Capture vulnerability of sea turtles on recreational fishing piers

Abstract Capture vulnerability of commercial and recreational fishes has been associated with behavioral, morphological, and life‐history traits; however, relationships with non‐target species, such as sea turtles, have not been adequately studied. We examined species composition, timing of captures, morphological variables including body size and head width, and body condition of sea turtles captured from a recreational fishing pier in the northern Gulf of Mexico and of sea turtles captured in the waters adjacent to the pier. From 2014 to 2019, 148 net captures and 112 pier captures of three sea turtle species were documented. Green turtles were captured most frequently in the net and on the pier. Turtles captured from the pier were larger than those captured in the net. There was no difference in head width between net‐caught and pier‐caught turtles; however, small sample sizes limited those comparisons. The body condition index was lower for pier‐caught than net‐caught Kemp’;s ridleys but did not differ with green turtles or loggerheads. Differences were also observed in the timing of capture on the pier as compared to in the net. Finally, the relationship between size, body condition, and pier‐capture vulnerability suggests these are complex interactions. Mortality of sea turtles captured from fishing piers could be selecting against bolder individuals, which may result in changes in sea turtle population demographics over a long time period.


| INTRODUC TI ON
Human population growth is increasing in coastal areas (Bounoua et al., 2018). This rapid rate of urbanization globally has led to dramatic changes to natural habitats (Gaston, 2010;Uchida et al., 2019).
Consequently, some individual animals have adjusted their behaviors by foraging in new habitats or on novel prey items (Breck et al., 2019;Garamszegi et al., 2009;Tuomainen & Candolin, 2011). Drastic environmental changes often encourage individual behavioral traits such as boldness (Breck et al., 2019;Kelleher et al., 2017;Klefoth et al., 2017), which results in indirect selection against morphological and life-history traits such as larger body size and faster growth rates (Alós et al., 2014;Enberg et al., 2012;Klefoth et al., 2017). These changes in individual traits can lead to increases in predation risk or vulnerability of capture in fishing activities of larger, bolder individuals (i.e., capture vulnerability; Klefoth et al., 2017;Phillip et al., 2009), thus ultimately leading to negative population-level effects. As anthropogenic activities increase in coastal habitats, the response of wildlife, particularly imperiled species, could have far-reaching consequences to ecology and conservation.
In the marine environment, one anthropogenic activity that has been shown to alter traits of target species is commercial and recreational fishing (Hamley, 1975;Klefoth et al., 2017;Lewin et al., 2006). In fact, commercial fishing is likely influencing the course of evolution for many target species (Enberg et al., 2012). Capture vulnerability has been shown to be a heritable trait in some fish species Phillip et al., 2009), and behavioral traits such as boldness and morphological traits such as body size can impact catchability (Enberg et al., 2012;Uusi-Heikkilä et al., 2008). Hookand-line fishing, in particular, is more likely to capture individuals that are exploratory or have higher activity, boldness, or aggression levels. These behavioral traits increase encounters with fishing gear or increase the probability of ingesting certain baits or lures (Alós et al., 2012;Arlinghaus et al., 2016Arlinghaus et al., , 2017Biro & Stamps, 2008;Diaz Pauli & Sih, 2017;Enberg et al., 2012;Lennox et al., 2017;Uusi-Heikkilä et al., 2017).
Although increased capture vulnerability has been documented for many commercially and recreationally valuable fish species, many non-target animals, including imperiled species such as sea turtles, are also often incidentally captured in fishing gear Pate & Marshall, 2020). It would be expected that a suite of traits similar to those documented in fishes would increase sea turtle capture vulnerability. Sea turtles often forage in neritic waters (Bolten, 2003). Increases in incidental sea turtle captures by hook-and-line anglers, particularly from fishing piers, have recently been reported . Studies of sea turtles captured in recreational fishing activities have focused almost exclusively on individuals once they are captured Rudloe & Rudloe, 2005;Seney, 2017) or after they have been released from subsequent rehabilitation . Other studies have shown that juvenile green turtles (Chelonia mydas) exhibit individual differences in boldness (Griffin et al., 2017;Kudo et al., 2021). Capture vulnerability may also be affected by variations in the morphology, life-history traits, and behavior of a species. Identifying such relationships may help managers design actions to reduce hook-and-line captures of protected species. For example, if a protected species co-occurs with fishing activities during a certain time of the year, fishing activities could potentially be restricted at those times to reduce unintentional bycatch.
Globally, all sea turtle species, except the olive ridley (Lepidochelys olivacea), are listed as threatened or endangered by the U.S. Endangered Species Act and CITES. Five species of sea turtles are found in the Gulf of Mexico (GOM), including the Kemp's ridley (Lepidochelys kempii), loggerhead (Caretta caretta), green turtle, leatherback (Dermochelys coriacea), and hawksbill (Eretmochelys imbricata) (Ward & Tunnell, 2017). The recovery plans for all of these species, except the leatherback (which is threatened by the pelagic longline fishery; Lewison et al., 2004), identify nearshore, recre-  Fish & Wildlife Service, 2008;Seminoff et al., 2015). Information on sea turtle bycatch and mortality in recreational fishing activities is minimal. We provide quantitative information on the vulnerability of sea turtles to recreational fishing, which can be used for developing management actions to reduce this source of mortality for sea turtle populations. The objectives of this project were to identify and compare characteristics that may affect capture vulnerability of sea turtles on a recreational fishing pier.

