Camera-trap studies in the wild record true-positive data, but data loss from false-negatives (i.e. an animal is present but not recorded) is likely to vary and widely impact data quality. Detection probability is defined as the probability of recording an animal if present in the study area. We propose a framework of sequential processes within detectiona pass, trigger, image registration, and images being of sufficient quality. Using closed-circuit television (CCTV) combined with camera-trap arrays we quantified variation in, and drivers of, these processes for three medium-sized mammal species. We also compared trigger success of wet and dry otter Lutra lutra, as an example of a semiaquatic species. Data loss from failed trigger, failed registration and poor capture quality varied between species, camera-trap model and settings, and were affected by different environmental and animal variables. Distance had a negative effect on trigger probability and a positive effect on registration probability. Faster animals had both reduced trigger and registration probabilities. Close passes (1 m) frequently did not generate triggers, resulting in over 20% data loss for all species. Our results, linked to the framework describing processes, can inform study design to minimize or account for data loss during analysis and interpretation.

No abstract

Unmanned aerial vehicles (UAVs) are increasingly used in wildlife surveying, including estimation of population densities. It is essential that we evaluate and test new survey methods to guide optimal sampling strategies. This study aimed to assess the accuracy of using a UAV-mounted thermal infrared (TIR) camera to count red deer Cervus elaphus populations, and how this was influenced by flight season, height and velocity, in order to help guide future census design. We flew 57 flights across a captive population of red deer in a 13 ha deer park enclosure of semi-natural habitat, representative of the species' range in northern Germany. Flights and image assessments were performed with no prior knowledge of actual population size. Accuracy was quantified by comparing real population size (known only to deer park staff) and independently estimated population sizes from UAV TIR images. Accuracy was significantly influenced by ecological season (early and late winter, spring and early summer) and height. Across all seasons, lower flights (100 m) performed better than higher ones (120 m), with lower flights in early winter and early summer being on average accurate to within 1% of actual population counts. For the season where we had the largest range of temperatures between flights (late winter) we found that accuracy was highest when temperatures were lowest. Flights were also able to identify all five stags (defined as a male deer ≥ 2 years old) present in early summer, but not in spring. Deer appeared to avoid the landing/take-off area, but there were no noted behavioural responses to drones flying over animals when at constant height and velocity during surveys. Our results indicate that UAV-mounted TIR camera have the potential to accurately count populations of large ungulate species, but that flight season, height and potentially temperature need to be taken into account to maximise accuracy. This approach has the potential to be scaled up to more accurately estimate densities of wild populations compared to existing approaches.

No abstract

No abstract