Reducing data processing effort in camera trap density estimation: Extending the REST model by explicitly modeling animal detection processes

Ecological communities are increasingly unstable in the Anthropocene, requiring continuous multi-species monitoring across broad spatial scales. While camera traps offer great potential for monitoring ground-dwelling mammal densities, labor-intensive data processing constrains their application. We developed the RAD-REST (Random Encounter and Staying Time model Relying on All Detections) model, an extension of the conventional REST model, to enable community-level density monitoring with substantially reduced effort. By modeling the probabilistic process of animals entering predefined focal areas using a Dirichlet compound multinomial distribution, our model estimates densities using detection record subsets (videos/photo sequences) while leveraging all available data. We created a user-friendly R package that performs full Bayesian parameter estimation, including activity levels. Using data in the Boso Peninsula, Japan, Monte Carlo simulations determined analytical workload requirements for standard precision (CV = 0.35) and stringent standards (CV = 0.2). RAD-REST produced unbiased estimates with appropriate coverage probabilities. For 200-camera arrays, analyzing 100 detection records per species achieved standard precision (CV = 0.35) with only one hour of analysis time. Achieving CV = 0.2 with 400-camera networks requires analyzing approximately 7% of total records. We present a specific protocol for integration into Snapshot, a globally expanding annual camera trap survey program.