Blanding's turtles (Emydoidea blandingii) have experienced widespread declines and are listed as threatened or endangered throughout most of their remaining range (Congdon et al. 2008). Recently, this species was petitioned for listing under the US Endangered Species Act and had a “substantial findings” result from the 90-day review (US Fish and Wildlife Service [USFWS] 2015). The next step in this review process will be for the USFWS to request status information from states where Blanding's turtles are known to occur. In Ohio, Blanding's turtles are listed as threatened and considered a Species of Greatest Conservation Need (Ohio Division of Wildlife [ODOW] 2015). The most recent conservation assessment identified 19 known, extant populations/metapopulations in Ohio. Of the remaining populations, 2 are thought to be declining, 2 are thought to be stable, 1 is thought to be increasing, and the rest are unknown (Midwest Partners in Amphibian and Reptile Conservation [MWPARC] 2010). In all known populations, there were insufficient data to determine viability (MWPARC 2010). Identifying and delineating priority conservation sites through surveys are crucial steps for determining Blanding's turtle status in the state (ODOW 2015).

As agencies are pressured to take conservation actions, more data are required to make informed decisions. Holistic conservation strategies for turtles, particularly Blanding's turtles, will include population demographics and regional distribution of populations and habitats (Mitchell and Klemens 2000; Congdon and Keinath 2006; Compton 2007; McGuire et al. 2013). Collecting meaningful, long-term data for long-lived species has a number of inherent difficulties (i.e., substantial time investments; Hof et al. 2017; Congdon et al. 2018) and presents challenges to conservation measures that require swift action.

Community and citizen science (also known as citizen science) can provide professional scientists with cost-effective data to augment research projects (Bonney et al. 2009; Dickinson et al. 2010; Gibbons 2019). Community and citizen science projects are increasingly being used to gather large quantities of data on sightings of wildlife while engaging local communities (Cohn 2008; Silver-town 2009). The main pitfall of using data from community and citizen science is that the quality of data is often unknown, stemming from issues such as inadequate training, changing protocols, lack of objectivity, and unfamiliarity with many effective and study-tested protocols, which will often result in unpublishable data (Dickinson et al. 2010; Conrad and Hilchey 2011; Hof et al. 2017). However, there are a myriad of benefits from using community and citizen science data produced by studies where research protocols are overseen by scientists (Cohn 2008; Bonney et al. 2009; Parsons et al. 2018). Benefits from such collaborations include, but are not limited to, increased scientific literacy, inclusion of the community in local conservation issues, low cost, and facilitation of otherwise under-resourced monitoring programs (Conrad and Hilchey 2011; Dickinson et al. 2012). Involving the local community can also facilitate management actions with added support and pressure for action (Danielsen et al. 2008).

One successful avenue of engaging community and citizen scientists has been soliciting sightings and photographs to identify individuals, determine abundance, and examine distributions (e.g., iNaturalist, https://www.inaturalist.org/; eBird, https://ebird.org/; HerpMapper, https://www.herpmapper.org/). Photos present a unique opportunity, as many animals possess distinctive, natural marking patterns that can be used to identify species or individuals within a population (Bolger et al. 2012). Photographic identification has been used in a number of studies on multiple species and taxa: polar bears (Anderson et al. 2010), otters (Gilkinson et al. 2007), cheetahs (Kelly 2001), dolphins and whales (Karczmarski and Cockcroft 1998; Schweder et al. 2010; Gomez-Salazar et al. 2011), sharks (Graham and Roberts 2007; Anderson et al. 2011; Hughes and Burghardt 2017), salamanders and newts (Gamble et al. 2008; Hoque et al. 2011), toads (Nordick et al. 2015), and turtles (Reisser et al. 2008; Cross et al. 2014).

Here we present a case study of community scientists collecting photographic mark–recapture data for Blanding's turtles that were used to generate preliminary population estimates for this species in an Ohio wetland.

METHODS

Study Site. — This study took place in a residential area near Lake Erie where a beachside community borders a 184-ha, privately owned, wetland complex (coordinates are omitted due to collection risk). A combination of seawalls, roads, and houses limit access to a 2.6-km stretch of beach with 0.8 km free of seawalls. Each year, female turtles leave the marsh and travel into the residential areas and beaches to find suitable nesting sites (T.B., pers. obs.).

