Editor’s Choice: One of the more interesting issues in data collection citizen science is how to know whether a “zero counts” measurement really means nothing was there. This paper describes a Bayesian inference method to obtain occupancy probabilities providing a potential solution to this issue. — LFF —
Occupancy monitoring is particularly suitable for freshwater turtles because many species are relatively easy to detect due to their basking behavior. However, the probability of detecting turtles can be highly variable. There are sophisticated methods available for accounting for detection probability in occupancy monitoring, but standard sampling designs involve surveying all sites several times. Here I illustrate a method whereby an accessible reference site was repeatedly surveyed to obtain models of detection probability for three turtle species, and these models then applied to surveys of 23 water bodies in two nearby conservation reserves with similar habitat. The explanatory variables for detection probability included the date, time of day, and a set of environmental measurements designed to capture the key factors likely to affect basking. The estimated probability of detecting a species present at a water body ranged from 0.10–0.99 for Painted Turtles (Chrysemys picta), 0.03–0.84 for Blanding’s Turtles (Emydoidea blandingii), and 0.01–0.60 for Snapping Turtles (Chelydra serpentina). All species were unlikely to be detected in overcast conditions, but the other factors affecting detection varied among species. I used Bayesian inference to estimate the posterior (post-survey) occupancy probabilities for water bodies where a species was not detected, illustrating that the surveys gave strong evidence of absence in some cases but provided little information in others. I believe the method could be usefully applied to regional monitoring programs for some turtle species, as the field surveying does not require specialist equipment or training, so lends itself to citizen science.