Statistical methods for modeling the movement and space-use of carnivores
Date
2017
Authors
Buderman, Frances E., author
Hooten, Mevin, advisor
Boone, Randall, committee member
Crooks, Kevin, committee member
Ivan, Jacob, committee member
Journal Title
Journal ISSN
Volume Title
Abstract
Recent advancements in the ability to monitor animal locations through time has led to a rapidly expanding field focused on statistical models for animal movement. However, many of the existing methods are computationally time-consuming to fit, restricting their application to a few individuals, and inaccessible to wildlife management practitioners. In addition, existing movement models were developed for contemporary animal location data. Many previously collected telemetry data sets may provide important information on animal movement, but there may be additional challenges that are not present in data collected explicitly for movement modeling. For example, telemetry data collected for survival studies may have large temporal gaps, and long-term studies may have used multiple data collection methods, resulting in data points with different error structures. My goal is to develop and expand on methods for modeling individual- and population-level animal movement in a flexible and computationally accessible framework. In Chapter 1, I discuss the role of carnivores in natural resource management and the habitat associations and movement ecology of two carnivores native to Colorado, Canada lynx and cougars. I describe the existing data sets, collected by Colorado Parks and Wildlife, that are available for analyzing Canada lynx and cougar movement ecology. I also discuss contemporary statistical methods for analyzing animal telemetry data. Finally, I conclude with my research objectives. Chapter 2 presents a new framework for modeling the unobserved paths of telemetered individuals while accounting for measurement error. Many available telemetry data sets were not collected for the purposes of movement modeling, making the use of existing methods challenging due to large temporal gaps and varying monitoring protocols. In contrast to the more traditional mechanistic movement models that appear in the literature, I propose a phenomenological functional model for animal movement. The movement process is approximated with basis functions (e.g., splines), which are an extremely flexible statistical tool that allows for complex, non-linear movement patterns at different temporal scales. In addition, the observed data contains complicated error structures that vary across telemetry type. I then apply this model to a case-study of two Canada lynx that were reintroduced to Colorado and show that inference about spatio-temporal movement behaviors can be obtained from the unobserved paths. For Chapter 3, I apply a population-level version of the functional movement model, developed in Chapter 1, to 153 Canada lynx that were released in Colorado as part of a state reintroduction program. Twelve offspring of the reintroduced individuals were also included in the analysis. I perform a post hoc analysis of movement paths using spatial visualizations and linear mixed models, allowing the different movement behaviors to vary as a function of season, sex, reproductive status, and reintroduction timeline. This chapter represents one of the most comprehensive analyses of Canada lynx movement in the continental United States. In Chapter 4, I discuss the fine-scale movement of cougars in the Colorado Front Range using a continuous-time discrete-space (CTDS) framework. The CTDS framework is computationally fast, flexible, and easily implemented in standard statistical programs. This chapter focuses on a population-level extension of the CTDS framework that can be used to model the population- and individual-level effect of landscape variables on movement rates and directionality. I use this model to determine potential drivers of cougar movement in the Colorado Front Range, a rapidly urbanizing area in the foothills of the Rocky Mountains. This work also uses the functional model I developed in Chapter 1, but with an error structure more appropriate for small-error GPS data. I conclude with a summary of findings, overarching themes, and potential future research directions in Chapter 5.