Mark–Recapture. — From 2014 to 2019, community and citizen scientists in the beachside community followed the recommendations of Cross et al. (2014) and began taking pictures of Blanding's turtles encountered on their properties during nesting season. Early in this study, there was concern surrounding use of community and citizen scientists to collect data on a state threatened species, particularly when it came to biosecurity and illegal collection. We addressed the concern by “deputizing” a lead community/citizen scientist (coauthor T.B.) who was appropriately trained in data collection and biosecurity and was permitted to handle the turtles.

Upon capture, observers would contact coauthor T.B., who would meet up with the reporting individuals, record the location, and take a picture of the turtle's plastron. Blanding's turtle carapaces are patterned, but often covered with algae; therefore, it was determined plastron photos would be better for identifying individuals and require less handling (i.e., shell cleaning). In some instances, observers would send pictures and locality information to T.B. For the purpose of this project, we consider the plastron patterns to be semipermanent marks, similar to the carapace and plastron patterns of eastern box turtles (Cross et al. 2014). Plastron color may change, but it is unlikely plastron patterns of adult Blanding's turtles would change appreciably over a 5-yr study period (J. Congdon, University of Georgia Savannah River Ecology Laboratory, pers. comm., 2018; R. Nagle, Juanita College, pers. comm., 2018; O. Kinney, Darlington School, pers. comm., 2018).

Sampling generally began in late May and ended in late July, coinciding with the Blanding's turtle nesting season (Congdon et al. 2008; Ernst and Lovich 2009). Annual effort (number of photographs) varied, but generally increased each year, presumably as word spread and neighbors became interested (Fig. 1).

View Figure

• Download as Powerpoint
• Download Figure

Data Analyses. — To identify individual Blanding's turtles, we used the pattern-recognition software HotSpotter (http://cs.rpi.edu/hotspotter/; Crall et al. 2013). HotSpotter is a freely available, open-source wildlife identification software package and offers several improvements over other pattern-recognition programs (e.g., Wild-ID), such as faster processing time and increased matching accuracy (Crall et al. 2013). HotSpotter has been used to identify individual zebras, lionfish, giraffes (Crall et al. 2013), and Wyoming toads (Anaxyrus baxteri; Morrison et al. 2016). HotSpotter workflow consists of preprocessing, matching, image scoring, spatial ranking, and label scoring. Briefly, HotSpotter uses 2 algorithms when comparing photos: 1) one is almost identical to the scale invariant feature transformation (SIFT; Lowe 2004) operator and compares the photos within the database; 2) the other is a “one vs. many” algorithm that compares all pattern descriptions from individuals within the database. Image scoring is based on the combination of both algorithms, where higher values indicate a stronger match. HotSpotter also allows users to visually confirm matching scores; all image matches in this study were visually verified, independently, by two coauthors.

We used the POPAN formulation of the Jolly-Seber model run in Program MARK (White and Burnham 1999) to estimate the population size of adult female Blanding's turtles at our study site. Individual capture histories from photographic recapture data were used as model input. The POPAN model estimates apparent survival between sampling occasions (Φ), capture probability (p), probability of entry (P_ENT) into the study area, population size during each sampling occasion (N_t), and super-population size ( number of individuals available to enter the population; Schwarz and Arnason 1996). We interpreted Φ as the turtles that had survived between sampling years and had not permanently emigrated out of the study area. Likewise, P_ENT was the probability of individuals from the super-population entering the population each year. We constructed and evaluated biologically justifiable models that included constant (.) and time (t) varying iterations of Φ, p, P_ENT.

To ensure our data met the assumptions of the Jolly-Seber model, we first fitted and tested a fully time-dependent global model for overdispersion using the bootstrap goodness-of-fit test and by estimating median ĉ using logistic regression in MARK. The most parsimonious model was selected based on minimizing Akaike's information criterion corrected for small sample sizes (AIC_c) values and ranked according to ΔAIC_c (AIC_c(i₎ –AIC_c(min)). Further strength and evidence for support of each model was provided through AIC_c weights (Burnham and Anderson 2002).

