Hierarchical Bayesian models for population ecology
Date
2017
Authors
Ketz, Alison C., author
Hobbs, N. Thompson, advisor
Hooten, Mevin, committee member
Wittemyer, George, committee member
Webb, Colleen, committee member
Journal Title
Journal ISSN
Volume Title
Abstract
Models, by their definition, are abstractions of the systems they describe and require a delicate balance of inclusion of information with reduction. Hierarchical Bayesian models are well suited for ecological problems, because they facilitate the decomposition of highly complex ecological systems into lower dimensional elements. We can partition variability that arises from the ecological processes separately from variability that arises from sampling error, thereby rigorously accounting for uncertainty. In this way, we can better answer questions pertaining to the ecology of populations and aid in better management of their ecosystems. Estimation of abundance is the central challenge in population ecology, and we begin this dissertation by addressing the problem of determining the population size of elk across multiple time and spatial scales during five winters. In Chapter 2, I build upon existing multi- state mark-recapture methods using a hierarchical Bayesian N-mixture model with multiple sources of commonly collected data on abundance, movement, and survival, to accurately estimate the abundance of a mobile population of elk on the winter range of Rocky Mountain National Park and Estes Park, CO. Classification data are used in ecology to examine population trends through model-based theoretical approaches. For ungulates such as elk, wildlife managers use sex-ratios and stable age or stage distributions to assess population growth or decline. However, physical ambiguities and observer skill can lead to biased results. In Chapter 3, I develop two hierarchical models to address the sample bias that results when data are missing-not-at-random, which occurs when individuals are observed but not classified. Forecasts are used to aid management to evaluate the probability that resource objectives will be met given different management actions. In Chapter 4, I develop a hierarchical model incorporating a discrete time, stage structured model assimilated with abundance and classification data, to provide forecasts under a variety of management actions to aid decision makers to meet objectives. I use Bayesian hierarchical models that incorporate multiple sources of information to address common estimation problems that arise in population ecology. We are frequently interested in constructs and latent processes that are not necessarily observable in ecological systems. I use theoretical models of the underlying processes to extract information pertaining to populations and management goals. Compounding the challenge is that we must rely upon survey samples rather than complete census. I illustrate the utility of hierarchical Bayesian models using data on the population of elk (Cervus elaphus nelsoni) on the winter range of Rocky Mountain National Park in Colorado, USA.