A1295
Title: Bayesian and empirical Bayes approaches for wildlife abundance estimation using aggregated camera-trap data
Authors: Woncheol Jang - Seoul National University (Korea, South) [presenting]
Abstract: Estimating wildlife abundance in complex terrains is challenging when individual identification is not feasible. Using camera-trap data for the endangered long-tailed goral in South Korea, a Bayesian framework for population size estimation based on temporally aggregated observations is developed. A spatially explicit Poisson model is first considered that links camera detections to latent activity centers, with parameters estimated via a Reversible Jump Markov Chain Monte Carlo (RJMCMC) algorithm, enabling direct inference on population size without superpopulation augmentation. Additionally, observations are aggregated at the camera level and a binomial model with a common population size and time-varying detection probabilities is proposed. To account for heterogeneity in detection probabilities, an empirical Bayes approach is adopted in which the prior distribution is modeled using an exponential family representation following Efrons method. By integrating out the latent probabilities, a marginal posterior distribution for the population size is obtained and inference is performed using stable numerical procedures. Simulation studies and real data analysis demonstrate that the proposed methods provide flexible and efficient tools for abundance estimation, while highlighting the importance of prior specification and identifiability.
