated precipitation into winter and summer seasons to determine if precipitation during either period affected the observed annual catch rates. We defined the winter precipitation parameter as the sum of precipitation calculated from 1 December through 31 March to account for snow accumulation prior to the sampling period. We defined summer precipitation as the sum of precipitation from 1 April through 30 September of each sampling year to account for precipitation from ice-out through the end of the sampling period. We calculated the mean summer air temperature parameter using monthly mean temperature from June through September to encompass the sampling period. We used winter severity to determine if overwinter mortality, relating to duration of ice cover, affected bycatch rates. We defined the winter severity parameter as the number of days with mean air temperature below 0°C from 1 November through 30 April prior to the sample year to encompass the entire period in which waters could be ice covered. We obtained weather data from the nearest weather station to each water body. We used three weather stations: the Weather Service Office at the Aberdeen, South Dakota, regional airport; the Webster, South Dakota, water department; and the Automated Surface Observing System Surface weather observation station at the Watertown, South Dakota, municipal airport.

We removed waters with a sample size of less than 10 turtles from sex ratio analysis. We used a χ² test to examine potential differences in sex ratios between sampling years. We used linear regression to assess variations in sex ratio by day of year. We assigned data to the first day of each fish population assessment to match the mean sex ratio to the day of year. We tested normality using the Shapiro–Wilk test. The data were not normally distributed; however, lack of normality is considered a minor violation of the assumptions for linear regression analysis of large samples (Kleinbaum et al. 1988, Sokal and Rohlf 1995). We tested differences in the bycatch size distribution in lakes and impoundments using the Kolmogorov–Smirnov test. The statistical tests listed above were completed in SYSTAT 13; alpha was set at 0.05.

We caught a total of 5,729 western painted turtles during 2,237 net nights. Turtle bycatch in individual nets ranged from 0 to 62 with the mean being 2.56 per net night (SE = 0.12). We observed a significant difference in catch rate (t_2,235 = 2.65; P = 0.008) between lakes and impoundments with the mean catch rate being higher for impoundments (mean = 3.48/net night; SE = 0.29) than lakes (mean = 2.36/net night; SE = 0.14). Annual mean catch rates for lakes ranged from 1.07 to 3.28 western painted turtles per net night (Table 1), and the difference between years was significant (F_5,1832 = 15.49; P < 0.001). Annual mean catch rates for impoundments ranged from 0.70 to 6.63 western painted turtles per net night (Table 1), and the difference between years was significant (F_5,391 = 9.94; P < 0.001).

Mean depth data were not available for 12 waters and we sampled three of those waters twice from 2007 to 2012 (Table S1). Mean depth data were available for 90 fish population assessments in lakes and 24 in impoundments and we knew surface area for 102 fish population assessments in lakes and 27 in impoundments. The mean depth and surface area did not differ significantly across years for lakes (F_5,84 = 1.15; P = 0.341 and F_5,96 = 0.54; P = 0.744, respectively) or impoundments (F_5,18 = 0.15; P = 0.976 and F_5,21 = 0.11; P = 0.989, respectively).

The relationship between mean depth and catch rate was significant for lakes (P = 0.040), but little variation in the catch rate was explained by mean depth (r² = 0.05; Figure 3). Turtle catch rates tended to decrease with increasing mean depth in lakes; however, we found no relationship for impoundments (P = 0.911). Catch rates were poorly related to lake surface area (P = 0.609) or impoundment surface area (P = 0.199). We observed a significant but weak relationship (r² = 0.28, P < 0.001; n = 102) for catch rate by day of year for lakes, but not impoundments (P = 0.448; n = 27). Western painted turtle catch rates in lakes increased from June to September (Figure 4).

Winter precipitation was the most supported model describing annual catch rates for lakes (w_i = 0.32, Δ_i = 0.00) and impoundments (w_i = 0.53, Δ_i = 0.00; Table 2). The confidence interval for the winter precipitation parameter did not include zero and had a negative parameter estimate, indicating that higher winter precipitation correlated to lower catch rates for lakes and impoundments (Table 3). The second most supported model differed between water types; winter severity being the second most supported model for lakes (w_i = 0.18, Δ_i = 1.12) and summer precipitation being the second most supported model for impoundments (w_i = 0.22, Δ_i = 1.78; Table 2). We observed negative parameter estimates for the top two models tested for each water type. Confidence intervals for all parameters tested, except winter precipitation, include zero. Air temperature and combined models showed little support in describing the observed variation in annual catch rates.

