Challenges with the Multitude of Fatality Estimators and the Need for a Generalized Estimator

Estimating fatality of bats and birds at wind energy facilities is a complicated endeavor and several correction factors must be combined with the number of raw carcasses observed to produce an unbiased estimate. Numerous... [ view full abstract ]

Estimating fatality of bats and birds at wind energy facilities is a complicated endeavor and several correction factors must be combined with the number of raw carcasses observed to produce an unbiased estimate. Numerous fatality estimators have been developed and advances to reduce bias have been made over time. Yet, the availability of multiple estimators has resulted in confusion as to which estimator is the most appropriate given certain conditions. Moreover, given the same data, different estimators can produce significantly different results. In cases where regulatory agencies require specific minimization actions to be implemented if fatality exceeds a certain threshold, the variability in results derived from multiple estimators are problematic and potentially expensive. Here we present a case study highlighting this issue and call for a generalized estimator to eliminate unnecessary confusion, and loss of time and money. In 2013, we conducted a post-construction fatality monitoring study at a wind energy facility in Pennsylvania, and estimated fatality using two different fatality estimators, the Erickson and Huso estimators. Based on a sample size of 10 turbines, the estimated number of bat fatalities was 29.22 bats/turbine (95% CI: 23.08–40.49) and 69.14 (95% CI: 48.77–124.65) using the Erickson and Huso estimators, respectively. The 95% confidence intervals of each estimator do not contain the mean of the other, suggesting a statistically significant different between estimators. Furthermore, the Erickson estimate is below the 30 bats/turbine threshold set by the Pennsylvania Game Commission, whereas the Huso estimate is not. The disparity among estimates are likely the result of differences in the assumptions of each estimator. For example, the Erickson estimator assumes carcass removal occurs at a constant rate (i.e., exponential distribution), whereas the Huso estimator allows for model selection of carcass persistence based on the data. Assuming an exponential distribution rather than the best model, which in this case was a log-logistic model, resulted in an overestimation of carcass persistence and consequently an underestimate of fatality. However, in situations where there are short search intervals, high carcass persistence and low searcher efficiency, the Huso estimator can overestimate fatality, unless the observed carcasses are restricted to those believed to have been killed in the preceding interval. To alleviate the confusion of which estimator is the most appropriate to use, we recommend the development of a generalized fatality estimator that allows the user to test assumptions regarding input parameters, and select the approach that best reflects their particular situation and data. Ultimately, this will 1) provide guidance on study design to increase efficiency and reduce costs of fatality studies, 2) standardize carcass searches and data analyses, and 3) reduce bias and thereby improve accuracy and precision of fatality estimates generated from carcass searches.