No juveniles were reported during the image-collection portion of the study, so we assumed all observations were of female Blanding's turtles because all sightings were reported during nesting season and were associated with sandy beachfronts.

We used locality data and the program Geospatial Modeling Environment (Beyer 2015) to estimate straight-line distance between sightings for turtles with more than one observation. For turtles with multiple, within-season encounters, we considered the midpoint or centroid of the sightings as the observation location, thus giving a general approximation of the area where the turtle was observed that year.

RESULTS

Community and citizen scientists at our study site gathered 200 photos of Blanding's turtles, resulting in 155 useable plastron images. Excluded images included carapace, side profile, and blurry photos deemed too distorted for use. From these pictures, HotSpotter identified 65 individual female Blanding's turtles. Visual assessment of HotSpotter results indicated there were no mismatches (i.e., false positives or false negatives) when analyzing our photo data set. Individual turtles were not encountered every year and the number of times turtles were captured throughout the study ranged from 2 to 5. Within-year sightings were common (n = 30), with some individuals encountered up to five times in a season. Average distance between yearly individual observations was 117 m and ranged from 11 to 1080 m. Of the reported sightings, 83% came from a 5.5-ha residential area.

There was no significant evidence of overdispersion (bootstrap p = 0.34), and the estimated overdispersion parameter was close to 1.0 (median ĉ = 0.92, set to 1.0 without applying a correction factor; Cooch and White 2019). The constant survivorship and time-dependent recapture and entrance model () was the most parsimonious model for our study site, with an AIC_c weight of 0.76, suggesting time-dependent recapture probability and probability of entrance (Tables 1 and 2). Apparent survival rate (constant) was estimated to be 0.836 (95% confidence interval [CI] = 0.693–0.920; 0.057 standard error [SE]). Using the results from this model, we estimated the super-population size () of adult female Blanding's turtle population at our study site to be 87 individuals (95% CI = 74–116; 10.1 SE). The estimated number of Blanding's turtles at each sampling occasion ranged from 36 to 50 (Table 2).

Table 1. Model comparisons (MARK, POPAN model) for adult female Blanding's turtles in an Ohio wetland identified from community and citizen scientists photographs, including Akaike values corrected for small sample size (AIC_c), the difference in AIC_c between a given model and the best supported model (ΔAIC_c), model weight (w), model likelihood, and number of parameters (K) for candidate models.

View Fullscreen

Table 2. Estimated annual population size (n; with 95% CI) and SE from community science photo surveys for Blanding's turtles in an Ohio wetland from 2014 to 2019. Estimates for some years were confounded using the POPAN formulation or were poorly estimated under the model and were omitted (Schwarz and Arnason 2019). — = not applicable.

View Fullscreen

DISCUSSION

Our study used community and citizen science–generated data to identify a previously unreported population of Blanding's turtles in Ohio and to generate preliminary estimates of number of adult females in the population. The use of community and citizen scientists enabled us to collect useful data over many years, without a considerable expenditure of effort on the researchers' part.

Community and citizen scientist–collected photographic data are a beneficial addition to survey efforts, but relying on individuals with different camera types can generate images of varying quality and usefulness. The most common quality issues were blurry images and “artsy” pictures, where the turtle was posing in such a manner where the plastron was not visible. In general, the lowest matching scores were of pairs of images where one image was poor quality. Cross et al. (2014) recommended removing low-quality (e.g., blurry or distorted) images; however, we found HotSpotter worked well, even with low-quality images.

It is difficult to determine the sex of Blanding's turtles from photographs; however, all turtles in the images lacked the concave plastron typical of males, and sightings coincided with the nesting season for this species (Congdon et al. 2008; Ernst and Lovich 2009). Therefore, it is likely that most, if not all, individuals photographed are female. Female-skewed sex ratios are often the result of sampling bias (Evrard and Canfield 2000; Tesche and Hodges 2015), as is the case in this study where sex bias is an inherent risk due to females comprising the portion of the population most visible to homeowners.

