Comparison of Two Methods to Detect the Northwestern Pond Turtle (Actinemys marmorata) and the Invasive American Bullfrog (Lithobates catesbeianus) in Interior Northern California
Abstract
Knowledge about the distributions of species and the variables influencing their occurrence is important for their management and conservation, but factors affecting occurrence can vary across the range of a species. Northwestern pond turtles (Actinemys marmorata) are widespread generalist turtles, but are nonetheless of conservation concern throughout their range. To better understand the distribution of northwestern pond turtles and introduced American bullfrogs (Lithobates catesbeianus), we surveyed streams on private timberlands of the interior foothills of northern California using visual encounter surveys and collecting samples of environmental DNA. We found that northwestern pond turtle occurrence was negatively related to elevation in our sampling frame. Detection probabilities with environmental DNA were approximately twice those of visual encounter surveys, but both methods were effective for detecting turtles in streams. American bullfrogs were detected in a single sample at each of 2 sites (one by environmental DNA, one by visual encounter surveys). Management for northwestern pond turtles in forest streams within our sample area will likely have the largest effect at lower elevation sites where turtles are most likely to occur.
Understanding the distribution of species is essential for their conservation and management. In landscapes dominated by human land uses, knowledge of species distributions and their probability of occurrence can help identify areas for species conservation measures (De Wan et al. 2009; McFarland et al. 2012). Moreover, knowledge about the variables related to species occurrence can guide further survey effort and habitat restoration as needed (Heard et al. 2013; Einoder et al. 2018). For widespread species, site and landscape characteristics affecting their occurrence can vary throughout their range (Barria et al. 2020; Chardon et al. 2020).
Northwestern pond turtles (Actinemys marmorata) are distributed from Washington to south-central California, mostly west of the Sierra–Cascade crest from sea level to 2000 m elevation (Bury et al. 2012c, 2012d). They are considered habitat generalists and occur in a wide variety of aquatic systems in diverse terrestrial settings (Germano 2010; Bury et al. 2012c; Horn and Gervais 2018; Agha et al. 2020; Fulton et al. 2022). Despite their generalist habits and broad distribution, northwestern pond turtles are considered a species of conservation concern throughout their range (Bury et al. 2012d) and have recently been proposed for listing as Threatened under the US Endangered Species Act (US Fish and Wildlife Service 2023). Primary threats to northwestern pond turtles include habitat loss and degradation and introduced species such as the American bullfrog (Lithobates catesbeianus). Natural disasters, contaminants, and disease are additional threats (Bury et al. 2012d; Manzo et al. 2021).
Although occurrence of northwestern pond turtles has been studied in portions of their range, knowledge about their distribution is limited in many places, including private timberlands. In the Umpqua River Basin, Oregon, northwestern pond turtle abundance and occurrence were found to be affected by different processes in lentic and lotic systems (Horn and Gervais 2018). In rivers, abundance was negatively related to surrounding pond area, but positively related to the amount of nearby wetlands and distance to the nearest pond (Horn and Gervais 2018). In ponds, however, occurrence was positively related to solar radiation (Horn and Gervais 2018). In the highly agricultural Sacramento Valley, California, northwestern pond turtle occurrence was associated with wider canals farther from urban areas that had more frogs (predominantly American bullfrogs; Fulton et al. 2022). The distribution and variables affecting occurrence of northwestern pond turtles in the interior foothills of northern California, where much land is privately owned and used for timber production, is less well known.
To better understand the distribution and occurrence of northwestern pond turtles on private forests, we surveyed for northwestern pond turtles in timberland streams of the interior foothills surrounding the northern portion of the Sacramento Valley, California. We combined 2 different survey techniques, collecting samples for environmental DNA (eDNA) analysis and conducting visual encounter surveys (VES), to account for imperfect detection and estimate detection probabilities of northwestern pond turtles. As a secondary objective, we concurrently surveyed for nonnative American bullfrogs by both methods to evaluate the potential threat these invasive frogs might pose to northwestern pond turtles in our study area. We quantified how combining these 2 survey techniques improved detection probability compared to either method used in isolation.
