How many sequences should I track when applying the random encounter model to camera trap data?

Abstract

The random encounter model (REM) is a camera trapping method to estimate population density (i.e. number of individuals per unit area) without the need for individual recognition. The REM can be applied considering camera trap data only by tracking the passages of animals in front of the camera (i.e. sequences). However, it has not been assessed how the number of sequences tracked (i.e. trajectory of the animal reconstructed) influences the REM estimates. In this context, we aimed to gain further insights into the relationship between the number of sequences tracked and reliability in REM estimates to optimize its applicability. We monitored multiple species using camera traps, and we applied REM to estimate population density. We considered red fox Vulpes vulpes, roe deer Capreolus capreolus, fallow deer Dama dama, red deer Cervus elaphus and wild boar Sus scrofa as model species. We tracked from a minimum of 154 (red fox) to a maximum of 527 (red deer) sequences per species, and we then sampled the dataset to simulate different scenarios in which a lower number of sequences were tracked (20, 40, 80 and 160). We also assessed the effect of adjusting the survey period to the minimum necessary to record the desired number of sequences. Our results suggest that tracking around 100 sequences returns a precision level equivalent to the one obtained by tracking a considerably higher number of sequences and reduced and optimized the human effort necessary to apply REM. Tracking less than 40 sequences could result in low precise density estimates. Our results also highlighted the relevance of considering study periods of ca. 2 months to increase the number of sequences recorded and tracking a random sample of them. Our results contribute to the optimization and harmonization of REM as a reference method to estimate wildlife population density without the need for individual identification. We make clear recommendations on the cost-effective sample size for estimating REM parameters, optimizing the human effort when applying REM, and discouraging REM applications based on low sample sizes.