Predictive Occurrence Models for Bat Species in California

The ability to accurately predict the potential occurrence of species of management concern is useful for wildlife managers, particularly for those whose management activities involve large areas where sampling is difficult due to logistical or financial constraints. During the summers of 2002 and 2003, we used mist nets to capture bats ( Myotis yumanensis, M. californicus, M. evotis, M. thysanodes, Eptesicus fuscus, Lasionycteris noctivagans, Tadarida brasiliensis, Antrozous pallidus, Lasiurus borealis , and Lasiurus cinereus ) in Whiskeytown National Recreation Area in north‐central California, USA. We used landscape‐scale variables, logistic regression, and Akaike's Information Criterion (AIC c ) to model species distributions and produce spatially discerning predictive occurrence maps. We developed a priori models that we used to determine which landscape‐scale variables best discriminated between capture sites and non‐capture sites. The odds of capturing a bat were 3.3 greater when total edge increased by 10,000 m, whereas for Yuma myotis ( Myotis yumanensis ), the odds of predicting presence were 0.2 greater when distance to lakes and ponds decreased by 2,000 m. Elevation was important in predicting the distribution of silver‐haired bats ( Lasionycteris noctivagans ) and big brown bats ( Eptesicus fuscus ). Increasing elevation by 400 m decreased the odds of capturing a silver‐haired bat by 0.1 and a big brown bat by 0.4. Classification accuracy for our models ranged from 80.9% for all bat species combined to 72.3% for Yuma myotis and silver‐haired bats. Predictive occurrence models can be valuable to bat conservation efforts because they provide spatial data important for evaluating the effects of management activities on species distributions.