| ME THODS
We captured sea turtles along 21 km of Santa Rosa Island (SRI) owned by Eglin Air Force Base (AFB, Figure 1). The nearshore sediments in this area are predominately fine silica sand (Williams et al., 2012). The Navarre Beach (NAV) Marine Sanctuary, an artificial reef that consists of 78 structures constructed of piling-mounted concrete disks located 340 feet offshore of the mean high tide line, lies approximately 0.5 km west of the study site. A fishing pier (NAV pier) is located on the GOM coast of SRI ~ 1 km west of the NAV Marine Sanctuary. This is the longest fishing pier along Florida's GOM coast (Clark, 2010).
At the SRI study site, we captured turtles (hereafter net-caught turtles) between March and November 2014-2019. Capture and sampling occurred following methods described by Lamont and Johnson (2021). Briefly, we surveyed for turtles from all-terrain vehicles ridden on the beach. Once observed, we captured turtles using a modified set-net technique in nearshore waters typically <2 m deep and within 100 m of shore. We marked all turtles with an Inconel tag in each front flipper and a passive integrated transponder (PIT) tag in one front flipper. We measured straight carapace length measured notch to tip (SCL) using metal calipers and curved carapace length measured notch to tip (CCL) using a cloth tape measure. Additionally, we used metal calipers to determine the straight head width following Price et al. (2017) and then calculated a relative head width (hereafter head width) by dividing head width by SCL. We determined weight (kg) by placing the turtle in a harness and hanging the harness from a hand-held Pesola spring scale.
Data for turtles captured from the NAV pier (hereafter piercaught turtles) were collected by participants in the Florida Sea Turtle Stranding and Salvage Network using standardized protocols described by Foley et al. (2005). Pier-caught individuals were taken to a local rehabilitation facility where they remained for varying lengths of time, depending on health status. The SCL, CCL, and weight were determined for pier-caught turtles, but head width was not measured. Most individuals were released in nearby waters; however, some were relocated at distances >200 km.
We used SCL in all analyses. If SCL was not gathered for an individual, we converted CCL to SCL using the following regression equations from Teas (1993), where r 2 was >.95 for all modeled relationships: We calculated body condition index (BCI) as Fulton's K (BCI = body mass/SCL 3 × 10 4 ; Bjorndal et al., 2000;Lamont & Johnson, 2021). The correlation between SCL and BCI was low (r = −.14).

