Assessing the density, demography, and resilience to commercial harvest of aquatic turtles in the Mississippi Delta region of Arkansas

1 Assessing the density, demography, and resilience to commercial harvest of aquatic turtles in the Mississippi Delta region of Arkansas Final Report Submitted to Arkansas Game and Fish Commission December 31, 2021 By Brett DeGregorio U.S. Geological Survey. Arkansas Cooperative Fish and Wildlife Research Unit. University of Arkansas, Fayetteville, USA J.D. Willson and Andrhea Massey Department of Biological Sciences, University of Arkansas, Fayetteville, USA . 2 Executive Summary • We conducted a capture-mark-recapture study across three summer field seasons (2019 – 2021) to assess the community composition, density, and demography of the most harvested freshwater turtle species in agricultural ditches (n=32) and aquaculture ponds (n=51) in Eastern Arkansas and to document the effects of harvest on turtle populations. • Although 10 turtle species can be legally harvested in Arkansas, the red-eared slider (Trachemys scripta elegans) and spiny softshell turtle (Apalone spinifera) dominated these anthropogenically altered wetlands collectively constituting 70% of all turtle captures in ditches and 89% of all turtle captures in aquaculture ponds. Common musk turtles (Sternotherus odoratus), eastern mud turtles (Kinosternon subrubrum), and common snapping turtles (Chelydra serpentina) were also captured regularly in these habitats. Collectively these 5 species represented over 99% of all captures in these two habitat types. • By extrapolating the full range of estimated turtle densities quantified in this study using a Monte Carlo randomization process, we projected that approximately 2.16 million sliders, 427,747 spiny softshells, 942,014 musk turtles, and 76,955 common snapping turtles occur in agricultural ditches and aquaculture ponds across the harvestable region of Eastern Arkansas. • Aquaculture ponds that had experienced commercial harvest 1 year prior to this study had significantly reduced densities of sliders (mean = 4.79 turtles / ha) compared to the two unharvested sites (means = 41.14 and 45.42 turtles / ha). Mean slider density at an aquaculture facility harvested 5 years prior to our study (mean = 15.7 turtles /ha) was still reduced below that of the two comparable unharvested sites. • Sliders constituted a smaller proportion of the turtle community in harvested ponds (38%) compared to unharvested ponds (74% ) potentially indicating that they are removed by commercial trapping at a higher rate than other species. • Spiny softshells also occurred in very low density in the aquaculture ponds harvested 1 year prior to this study (mean = 2.45 turtles /ha). However, at the facility harvested five years prior, softshell densities were higher (mean = 27.83) than at either of the unharvested aquaculture facilities. This finding supports model predictions that softshells rebound more rapidly after a single harvest event than do sliders. • We found relatively few consistent differences in turtle density or demography in harvested and unharvested agricultural ditches. Evidence indicates that the extensive connectivity and dynamic hydrology of these ditches (associated with farming practices, not rainfall) results in unstable turtle communities that are constantly changing due to immigration and emigration of individuals. Turtle movement in ditch networks may mask impacts of local harvest and potentially create population sinks that pull turtles from natural water bodies. 3 • Sensitivity analyses of matrix models parameterized based on our data and the literature suggested that sliders, softshells, and snapping turtles all have relatively slow life histories that rely on adult survival >70% to maintain viable populations. All species exhibited positive population growth (λ > 1) under unharvested conditions. Spiny softshells were least resilient to harvest, with deterministic models suggesting that annual harvest of >5% of adults (>19,500 turtles) would lead to population decline. Red-eared sliders and common snapping turtles were more resilient, with both species predicted to be able to withstand removal of up to 11% of adults annually (244,600 and 8,400, respectively). Common musk turtles were the most resilient, and could withstand the highest proportional rate of annual harvest (~20%; 185,700) of the four species. • Stochastic models for sliders and spiny softshells allowed us to examine probability of local decline, considering chance variation and uncertainty in vital rates. Annual removal of less than 90,000 and 11,000 adults and large juveniles was needed to keep probability of decline below 5% for red-eared sliders and spiny softshells, respectively. • Although turtle populations in the Arkansas Delta are extensive, models demonstrated that local populations are easily overharvested. We estimated that for an average aquaculture pond, annual harvest of just 18 red-eared sliders, 3 softshells, 1 snapping turtle, or 10 musk turtles would be sufficient to result in local population declines; this level of removal could be achieved with just 4 traps set for one day per year. • Models predicted that local populations take years to recover from a single harvest event. In the absence of immigration, adult populations took 7 to 14 years to recover from a single harvest of 10 to 15 trap-nights, but softshells were predicted to recover more rapidly than sliders. • Models evaluating the impact of size restrictions on population growth suggested that minimum size limits were ineffective for sliders and softshell turtles because they shifted harvest pressure to larger females and reduced population growth rate. Maximum size limits were effective in reducing effects of harvest on population growth for both species. Careful selection of species-specific maximum size limits may be necessary to minimize impacts on population growth rates. • We conclude by identifying several key uncertainties and future research needs that might further improve management of turtle populations in eastern Arkansas. 4 Background Freshwater turtle populations are declining worldwide and harvest for food markets and the pet trade have frequently been identified as drivers of these declines (Gibbons et al. 2000; Jensen and Birkhead 2003). Commercial harvest is particularly troubling for turtle populations (Sung et al. 2013) because the industry tends to be under regulated or at times, unregulated (Coulteaux and Johnson 2017). The export of wild turtles from the United States has been increasing rapidly over the last two decades, likely to meet the demand for Asian markets that have depleted local turtle populations (Mali et al. 2014). Scientifically sound, evidence-based regulations may be crucial for protecting the viability of turtle populations in the United States in the face of intensifying harvest pressure. Several aspects of turtle ecology and life history make them especially vulnerable to harvesting, even at modest levels (Congdon et al. 1993, 1994; Stayermark et al. 2008). Most turtles do not reach sexual maturity until an advanced age and survival of hatchlings is often very low (Crouse et al. 1987; Congdon et al. 1994). Demographic analyses of several turtle species indicate that population viability is extremely sensitive to adult mortality. Even modest increases in adult mortality (10%) can have dramatic long-term effects on population persistence (Galbraith and Brooks 1989; Congdon et al. 1993, 1994; Zimmer-Shaffer et al. 2014). Not only does harvest reduce overall abundance and population viability but it can reduce the density of turtles relative to reference sites (Gamble and Simons 2004) and can alter the size and age class distribution of turtles (Close and Seigel 1997). In response to concerns regarding overharvest of turtles, several states have enacted regulations, including minimum size limits, maximize size limits, or slot limits (summarized in Coulteaux and Johnson 2017). Most states that enact regulation adopt size limits such that only larger individuals can be harvested, although a small number of states have done the opposite and only allow the harvest of individuals below a certain size (Coulteaux and Johnson 2017). Other states have yet to enact any regulation (e.g., Louisiana) or allow unrestricted harvest after attaining a permit (e.g., Arkansas). As some states restrict harvest, the pressure on surrounding states that lack regulations may intensify. Even the presence of harvest regulations may not be enough to save turtles. Coulteaux and Johnson (2017) suggest that size restrictions are enough to reduce harvest of snapping turtles by 30-87%, but that the number of turtles harvested is likely to have long-term impacts on populations. In Arkansas, an estimated 1.3 million wild aquatic turtles were harvested, primarily from the Mississippi Delta region, from 2004 to 2017. To date, we lack a comprehensive understanding of current turtle population levels or species-specific sensitivity to harvest that are needed to guide management that ensures persistence of turtle populations in the state. The long-term viability of harvested turtle species will rely upon a scientifically sound understanding of the effects of harvest on turtles, the resilience of different turtle species to harvest, and sensitivity analyses identifying which sexes, ages, and sizes of turtles are most critical for population persistence and thus in the most need of protection. 5 Project Objectives Our objectives were to: 1. Quantify the community composition and density of freshwater turtles inhabiting agricultural irrigation ditches and aquaculture ponds in eastern Arkansas. 2. Use capture-mark-recapture (CMR) techniques to quantify the density of commerciallyharvested freshwater turtle species inhabiting irrigation ditches and aquaculture ponds in eastern Arkansas. 3. Extrapolate density estimates to all available irrigation ditch and aquaculture pond habitat in the region to estimate the total abundance of each species of turtle occurring in the region in these habitats. 4. Compare the density and demography of turtles from sites with known harvest pressure to those that are structurally similar but free of harvest pressure. 