Bayesian hierarchical latent variable models for ecological data types

Abstract

Ecologists and environmental scientists employ increasingly complicated sampling designs to address research questions that can help explain the impacts of climate change, disease, and other emerging threats. To understand these impacts, statistical methodology must be developed to address the nuance of the sampling design and provide inferences about the quantities of interest; this methodology must also be accessible and easily implemented by scientists. Recently, hierarchical latent variable modeling has emerged as a comprehensive framework for modeling a variety of ecological data types. In this dissertation, we discuss hierarchical modeling of multi-scale occupancy data and multi-species abundance data. Within the multi-scale occupancy framework, we propose new methodology to improve computational performance of existing modeling approaches, resulting in a 98% decrease in computation time. This methodology is implemented in an R package developed to encourage community uptake of our method. Additionally, we propose a new modeling framework capable of simultaneous clustering and ordination of ecological abundance data that allows for estimation of the number of clusters present in the latent ordination space. This modeling framework is also extended to accommodate hierarchical sampling designs. The proposed modeling framework is applied to two data sets and code to fit our model is provided. The software and statistical methodology proposed in this dissertation illustrate the flexibility of hierarchical latent variable modeling to accommodate a variety of data types.