Abstract
Fatality monitoring at wind projects requires carcass detection trials to adjust fatality estimates for the proportion of fatalities not found. However, detection trials vary greatly in metric, duration, carcass monitoring schedule, species, number placed, state of decomposition, whether placed within or outside search areas, and other factors. We introduce a new approach for estimating fatalities by quantifying overall detection rates rather than separate rates for searcher detection error and carcass persistence, and by leaving placed and found fatality carcasses undisturbed throughout monitoring. We placed 2 fresh‐frozen bird carcasses weekly at random sites within fatality search areas and on randomized days Monday–Friday at Sand Hill and Santa Clara wind projects, Altamont Pass Wind Resource Area, California, USA. To estimate detection rates, we used logit regression to fit detection outcomes on body mass, which served as an axis of similitude between placed trial carcasses and fatality finds. Adjusted carcass placement rates among species detected by searchers regressed on true placement rates with a slope of 1.0 so long as sufficient numbers of trial carcasses were placed, thus validating our approach as an unbiased estimator. Our approach generally estimated lower fatality rates than did conventional approaches, the latter of which demonstrated biases in searcher detection rates and carcass persistence rates whether based on proportion of carcasses remaining or mean days to removal. Our approach also revealed detection errors that highlight the difficulty of finding and identifying the remains of dead animals, and which warrant routine reporting. Despite averaging only a 5‐day search interval on intensively grazed annual grasslands where ground visibility was usually high, our experienced fatality monitors averaged 4.3 searches/first carcass detection, failed to detect 25% of 75 species represented by placed carcasses, and misidentified carcasses to species among 44% of species detected. Estimates of time since death also suffered bias and large error. Our approach more realistically simulates carcass detection probabilities associated with fatality monitoring, is less costly, facilitates hypothesis testing, eliminates multiple sources of error and bias suspected of conventional methods, and enables quantification of errors in estimated time since death, species identifications, and false negative findings. © 2018 The Wildlife Society.