Improved Bayesian analysis of metapopulation data with an application to a tree frog metapopulation.

Metapopulation models are important tools to predict whether a species can persist in a landscape consisting of habitat patches. Here a Bayesian method is presented for estimating parameters of such models from data on patch occupancy in one or more years. Earlier methods were either ad hoc, produced only point estimates, or could only use turnover information. The new method is based on the assumption of quasi-stationarity. which enables it to use not only turnover data, but also snapshot data. Being Bayesian, the method produces reliable information about the uncertainty of the parameters and model-based predictions in the form of posterior distributions. It is computationally demanding, but considerably faster than a recently developed Bayesian method extended beyond turnover data. The method is compared with existing methods (placed in a Bayesian framework) by fitting an extended incidence function model to a data set on a tree frog metapopulation with many missing values and by predicting its viability, mean occupancy, and turnover rate after 100 yr.