Monitoring amphibian populations with incomplete survey information using a Bayesian probabilistic model

Abstract

Bayesian networks (BNs) are a probabilistic modeling platform that connect variables through a series of conditional dependences. We demonstrate their utility for broad-scale conservation of amphibian populations where different types of information may be available within the region. Wildlife conservation decisions for most species are made jointly with other objectives and are tightly constrained by finances. Bayesian networks allow the use of all available information in predictions, and can provide managers with the best available information for making decisions. Habitat models were developed as a hierarchical Bayesian (HB) model for aquatic amphibian populations in the temperate Oregon Coast Range, USA. Predictions for new streams sections were made jointly using a Bayesian network to allow the inclusion of different types of available information. Missing habitat variables were modeled based on habitat survey information. Uncertainty in the true (but unknown) habitat variables were incorporated into the prediction intervals. Further, the probabilistic approach allowed us to incorporate survey information for co-occurring species to help make better predictions. Such species information was connected through the Bayesian network by the conditional dependence that arises from shared habitat variables. The utility of Bayesian networks was shown for these populations for broad-scale risk management. In contrast to deterministic models, the probabilistic nature of Bayesian networks is a natural platform for incorporating uncertainty in predictions and inference.