5. Use demographic data from our field studies and the scientific literature to parameterize structured population models for each species, and to conduct sensitivity analyses and simulations of these models to: a. Evaluate the resilience of each species to simulated harvest b. Identify sustainable levels of harvest for each species, given estimated total abundances across the Mississippi Alluvial Valley (MAV) of eastern Arkansas c. Evaluate the efficacy of different management scenarios to mitigate negative effects of harvest 6 Study Sites and Turtle Trapping Our study took place widely across 27 counties in Eastern Arkansas where commercial turtle harvest is allowable. These counties are in the Mississippi Alluvial Valley (MAV) ecoregion of Arkansas also referred to as the Delta. Each component of our study relied upon trapping, capturing, and marking turtles in irrigation ditches and aquaculture ponds. The trapping methods used to inform each component of the project did not change and consisted of trapping turtles in 7 counties across eastern Arkansas (Fig 1). We selected sites with and without known commercial harvest history. Our goal was to trap in enough wetlands to be representative of the full range of densities of turtles occurring in the region. We considered sites to be harvested if turtles had been commercially removed from the location within five years prior to our first sampling period, as reported by landowners, Arkansas Game and Fish Commission biologists, or local turtle harvesters themselves. We considered sites to be unharvested if they were located on protected land (i.e., state fish hatcheries or management areas explicitly forbidding turtle harvest). We selected sites that were representative of available habitat across our study region with the intention of selecting a broad range of habitats within the constraints of accessibility. We conducted short, intense trapping sessions, marking turtles to estimate populations at a large number of harvested and unharvested sites, some of which were visited repeatedly in a given year. Given high inter-annual variation in habitat and environmental conditions, as well as turtle densities, we treated the site-year as the replicate in our analyses. We trapped turtles from late April or early May to early August for three consecutive summers (2019 – 2021), capturing turtles with baited hoop net traps (Memphis Net and Twine, Memphis, Tenn.) set approximately 15-30 m apart and left in place for five days (i.e., set Monday, checked daily, and removed Friday). Number of traps set per site varied from 5 to 35 traps based on wetland size in an attempt to standardize trap density across sites. We used custom hoop nets replicating the trap most often used by local commercial turtle harvesters such that traps had two ‘Arkansas style’ funnel throats with mesh size of 3.18 cm and 50.8 cm diameter. The front of the trap was held open with a wooden stake while the back of the trap was held with PVC, hammered into the substrate so that a minimum of approximately 40% of the trap remained above water, allowing captured turtles to breathe. To avoid traps becoming submerged in the event of unexpected rapid water level rise, a common occurrence in agricultural ditches, we fitted each trap with a flotation device in the back section of the trap. In the event of rising water levels, the PVC allowed the trap to slide up, lifted by the float, avoiding the drowning of captured turtles. We baited traps with raw chicken or fish. To avoid spreading disease in aquaculture farms, we only used chicken as bait and decontaminated traps with a 1% Virkon® Aquatic solution between each trapping session. We checked traps daily to remove all captured turtles and to replace bait. We identified the species of each turtles and recorded the mass, straight-line carapace length (SCL) and straightline plastron length (SPL) of each captured turtle. We recorded the sex of each turtle by examining secondary sexual characteristics specific to each species (Ernst et al. 1994; Ernst and Lovich 2009). We individually marked all captured turtles. We marked hard-shelled turtles using a Dremel or hand file to notch the marginal scutes with a unique code, a modified version of the 7 Nagle et al. (2017) marking schematic. We marked soft-shelled turtles with a 2.5 cm metal clip tag displaying a unique number attached to the posterior edge of the carapace (Ostovar et al. 2021; National Band and Tag Company, Newport, KY, USA). We released all turtles near the trap they had been captured in immediately after processing. Figure 1: Location of turtle trapping sites in Eastern Arkansas, USA including a delineation of the Mississippi Alluvial Valley (MAV). Community Composition of Turtles in Irrigation Ditches and Aquaculture Ponds Diversity and Community Composition – In order to describe the turtle assemblage within each habitat type we calculated several indices of diversity. First, we calculated the raw species richness for each site (the number of species captured at that location). We also calculated the percent composition of each species captured at each site. Finally, we calculated a value for the Simpson’s Diversity Index for each site. Simpson’s Diversity Index is a measure of diversity which takes into account both the number of species present (richness) and the relative abundance of each species (evenness) through the following equation: 8 D = 1 – (Ʃn(n-1)/N(N-1) where n is the number of individuals of each species and N is the total number of individuals of all species (Simpson 1949). Values range from 0 (no diversity) to 1 (infinite diversity). We compared percent composition, diversity values and species richness between habitat types using two sample t-tests. We trapped at numerous agricultural ditches (n=32) and aquaculture ponds (n=51). In total, we captured 4,007 individuals of nine species: spiny softshell turtle (Apalone spinifera), red-eared slider (Trachemys scripta elegans), common snapping turtle (Chelydra serpentina), eastern mud turtle (Kinosternon subrubrum), common musk turtle (Sternotherus odoratus), southern painted turtle (Chrysemys dorsalis), Mississippi map turtle (Graptemys pseudogeographica kohnii), river cooter (Pseudemys concinna), and alligator snapping turtle (Macrochelys temminckii). The most commonly captured species were red-eared sliders (n = 2,695), spiny softshells (n = 640), common musk turtles (n = 508), mud turtles (n = 81) and common snapping turtles (n = 56). (Table 1). Table 1: Total number of captures and recaptures of turtles from irrigation ditches and aquaculture ponds in Eastern Arkansas, USA. Although most species occurred in both habitat types, alligator snapping turtles were only captured in one agricultural ditch and all but one of the eleven cooters were also captured in agricultural ditches. Four of the five most commonly captured species (sliders, musk turtles, mud turtles and common snapping turtles) comprised a greater percentage of captures in agricultural ditches than in aquaculture ponds, with one exception: spiny softshells comprised a greater percentage of captures in aquaculture ponds than in agricultural ditches (Fig 2). Sliders accounted for a mean (± 1 SD) of 66.3% (± 22%) of all captures in agricultural ditches and 62.9% (± 32.4%) of all captures in aquaculture ponds. Percent composition of spiny softshells was greater in ponds (mean = 26.4 ± 28.8%) than in ditches (6.1 ± 5.8%; t = -4.26, P > 0.001) (Fig 2). Musk turtles were the second most frequently captured turtle in agricultural ditches and Species Individuals captured Recaptures Total Captures Red-eared Slider (T. s. elegans) 2695 879 3574 Spiny Softshell turtle (A. spinifera) 640 254 894 Common Musk turtle (S. odoratus) 508 29 537 Eastern Mud turtle (K. subrubrum) 81 19 100 Common Snapping turtle (C. serpentina) 56 4 60 River Cooter (P. concinna) 11 0 11 Southern Painted turtle (C. dorsalis) 7 0 7 Mississippi Map turtle (G. p. kohnii) 7 0 7 Alligator Snapping turtle (M. temminckii) 2 0 2 Grand Total 4007 1185 5192 9 accounted for a larger proportion of the turtle community in ditches (mean = 15.7 ± 15.5%) than in ponds (5.2 ± 7.4%; t =2.67, P = 0.015) (Fig 2). Mud turtles accounted for an average 5.8% (± 8.6%) of captures in agricultural ditches and 2.36% (± 8.5%) of captures in aquaculture ponds. Common snapping turtles accounted for an average of 5.7% (± 17.1%) of captures in agricultural ditches and 2.7% (± 10.6%) of captures in aquaculture ponds. Both common snapping and mud turtles were caught in similar proportions in each habitat (P > 0.176). All other species collectively accounted for only 0.8% of all captures in aquaculture ponds and 0.5% of all captures in agricultural ditches. Figure 2: The proportional composition of each of the most commonly captured turtle species within irrigation ditches and aquaculture ponds. Spiny softshells constituted a larger proportion of the turtle community in aquaculture ponds while Common Musk Turtles and Eastern Mud Turtles constituted a larger proportion of the community in agricultural ditches. * indicate significant differences between ponds and ditches at P > 0.001 and P = 0.015 for softshell and musk turtles, respectively). Species richness of turtles in agricultural ditches ranged from 0 to 8, with an average of 3.76 and richness for aquaculture ponds ranged from 0 to 7, with an average of 3.19. Richness was not significantly different (t = 0.81, P = 0.42) between habitat types. Simpson’s Diversity values for agricultural ditches ranged from 0 to 0.68, with an average of 0.35 and for aquaculture ponds ranged from 0 to 0.71 with an average of 0.35. Simpson’s Diversity Index (t = -0.32, P = 0.75) also did not vary between wetland types. 