METHODS
Field Methods. —
We studied the distribution of northwestern pond turtles in Lassen, Plumas, Shasta, Tehama, and Trinity counties in interior northern California (Fig. 1). Our study area was selected as an area of interest by project partners including the US Fish and Wildlife Service, the National Alliance of Forest Owners (NAFO) Wildlife Conservation Initiative (https://nafoalliance.org/issues/wildlife-conservation/), and the National Council for Air and Stream Improvement, Inc. (https://www.ncasi.org/). Within the study area, our sampling frame was defined by several constraints. First, all sampled areas were timberlands owned or managed by NAFO members. Second, to ensure sampling of a reasonable number of sites within time and budget constraints, we defined study sites as intersections of streams and roads or trails on private timberlands below 1500 m elevation. Sites extended from the intersection upstream for 500 m or until the first barrier impassable to technicians, whichever came first. Because stakeholders were interested in persistence near historical records, we used multiple-frame sampling (Hankin et al. 2019) to include equal representation of a list frame of sites < 10 km from California Natural Diversity Database (CNDDB; California Department of Fish and Wildlife 2021) or Global Biodiversity Information Facility (2021) records of northwestern pond turtle occurrence and a random frame of sites > 10 km from CNDDB northwestern pond turtle records. To minimize sampling of ephemeral or very small streams unlikely to support northwestern pond turtles, we limited our sample to only include streams with perennial flow as indicated by the National Hydrography Dataset (US Geological Survey 2019). After applying these constraints and filters, we selected sites using generalized random tessellation-stratified sampling to result in a spatially balanced sample of available sites (Stevens and Olsen 2004; Hankin et al. 2019). We selected 30 sites plus an oversample of an additional 30 sites to a priori accommodate for sites that were dry (our study occurred during historic drought) or inaccessible in the field. If a selected site was inaccessible or dry, we attempted to sample nearby sites from the oversample list.
Citation: Chelonian Conservation and Biology: Celebrating 25 Years as the World’s Turtle and Tortoise Journal 23, 1; 10.2744/CCB-1591.1
We surveyed sites from June to September in 2021 and June to July in 2022. At each visited site that had water, we concurrently conducted VES and eDNA sampling. We expected the 2 survey methods to be complementary in their abilities to detect northwestern pond turtles, with VES being more effective for basking turtles and eDNA working better for turtles (or more precisely, turtle DNA) in the water. To conduct surveys, a crew of 2 biological technicians slowly waded upstream, with 1 technician scanning likely basking locations and the water for turtles and frogs, while the other technician filtered water for eDNA. Creeks are difficult to survey for western pond turtles (Bury et al. 2012b), and VES as we conducted them have been used infrequently for western pond turtle surveys. We chose to use VES for several reasons: 1) our primary interest was in the occurrence of northwestern pond turtles inhabiting streams in working timberlands, so our sampling frame necessarily included small creeks; 2) many stream segments did not have obvious vantage points from which to scan for turtles prior to conducting surveys (e.g., Bury et al. 2012b; Horn and Gervais 2018); 3) snorkeling was not practical in many streams, especially in combination with the continuous eDNA sampling, and funding and time constraints did not allow the effort required for proper training and safety of field crews (Bury et al. 2012a); and 4) trapping was deemed infeasible, given stream characteristics (often shallow or fast-moving with no pools in the sampled reach) and funding and time constraints that made detecting turtles in a single or few visits paramount. Importantly, despite the reported difficulty detecting western pond turtles in creeks and the use of an infrequently used survey method, we quantified detection probabilities of northwestern pond turtles by this technique so that researchers and managers can quantitatively compare this method of detection to other methods in similar habitats.
To conduct eDNA sampling, we filtered water through a 5.0-μm self-preserving eDNA filter using a backpack eDNA sampler (Smith-Root, Inc). Both technicians were careful to always remain downstream of the intake of the eDNA sampler. We changed filters when the unit indicated clogging and the transect continued with a new filter from that point. Field blanks were taken using distilled water at least once per day.
