Multi-species occupancy models (MSOMs) have been gaining in popularity in recent years as a means to gain inference on an ecological community, but formal checking and validation approaches for this class of models have lagged behind. Unlike single-species occupancy models that maybe written as integrated likelihoods, MSOMs usually have a Bayesian framework and contain multiple levels. Standard model checking and selection methods are often unreliable in this setting and there is only limited guidance in the ecological literature for this class of models. We examined several different approaches for checking and validating MSOMs and applied these methods to an assemblage of freshwater fishes in Colorado, USA, to better understand their diversity and distributions. Additionally, we motivate the work through its context in a model-based sampling design.
Results/Conclusions
Our findings indicated distinct differences among model selection approaches. The within-sample model selection criteria, Watanabe-Akaike information criterion (WAIC) and conditional predictive ordinate criterion (CPO), ranked models based on more parsimonious models given larger, and therefore worse, scores. In addition, CPO was unable to discriminate against superfluous variables. Cross-validation techniques performed the best in terms of prediction, and were used to identify the best-fitting model. Examination of its associated residual plots showed no clear lack of fit but did highlight sites with sparse data. The model was then used to select future sampling locations through a model-based, adaptive sampling design for continued monitoring of the system.