Mammal population density estimation using camera traps based on a random encounter model: theoretical basis and practical recommendations

Estimating the population density of mammals has long been one of the problematic tasks of both fundamental population ecology and practical programmes for their conservation and management. The majority of methods for population density estimation using camera traps are focused on individually marked species. This review paper presents the theoretical and practical foundations of a method, Random Encounter Model (REM), used for estimating the population density of unmarked mammal species using camera traps. Based on an extensive analysis of the literature and our personal practical experience, we discussed the theory and practice for the application of this method, as well as its strengths and weaknesses. In this method, if we know parameters of the effective detection zone of a camera trap (radius and angle), and the length of the day range, it is possible to correct the trapping rate (i.e. the number of independent trap events per total number of camera traps-nights) in order to calculate the population density of species. The effective detection zone of a camera trap is determined through modelling using computer vision algorithms. The mammal day range is calculated based on its activity level and travel speed, taking into account behavioural patterns based on machine learning models. For REM, a random or systematic design for the camera trap placements should be used. If camera traps are installed against trails or roads, appropriate correction factors must be applied. The effectiveness and reliability of REM has been confirmed by many independent population density estimates, including capture-recapture analyses, visual transect counts, and scat counts. To date, the implementation of REM and its extensions is presented in the R programming environment. It has been established that the main difficulties in the use of the REM are technical imperfections of the camera traps themselves, the relatively large required number of their stations (at least 50 or more), as well as long calibration work. For all these difficulties, possible solutions are proposed. In conclusion, practical recommendations are provided for the use of REM in studies in Protected Areas.