The western painted turtle bycatch comprised 2,141 female, 3,436 male, and 152 undetermined individuals. The annual sex ratios (female : male) ranged from 0.59:1 to 0.64:1 for lakes and 0.50:1 to 0.76:1 for impoundments (Table 1). The bycatch sex ratio did not differ significantly across years for lakes (χ²₅ = 0.82; P = 0.976) or impoundments (χ²₅ = 4.89; P = 0.430). We observed a significant but weak relationship between the sex ratio and day-of year for lakes (r² = 0.084, P = 0.005; n = 70), but not impoundments (P = 0.145; n = 21); the proportion of female western painted turtles captured decreased from June to September (Figure 5).

The size distribution for female western painted turtles was significantly larger than males for lakes (P < 0.001) and impoundments (P < 0.001; Figure 6). Males in lakes had a maximum CCL of 22.0 cm and an overall mean CCL of 16.0 cm (SE = 0.41). Males in impoundments had a maximum CCL of 21.8 cm, and an overall mean CCL of 16.7 cm (SE = 0.75). The maximum CCL for females in lakes was 24.9 cm, and the mean CCL was 17.4 cm (SE = 0.80). Females in impoundments had a maximum CCL of 24.5 cm, and a mean CCL of 18.2 cm (SE = 1.43). Western painted turtles classified as undetermined sex ranged in CCL from 5.4 to 9.9 cm for lakes and 5.2 to 9.9 cm for impoundments (Figure 6). The overall size structure was skewed toward larger individuals with the mean CCL of all captured turtles being 16.5 cm (SE = 0.39; Figure 6).

Passive sampling gears are extensively used by scientific researchers of fishes in freshwater ecosystems. However, researchers often inadvertently capture nontarget species, such as turtles, in passive sampling gear. Understanding bycatch during biological sampling is a critical first step to minimize the impact that fish sampling may be having on nontarget species.

The strongest conclusion that we can make about modified fyke net bycatch rates of western painted turtles pertains to variations in annual catch rate and the associated models. The absence of variation in annual mean depth and surface area for both water types precludes either from explaining the observed variation in annual catch rates. Therefore, the strong support for the winter precipitation model for each water type indicates that annual modified fyke net bycatch rates are highly influenced by the previous winter's precipitation.

We believe there may be two reasons for this observation. First, it is likely that a greater proportion of western painted turtles migrate from permanent waters to adjacent waters with increasing water levels during the spring due to increased winter precipitation. In northeastern South Dakota water level can fluctuate substantially because many of the wetland and lake basins are classified as closed or transitional. Water level is highly dependent on precipitation and evaporation in closed and transitional lake and wetland basins (Kalff 2002). McAuliffe (1978) found western painted turtles in oxbow lakes in Nebraska to exhibit patterns of emigration and immigration in response to increasing and decreasing water levels. Emigration from permanent waters occurred when water levels increased and immigration back to permanent waters occurred with decreasing water levels. Sexton (1959) observed painted turtles moving from hibernation ponds to outlying bodies of water in the spring. We contend that increased runoff enables a greater proportion of turtles to migrate from permanent waters to nearby nesting and summer ponds, resulting in decreased catch rates during fish population assessments in permanent waters.