16 Comparison of Harvested to Unharvested Wetlands Density: We compared estimated densities of red-eared sliders, spiny softshell turtles, and common musk turtles between harvested and unharvested ditch sites using two-tailed t-tests. For aquaculture pond sites, we compared estimated densities of the three species among the four aquaculture facilities rather than simply grouping into harvested and unharvested. We chose this approach because the two harvested aquaculture facilities experienced harvest at very different times prior to our sampling (~5 yrs. versus 1 yr.). We were unable to attain this level of resolution for ditch sites and thus grouped them as simply harvested or unharvested. We used Kruskall-Wallis tests (chi-square distribution) to look for differences in density among ponds at the four aquaculture facilities and then used post-hoc Wilcoxon Rank Sum tests to make pairwise comparisons between the different aquaculture facilities. We found significantly higher densities of red-eared sliders in unharvested ponds than in harvested ponds (t = -3.41, P < 0.01). Red-eared slider density estimates in unharvested ditches ranged from 0 – 500.1 (mean = 86.34) turtles/linear km and 0 – 177.48 (mean = 85.23) turtles/linear km in harvested ditches, but we found no significant difference between the two harvest categories (t = -0.04, P = 0.98). When comparing density estimates among the four aquaculture farms, we found significant differences in red-eared slider density among farms (χ² (3) = 12.57, P < 0.01; Fig 6). Aquaculture ponds that had experienced commercial harvest 1 year prior to this study had significantly reduced densities of sliders (mean = 4.79 turtles / ha) compared to unharvested sites (mean = 41.14 and 45.42 turtles / ha). Mean slider density at an aquaculture facility harvested 5 years prior to our study was still reduced below that of comparable unharvested sites (mean = 15.7 turtles /ha). Estimated density of spiny softshells did not differ between unharvested and harvested ponds (t = 1.69, P = 0.63) or ditches (t = 1.49, P = 0.15). However, when exploring differences among the four aquaculture facilities, we found that softshell densities varied among the farms (χ² (3) = 26.09, P < 0.01), driven by low densities at the farm harvested one year prior and one of the unharvested sites (Fig 6). Spiny softshells also occurred in very low density in the aquaculture ponds harvested 1 year prior to this study (mean = 2.45 turtles /ha). However, at the facility harvested five years prior, softshell densities were higher (mean = 27.83) than at either of the unharvested aquaculture facilities. This finding supports model predictions that softshells rebound more rapidly after a single harvest event than do sliders 17 Figure 6: Mean density (± 1 SE) of red-eared slider turtles (top) and spiny softshell turtles (bottom) in ponds at four aquaculture facilities, two of which are protected from harvest and two of which experienced harvest either 1 year prior to our sampling or 5 years prior to our sampling. We found no significant difference in density of common musk turtles between harvested and unharvested agricultural ditches (t = 1.92, P = 0.08) or between harvested and unharvested aquaculture ponds (t = -1.08, P = 0.29), although we caught no common musk turtles in ponds at one unharvested farm. Capture-recapture events did not occur frequently enough to draw many conclusions for common snapping turtles – however, it is important to note that of our 23 captures in aquaculture ponds, only four (17%) occurred in ponds with history of harvest. Of our 30 captures in ditch systems, only eight (27%) occurred in ditches with a history of harvest. 0 10 20 30 40 50 60 70 Unharvested Farm #1 Unharvested Farm #2 Harvested 1yr Prior Harvested 5 yr Prior Turtles per Ha 0 5 10 15 20 25 30 35 40 Unharvested Farm #1 Unharvested Farm #2 Harvested 1yr Prior Harvested 5 yr Prior Turtles per Ha 18 Community Composition: We compared the percent composition of red-eared sliders and spiny softshells captured in harvested versus unharvested wetlands using t-tests. Red-eared sliders dominated the turtle community in three of the four categories, comprising a mean (± 1 SD) of 61% (± 22%) of individuals captured in harvested ditches, 68% (± 24%) of individuals captured in unharvested ditches, and 74% (± 27%) of individuals captured in unharvested ponds. However, in harvested ponds, red-eared sliders comprised only 38% (± 35%) of individuals captured. Percent community composition of red-eared sliders was significantly higher at unharvested pond sites (t = -2.94, P < 0.01) than at harvested pond sites, but we detected no statistically significant difference between harvested and unharvested agricultural ditch sites (t = -0.56, P = 0.59). Spiny softshell turtles were the most frequently captured turtle in harvested ponds, comprising 49% (± 38%) of individuals captured, which was significantly higher than that of unharvested ponds (16% (± 16%); t = 2.73, P = 0.02). While the percent community composition of softshells was not significantly different between harvested and unharvested ditches (t = -0.33, P = 0.75), softshells only comprised about 6.5% (± 6%) of captures in agricultural ditches overall. Size Differences: Both male and female red-eared sliders captured in unharvested ditches had significantly larger mean SCL than those captured in harvested ditches (Females: t = -6.27, P <0.01; Males: t = -7.38, P <0.01). We found no significant differences in the size of male or female red-eared sliders captured in harvested versus unharvested aquaculture facilities (Male: t = 0.83, P = 0.40; Female: t = -.053, P = 0.59). Female spiny softshells captured in harvested aquaculture ponds had larger mean SCL than individuals captured in unharvested aquaculture ponds (t = 2.15, P = 0.03), but we found no significant difference in size for females in harvested versus unharvested ditches (t = 0.18, P = 0.86). We found no significant difference in male SCL in aquaculture ponds (t = 1.76, P = 0.08) or ditches (t = 0.30, P = 0.76). Sex Ratios: We found that sex ratios were not significantly different between harvested and unharvested sites in either habitat type for red-eared sliders (ponds: t = -0.77, P = 0.46; ditches: t = -1.50, P = 0.16), musk turtles (ponds: t = 1.21, P = 0.26; ditches: t = 1.35, P = 0.22), or mud turtles (ponds: t =1.63, P = 0.20; ditches: t = 0.42, P = 0.69). We found that sex ratios were more strongly male biased for softshell turtles in unharvested ditches (0.87M:1F; t = 2.85, P = 0.02), but we found no significant difference in sex ratio between harvested and unharvested ponds (t = -1.94, P = 0.09). While sample sizes at both the aquaculture facility harvested 1 year prior and one of the unharvested farms were not large enough for testing, spiny softshell sex ratios at the other two sites were not significantly different (t = -1.77, P = 0.11). We captured too few common snapping turtles in harvested ponds (one male, two females) to meaningfully compare sex ratios, but sex ratios did not differ significantly between harvested and unharvested ditch habitats (t = 0.92, P = 0.43). 19 Demographic Modeling Model Formulation: We constructed and projected a series of discrete-time matrix population models (Caswell 2001) to examine the population dynamics of our four most commonly harvested turtle species and to explore the effects of harvest on population growth. For our two most frequently harvested and data-rich species, red-eared slider and spiny softshell, we constructed age structured matrix models with 20 age classes and 25 age classes, respectively (e.g., Fig. S4), and a Lefkovitch (1965) modification that retained adults in the oldest age class. Matrix dimensions for these species were selected based on the ages at which adult females reach asymptotic size for each species (Figs. S2B, S3B). Much simpler stage-structured models are possible (e.g., Zimmer-Shaffer et al. 2014). Although these models have fewer parameters and can be more analytically tractable (i.e., allowing for more straightforward sensitivity/elasticity analyses), we chose to use age structured models for several reasons: 1) Most turtle species, including our focal species, exhibit indeterminate growth with body size strongly tied to age for much of their life span; 2) many aspects of life history, especially reproductive characteristics, increase strongly with body size and thus lumping ages into a few stages can inaccurately inflate reproductive capacity of young individuals; 3) fully age structured models more accurately capture transient dynamics that are important drivers in stochastic models and can alter recovery time from a one-time or pulsed population perturbation (Crowder et al. 1994); 4) assigning body sizes to our age structured models based on observed growth rates allowed us to simulate the effects of size-specific removal (i.e., size limits that exclude harvest of certain size classes) on population dynamics. Models were formulated as a pre-breeding census, thus the youngest age class represents one-year old individuals and first year (i.e., nest and hatchling) survival is included along with age-specific reproduction in the top row of the matrix (Fig. S4). Lacking growth data specific to our system for common snapping turtles and musk turtles, we constructed models for these species that were age structured up to age at maturity and then lumped adults into a single stage with a Lefkovitch (1965) modification, resulting in 4x4, and 8x8 matrices for these species. In these models, reproduction is collapsed into a single matrix element in the top right position, representing adult reproduction, nest, and hatchling survival. Model Parameterization We reviewed the literature and used our own data to generate the most plausible set of demographic parameters for each of our species, as well as ranges for parameters that were varied in stochastic simulations (Table 3). For detailed description of parameterization see Section S2. 20 Table 3. Most plausible demographic parameter values for four species of commercially harvested turtles in the Mississippi Alluvial Valley (MAV) of Arkansas derived from our data or the literature and used to model effects of harvest on population growth. Values used in deterministic models are listed along with ranges (in parentheses) for parameters varied in stochastic models. Note that only deterministic models were used for snapping turtles and musk turtles. Clutch size varied by size/age for red-eared sliders and spiny softshells. Literature sources and location of studies are noted. PL=plastron length. NA = data not available Parameter Red-eared slider Spiny softshell Common snapping turtle Common Musk turtle Size at maturity (PL in mm) 1701 1802 NA NA Age at maturity (year) 61 61 79 410 Survival to age 1 (SN*SH) 0.11 (0.01-0.28)3 0.15 (0.06-0.31)4 0.05813 0.2611 Juvenile survival (SJ) 0.87 (0.79-0.94)7 0.78 (0.64-0.79)8 0.7713 0.8610 Adult survival (SA) 0.86 (0.67-1.00)7 0.84 (0.64-0.79)8 0.9413 0.8410 Clutch size 5.3-8.25 7.6-21.56 36.913 2.710 Clutches per year 212 1.513 113 210 Proportion reproductive 0.914 0.914 0.914 0.914 Sex ratio 0.5 0.5 0.5 0.5 1This study (Mississippi Delta of Arkansas) 2Robinson and Murphy 1978; Plummer and Mills 2015; Ernst and Lovich 2009 3Frazier et al. 1990 (South Carolina); mean and range across 5 years, includes nest and hatchling survival 4SN: Kolbe and Janzen 2002, Strickland and Janzen 2010 (Illinois): 6 years of nest survival for Chyrsemys picta mean = 0.46, range = 0.20–0.96; SH: Zimmer-Shaffer et al. 2014 (average across multiple studies). 