A multi-method hierarchical modeling approach to quantifying bias in occupancy from noninvasive genetic tagging studies

Abstract

Detecting ecological change in mammalian communities requires large volumes of data collected across large landscapes. Noninvasive genetic tagging (NGT) can provide such data and is a common method of estimating species occupancy, distribution, and abundance. Methods to correct errors in genetic analysis exist, but estimating and accounting for detection error in NGT studies—specifically, detection error during hair trapping—has received less attention. If detectability varies with an ecological characteristic that is instead attributed to occupancy, detection error may lead to spurious conclusions. We demonstrate how multi-method occupancy models informed by camera-trapping data can quantify error, partition sources of variability, and estimate occupancy from NGT studies. We surveyed marten (Martes americana), fisher (Pekania pennanti), and wolverine (Gulo gulo) occurrence via hair trapping and camera trapping for 3 repeated monthly surveys at 66 sites in the Rocky Mountains of Alberta, Canada. Camera traps photographed the hair traps and surrounding sampling site. We used multi-method occupancy models to estimate site occupancy, probability of detection, and conditional occupancy at a hair trap. We tested the predictions that detection error in NGT studies could be induced by temporal variability and behavioral variability arising from the presence of heterospecifics. Models indicated that NGT via hair trapping consistently underestimated occupancy of marten, fisher, and wolverine compared to cameras. The magnitude of bias varied, but at worst underestimated occupancy by half. The presence of a heterospecific affected a species' detection at hair traps. Wolverine presence increased marten detection, whereas marten decreased fisher detection at NGT traps. For all species, detection error was reduced through successive monthly surveys, although heterogeneity induced by other mustelids remained. Detection error from multiple sources can bias NGT occupancy estimates, with implications for population estimation, ecological inference, species' legal status assessments, and conservation decisions. We recommend that NGT studies quantify and correct for detection error using independent survey methods and multi-method occupancy models, to improve the ability to answer ecological questions and make conservation decisions in the face of ecological change. © 2014 The Wildlife Society.