Before, during, and after surveys, technicians recorded variables expected to affect the occurrence or detection of northwestern pond turtles or American bullfrogs. In particular, we visually estimated reach width (meters), mean and maximum depth (meters), substrate (categorical: bedrock, boulder, cobble, gravel, sand, silt or organic), flow rate (ordered categorical: still, slow, medium, fast), and water clarity (1 = perfectly clear; 5 = completely opaque), and measured water temperature (°C) in the sampled reach. We also recorded weather conditions, such as air temperature (°C), cloud cover (clear, partly cloudy, mostly cloudy, overcast), and wind speed (calm, light, moderate), that likely affect northwestern pond turtle basking behavior. A visual estimate of shading caused by canopy cover (percentage) in the sampled reach also was recorded. For eDNA samples, we recorded the volume of water filtered and number of filters used in each sampled reach. Data generated during this study, including metadata with additional detail about data collection and data definitions, are available as a US Geological Survey data release (Halstead et al. 2024a).
Laboratory Methods. —
We extracted eDNA from filters in a restricted-access lab dedicated to eDNA sample handling using the Qiashredder + DNeasy Blood & Tissue Kit (Qiagen©) protocol from Goldberg et al. (2011). We then analyzed samples using a species-specific quantitative PCR (qPCR) assay designed and validated for northwestern pond turtles (Kamoroff et al. 2023). We created an extraction blank with each set of extractions and a qPCR negative control with every plate of qPCR reactions. A standard curve was included on each plate consisting of 4 levels of 10-fold dilutions in duplicate, a tissue sample diluted from 10−3 to 10−6 for the northwestern pond turtle and synthetic DNA (gBlock™, Integrated DNA Technologies™) from 10,000 to 10 copies for the American bullfrog. Each reaction included an internal positive control (IPC, Thermo Fisher Scientific™) to detect reaction inhibition. We considered a ≥ 3-Cq shift in the IPC to indicate inhibition. Inhibited samples were cleaned using a OneStep PCR Inhibitor Removal Kit (Zymo Research). Samples still testing as inhibited were diluted 1:10 and reanalyzed; any sample still testing as inhibited was diluted 1:100 and reanalyzed. We analyzed each sample in triplicate and reanalyzed samples if there were any inconsistencies in the results. Final determination of a positive for inconsistent samples was made if ≥ 0.33 of samples tested positive across plates.
Analytical Methods. —
We analyzed all data using single-season occupancy models (MacKenzie et al. 2002, 2006; Tyre et al. 2003; Kéry and Schaub 2012; Kéry and Royle 2016) that correct estimates of the probability of occurrence for false absences, or failing to detect the target species when they are present. We assumed that occupancy was constant across the 2 yrs (i.e., if turtles occurred at a site in 1 yr, they occurred at that site in both years), and used each VES and eDNA sample as an independent survey to estimate detection probabilities for each survey method and to estimate occupancy probabilities (Rose et al. 2019). We used indicator variables (Kuo and Mallick 1998; Hooten and Hobbs 2015) with hierarchical shrinkage priors (Kruschke 2015) for model and variable selection. Our full occupancy model evaluated evidence for the effects of proximity to historical records (binary to reflect the sampling process), elevation, mean stream width, maximum stream depth, and percent of stream shaded as follows:
where ψi is the probability of occurrence at site i, β0 is the logit-scale intercept, ωk is an indicator variable for the inclusion of explanatory variable xk on ψ, and βk is a coefficient for the effect of explanatory variable xk on ψ; zi is an indicator variable for the occurrence status of site i, where zi = 1 for occupied sites and 0 for unoccupied sites.
Our full detection model, conditional on occurrence, included effects of survey method (VES or eDNA; fixed to always be included in the model), temperature (air temperature for VES; water temperature for eDNA), and day of year as follows:
where pi,j is the probability of detection at site i during survey j, αm is the intercept for survey method m (i.e., eDNA or VES), υk is an indicator variable for the inclusion of explanatory variable xk on p, and αm,k is a coefficient for the effect of explanatory variable xk on p for each detection method; yi,j is the indicator for observed detections at site i during survey j, where yi,j = 1 for detections and 0 for nondetections.
We analyzed the occupancy model using Bayesian methods (Kéry and Schaub 2012; Kéry and Royle 2016). All priors for the model were selected to be vague (Table 1). We analyzed the model using Nimble (de Valpine et al. 2017) in R version 4.1.3 (R Core Team 2022). We ran the model on 5 independent chains of 20,000 iterations each after a burn-in period of 10,000 iterations for 100,000 retained posterior samples. We assessed convergence with the R-hat diagnostic (Brooks and Gelman 1998) and visual examination of history plots, which appeared well mixed; all R-hat = 1.00 and the minimum effective sample size across all monitored parameters was 4893. The model code to reproduce analyses is available on GitLab (Halstead et al. 2024b). Unless otherwise indicated, we report results as model-averaged median (95% equal-tailed interval [ETI]).
