A Bayesian hierarchical occupancy model for track surveys conducted in a series of linear, spatially correlated, sites

摘要

1. Natural resource agencies often rely on surveys of animal sign (e.g. scat, scent marks, tracks) for population assessment, with repeat surveys required to model and account for uncertain detection. Using river otter Lontra canadensis snow-track survey data as a motivating example, we develop a 3-level occupancy model with parameters that describe (i) site-level occupancy probabilities, (ii) otter movement (and thus, track availability) and (iii) recorded presence-absence of tracks (conditional on the availability of tracks for detection). 2. We incorporated several recent developments in occupancy modelling, including the presence of both false negatives and false positives, spatial and temporal correlation and repeated sampling across distinct observers. 3. We investigated optimal allocation of sampling effort (e.g. within and among snowfall events) using simulations. We also compared models that allowed site-level occupancy and track-laying processes to be spatially correlated with models that assumed independence among sites. 4. Both types of models (independence and spatial) performed well across a range of simulated parameter values, but the spatial model resulted in more accurate point estimates for detection parameters and credibility intervals with better coverage rates when data were spatially correlated. When applied to real data, the spatial model resulted in a higher estimate of the occupancy rate than the baseline model (0·82 vs. 0·59). A minimum of 15-20 helicopter flights, distributed among at least three unique snow events, were needed to meet precision goals (standard error). 5. Synthesis and applications. We describe a flexible and robust occupancy modelling framework that accounts for heterogeneous detection rates in surveys of animal sign. The method allows for spatially correlated sites and should have broad relevance to surveys conducted by many natural resource agencies.