An investigation of study design and prior selection in Bayesian occupancy models

Abstract

Ecological data are inherently complex and often require making inferences about quantities that are not directly observed. To estimate these unobserved quantities, statisticians typically employ hierarchical modeling techniques. Detection/non-detection data, often referred to as occupancy data, are collected by ecologists and biologists to address a wide range of ecological questions; these inquiries may include, but are not limited to, questions about species distributions, changes in species distributions over time, or species detectability. Occupancy data are particularly well suited for analysis using hierarchical occupancy models. In this work, we present novel methods and recommendations for a variety of Bayesian occupancy models, focusing on improving accessibility for practicing scientists; in our work, we emphasize both study design and prior selection as critical components for enhancing analyses to better address ecological questions. Specifically, we identify methodological modifications and sampling recommendations to guide early detection monitoring of invasive dreissenid mussels in the western United States. We provide sampling and modeling recommendations for long-term ecological monitoring projects in the presence of incomplete data. We offer a flexible method for implementing Bayesian regularizing priors in the occupancy modeling framework, and offer an accessible tool for implementing the technique in R.