Stationary Point Acoustic Surveys of Bats on Conservation Lands

Neece, Ben; Gates, J.

Abstract. The North American Bat Monitoring Program (NABat) has provided a framework for regional biologists to conduct large scale bat acoustic monitoring. Acoustic surveys of this scale often produce large datasets that can assist with conservation and management decisions, but the calls must first be processed and classified to species. Automated classification software is commonly used, but can produce misclassifications. Thus, our objective was to examine different methods of verifying automated classification results to provide useful species presence data. In summer 2018, we conducted stationary point acoustic bat surveys on conservation lands at 158 points within 57 NABat grid cells across Maryland. We used SonoBat 4.2.2 to classify calls to species. To determine species presence, we used three approaches: 1) accepted species classifications, 2) the probability of presence of each species from maximum-likelihood estimates (MLEs), and 3) manually vetted classifications. We then examined differences in the number of cells in which each species was determined to be present by each approach. We collected 178,117 recordings of potential bat calls, 93,983 of which were accepted species classifications in SonoBat. However, even with the lowest MLE confidence level (50%), four WNS-susceptible species and one other species were determined to be present in fewer than 25% of the cells in which their calls were identified. Ten species were present in a greater number of cells based on manual vetting than MLEs. We believe it is necessary to verify automated species classifications. Due to different levels of effort and variation in species presence among the three approaches we tested, it may be best to use a combination such as MLEs with 95% confidence and manual vetting, or false positive occupancy models, to inform conservation and management decisions.