Experimental assessment of large mammal population estimates from airborne thermal videography

Wildlife resource management requires reliable, fast, and affordable methods of surveying wildlife populations to develop and adaptively adjust policies. Many methods are used to survey populations of large mammals, and thermal video from drones can yield high rates of detection over large extents with relative speed and safety. In wild populations it can be difficult to estimate detection rates and the accuracy of resulting abundance estimates, especially because many factors of study design and natural systems can influence counts from thermal surveillance. We surveyed a captive white-tailed deer (Odocoileus virginianus) population of known size at the 174-ha Auburn Deer Facility in Auburn, AL, USA, and used a designed experiment to investigate the effects of observers, time-of-day, and day-to-day variation on the accuracy of abundance estimates from trial flights using drone-based airborne thermal videography. The experiment consisted of 20 full-census trials, occurring over 4 days, at times-of-day before sunset, after sunset, near midnight, before sunrise, and after sunrise, with each trial video counted by three trained observers. Counts across all trials and observers yielded a mean point estimate of 77 (95% CI 71-83) deer representing 81-97% of the known population range. Flights near sunset yielded estimates of the highest accuracy, within or close to the true range of the population. Variability in estimates was primarily influenced by daily climatic conditions and time-of-day, with only minor observer effects. Results highlight the importance of day-to-day variability in environmental conditions and diel processes, such as nocturnal cooling of the environment and crepuscular peaks in deer activity. The estimates from our experiment and resulting model illustrate that censuses conducted at optimal times-of-day and averaged across multiple days were able to achieve accurate estimates within the known range without correction, and that by averaging across flights, even including suboptimal conditions, thermal videography returned estimates within 4% of the known range with substantial overlap of the estimate. Conversely, single estimates chosen at random from any time-of-day varied widely around the known value, highlighting the need to plan flights with respect to diurnal activity patterns and landscape thermal properties, or, in the absence of information, average multiple surveys across times and days.