5Congdon and Gibbons 1983 (South Carolina) varied with body size based on linear relationship between PL and clutch size (clutch size = 0.07*PL-7.31; r2=0.62) derived from 6 populations 6Steen et al. 2015 (Virginia); varied with body size based on linear relationship between PL and clutch size (clutch size = 0.113*PL-14.19; r2=0.24) 7Parker 1996 (Mississippi), mean and range of 5 years (juvenile females), and 10 cohort-years (adult females) 8Plummer et al. 2008 (Arkansas); mean (95% CI) 9Christiansen and Burken 1979 (Iowa), Aresco and Gunzburger 2007 (Florida) 10Mitchell 1988 (Virginia) 11Heppell 1998 (multiple studies); used nest and hatchling survival data for Kinosternon subrubrum 12Trauth et al. 2004 (Arkansas); Ernst and Lovich 2009 (location unspecified) 13Zimmer-Shaffer et al. 2014 (average across multiple studies) 14see methods 21 Model Simulations We conducted several analyses to examine the resilience of our four focal species to harvest and the potential effects of harvest on turtle populations at the local (pond) scale and across the MAV region of Arkansas. All simulations were constructed and simulated in Program R (R Core Team 2017). Sensitivity of Population Growth to Harvest We examined the sensitivity of each model to changes in adult survival (SA) by holding all other parameters constant and varying SA continuously from 0.2 to 1.0 in the matrix, calculating the intrinsic rate of population increase, λ, as the right (dominant) eigenvalue of the matrix, and comparing empirical estimates of SA for each species to the threshold value of SA below which the population was expected to decline (i.e., λ <1). Then, we visualized this result by starting with the baseline (mean) value for adult female survival and plotting the decline in λ as SA was reduced sequentially by 0 to 0.5 (i.e., simulating harvest of up to 50% of adult females). Next, we conducted simulations evaluating the effect of different levels of harvest on population growth rate for each species, given current estimated total abundances across the MAV region. For each species we explored a range of annual harvest levels. For each harvest level, we divided the total estimated catchable (i.e., adult and large juvenile) population of each species (see above) by the simulated harvest level to determine the proportion of the population that would be removed annually at that harvest level (Harvest Proportion). Although our matrix models were constructed for females and do not model males directly, our density estimates include both males and females; thus, our calculation of Harvest Proportion at a given harvest level is unbiased with respect to sex assuming an equal sex ratio and similar individual capture probability among the sexes. We reduced adult survival by subtracting the Harvest Proportion from the proportional value for adult survival, under the assumption that age classes of adult females would be removed in proportion to their relative abundance within the population. Juvenile turtles are under-sampled relative to adults (Fig. S1). We accounted for the fact that juveniles of each species are harvested, but in low proportion relative to their true abundance by dividing the Harvest Proportion by 3 for sliders and by 4 for softshells, snapping turtles, and musk turtles and subtracting the resulting proportion from the value for juvenile survival for all but the 1 year-old age class. This is equivalent to saying that females of each species are not harvested up to age 2 and thereafter are 33% and 25% as likely to be captured, compared to adults, for sliders and the other species, respectively. Having constructed a new matrix that included harvest-induced mortality (i.e., reduced SA and SJ), we calculated λ and plotted λ against the simulated harvest level for each species. As in the previous simulation, simulated harvest levels resulting in λ >1 for each species/parameterization were deemed sustainable because harvest levels at such population growth rates would not be expected to result in population decline. 22 Sensitivity analyses of matrix models parameterized with the most plausible parameter values for our region suggest that sliders, softshells, and snapping turtles all have relatively slow life histories that rely on adult survival >70% to maintain viable populations (Fig. 7). Spiny softshell and common snapping turtles in particular are predicted to require adult survival >80% to maintain positive population growth. However, baseline (unharvested) estimates of adult survival are much greater for snapping turtles (>90%; Congdon et al. 1994, Zimmer-Shaffer et al. 2014) than for softshells (84%; Plummer et al. 2008). As a result, softshells are predicted to be most sensitive to harvest, with a baseline λ of 1.03 and annual removal of >5% of adult females expected to result in population decline (Fig. 8). Sliders have a faster life-history, but lower baseline adult survival than snapping turtles, resulting in similar rates of population growth under unharvested conditions (λ = 1.06 and 1.08, respectively; Fig. 7). Both species are predicted to be able to withstand higher levels of harvest (removal of up to 11% of adult females annually) than softshells while maintaining positive population growth (Fig. 7). Our models suggest that musk turtles have the fastest life history, a high baseline rate of population growth (λ = 1.1) and can withstand the highest proportional rate of annual harvest (~20%) of the four species (Figs., 7, 8). 23 Figure 7. Sensitivity of population growth rate (λ) to changes in adult survival for four common species of freshwater turtles in ditches and aquaculture ponds in the Mississippi Delta of Arkansas, with λ < 1 (below horizontal red line) indicating a declining population and λ > 1 representing a growing population. Baseline adult survival values without harvest for each species are indicated by vertical dashed lines. For each species, the reduction in survival between where the dashed line and the curved line intersect the horizontal red line represents the estimated amount that adult female survival can be reduced without inducing population decline. Thus, for spiny softshells, reducing S A from the baseline value (0.84) to 0.79 would result in a declining population. 24 Figure 8. Sustainability of harvest of four commercially harvested turtle species. Curves represent the effect of removal of adult female turtles on population growth rate (λ). Sections of the curve with λ<1 (below red horizontal line) represent levels of harvest predicted to result in population decline for each species. For example, removal of >6% of adult female softshells annually would lead to population decline. Extrapolating rates of harvest to our mean estimated abundances of each species in ditch and aquaculture pond habitats across the MAV of Arkansas and simulating removal of some juveniles allowed us to examine population growth rate under a range of harvest levels and estimate the maximum level of harvest expected to maintain a stable level of abundance across the region (Harvest Threshold) under baseline conditions (i.e., deterministic model using mean parameter values) (Fig. 9; Table 4). Specifically, models predict that annual removal of greater than approximately 244,000, 19,000, 8,500 or 186,000 turtles would result in population declines for sliders, softshells, snapping turtles, and musk turtles, respectively (Table 4). 25 Figure 9. Population growth rates (λ) under varying levels of annual harvest for the four most commonly harvested species of freshwater turtles in the Mississippi Alluvial Valley of Arkansas, based on mean total abundance estimates across the region (red-eared slider = 2,168,803; spiny softshell = 427,747; common snapping turtle = 76,955, common musk turtle = 942,014) and removal of adults and large juveniles age 2+. Harvest levels resulting in population decline (λ <1; below red line) are considered unsustainable. Stochastic Simulations To explore how environmental stochasticity and uncertainty in model parameters influenced the sensitivity of turtle populations to harvesting across the MAV, we conducted stochastic simulations for sliders and softshells in which survival parameters (SA, SJ, and S0) were varied annually (i.e., across time steps), resulting in fluctuating population trajectories (e.g., Fig. 10). Our initial population vectors were constructed by multiplying our total estimated abundances of each species across the MAV by the size frequency distribution of our captures, thereby allocating our estimated densities across the same age/size distribution represented by our capture records. Because our trapping under-samples small turtles (young age classes), we added additional juveniles that were not included in our population estimates by multiplying our total population estimate by the proportion of the population in the age 1 and age 2 classes predicted 32 unique individual male and female red-eared sliders and spiny softshells captured in unharvested sites during our sampling (Figs. S5, S6) and for each size restriction scenario, used the histograms to delineate the age classes that would be protected in our model, the sex ratio of turtles that would be harvested based on our captures, and the proportion of turtles that we would have had to release (Tables 6, 7). Our baseline scenario assumed an equal sex ratio of captures (which approximated our field captures) and that individuals were removed in proportion to their frequency within the population. We used the same approach here, but modified the harvest by adjusting the sex ratio of harvested individuals and spreading removal only across those age classes that were not protected under each size restriction scenario. As in previous models, this allowed us to reduce survival parameters of the appropriate age classes, constrict a new Leslie matrix that included harvest-induced mortality, and plot λ against the simulated harvest level for each species. Table 6. Influence of different size limits on the composition of captures for red-eared sliders, based on our captures at unharvested sites in the MAV of Arkansas (Fig. S5). % Released indicates the percentages of our captured turtles that would need to be released under each size limit and could be viewed as the degree to which capture rate of harvestable turtles would be reduced. However, it is important to note that although a smaller percentage of turtles would be released under maximum size limits (<13, <14 in), those released would be the largest individuals. Size Limit (CL in inches) Female Ages Protected Sex Ratio of Harvest (%F) % Released >7 <6 50% 31% >8 <7 69% 62% >9 <11 95% 91% <8 >6 36% 38% <9 >10 43% 9% 33 Table 7. Influence of different size limits on the composition of captures for spiny softshells, based on our captures at unharvested sites in the MAV of Arkansas (Fig. S6). % Released indicates the percentages of our captured turtles that would need to be released under each size limit and could be viewed as the degree to which capture rate of harvestable turtles would be reduced. However, it is important to note that although a smaller percentage of turtles would be released under maximum size limits (<13, <14 in), those released would be the largest individuals. Size Limit (CL in inches) Female Ages Protected Sex Ratio of Harvest (%F) % Released >11 <7 100% 63% >12 <8 100% 68% >13 <10 100% 76% <14 >12 50% 15% <13 >9 25% 13% For red-eared sliders (Fig. 13), adopting minimum size restrictions of 7 or 8 inches carapace length did not improve population growth rate and would require harvesters to release 31% and 62% of their catch, respectively, thereby increasing the effort required to achieve a given level of harvest. A minimum size limit of 9 inches protected most turtles, including all males and females less than 11 years old. This was sufficient to prevent population decline, but harvesters would need to release 91% of their catch and would eventually be unable to achieve targeted levels of catch, because turtles >9 inches would be completely depleted. Maximum size limits of 8 or 9 inches were similarly effective in increasing population growth rate, allowing nearly twice the level of harvest as unrestricted harvest, while preventing population decline, and allowing commercial harvesters to retain 62% and 91% of their catch, respectively. Functionally, these regulations work by protecting large breeding females and shifting harvest pressure to males. It is important to note, however, that these maximum size limits would require return of the largest, presumably most commercially valuable, individuals. 