A spatiotemporal model for multivariate occupancy data

We present a multivariate occupancy model to simultaneously model the presence/absence of multiple species, and demonstrate its use with a goal of estimating parameters related to occupancy. The proposed model accounts for both spatial and temporal dependence within each species, as well as dependence across all species. These dependencies are addressed through random effects, defined so there is no confounding with estimating occupancy covariate effects. Data augmentation and specific choices for the random effects permit all Gibbs updates in the Markov chain Monte Carlo algorithm, making the model computationally efficient and scalable with the number of species and size of spatial domain. A simulation study shows that the model outperforms single‐species spatiotemporal occupancy models with regard to estimating occupancy parameters. We demonstrate the model with a three species camera trap study on Thomson's gazelle, wildebeest, and zebra in the Serengeti National Park of Tanzania, Africa.