| Body size and condition
We used linear modeling to test for differences in SCL and BCI between pier-caught and net-caught sea turtles. We fit separate models with the same structure with SCL and BCI as the respective response variable Y. Each SCL and BCI observation was a measurement at the time of capture. Thus, turtles with more than one capture had multiple observations, each with unique measurements associated with either a net or pier capture. We could not directly account for individual fidelity in net or pier captures with SCL and BCL as Y because any lack of independence was at the observation, not turtle, level. Data plots supported a normal distribution assumption for BCI, but SCL was right-skewed. Thus, we used a log-normal distribution for SCL (see linear equations below). We treated species j as a factor with three levels (green turtle, Kemp's ridley, and loggerhead) using a means parameterization (Gelman & Hill, 2007;Kéry & Royle, 2016). Pier caught or net caught was treated as an indicator (dummy) variable (pier caught = 0, net caught = 1). We accounted for variation in SCL and BCI due to time of year (hereafter day) using a covariate (i.e., continuous predictor variable). Day was quantified as an integer that coincided with the calendar day of observation i ranging from March 23 (day 82) to November 29 (day 334). We standardized day to a mean of zero and standard deviation (SD) of one.
We also included a year k grouping factor (i.e., "random intercept", Gelman & Hill, 2007) to account for unexplained annual variation in SCL and BCI. Standard deviations for both SCL and BCI were F I G U R E 1 Locations where sea turtles were captured from 2014 to 2019 including the Navarre Beach fishing pier and surrounding waters off of Navarre Beach and Santa Rosa Island, Florida similar between pier-caught and net-caught turtles for each species ( Figure 2). Thus, we assumed equal variances for both models.

The linear equation can be written as:
where α is pier-caught, α 1 is net-caught, β 1 is the day slope, µ is mean, and σ 2 is variance. We report SCL and BCI relationships for each species as the mode (effect size) of the difference between pier-caught and net-caught turtles with a 90% highest density interval (HDI).
Differences were considered significant if the HDI did not overlap zero (Kruschke & Liddell, 2019). The complete linear model estimates are included in Table 1. We assessed model fit using a posterior predictive check of observed versus simulated Y values (Kéry & Royle, 2016).
In addition to a visual examination, we formally assessed fit using a Bayesian p-value. A Bayesian p-value .10-.90 supports adequate fit (Conn et al., 2018;Kéry & Royle, 2016).
We fit the models using the program JAGS (Plummer, 2003) called from the package jagsUI (Kellner, 2018) within the statistical software R (version 3.5.3; R Development Team, 2019). Posterior distributions for coefficients were estimated with Markov chain Monte Carlo (MCMC) methods using two chains of 15,000 iterations each after a 5000-iteration burn-in phase (no thinning). We assessed model convergence using the Brooks-Gelman-Rubin statistic (R, Gelman & Rubin, 1992). R < 1.1 for all parameters indicates adequate chain mixing (Kruschke, 2015). We also examined parameter trace plots to confirm good convergence.

| Head width and body condition
We used the same general model structure as the pier-caught and net-caught SCL and BCI comparisons to test for differences in head width and body condition between groups of green turtles with different numbers of pier captures. Because head width was only determined for turtles that were net-captured, we could only use pier-captured turtles that were initially captured by net. For turtles that were captured multiple times by net, we used the mean for head width. Loggerheads and Kemp's ridleys were not included in this analysis because head width was not measured on the majority of individuals. We fit identical models treating head width and BCI as the respective Y variable. A plot of head width supported a normal distribution assumption. A recapture factor comprised three levels associated with each net-caught green turtle: no pier captures (i.e., only net caught), one pier recapture of a net-caught individual, and multiple pier recaptures of net-caught individuals. Standard deviations for both head width and BCI were similar among the recapture levels ( Figure 3). We used multiple comparisons (Kruschke, 2015;Kruschke & Liddell, 2019) to test for differences in green turtle head width and BCI between the following groups of net-caught turtles: no pier captures versus one or multiple pier captures, no pier captures versus one pier capture, no pier captures versus multiple

| Pier-capture vulnerability
We examined green turtle pier-capture vulnerability in relation to SCL, BCI, and day. Day provided a surrogate for both general tendencies of being pier caught at certain times of year (e.g., increased boldness) and seasonal variation in fishing effort. We did not include loggerhead or Kemp's ridley in this analysis due to the small number of captures from the pier. Pier-capture vulnerability Ψ was treated as a Bernoulli process (Y = 0 if net caught, Y = 1 if pier caught) and modeled as a linear function of covariates. Straight carapace length was natural-log transformed due to a right-skewed distribution. All covariates were standardized to a mean of zero and SD of one. We also included an individual s grouping factor to account for pseudoreplication and unexplained individual variation associated with repeat measurements of individuals (Wagner et al., 2006). The linear model can be written as: We considered a covariate significant if the 90% HDI for the slope did not overlap zero. We fit the model using JAGS with MCMC settings as four chains of 50,000 iterations each after a 25,000-iteration burn-in phase (thinning = 10). We assessed model convergence using R and parameter trace plots. (5)

