Integrating multiple sign types to improve occupancy estimation for inconspicuous species: a case study of American pika in the Pacific Northwest
Loading...
Authors
Goldman, Mia
Issue Date
2022
Type
Thesis
Language
Keywords
American pika , Bayesian hierarchical model , Detection probability , False positives , Ochotona princeps
Alternative Title
Abstract
As rapid global environmental change continues to affect populations through a range of mechanisms, a clear understanding of species responses to these threats is increasingly critical for conservation and management. Occupancy models have emerged as one of the most powerful tools to investigate regional population trends and dynamics, range shifts and habitat associations. Standard occupancy models are capable of producing unbiased estimates of occupancy and its environmental drivers by allowing for observation errors �" in particular, the failure to observe a target species when present (false-negative detection error). Use of indirect sign (e.g., scat, tracks) can vastly improve sample size and reduce costs for inconspicuous focal species, but can also introduce additional sources of observation error that may bias estimates of occupancy, including incorrect identification (false-positive detection error). Reliance on multiple unique sign types may introduce additional sources of bias if these sign types vary in re-liability or if their relative reliability changes under differing environmental conditions. A ‘multi-sign’ occupancy approach, which models the detection process separately for each unique sign type, may therefore improve our ability to generate unbiased estimates of occupancy and its environmental drivers for inconspicuous species like the American pika (Ochotona princeps). In this study, we modeled occupancy dynamics for American pika using multiple direct and indirect indicators of pika presence (fresh scat, fresh haypiles, pika calls and pika sightings) collected from 2010 to 2021 at five national parks in the Pacific Northwest. Furthermore, we investigated how estimates of pika occupancy trends and environmental drivers differ under three increasingly realistic representations of the pika observation process: (1) perfect detection (a common assumption for modeling pika occupancy), (2) standard occupancy model (single observation process with no possibility of false detection), (3) multi-sign with no false detections (non-false positive model), and (4) multi-sign with false detections (full model). For the multi-sign occupancy models, we modeled each observation process separately as a function of climatic and environmental covariates including substrate complexity, season, survey period and vegetation cover. In addition, we modeled each occupancy process (initial patch occupancy, colonization, and extinction) separately as a function of temperature, precipitation, forb, rock and shrub cover. Results from all three models indicated that annual occupancy across parks was relatively stable during our study period, ranging from 32% to 40% and exhibited a general increase from 2010 to 2014 and a weak decline from 2017 to 2021. All models also indicated that that forb, shrub and rock cover positively influenced colonization rates at the plot level and forb cover negatively influenced extinction. Consistent with previous research, detection rates were high in our study, averaging 82% across all study sites. False detection rates were higher than expected, averaging 6.8%. Both true and false detection probabilities exhibited substantial variation across the five national parks in our study, with true detection rates varying regionally from 77.1% to 92.4% and false positive rates varying from 11.7% to 2.8%. Estimates of occupancy processes and their environmental drivers were highly sensitive to different representations of the detection process. For example, initial occupancy rate for the non-false positive model was 8% higher than the full model, and the effect of forb cover was much stronger in the full model versus the perfect detection and non-false positive models. Overall, we demonstrated that a “multi-sign” approach to dynamic occupancy modeling and the inclusion of false detection errors have strong potential to generate more robust estimates of occupancy dynamics for inconspicuous species than standard occupancy modeling approaches.