Evaluation of replicate sampling using hierarchical spatial modeling of population surveys accounting for imperfect detectability

Effective species management and conservation benefit from knowledge of species distribution and status. Surveys to obtain that information often involve replicate sampling, which increases survey effort and costs. We simultaneously modeled species distribution, abundance and spatial correlation, and compared the uncertainty in replicate abundance estimates of the endangered palila (Loxioides bailleui) using hierarchical generalized additive models with a soap film smoother that incorporated random effects for visit. Based on survey coverage and detections, we selected the 2017 point-transect distance sampling survey on Mauna Kea, Hawai‘i Island, for our modeling. Our modeling approach allowed us to account for imperfect detections, control the effects of boundary features, and generate visit-specific density surface maps. We found that visit-specific smooths were nearly identical, indicating that little information was gained from a subsequent visit, and that most of the estimator uncertainty was derived from within-visit variability. Scaling back the palila survey to a single visit would halve the survey effort and logistical costs and increase efficiencies in data management and processing. Changing the sampling protocol warrants careful consideration and our findings may help management and regulatory agencies by maximizing efficiency and minimizing costs of surveying protocols, while providing guidelines on how to best collect information critical to species' conservation.