| Net-caught turtles
From October 2014 to October 2019, 148 captures of 102 unique individuals occurred in the nearshore waters at our study site (Figure 2).
Captures were attempted in every month from May to October.
Sampling effort was greatest in September (27% of sampling days), but most turtles were captured in October (44%; Figure 4). Green turtles were caught most frequently followed by Kemp's ridleys and loggerheads.
Twelve net-caught turtles were also caught (either before or after) on the pier. Our overall recapture rate (i.e., either net or pier  Kemp's ridleys were captured most frequently in April (33%) and
Mean weight ± SD for pier-caught turtles was 15.04 ± 12.89 kg

| Pier-caught versus net-caught size and body condition
Both SCL and BCI differed significantly between pier-caught and net-caught turtles for at least one species (Table 3)

| Pier recaptures head width and body condition
There were no significant differences in either head width or BCI among green turtles with different numbers of pier recaptures (

| Pier-capture vulnerability
Green turtle pier-capture vulnerability was significantly related to both size and time of year, with strong linear relationships (Table 5, Figure 5). Ψ increased significantly with increasing SCL and decreased significantly with increasing day. The HDI for the BCI slope overlapped zero; however, the posterior distribution supported a negative relationship with Ψ ( Figure 6). R was <1.05 for all model parameters, and trace plots confirmed adequate convergence.

| DISCUSS ION
Morphological and behavioral traits are commonly used to assess Kemp's ridleys are commonly documented as incidental captures on fishing piers Cook et al., 2020;Seney, 2017); however, during our study, green turtles dominated pier captures. Species composition of pier captures most likely Note: Coefficients for SCL and BCI reported as the mode (effect size) of the difference with a 90% highest density interval (HDI). The direction of the tests is left minus right. *Highlights HDIs that did not overlap zero. For example, the finding for "green turtle-pier versus net" is that mean SCL was greater for pier-caught green turtles than net caught, and the difference was significant. Mean BCI was greater for net-caught green turtles; there it was not significant difference from piercaught turtles.

TA B L E 3 Comparisons of straight carapace length (SCL) and body condition index (BCI) between pier-caught (pier)
and net-caught individuals (net) for green turtle, Kemp's ridley, and loggerhead (also see Table 1) reflects the composition and abundance of species using surrounding waters; green turtles are frequently captured in nearshore waters off of northwest Florida (Lamont & Johnson, 2021).
Additionally, although the genetic origin of juvenile green turtles captured on the pier is unknown, green turtle nesting has increased exponentially in Florida in recent years (Ceriani et al., 2019) while loggerhead nesting has remained stable (Ceriani et al., 2019) and Kemp's ridley nesting appears to have declined (Caillouet et al., 2018;Gallaway et al., 2016). Capture of Kemp's ridleys and loggerheads on recreational fishing piers may not be unexpected considering that the bait typically used by recreational anglers (e.g., fish, shrimp) are known diet items of these species (Molter et al., 2021;Ramirez et al., 2020;Shaver, 1991). However, capture of green turtles on the NAV pier was unexpected as juvenile green turtles in neritic habitats are generally considered herbivores (Williams et al., 2014). Most of the pier-caught green turtles were foulhooked (67%), which may suggest turtles were foraging on algae growing on the pier pilings rather than targeting bait. However, turtles may have also been foul-hooked while attempting to take bait. Probability of pier capture for green turtles increased with size, which may simply reflect a larger surface area for foul hooking to occur. The six pier-caught Kemp's ridleys were also larger than our net-caught individuals and were also larger than Kemp's ridleys captured from piers in Mississippi Sound (mean SCL ± SD: 36.0 ± 7.5 cm, Coleman et al., 2016).  et al., 2015). In our study, green turtles captured both from the pier and in the net were on average 14.5 cm larger when they were pier-captured (n = 11). It is difficult to directly examine boldness in juvenile sea turtles either in a laboratory  but see Kudo et al., 2021) or in the wild (Breck et al., 2019;Hertel et al., 2019). Griffin et al. (2017) were able to observe green turtle behavior in relation to snorkelers; however, in general, sea turtles move large distances, inhabit deep waters, and may remain submerged at depth for several hours. Fine-scale tracking of individuals has been used to assess boldness, primarily with terrestrial species (Breck et al., 2019;Hertel et al., 2019); however, satellite tags that transmit GPS-quality locations are not readily available in sizes small enough for juvenile sea turtles. In addition, tracking durations for small turtles are not long enough to assess behavior due to rapid carapace growth (Lamont & Iverson, 2018). Acoustic telemetry using a passive receiver array installed on and in the vicinity of an active fishing pier (such as NAV pier) would provide useful data, particularly when installed as part of a cooperative Note: Coefficients are reported on the logit scale as the mode (effect size) with a 90% highest density interval (HDI). The intercept is interpreted as estimated capture vulnerability at mean straight carapace length (SCL, 36 cm), body condition index (BCI, 1.29), and day of year (day, August 9). *Highlights covariate slope HDIs that did not overlap zero. Individual standard deviation (SD) represents unexplained individual variation in capture vulnerability.