Our population estimate of adult female Blanding's turtles was low compared with estimates from other studies (Herman et al. 1995; Pappas et al. 2000; Congdon et al. 2001; Rubin et al. 2004; Ruane et al. 2008; Hasler et al. 2015). However, direct comparisons between studies may not be appropriate because we used a sex-biased sample for our estimates. Based on known sex ratios, the estimate presented here is likely much lower than the true population size. It also is important to note community and citizen scientists in our study were only permitted access to ∼ 35% of the perimeter of the available wetland. Such limited access may have underrepresented the population and potentially contributed to the female-biased captures. Our estimate should therefore be considered preliminary, as different sexes and age classes were not represented in this study. Additionally, our study site is part of a much larger wetland complex where Blanding's turtles are known to occur throughout. These factors suggest the Blanding's turtle population at our site is likely larger than predicted.

Survivorship in this study was similar to findings in other studies (Congdon et al. 1993; Rubin et al. 2004; Hasler et al. 2015; Reid et al. 2016a, 2016b) and consistent with survival estimates of adult turtles (Iverson 1991). Despite high adult survival, these results still suggest several adult females die each year, as evidenced by the 1–2 Blanding's turtles reported dead annually on the roads surrounding our study site. Half of the wetland is bordered by residential areas and beaches, whereas the other half is bordered by roads and agricultural fields. It is feasible an entire subset of this female population nests in these agricultural fields and are exposed to mortality from roads.

Our estimated probability of recapture was lower than reported by Hasler et al. (2015), but similar to Reid et al. (2016b). These discrepancies with other work are likely due to the nature of our sampling and variation among studies (e.g., location, habitat quality, methods). Whereas most studies benefit from a rigorous sampling design, our project relied on community and citizen scientist surveys and mostly chance encounters; this method is likely to miss a number of nesting females each year. Similarly, our estimated probability of entering the population was low, but also expected given our definition (i.e., adults immigrating). This is consistent with the general conclusion that nesting site fidelity for Blanding's turtles appears to be variable with some, if not most, of the individuals returning to the same area in successive years while others make use of new areas (Congdon et al. 1983; Joyal et al. 2001; Beaudry et al. 2010; Congdon et al. 2011; McGuire et al. 2013).

Despite the limitations of using sex- and age-biased data for population estimates, this article highlights the utility of community and citizen science data for identifying priority management areas and generating preliminary abundance estimates. At our study site, community and citizen scientists documented the greatest number of female Blanding's turtles in one area within the state of Ohio (MWPARC 2010). The value of such information cannot be overstated, especially when dealing with rare and elusive species. Comparatively, at a nearby wildlife refuge along Lake Erie, Blanding's turtle trapping efforts (447 trap days) since 2013 have resulted in capture of 28 individuals (19 males, 9 females) with no recaptures (K.B., unpubl. data, 2018). The observations of hatchling turtles at our study site further contribute to the status assessment within the state by adding an additional location where recruitment has been documented in the past 10 yrs (MWPARC 2010).

Several community and citizen scientists named the turtles they repeatedly saw, adding an unanticipated level of personalization to their involvement. Participants in the project would take note when their “favorite turtles” were present or absent that survey year and ask questions about those individuals. The continued commitment and level of involvement by the participants in the annual surveys will help in early indications of population declines, should they occur. Furthermore, upon learning of the importance of the population they were monitoring, several of the community and citizen scientists expressed interest in pursuing mitigation actions, such as creating openings in the sea walls or artificial nesting habitat closer to the marsh. Having the support of the community will undoubtedly be beneficial when proposing management actions.

Efforts of carefully trained community and citizen scientists can provide the basis upon which to conduct in-depth trapping surveys to further elucidate population demographics at this and other sites throughout the state. Issues related to permanently marking individuals (i.e., permitting, interpretation of marks, and notch wear) and collecting morphometric data are easily rectified through increased training. Outfitting trained field technicians (i.e., students and community/citizen scientists coupled with a central repository for data on sightings and photographs) could be further replicated to provide much-needed, large-scale status assessments.