34 Figure 13. Influence of different size limits on population growth rate of red-eared sliders under different harvest levels, assuming fishers increase effort to offset lower retention rate of turtles. Analyses assumed a current abundance of 2,168,803 catchable individuals with an even sex ratio and reduced survival according to age classes and sex ratios noted in Table 6. Sections of the curve with λ<1 (below red horizontal line) represent levels of harvest predicted to result in population decline. Note that in the >9 in scenario, the complete level of harvest cannot be achieved at high harvest rates (i.e., turtles >9 in are depleted from the population). Size restriction analyses yielded similar results for spiny softshells (Fig. 14). Minimum size limits were ineffective, slowing population growth at a given level of harvest, while requiring fishers to release >60% of catch. This is not surprising, given the extreme sexual size dimorphism seen in this species; minimum size restrictions shift harvest pressure completely to females. A maximum size limit of 13 inches carapace length was effective, substantially increasing population growth rate at a given level of harvest by shifting harvest pressure to males and smaller females. However, a larger maximum size (14 in) did not protect a sufficient number of adult females to improve population growth. This result can be explained by the faster life history (lower adult survival) and less right-shifted stable age distribution expected for this species. 35 Figure 14. Influence of different size limits on population growth rate of spiny softshells under different harvest levels, assuming fishers increase effort to offset lower retention rate of turtles. Analyses assumed a current abundance of 427,747 catchable individuals with an even sex ratio and reduced survival according to age classes and sex ratios noted in Table 7. Sections of the curve with λ<1 (below red horizontal line) represent levels of harvest predicted to result in population decline. Note that only a maximum size limit of 13 in CL would allow for greater harvest of turtles while maintaining positive population growth. 36 Overall Conclusions Widespread conversion of bottomland forest to agriculture and aquaculture in the Mississippi Alluvial Valley of Arkansas has led to the creation of extensive novel wetland habitats including agricultural irrigation ditches and aquaculture ponds. Using a GIS, we estimated that the area of East Arkansas open to commercial turtle harvest contains more than 18,000 linear kilometers of irrigation ditch and 4,723 aquaculture ponds totaling over 22,000 ha. These anthropogenically created wetlands experience commercial turtle harvest with 10 species of freshwater turtles being legal to harvest from this area. Our study indicated that although at least 9 species of turtles were captured inhabiting these habitats, only 5 species (red-eared slider, spiny softshell, common musk turtle, eastern mud turtle, and common snapping turtle) comprised more than 99% of the turtles occurring in these habitats. By extrapolating the density estimates for the most commonly captured turtles in these habitat types we estimated a mean total abundance of 2,168,803 (95% CI = 2,167,081 – 2,170,524) red-eared sliders, 427,747 (95% CI = 426,976 – 428,517) spiny softshells, and 942,014 (95% CI = 940,827 – 943,200) musk turtles and 76,955 (95% CI = 76,614 – 77,297) common snapping turtles. We estimate that 83% of all spiny softshell turtles in these habitats across the MAV region occur in aquaculture ponds and 17% occur in agricultural ditches while 66% of musk turtles occur in ditches. While 10 species of turtles are legally harvestable in Eastern Arkansas, only these 5 species likely occur regularly enough in agricultural ditches and aquaculture ponds to warrant management attention. Our estimates show a large overall population of turtles occurring in agricultural ditches and aquaculture ponds across the Arkansas MAV, however, the estimated average densities of turtles in each of these habitats are relatively modest compared to estimates for the same species in other similar systems. Although some of our ponds had high slider density, our average densities were lower than published studies from sliders inhabiting golf course ponds and farm ponds (e.g., Rose and Manning 1996, Beshara 2009, DeGregorio et al. 2012). While there are fewer estimates of sliders inhabiting other analogous ditch systems, a similar study showed much higher density of sliders occurring in an urban ditch system in Arkansas (Elston et al. 2016). Our estimates of softshell density in ditches were amongst the lowest density estimates for this species in the published literature indicating that irrigation ditches are not suitable habitat for this species (reviewed in Massey 2021). Conversely, the softshell density estimates from aquaculture ponds in this study are amongst the highest in the literature indicating that aquaculture ponds are very suitable for softshells. By comparing turtle population density and demography between harvested and unharvested aquaculture facilities we found red-eared slider densities were lower in facilities with recent harvest history suggesting that harvest substantially reduced these populations. The reduced density persisted for at least 5 years after harvest. Softshells were similarly much reduced in density a year after harvest in aquaculture ponds. However, evidence from an aquaculture facility that had been harvested five years before our study showed a softshell density equivalent to or higher than at unharvested ponds. Our demographic modeling results indicate that this relatively rapid recovery is due to a large proportion of juveniles in these populations that mature and lead to recovery following a one-time harvest event. 37 Our comparisons of density and demography between harvested and unharvested ditches largely failed to elucidate any differences between sites due to harvest. Turtles in these ditches were found to move large distances and thus the population present in ditches at any given time was likely due more to immigration and emigration than to the actions of harvest. It is likely that harvest in ditches removes a large number of turtles that are present but that new individuals rapidly move in from either adjacent ditch or connected river and stream systems. A better understanding of turtle movement between natural habitats and these linear ditch systems will be important in understanding where turtles move in from and what impacts large-scale harvest might have on adjacent populations occurring in more natural habitats (i.e., rivers and creeks). We should also note that agricultural ditches in this system are used for rice irrigation and as a result the hydrology of these wetlands was extremely dynamic. It was not uncommon to arrive at our ditch sites to check our traps and to find the entire ditch drained of water. These fluctuations are unrelated to weather and are instead driven by the farm management. This dynamic hydrology further facilitates rapid immigration and emigration from these ditch segments making it even more difficult to discern differences in harvested and unharvested areas. We used demographic data from our field studies and those reported from the most comparable field systems in the literature to parameterize a series of structured population models for the four most frequently harvested turtle species across the Mississippi Delta region of Arkansas. Turtles exhibited rapid growth in our systems, which is not surprising given the relatively warm climate of our study region, eutrophic conditions of ditch and pond habitats, and supplemental feeding in aquaculture ponds. As a result, our analyses suggested that all four species exhibited positive population growth under unharvested conditions, and that some species were relatively resilient to harvest. Of our four species, spiny softshells were least resilient to harvest, with deterministic models suggesting that annual harvest of >5% of adults (>19,500) would lead to population decline. Red-eared sliders and common snapping turtles were more resilient, with both species predicted to be able to withstand removal of up to 11% of adults annually (244,600 and 8,400, respectively). Common musk turtles were most resilient and could withstand the highest proportional rate of annual harvest (~20%; 185,700) of the four species. Simulation of stochastic models for sliders and spiny softshells allowed us to examine probability of local decline, considering chance variation and uncertainty in vital rates. Annual removal of less than 90,000 and 11,000 adults and large juveniles was needed to keep probability of decline below 5% for red-eared sliders and spiny softshells, respectively. We lacked sufficient vital rate data to reasonably parameterize stochastic models for common snapping turtles, but it is likely that those would also suggest that the harvest threshold identified by deterministic models would incur significant risk of decline in light of natural variability in these systems. Our