RESULTS
We detected northwestern pond turtles 13 times at 6 of 39 surveyed sites across 133 surveys (68 VES; 65 eDNA; Table 2; Fig. 1). Twenty-five sites were surveyed twice by each method, 11 sites were surveyed once by each method, 1 site was surveyed 3 times by each method, 1 site was sampled for eDNA once and VES 3 times, and 1 site was surveyed once by VES. Of the 6 sites at which northwestern pond turtles were detected, 4 had detections from both methods, 2 were eDNA detections only, and none had VES detections only (Fig. 1). All field and lab blanks tested negative. Reaction efficiencies ranged from 86% to 108% and r2 values were > 0.98. We detected inhibition in 4 of the samples, all from 2022.
The occurrence model with the greatest posterior support included an effect of elevation, but no other explanatory variables (Table 3). The posterior probability of inclusion of elevation was 0.672; all other explanatory variables for occurrence had lower posterior than prior probabilities of inclusion (Table 1). At the mean surveyed elevation (910 m), the probability of occurrence of northwestern pond turtles was median = 0.16 (95% ETI = 0.05–0.33; Table 1); for every 300-m (1 SD) decrease in elevation, northwestern pond turtles were 1.26 (0.94–7.28) times more likely to occur (Fig. 2). Of the 39 surveyed sites, we estimate that 6 (6–9) were occupied by northwestern pond turtles (Table 1).
Citation: Chelonian Conservation and Biology: Celebrating 25 Years as the World’s Turtle and Tortoise Journal 23, 1; 10.2744/CCB-1591.1
Detection probabilities of northwestern pond turtles differed by method (eDNA vs. VES), with a model including only an effect of method on detection receiving the greatest posterior support (Table 4). The posterior inclusion probabilities for temperature and day of year were lower than their prior inclusion probabilities (Table 1). The mean probability of detection for eDNA was 0.72 (0.34–0.93), which was 2.38 (0.24–18.7) times higher than the mean probability of detection for VES (0.52 [0.22–0.86]; Table 1; Fig. 3). The probability that eDNA detection probability was greater than VES detection probability was 0.79, with a difference of 0.19 (−0.29 to 0.57) on the probability scale. The cumulative probability of detection at a single site visit that incorporated both VES and eDNA sampling was 0.88 (0.60–0.98). Sites visited twice and sampled by both methods on each visit had a cumulative detection probability of 0.98 (0.84–>0.99).
Citation: Chelonian Conservation and Biology: Celebrating 25 Years as the World’s Turtle and Tortoise Journal 23, 1; 10.2744/CCB-1591.1
We detected American bullfrogs twice, once at each of 2 sites. One detection of American bullfrogs was during VES and the other detection was from an eDNA sample.
DISCUSSION
In timberland streams of interior northern California, northwestern pond turtles were more likely to occur at lower elevations. This result is perhaps unsurprising because our sampling frame approached the upper elevational limit for northwestern pond turtles (Ashton et al. 2012). Whether the relationship between elevation and occurrence of northwestern pond turtles was strictly caused by elevation or by environmental correlates of elevation is unknown. For example, rivers tend to be larger at lower elevations, and in the Sacramento Valley of California, northwestern pond turtle occurrence was positively related to the width of irrigation canals (Fulton et al. 2022). We did not find evidence that reach width was important for explaining variation in our model set compared with elevation, however. Active seasons are also longer at lower elevations and might have been prohibitively short at some of our highest elevation sites. Of course, these and other potential mechanisms do not act in isolation and several different mechanisms could contribute to the negative relationship between elevation and northwestern pond turtle occurrence in our study area.