The second reason annual catch rates may be influenced by winter precipitation is that western painted turtles likely remained in some of the nesting and summer ponds because these habitats, which may not be suitable for hibernation during low-water years, could provide the necessary conditions for hibernation following winters with above-average precipitation. Johnson et al. (2004) observed a shift in the time of inundation by approximately one permanence class for wetlands within the Prairie Coteau during periods of drought and deluge. Increased precipitation resulted in seasonal wetlands behaving as semipermanent wetlands and semipermanent wetlands behaving as shallow lakes. Kahara et al. (2009) observed an increase in the number of semipermanent wetlands and a decrease in the number of temporary and seasonal wetlands with above-average precipitation in areas with high wetland density. They attributed the observed difference to wetlands expanding and merging with nearby wetlands. Kahara et al. (2009) indicated that expansion of semipermanent wetlands may benefit species suited to larger and deeper wetlands. In South Dakota, western painted turtles have been found in all wetland types but most often in lakes (Bandas 2003), which potentially indicates a preference for lacustrine over palustrine wetlands. As a result of an increase in prevalence of larger and deeper wetlands during times of increased precipitation, fewer western painted turtles returned to the permanent lakes prior to hibernation, resulting in lower bycatch rates.

Dispersal patterns and how they relate to precipitation likely differ across the range of western painted turtles varying by latitude, regional hydrology, and climate. However, variations in fyke net bycatch relating to turtle dispersal patterns may be applicable to populations in northern latitudes and areas with a high density of closed or transitional basin wetlands. A mark–recapture or telemetry study would help confirm our conclusions and would likely provide further insight into turtle migration patterns and how they relate to bycatch rates.

Winter severity was the second strongest model explaining annual catch rate variation for lakes and the third strongest for impoundments. Painted turtles develop metabolic acidosis during prolonged submergence associated with hibernation. Acidosis can result in death in painted turtles if blood pH falls by one unit (Ultsch et al. 1984). Ultsch et al. (1999) observed that greater duration of submergence and lower oxygen content of the water resulted in more severe acidosis. The winter severity model parameter estimate for each water type was negative with decreasing catch rate as winter severity increased. However, based on the weak model parameter estimate, overwinter mortality likely has a minimal impact on annual modified fyke net bycatch rates.

Summer precipitation appears to have an impact on bycatch rates in impoundments but not in lakes. The reason for the difference between water types is unknown and cannot be accounted for by data collected during this study. Further research is needed because fisheries management occurs in impoundments throughout the country.

We observed significant relationships in comparisons of catch rate in lakes vs. impoundments, catch rate and mean depth for lakes, and catch rate and day of year for lakes. However, little variation in catch rates was explained by the tested variable for these three significant relationships. Variables that were unaccounted for in this study likely influenced the observed variability. Additional research is needed to understand the factors that influence bycatch rates. For example, identifying variations in catch rate could be useful in determining time periods when biological and commercial use of modified fyke nets would have the lowest bycatch rate of western painted turtles.

Water surface area was not related to catch rates in lakes or impoundments and, therefore, should not be used as a predictor of bycatch rates for either water type. In addition, no significant relationships in catch rate were evident for mean depth of impoundments and day of year for impoundments. It is possible that the influence of mean depth and day of year on modified fyke net catch rates differ between the two water types. However, the substantially smaller sample size for impoundments and the high degree of unexplained variability among significant relationships for lakes prevent substantial comparison.

Skewed sex ratios can be the result of differences within the population or biased sampling methods. Unfortunately, we found no prior research on sex ratios in unbaited modified fyke nets. Ream and Ream (1966) observed a male-biased sample in baited hoop nets with catches comprised of 33% females. Bandas (2003) observed a sex ratio nearer to 1:1, comprised of 44% females in South Dakota. Approximately 95% of the western painted turtles captured during their study were caught in baited hoop nets like those used by Ream and Ream (1966). We observed annual modified fyke net bycatch sex ratios between those observed in hoop nets by Ream and Ream (1966) and Bandas (2003), ranging from 37 to 39% females among lakes and 33 to 43% females in impoundments. Comparisons of painted turtle sex ratios in baited hoop nets indicate varying results; therefore, we cannot determine if the skewed sex ratio we observed is a result of a biased sampling method or is an indication of a male-dominated sex ratio within the populations.

Further research is warranted because a skewed sex ratio within modified fyke net bycatch may indicate that bycatch mortality rates are higher for one gender, which can affect turtle populations. High rates of female turtle mortality relating to road density has resulted in male-dominated sex ratios among populations in areas of high road density and has called into question the survival of some populations (Gibbs and Shriver 2002; Gibbs and Steen 2005; Steen et al. 2006). A source of gender-specific mortality, cumulatively with other causes of mortality, could have substantial impacts on turtle populations.