study is the first to estimate turtle densities across an extensive region and interpret those densities relative to harvest thresholds identified by models. Taken as a whole, extensive habitat availability and some relatively dense populations of each species suggest that some harvest of turtles from ditch and aquaculture pond habitats in the region is sustainable. Our deterministic results should be viewed as caps to harvest which, if exceeded for extended periods, would likely lead to widespread declines. Our stochastic results are more conservative and suggest that lower harvest rates for sliders and softshells would be prudent to keep risk of decline acceptably low. 38 Maintaining harvest within these guidelines will require accurate and complete reporting by commercial harvesters as well as monitoring of the resource. Although turtle populations in the Arkansas Delta are extensive, models demonstrated that local populations are easily overharvested, as has been noted by many other studies in the scientific literature (e.g., Congdon et al. 1994; Zimmer-Shaffer 2014). We estimated that for an average aquaculture pond, annual harvest of just 18 red-eared sliders, 3 softshells, 1 snapping turtle, or 10 musk turtles would be sufficient to result in local population declines; this level of removal could be achieved with just 4 traps set for one day per year. Further, models predicted that local populations take years to recover from a single harvest event. In the absence of immigration, adult populations took 7-14 years to recover from a single harvest of 10 to 15 trap-nights. Thus, even modest level of harvest could easily deplete individual populations. Given that turtles are known to occur at high densities (reviewed in Massey 2021), serve as prey and predators in many food webs (Ernst and Lovich 2009), and are thought to preform valuable ecosystem services (Lovich et al. 2018), turtle harvest is unlikely to be compatible with conservation on wildlife management areas or wildlife refuges. The more rapid predicted recovery of softshell populations from a single harvest event is interesting and may explain why we observed depressed populations of sliders, but not softshells, at a farm that was harvested five years prior. Future modeling of continual versus pulsed harvest of these species would be interesting from both a conceptual and management perspective. Several management actions that have been implemented in other states (Table S1) could be taken to ease harvest pressure on freshwater turtles in the Mississippi Delta region of Arkansas. Limiting the number of licenses issued, number of traps, season length, regions or habitats open to harvest, or implementing possession limits would all limit total overall take of turtles, and the effects of those actions could be compared to the results of our sensitivity and simulation analyses presented here. Size restrictions, however, operate differently, by altering the demography of harvested individuals without necessarily directly limiting total harvest. Our models evaluating the impact of size restrictions on population growth suggested that minimum size limits were ineffective for sliders and softshell turtles. Because both of these species exhibit female-biased sexual size dimorphism (Fig. S5, S6), minimum size limits shift harvest pressure to larger females and reduce population growth rate by removing these naturally long-lived and fecund individuals. Minimum size limits large enough to protect sufficient smaller females to maintain the population viability severely limit take and could lead to complete depletion of the largest individuals as a result of increased effort by harvesters. Maximum size limits were far more effective and for both species some maximum size limits functioned to improve population growth rate at a given level of harvest, while allowing fishers to retain much of their catch. However, it is worth noting that these strategies require release of the largest (and presumably most commercially valuable) individuals. Maximum size limits for these species function by shifting harvest pressure preferentially to smaller males and protecting large females. Although our data for snapping turtles were insufficient to build the fully age structured models needed to examine size limits, it should be noted that snapping turtles exhibit male-biased sexual size dimorphism (Ernst and Lovich 2009) and thus maximum size limits would likely perform differently for this species. 39 Our research accomplished our objectives of estimating turtle populations in ditch and aquaculture pond habitats across the Arkansas Delta, comparing harvested and unharvested sites, and creating models that delineate sustainable harvest levels and evaluate management actions. However, we also identified several areas of uncertainty where future research would improve management of turtle populations in the region. Key uncertainties and future research needs: • How do agricultural ditch and aquaculture pond habitats compare to more natural habitats in the region in terms of turtle densities and species composition? • Better understanding of turtle movements, especially in ditch habitats and between natural and anthropogenic habitat types. Of particular interest is whether harvest from ditches may negatively affect nearby streams and river systems, given that we documented numerous instances of multi-kilometer movements by turtles in ditch habitats. • Better understanding of common snapping turtle densities and life history in our region. Snapping turtles have lower capture probability than other species, which hampered CMR, and we failed to generate sufficient recaptures to examine individual growth in this species. Given that most of the research on this species has occurred in the far north (e.g., Michigan, Ontario), improved understanding of biology and vital rates in our study systems is needed. • Better understanding of nesting and hatchling biology and survival. Surprisingly few studies have examined nesting biology (clutch frequency, proportion of females reproductive) or nest survival rates in systems analogous to ours and almost none have estimated first-year survival. These are key uncertainties in turtle biology in general and our systems could be ideal for answering these questions. Turtles inhabiting aquaculture ponds and irrigation ditches almost certainly nest along pond levees and roadsides and it is unclear what survival rates for nests in these microhabitats are. • Long-term studies that rigorously estimate mean and variability in adult vital rates in our systems. Rigorous estimation of vital rates for long-lived species requires long-term research with high intensity sampling of focal populations. We have laid the groundwork for this kind of research at several sites, particularly at the state owned fish hatcheries used in this study. 40 Literature Cited Aresco, M.J. & Gunzburger, M.S., 2007. Ecology and morphology of Chelydra serpentina in northwestern Florida. Southeastern Naturalist, 6(3), 435-448. Beshara, Andrea L. 2009. Determining population boundaries and densities of red-eared slider turtles (Trachemys scripta): an empirical field test. University of Central Oklahoma: Jackson College of Graduate Studies: Masters' Theses. Brown, G.P. & Brooks, R.J., 1994. Characteristics of and fidelity to hibernacula in a northern population of snapping turtles, Chelydra serpentina. Copeia, 1994(1), 222-226. Caswell, H. 2001. Matrix population models, 2nd ed. Sinauer Associates, Inc. Publ., Sunderland: MA. Christiansen, J.L., & R.R. Burken. 1979. Growth and maturity of the snapping turtle (Chelydra serpentina) in Iowa. Herpetologica 35: 261-266 Close, L. M. & R. A. Seigel. 1997. Differences in body size among populations of red-eared sliders (Trachemys scripta elegans) subjected to different levels of harvesting. Chelonian Conservation and Biology. 2(4): 563-566. Colteaux, B. C., & Johnson, D. M. 2017. Commercial harvest and export of snapping turtles (Chelydra serpentina) in the United States: trends and the efficacy of size limits at reducing harvest. Journal for nature conservation, 35, 13-19. Congdon, J.D. & Gibbons, J.W., 1983. Relationships of reproductive characteristics to body size in Pseudemys scripta. Herpetologica, 147-151. Congdon, J. D., Breitenbach, G. L., van Loben Sels, R. C., & Tinkle, D. W. 1987. Reproduction and nesting ecology of snapping turtles (Chelydra serpentina) in southeastern Michigan. Herpetologica, 39-54. Congdon, J. D., Dunham, A. E., & van Loben Sels, R. C. 1993. Delayed sexual maturity and demographics of Blanding's turtles (Emydoidea blandingii): implications for conservation and management of long‐lived organisms. Conservation Biology, 7(4), 826-833. Congdon, J. D., Dunham, A. E., & Sels, R. V. L. 1994. Demographics of common snapping turtles (Chelydra serpentina): implications for conservation and management of long-lived organisms. American Zoologist, 34(3), 397-408. Cooch E.G. & G.C. White. 2012. Program MARK: A gentle introduction. Colorado State University, Fort Collins, Colorado. Crouse, D. T., Crowder, L. B., & Caswell, H. 1987. A stage‐based population model for loggerhead sea turtles and implications for conservation. Ecology, 68(5), 1412-1423. Crowder, L. B., Crouse, D. T., Heppell, S. S., & Martin, T. H. 1994. Predicting the impact of turtle excluder devices on loggerhead sea turtle populations. Ecological Applications, 4(3), 437445. 41 DeGregorio, B. A., Grosse, A. M. & J.W. Gibbons. 2012. Density and size class distribution of Yellow-bellied sliders (Trachemys scripta scripta) inhabiting two barrier island wetlands. Herpetological Conservation and Biology 7(3):306-312 Dunham, A.E. & Gibbons, J.W., 1990. Growth of the slider turtle. Life history and ecology of the slider turtle, 135-145. Elston, J., Rolland, V. & Trauth, S.E., 2016. Urban ditch characteristics associated with turtle abundance and species richness. Herpetological Conservation and Biology, 11(1), 132-141. Ernst, C.H., R.W. Barbour, & J.E. Lovich. 1994. Turtles of the United States and Canada. Smithsonian Institute Press, Washington, DC. 578 pp Ernst, C.H. & Lovich, J.E., 2009. Turtles of the United States and Canada. JHU Press. Fitch, H.S. & Plummer, M.V., 1975. A preliminary ecological study of the soft-shelled turtle Trionyx muticus in the Kansas River. Israel Journal of Ecology and Evolution, 24(1-2), 28-42. Frazer, N.B., Gibbons, J.W. & Owens, T.J., 1990. Turtle trapping: preliminary tests of conventional wisdom. Copeia, 1990(4), 1150-1152. Galbraith, D. A., & Brooks, R. J. 1989. Age estimates for snapping turtles. The Journal of wildlife management, 502-508. Gamble, T., & Simons, A. M. 2004. Comparison of harvested and nonharvested painted turtle populations. Wildlife Society Bulletin, 32(4), 1269-1277. Gibbons, J.W. & Semlitsch, R.D., 1982. Survivorship and longevity of a long-lived vertebrate species: how long do turtles live?. The Journal of Animal Ecology, 523-527. Gibbons, J. W., D. E. Scott, T. J. Ryan, K. A. Buhlmann, T. D. Tuberville, B. S. Metts, J. L. Greene, T. Mills, Y. Leiden, S. Poppy, and C. T. Winne. 