The lack of evidence for other variables affecting northwestern pond turtle occurrence in our study is either because the relatively small number of sites reduced statistical power to detect effects of these variables or because the response of northwestern pond turtles to environmental variables is context-specific. For example, in the proximate Sacramento Valley, northwestern pond turtle occurrence was positively related to distance to urbanized area, frog capture rates, and canal width (Fulton et al. 2022). Elevation was not considered in models of occurrence by Fulton et al. (2022) because it varied little among sites on the valley floor. Similarly, counts of northwestern pond turtles in streams of the Umpqua River basin in Oregon were positively related to distance from pond and wetland area and negatively related to pond area (Horn and Gervais 2018), suggesting that the availability of aquatic systems in the surrounding landscape affects counts of northwestern pond turtles at focal stream reaches.
Detection probabilities for northwestern pond turtles were about twice as high with eDNA as with VES, but estimates of detection probability for both methods were imprecise. VES for northwestern pond turtles in small creeks are difficult because of limited visibility and difficulty observing turtles prior to disturbing them (Bury et al. 2012b). Nonetheless, by incorporating these 2 survey methods, we achieved high cumulative detection probabilities with a single visit to a stream reach. Combining eDNA and VES is therefore an efficient method of sampling small streams for occurrence of northwestern pond turtles. The detection probability for a single site visit achieved with eDNA (0.72 [0.34–0.93]) and VES (0.52 [0.22–0.86]), both individually and in combination, was higher than daily detection probability using 2 baited hoop nets in Sacramento Valley irrigation and drainage canals (0.12 [0.09–0.17]; Fulton et al. 2022) and higher than visual surveys for northwestern pond turtles at streams in the Umpqua River basin (approximately 0.20 [0.10–0.35] based on fig. 5 in Horn and Gervais 2018, although these figures appear to be individual detection probabilities from an N-mixture model), but lower than visual surveys at ponds there (approximately 0.70 [0.60–0.80] based on Fig. 5 in Horn and Gervais 2018). Given the high proportion of time northwestern pond turtles can spend in and underwater in ephemeral and perennial streams (Ruso et al. 2017), incorporating snorkeling into VES might increase detection probabilities if surveys must be accomplished in a single visit, but will likely increase the time required to conduct surveys and require more extensive training of personnel (Bury et al. 2012a). Estimating detection probabilities of different survey techniques under varied conditions would improve the ability of managers and researchers to choose efficient survey methods and interpret the results of surveys that fail to detect turtles.
American bullfrogs were found only during 1 survey each at 2 sites. Because American bullfrogs are introduced and widespread in areas downstream of our study area, we suspect that our rare detections might indicate isolated individuals or low-abundance, nascent populations near the edge of the introduced range of the species in northern California. Detection probability is often correlated with abundance (Royle and Nichols 2003) and surveillance as part of an early detection–rapid response (EDRR) effort for American bullfrogs might be warranted in the region given the negative effects of this invasive species on pond turtles elsewhere (Nicholson et al. 2020).
In interior foothill streams of northern California, northwestern pond turtle occurrence was negatively related to elevation. Conservation actions to increase the likelihood of persistence or abundance of northwestern pond turtles might best be implemented at lower elevations in this landscape; surveillance and EDRR for invasive species like American bullfrogs might be particularly useful at sites occupied by northwestern pond turtles. Incorporating eDNA and VES as complementary detection methods was successful for increasing overall detection probabilities by providing 2 means of detecting northwestern pond turtles during a single site visit. Improving knowledge about where northwestern pond turtles occur and identifying variables related to their occurrence and detection will likely increase efficiency of management actions for conserving these turtles.
Location and detection status of sites sampled for northwestern pond turtles (NWPT; Actinemys marmorata) in private timberland streams in the foothills of interior northern California, 2021–2022. Note that no sites had visual detections only.
Effect of elevation on probability of occurrence (ψ) of northwestern pond turtles (Actinemys marmorata) in private timberland streams in the foothills of interior northern California, 2021–2022. The bold line represents the posterior median; shading represents the posterior probability density in 0.05 quantile bands, with the outermost band representing the 95% equal-tailed interval.
Posterior distributions of detection probabilities for northwestern pond turtles (Actinemys marmorata) using environmental DNA (eDNA) and visual encounter surveys (VES) in private timberland streams in the foothills of interior northern California, 2021–2022. Points represent posterior medians, vertical lines represent 95% equal-tailed intervals, and violin plots represent full posterior distributions.
Contributor Notes
Handling Editor: Peter V. Lindeman