F I G U R E 5
Posterior distribution of the body condition index (BCI) slope with a 90% highest density interval vertical lines from the green turtle pier-use model acoustic network such as iTAG (Friess et al., 2021;Hertel et al., 2019).
Body condition of pier-caught green turtles and loggerheads did not differ from net-caught turtles. After being caught on the pier, turtles go to a rehabilitation center where they are examined and treated by a veterinarian (sometimes for weeks or months) before being released, and this care may improve their BCI (Hughes et al., 2019 Head width is a function of bite force (Marshall et al., 2014), and as such, it is reasonable to suspect differences among individuals that forage in different habitats, as has been shown for nesting loggerheads (Price et al., 2017). For example, turtles foraging on algae along rock jetties may have narrower head widths that those foraging in seagrass habitat because seagrass blades are relatively tough and must be torn by the turtle (Marshall et al., 2014). Head width is easy to measure, and we suggest it might be beneficial to add to the suite of data collected from turtles captured from fishing piers.
In addition to body size, pier-capture vulnerability was also in- June (Rudloe & Rudloe, 2005;Seney, 2017). However, anglers on piers reported capturing most turtles between June and August in Mississippi . Seasonal variation in pier captures could also reflect movements of juvenile turtles into neritic, summer foraging areas from deep-water, wintering home ranges (Lamont & Iverson, 2018;Metz et al., 2020). This type of movement was not reflected in our net captures, however.
Although we identified characteristics that differed between

| Conservation implications
It has been suggested that mortality in recreational fishing activities is resulting in selection against specific morphological traits, such as body depth and mouth size in target fishes (Alós et al., 2014).
Mortality of sea turtles captured from fishing piers could be having a similar impact on sea turtle populations, albeit over a much longer time scale. Additionally, sublethal injuries and stress from capture and handling may have population-level impacts. In a 6-year period (2010-2015), more than 1000 sea turtles were captured by recreational anglers in Mississippi alone . Capture in recreational fishing activities may result in selection against bolder turtles and that personality trait could impact the overall population, not just pier-captured turtles, by increasing exposure to threats such as predators (Griffin et al., 2017) or altering a turtle's ability to adjust to changing temperatures (Clark et al., 2017;Pich et al., 2019). On the other hand, bolder turtles may be more exploratory and that could benefit range expansion in populations as they adapt to a changing climate (Osland et al., 2021). Populations that contain a mix of behavior types may be most resilient to anthropogenic and natural pressures that affect these species over time (Griffin et al., 2017;Schindler et al., 2010).

ACK N OWLED G M ENTS
We are grateful to Kathy

CO N FLI C T O F I NTE R E S T
The authors declare no conflicts of interest. Any use of trade, product, or firm names is for descriptive purposes only and does not imply endorsement by the U.S. government.

D I SCLOS U R E
Any use of trade, product, or firm names is for descriptive purposes only and does not imply endorsement by the U.S. government. This draft manuscript is distributed solely for purposes of scientific peer review. Its content is deliberative and pre-decisional, so it must not be disclosed or released by reviewers. Because the manuscript has not yet been approved for publication by the U.S. Geological Survey (USGS), it does not represent any official USGS finding or policy.

DATA AVA I L A B I L I T Y S TAT E M E N T
The datasets generated for this study will not be made publicly available. Restrictions apply to the datasets. Raw data are exempt from publication due to the sensitivity of endangered species information. Requests to access the datasets should be directed to the corresponding author. All other data used for analyses are presented in the manuscript.