2000. The global decline of reptiles, déjà vu amphibians. BioScience 50:653–666. Godwin, C.D., Doody, J.S. & Crother, B.I., 2021. The Impact of ATVs on Survival of Softshell Turtle (Apalone spp.) Nests. Journal of Herpetology, 55(2), 201-207. Heppell, S.S., 1998. Application of life-history theory and population model analysis to turtle conservation. Copeia, 367-375. Jensen, J. B., & Birkhead, W. S. 2003. Distribution and status of the alligator snapping turtle (Macrochelys temminckii) in Georgia. Southeastern Naturalist, 2(1), 25-34. Kolbe, J.J. & Janzen, F.J., 2002. Spatial and temporal dynamics of turtle nest predation: edge effects. Oikos, 99(3), 538-544. Lefkovitch, L.P., 1965. The study of population growth in organisms grouped by stages. Biometrics, 1-18. 48 slider both the model and the capture data suggested most adults in the 180-220 mm PL range. Similarly, for spiny softshell the most prevalent age class in captured adults (240-270 mm PL) also showed a peak in the model-generated stable age distribution. These suggest that our estimates of adult survival were appropriate; if they were too low we would see larger adults in the capture data that were missing from the model. Size at Maturity Through opportunistic palpation of captured females, we identified 52 gravid red-eared slider turtles. The 5 smallest of these gravid females measured 155, 172, 172, 175, and 186 mm PL. Thus, we set the size at sexual maturity at 170 mm PL for female Red-eared slider in our populations. Growth We used capture-mark-recapture data to examine growth and age at maturity for red-eared slider and spiny softshell in our study systems. For each species we amassed all multi-year recaptures and calculated growth in plastron length (PL; mm) and time interval between recaptures. We then plotted growth rate in mm/day versus PL for males and females separately and fit those relationships with linear best fit lines. Some individuals exhibited slightly negative growth rates due to measurement error; these values were retained to avoid bias towards faster growth. We then used the program Growth II (Pisces-Conservation Ltd, Lymington, Hants, UK) to fit incremental growth data for females to a von Bertalanffy growth model, PLt = PL∞ (1 – be-rt), where PLt is the plastron length at time t, PL∞ is the asymptotic plastron length, PL0 is the hatchling plastron length, b is a parameter related to hatchling plastron length (~1 – PL0/PL∞), r is the growth rate, and t is the time since hatching. We set PL0 at 30 mm for red-eared slider and 30.4 mm for spiny softshell (Ernst and Lovich 2009). We plotted age versus PL curves generated using the mean and upper and lower bounds of the 95% CI of the growth parameter r, while holding PL∞ constant at the size dictated by the largest 5% of females captured for each species (223 mm PL for red-eared slider; 293 mm PL for spiny softshell). In addition to females with multiannual recaptures, for red-eared slider we included one unsexed juvenile and two juveniles with less than annual recapture intervals (mean = 43 days) to improve our approximation of juvenile growth, given that males and females grow rapidly and at similar rates as juveniles (Ernst and Lovich 2009). For spiny softshell, we excluded one female that exhibited unusually slow growth because she had recently lost a limb, which undoubtedly affected her growth. For common snapping turtle and musk turtle, we lacked sufficient multiannual recapture data to model growth in our systems. 49 Figure S2. Growth of red-eared slider in aquaculture pond and agricultural ditch habitats in the Mississippi Alluvial Valley (MAV) of Arkansas, based on capture-mark-recapture data (N=142 recaptures; mean recapture interval = 452 days). A) Growth rates of recaptured individual female and male red-eared slider, with linear trendlines. B) Von Bertalanffy growth curve for female red-eared slider, generated from data in A). Lines represent size at age for mean (solid black) and upper and lower bounds of the 95%CI (solid gray) of the growth parameter (r) generated by the von Bertalanffy growth model, holding asymptotic size (PL∞) constant at 223 mm. Note that size at sexual maturity (red dashed line; 170 mm PL) is attained in the 5th year of growth for females, and thus first reproduction occurs at age 6 years. A substantial dataset of multiannual recaptures for red-eared slider (N=142 recaptures; mean recapture interval = 452 days) allowed us to assess growth and age at maturity for our study system (Fig. S2). There was a strong negative correlation between body size and growth rate for both male and female red-eared slider, with females generally growing much more rapidly than males (Fig. S2A). Our sample sizes for juveniles were small, but growth rates of individuals <100 mm PL were high and appeared to be similar among the sexes, as has been noted in other studies of this species (Ernst and Lovich 2009). Fitting a von Bertalanffy growth curve to the recapture data for female red-eared slider yielded a mean estimate of the growth parameter (r) of 0.251 (upper and lower 95% confidence intervals (CI) = 0.200 and 0.303). The growth curve generated by using this growth rate and holding asymptotic size (PL∞) constant at 223 mm (the size encompassing 95% of our female red-eared slider captures) is shown in Fig S2B. Based on the mean curve, females attain mature size early in their 5th year and thus first reproduce at age 6. By age 10, females surpass 200 mm PL and thereafter slow dramatically in growth. Spiny softshell grew rapidly in our system (Fig. S3). Analysis of multiannual recapture data (N=45 recaptures; mean recapture interval = 368 days) suggested substantially faster growth of females in our population than those studied by Plummer and Mills (2015) in an urban stream in Arkansas (Fig. S3A). Fitting a von Bertalanffy growth curve to the recapture data for female spiny softshell yielded a mean estimate of the growth parameter (r) of 0.164 (upper and lower 95% CI = 0.068 and 0.206). The growth curve generated by using this growth rate and holding 50 asymptotic size (PL∞) constant at 293 mm (the size encompassing 95% of our female spiny softshell captures) is shown in Fig S3B. Based on the mean curve, females attain mature size early in their 5th year and thus first reproduce at age 6. In contrast to red-eared slider, however, female spiny softshell continue to grow later in life, not approaching PL∞ until age 20, based on the mean growth curve. Figure S3. Growth of spiny softshell in aquaculture pond and agricultural ditch habitats in the Mississippi Alluvial Valley (MAV) of Arkansas, based on capture-mark-recapture data (N=45 recaptures; mean recapture interval = 368 days). A) Growth rates of recaptured individual female and male spiny softshell, with linear trendlines. Dotted blue line is the best fit line for female growth rates from an urban stream in Arkansas (Plummer and Mills 2015), for comparison. B) Von Bertalanffy growth curve for female Spiny softshell, generated from data in A). Lines represent size at age for mean (solid black) and upper and lower bounds of the 95% CI (solid gray) of the growth parameter (r) generated by the von Bertalanffy growth model, holding asymptotic size (PL∞) constant at 293 mm. Note that size at sexual maturity (red dashed line; 180 mm PL [Robinson and Murphy 1978]) in the 5th year of growth for females, and thus first reproduction occurs at age 6 years. 51 S2. Model Formulation Figure S4. Example Leslie Matrix representation of age-structured discrete-time popualtion model for female red-eared slider, following the most plausible paramiterization (Table 3) used in deterministic projections. Females first reproduce at age 6 years. Annual survival parameters are denoted (S); S0 (first year survival; the product of nest (SN) and hatchling (SH) survival), juveniles (SJ), and Adults (SA). Fage represents age-specific female fecundity (# females eggs per female per year), and is a product of proportion of females reproductive, clutch size, # clutches per season, and sex ratio. 52 Table S2. Life table for female red-eared slider turtles featuring most plausible parameter values (Table 3) used to parameterize deterministic simulations. First year survival (S0) includes nest (SN) and hatchling (SH) survival. Age-specifc fecundity (F – female eggs laid per female per year) is the product of average clutch size, # clutches per year, and sex ratio. Age Class (years) Plastron Length (mm) Clutch Size F Survival (Sx) 0 30 0 0 0.10 1 73 0 0 0.874 2 106 0 0 0.874 3 132 0 0 0.874 4 152 0 0 0.874 5 168 0 0 0.874 6 180 5.3 4.78 0.859 7 190 6.0 5.38 0.859 8 197 6.5 5.84 0.859 9 203 6.9 6.21 0.859 10 207 7.2 6.49 0.859 11 211 7.5 6.71 0.859 12 214 7.6 6.88 0.859 13 216 7.8 7.01 0.859 14 217 7.9 7.11 0.859 15 219 8.0 7.19 0.859 16 220 8.1 7.25 0.859 17 220 8.1 7.30 0.859 18 221 8.2 7.34 0.859 19 221 8.2 7.37 0.859 20 222 8.2 7.39 0.859 53 Table S3. Life table for female spiny softshell turtles featuring most plausible parameter values (Table 3) used to parameterize deterministic simulations. First year survival (S0) includes nest (SN) and hatchling (SH) survival. Age-specifc fecundity (F – female eggs laid per female per year) is the product of average clutch size, # clutches per year, and sex ratio. Age Class (years) Plastron Length (mm) Clutch Size F Survival (Sx) 0 30 0 0 0.147 1 67 0 0 0.718 2 102 0 0 0.718 3 130 0 0 0.718 4 155 0 0 0.718 5 176 0 0 0.718 6 194 7.7 5.18 0.837 7 209 9.4 6.33 0.837 8 221 10.8 7.30 0.837 9 232 12.0 8.13 0.837 10 241 13.1 8.83 0.837 11 249 14.0 9.42 0.837 12 256 14.7 9.93 0.837 13 261 15.3 10.36 0.837 14 266 15.9 10.72 0.837 15 270 16.3 11.03 0.837 16 274 16.7 11.30 0.837 17 277 17.1 11.52 0.837 18 279 17.3 11.71 0.837 19 281 17.6 11.87 0.837 20 283 17.8 12.01 0.837 21 284 18.0 12.12 0.837 22 286 18.1 12.22 0.837 23 287 18.2 12.30 0.837 24 288 18.3 12.37 0.837 25 289 18.4 12.43 0.837 54 S3. Model Parameterization Red-eared Slider Life history parameters used to model slider population dynamics are listed in Table 3 and displayed in life table form in Table S2. Based on our data (Section S1), we set size at maturity for female slider at 170 mm PL for all simulations and used the growth curve