Accounting for heterogeneity in false‐positive detection rate in southeastern beach mouse habitat occupancy models

Abstract

Habitat occupancy models, designed to deal with non‐detection of a target species in occupied sites, have been expanded to allow for false‐positive detections when species are mistakenly detected in unoccupied sites. When a subset of the data are unambiguous detections, such occupancy models can produce reliable results. However, if not properly accounted for, heterogeneity in the rate of false‐positive detection between sites may bias estimates of habitat occupancy. We studied habitat occupancy of the southeastern beach mouse (Peromyscus polionotus niveiventris), a species along Florida's Atlantic coast that is threatened due to the reduction of its core range by over 79% since 1950. Southeastern beach mice were detected based on their footprints in track tubes, but because of co‐occurrence of the cotton mouse (Peromyscus gossypinus) with overlapping footprint size, we had to allow for possible false‐positive detections. Differences in the relative abundance of these two species between habitats may have resulted in heterogeneity in the false‐positive rate. By combining uncertain (track tube) and certain (live capture) detection methods, we were able to use covariates to account for heterogeneity in false‐positive detection rates between habitats. Southeastern beach mouse habitat occupancy was higher in coastal dune and strand habitat than in coastal scrub or interior scrub habitats. The rate of false‐positive detections was much higher in coastal scrub habitat than in coastal dune and strand habitat or interior scrub habitat, reinforcing the need for methods that accommodate heterogeneity in the rate of false positives to reduce bias in estimates of habitat occupancy. Using simulations, we show that heterogeneity in the false‐positive detection rate leads to bias in habitat occupancy estimates unless properly accounted for. We also discuss how differences in relative abundance between the target species and a confusing species can directly impact the heterogeneity of false‐positive detections.

Hydrology is a major influence on amphibian abundance in a large European floodplain

“…The best‐supported model had the lowest WAIC value (Burnham & Anderson, 2002 ). Bayesian p ‐values were used to assess model fit by calculating the Freeman–Tukey fit statistic (see Stolen et al, 2019 ). Values close to 0.5 indicate acceptable model fit whereas p <0.1 indicates a potential lack‐of‐fit (Conn et al, 2018 ; Gelman, 2013 ).…”

“…The best-supported model had the lowest WAIC value (Burnham & Anderson, 2002). Bayesian p-values were used to assess model fit by calculating the Freeman-Tukey fit statistic (see Stolen et al, 2019).…”

Implications of tree expansion in shrubland ecosystems for two generalist avian predators

“…We assessed model convergence using r values generated by JAGS [55] and model convergence using visual inspections of traceplots [56]. We compared the predictive power of each model in the set using leave-one-out cross-validation [57,58] and assessed model fit using the Freeman-Tukey test statistic to generate a Bayesian p-value [59]. Models with p-values close to 0.5 adequately fit the data, while extreme values close to 0 or 1 indicate a lack of model fit due to over-or under-dispersion.…”