Consequences of modelling procedures on detecting environmental effects on species distribution from camera-trap data: implications for wildlife conservation

Abstract Camera traps have revolutionised wildlife monitoring. However, no consensus method exists for analysing these data. We investigated how commonly used modelling procedures affect the detection of environmental effects and quantified how this affected species distribution maps, which are essential tools for conservation planning. We used the tapeti Sylvilagus brasiliensis sensu lato , monitored using camera traps in a Brazilian indigenous reserve. We compared the ability of two commonly used modelling procedures (occurrence- vs abundance-based models, controlling or not for imperfect detection, using or not time-to-independence thresholds) to detect species responses to environmental variables. We then compared the species distribution predicted from each modelling procedure. Abundance models detected additional effects compared with occurrence models. Occurrence models detected the same environmental effects whether or not they accounted for imperfect detection. In contrast, abundance models were sensitive to imperfect detection. N-mixture models that controlled for detection provided consistent results regarding the nature, sign, and magnitude of effects, whether no time-to-independence, 30-min or 60-min thresholds were applied. Ignoring imperfect detection should not be an option for analysing camera-trap data of unmarked individuals. Hierarchical modelling, allowing detection and ecological processes to be modelled separately, should be preferred. We advocate for developing guidelines for analysing camera-trap data.