We observed a significant relationship between sex ratio and day of year for lakes but not impoundments. It is possible that the influence of day of year on bycatch sex ratio differs between water types. However, as with catch rates, the difference in sample size between water types and unexplained variation prevent comparison.

We were unable to find any previous research on the size structure of western painted turtles collected in unbaited modified fyke nets. The size structure of western painted turtles collected in our modified fyke nets was similar to that reported for baited hoop nets (Ream and Ream 1966). Ream and Ream (1966) observed a capture bias skewed toward larger individuals with 96% adult and 4% juvenile painted turtles. We did not identify sexual maturity of captured turtles, but the juveniles captured in hoop nets reported by Ream and Ream (1966) ranged in size from 6 to 11 cm in carapace length, which is similar to the size structure we observed for undetermined sex turtles (Figure 6). The bycatch of undetermined sex turtles in unbaited modified fyke nets was 3% and is similar to the 4% of juveniles observed by Ream and Ream (1966) in baited hoop nets. Bandas (2003), primarily using baited hoop nets, observed 3% juveniles in a study in South Dakota. The low proportion of juveniles reported by Bandas (2003) and Ream and Ream (1966) support the size bias we observed in unbaited modified fyke nets.

It is important to note that inherent sampling biases may result in substantial effects on the at-large population. The skewed size structure we observed could indicate that most turtle mortality in modified fyke nets is directed toward large, mature turtles. Among painted turtles, Mitchell (1988) indicated that natural mortality rates were higher for juveniles than adults. This variation in mortality, coupled with long life spans of painted turtles (Congdon et al. 2003), often skews the population size structure toward large, adult turtles. Brooks et al. (1991) noted that increased mortality of adults, among species that exhibit high juvenile mortality and long life spans, can have substantial effects on the population.

This study has provided insight into western painted turtle bycatch during fish sampling with modified fyke nets. However, this study highlights the need for more research to better understand bycatch and the impacts of biological sampling on nontarget species. Future research focused on mortality rates of turtle bycatch could indicate negative impacts associated with fish sampling and could lead to research designed to mitigate negative impacts on turtle bycatch.

Please note: The Journal of Fish and Wildlife Management is not responsible for the content or functionality of any supplemental material. Queries should be directed to the corresponding authors for the article.

Table S1. Site description for 39 lakes and nine impoundments where we conducted western painted turtle Chrysemys picta bellii bycatch assessments from 2007 to 2012 in northeastern South Dakota. Data include surface area (ha), maximum depth (m), mean depth (m), and site coordinates (UTM). Parenthesis indicate county. All UTM coordinates are in zone 14T.

Found at DOI: http://dx.doi.org/10.3996/102014-JFWM-077.S1 (12 KB XLSX).

Table S2. Netting effort (number of net nights) by waterbody for each year of sampling during western painted turtle Chrysemys picta bellii bycatch assessments conducted from 2007 to 2012 in northeastern South Dakota.

Found at DOI: http://dx.doi.org/10.3996/102014-JFWM-077.S2 (12 KB XLSX).

Table S3. Catch rate of western painted turtles Chrysemys picta bellii (mean number per net night) by waterbody for each year of sampling during bycatch assessments conducted from 2007 to 2012 in northeastern South Dakota.

Found at DOI: http://dx.doi.org/10.3996/102014-JFWM-077.S3 (11 KB XLSX).

We want to thank all the permanent and seasonal South Dakota Game, Fish and Parks employees that assisted with the collection of data, especially R. Braun and S. Kennedy. We also want to thank T. Kaufman for advice on data analysis and G. Adams and C. Pasbrig for reviewing an earlier draft of this paper. Substantial improvements to the manuscript were possible due to excellent comments and critiques by the Associate Editor and the anonymous reviewers of this manuscript.

Any use of trade, product, or firm names is for descriptive purposes only and does not imply endorsement by the U.S. Government.