generated by our capture-mark-recapture (CMR)-based von Bertalanffy growth model (Fig. S2B) to set PL of individuals in each age class (Tables S2). Specifically, females grew from 30 mm PL at hatching to 168 mm PL at age 5, first reproduced at age 6 (180 mm PL), and surpassed 200 mm PL by age 9. By age 20, females neared the asymptotic PL for our population (PL∞) of 230 mm. Growth rates of our Most Plausible parameterization are similar to those of females in a large reservoir in South Carolina (Dunham and Gibbons 1990) and an Illinois lake (Thornhill 1982), but faster than those of females in declining population in a semi-permanent isolated wetland in South Carolina (Gibbons and Semlitsch 1982; Dunham and Gibbons 1990). Female sliders may lay clutches of up to 30 eggs (Ernst and Lovich 2009). Clutch size is highly variable among individuals and populations, but much of this variability is attributable to variation in body size of females; the relationship between PL and clutch size is positive and relatively consistent among populations (Ernst and Lovich 2009; Congdon and Gibbons 1983). Congdon and Gibbons (1983) examined reproduction of sliders across six populations in South Carolina and found a strong positive correlation between PL and clutch size (r2=0.62), that explained most interpopulation variation in clutch size. We used their linear relationship (clutch size = 0.07*PL-7.31) to set clutch size for each age class of mature females in our models based on body size, resulting in mean clutch sizes of 5.3 to8.2 across age classes (Table S2). Overall, our clutch sizes were smaller than the overall average of 10.5 reported by Ernst and Lovich (2009) from 234 clutches from across the species’ range, but this is not necessarily surprising, given that females attain larger body size in eastern United States populations. We were lacking data on clutch frequency and annual proportion of reproductive females, and surprisingly few studies have quantified these parameters for sliders. Ernst and Lovich (2009) report 1 to 5 clutches laid annually, with a mean of 2.5 clutches per female, based on 325 clutches. Trauth et al. (2004) reported 2 to 3 clutches per year, based on dissection of 11 Arkansas specimens. Based on the high resource availability in our systems, as indicated by rapid growth, we assumed that 90% of females lay 2 clutches annually. Taking into account average clutch size, reproductive frequency, and an equal sex ratio of hatchlings, annual fecundity (F; # female eggs laid per female annually) ranged from 4.78 to 7.39 across age classes. Survival of nests, hatchling and juvenile sliders has also seldom been studied. In the declining population that has been the subject of intense study at Ellenton Bay, SC, the proportion of eggs known to have been carried by nesting females that were captured entering the wetland the 55 following spring (i.e., incorporating nest and hatchling survival and potentially emigration of nesting females or hatchlings) ranged from 1% to 27.5% over 5 years, with an average of 10.5% (Frazer et al. 1990). Given that this is the most comprehensive study of slider nest and hatchling survival, we used the mean (10.5%) value from Ellenton Bay in deterministic simulations and drew randomly from the 5 annual values in stochastic simulations. Sliders are thought to have a Type II survivorship curve, with relatively constant survival of juveniles and adults (Ernst and Lovich 2009; Gibbons and Semlitsch 1982). Studies in South Carolina (Frazer et al. 1990), Illinois (Tucker and Sloan 1997), and Mississippi (Parker 1996) all found annual survival rates of 80-90% that did not differ substantially among age classes or sexes. Of these studies, Parker’s (1996) study in Mississippi farm ponds is most similar to our study systems and contains particularly robust data for juveniles. Thus, we used Parker’s (1996) mean estimates of juvenile (87.4%) and adult female (85.9%) annual survival for deterministic simulations and drew randomly from the empirical survival rates for juvenile females (5 years: range = 0.79 to 0.94), and adult females (10 cohort-years: range = 0.67 to 1) in stochastic simulations. Spiny Softshells We parameterized our models for spiny softshell (Table 3; Table S3) based on our own data and the literature, with particular reliance on the comprehensive population studies conducted in an Arkansas stream by Plummer and colleagues (Plummer and Mills 2015; Plummer et al. 2008). Our analysis of female growth based on CMR suggested that Spiny softshell in our study systems grow substantially faster than those studied by Plummer and Mills (2015), which is not surprising given that our sites are likely warmer and more productive (e.g., receiving supplemental food at aquaculture facilities) than their urban stream study site. Specifically, mean growth rates generated by our von Bertalanffy growth model (Fig. 3B) suggested that females reach size at maturity (>180 mm PL; Robinson and Murphy 1978; Plummer and Mills 2015; Ernst and Lovich 2009) in their 5th year of growth, first reproduce at age 6, and attain 240 mm PL by age 10. We based our age-structured matrix model on the body sizes predicted by this model (Table S3). Few studies have characterized reproductive biology of spiny softshell. Ernst and Lovich (2009) reported clutch size of 3 to 39 with a mean of 17.9. Previous researchers have noted that larger females have larger clutches than smaller females (Robinson and Murphy 1978; Ernst and Lovich 2009; Godwin et al. 2021), but statistical analyses of body size-clutch size relationships have been hampered by small sample sizes. We allowed clutch size to vary with female body size based on spiny softshell clutches (n=11) from southwestern Virginia (data derived from Steen et al. 2015). Specifically, we used the linear relationship between PL and clutch size (clutch size = 0.113*PL-14.19; r2=0.24) to assign mean clutch sizes for each age class, yielding mean clutch sizes of 7.7 to 18.4 with increasing age (Table S3). No studies have measured frequency of multiple clutching or proportion of females reproducing annually directly in spiny softshell. Ernst and Lovich (2009) state that “one or two, possibly three clutches are produced 56 each season” and Robinson and Murphy (1978) suggested double-clutching based on disparity in the size of follicles in dissected females from Tennessee. Based on the productivity of the habitats we studied and the relatively rapid mean growth rates indicated by our CMR data, we assumed 90% of females reproduce annually in all models, and that females lay an average of 1.5 clutches annually. Taking into account average clutch size, reproductive frequency, and an equal sex ratio of hatchlings (Ernst and Lovich 2009), annual fecundity (F; # female eggs laid per female annually) ranged from about 5.18 to 12.43 across age classes. Survival of softshell turtle (Apalone spp.) nests varies considerably across sites and years, with primary sources of mortality including nest predation, flooding (along rivers), and disturbance by vehicles (Fitch and Plummer 1975; Plummer 1976; Tornabene et al. 2018; Godwin et al. 2021). Unfortunately, these studies were conducted in regions or habitats dissimilar to ours (i.e., along rivers where flooding is the primary source of mortality [Godwin et al. 2021] or in very different climates [Lake Champlain, Vermont; Tornabene et al. 2018). Alternatively, it is likely that nest success rates are more strongly tied to habitat and geography (i.e., predator community, environmental conditions, soils, etc.) than to species. The most analogous system to our own where freshwater turtle nesting success has been monitored rigorously has been nests of Chrysemys picta on an anthropogenic causeway along the Mississippi River in Illinois (Kolbe and Janzen 2002; Strickland and Janzen 2010). Nest success in this system varied from 20 to 96% across 11 years, with an average of 46%. We used this mean in our deterministic simulations and drew randomly from the 11 annual empirical values in stochastic simulations. Survival of hatchling Spiny softshell to age 1 has not been estimated. Thus, following ZimmerShaffer et al. (2014), we used mean hatchling survival values for common snapping turtles in Michigan (0.32; Congdon et al. 1994), yielding a cumulative probability of survival to age 1 of 0.15 in deterministic models and 0.06 to 0.31 in stochastic simulations (Table 3). The most comprehensive studies of survival of spiny softshell are those of Plummer et al. (2008) in an Arkansas stream. We used the mean survival estimates for juveniles (mean = 0.72, upper and lower 95% CI = 0.63 and 0.79) and adults (mean = 0.84, 95%CI = 0.78 and0.88) in our deterministic models and drew randomly from an even distribution across the 95% CI for each parameter in stochastic simulations (Table 3). Common snapping turtle Our field data were too limited to derive life history parameters for common snapping turtles specific to our study systems. Thus, parameterization of our stage-based model for snapping turtles (Table 3) primarily followed Zimmer-Shaffer et al. (2014) who reviewed the available literature for this species and extracted mean parameter values across studies, the most comprehensive of which are based on Michigan (Congdon et al. 1987; 1994) and Ontario (Galbraith and Brooks 1989; Brown and Brooks 1994) populations. We deviated from Zimmer-Shaffer (2014) in one important way: by setting age at maturity at 7 years, instead of 9 years. Aresco and Gunzburger (2007) found that female common snapping 57 turtles reach maturity in 6 to 8 years in northwestern Florida, and Christiansen and Burken (1979) concluded that “Ovulations first occur during the sixth or more often seventh full year of growth or even later.” Based on our geographic location and rapid growth of softshell and sliders in our study systems, we expect that females mature at earlier ages in our systems than those studied further north. Likewise, we assume that 90% of females are reproductive annually in our system (versus 85% in Michigan; Congdon et al. 1994). Common musk turtles We parameterized our stage-based model for musk turtles (Table 3) based primarily on population studies of this species in Virginia (Mitchell 1988). Females matured at age 4, and laid 2 clutches of 2.7 eggs annually. Mean juvenile and adult survival were 0.84 and 0.86, respectively. As with other species, we assume that most (90%) females reproduce annually in our system. Lacking data on nest and hatchling survival for this species, we used the average value (0.26) reported for mud turtles at Ellenton Bay in South Carolina (Frazer et al. 1991).