Evaluating Great Lakes bald eagle nesting habitat with Bayesian inference

Abstract

Bayesian inference facilitated structured interpretation of a nonreplicated, experience-based survey of potential nesting habitat for bald eagles (Haliaeetus leucocephalus) along the five Great Lakes shorelines. We developed a pattern recognition (PATREC) model of our aerial search image with six habitat attributes: (a) tree cover, (b) proximity and (c) type/amount of human disturbance, (d) potential foraging habitat/shoreline irregularity, and suitable trees for (e) perching and (f) nesting. Tree cover greater than 10 percent, human disturbance more than 0.8 km away, a ratio of total to linear shoreline distance greater than 2.0, and suitable perch and nest trees were prerequisite for good eagle habitat (having sufficient physical attributes for bald eagle nesting). The estimated probability of good habitat was high (96 percent) when all attributes were optimal, and nonexistent (0 percent) when none of the model attributes were present. Of the 117 active bald eagle nests along the Great Lakes shorelines in 1992, 82 percent were in habitat classified as good. While our PATREC model provides a method for consistent interpretation of subjective surveyor experience, it also facilitates future management of bald eagle nesting habitat along Great Lakes shorelines by providing insight into the number, type, and relative importance of key habitat attributes. This practical application of Bayesian inference demonstrates the technique's advantages for effectively incorporating available expertise, detailing model development processes, enabling exploratory simulations, and facilitating long-term ecosystem monitoring.