Method matters: Use of thermal-imaging drones to assess the assumptions of density estimation techniques

Techniques to estimate the density of unmarked animals are widely used by ecologists, but accurate estimates from these methods rely on assumptions about the study system. We conducted thermal-imaging drone surveys to test the validity of three assumptions for conducting distance sampling on white-tailed deer (Odocoileus virginianus) via nocturnal spotlight surveys in Iowa, USA. We found that the proportion of the population that occurred within forests that are unsamplable (i.e., availability bias) was negligible when vegetative green-up was sparse but increased to more than 50% as spring green-up progressed. The proportion of deer that were bedded, which are less detectable than standing or walking deer, depended on the day of year and time of night, suggesting these variables should be modeled on detection probability to reduce bias in parameter estimates. Lastly, we found evidence of road avoidance which influences how we analyze distance sampling data from road-based survey designs. Each of these deviations from the assumptions of conventional distance sampling informs future sampling design and analysis and will improve the accuracy of density estimates in our system. More generally, our study provides an example of how drone surveys can be conducted to improve density estimation techniques for a range of animal systems.