Abstract
Adaptive management is often invoked, but infrequently implemented, as a management strategy for natural resource managers to learn from the decision-making process and reduce uncertainty over time. The U.S. Fish and Wildlife Service (Service) incorporates an adaptive management framework directly into their eagle permitting process to balance energy development with managing incidental eagle fatalities resulting from otherwise lawful activities. Specifically, the Service uses a collision risk model that combines prior probability distributions describing exposure and collision risk with site-specific data to predict eagle fatalities from wind energy facilities and prescribe compensatory mitigation. As new site-specific data become available, they can be incorporated into the existing prior probability distributions, allowing for more informed management decisions in the future. Here, we present and demonstrate flexibility of the adaptive management framework by exploring stratification of existing exposure prior distributions to better capture spatial variation in the abundance of bald and golden eagles within the coterminous U.S. We developed a low and high relative abundance exposure prior distribution for each species by binning site-specific exposure data according to the modeled year-round eBird relative abundance (i.e., two relative abundance strata) for each wind facility. We then used a leave one out cross-validation approach to determine how well exposure probability distributions captured site-specific eagle exposure data. For both species, the exposure rate within the stratum describing low relative abundance was lower compared to the previously developed nationwide prior distribution. The exposure rate for the high relative abundance stratum was similar to the previously developed nationwide prior distribution for both species. For both species and strata, the variance in exposure rate was greater than the variance of the previously developed nationwide priors. Results from our cross-validation suggest that our exposure prior distributions adequately capture the variation in eagle exposure rates within each stratum. Only 7% of the golden eagle exposure data-sets and 4% of the bald eagle data-sets within the high relative abundance stratum were not captured below the 95th quantile of the exposure distributions. By using stratified exposure probability distributions that vary with spatial variation in eagle density, the Service can provide more realistic pre-construction predictions of eagle fatalities, possibly guiding facility siting towards areas that represent a lower risk to eagles.