Multi-state occupancy modeling and optimal allocation of survey resources for Common Loons in Washington State

Abstract

Common Loons (Gavia immer) are a state listed sensitive species in Washington State; however, little is known about the distribution of the Common Loon or the habitat associations of this species. This is complicated by the limited resources available for management to monitor Common Loons in Washington. In Chapter 1, I develop a novel multi-state occupancy model to integrate citizen science eBird data with monitoring data. This framework was then applied to Common Loons in Washington State. Occupancy probabilities were influenced by level of human disturbance and physical lake characteristics. There was temporal autocorrelation in reproduction, such that reproduction at a site in a given year was positively associated with reproduction in the previous year. For eBird observers, detection of Common Loons at sites with reproduction was negatively related to site area and distance travelled during an observation bout, and positively related to time spent surveying. Washington Department of Fish and Wildlife observers were more likely to detect a Common Loon at sites with and without reproduction than were eBird observers. My results provide a better understanding of the distribution and breeding habitat requirements of Common Loons. In Chapter 2, I develop a framework to identify a study design that will optimally allocate limited survey resources to maximize the information gained from occupancy analyses. Placing study design within the broader framework of optimal decision making provides a structured and transparent approach to optimal survey design, while Bayesian analytical methods provide the opportunity to leverage Bayesian updating in the evaluation of candidate survey designs. The Common Loon is monitored by the Washington Department of Fish and Wildlife, but there are limited personnel hours available to conduct surveys each season. I formulated optimal survey design for the Common Loon in Washington as a resource allocation problem. Alternative designs were built through application of alternative decision rules wherein sites were selected based on various estimates from an initial occupancy analysis. The decision rule that minimized the predicted state-wide mean uncertainty in occupancy probability selected sites based on the site-specific uncertainty in covariate relationships. Together, the frameworks developed provide methods for utilizing multiple data sources and identifying a decision rule that can be used for survey design planning. I also demonstrate that a framework for including citizen science data with traditional monitoring data can lead to an expanded scope of inference in understanding the ecology and conservation needs of species. This optimal design framework is applicable to occupancy models generally, and provides a quantitative assessment of the outcome of potential survey designs with respect to a given monitoring objective.