Modeling Multi-Scale Occupancy For Monitoring Rare And Highly Mobile Species

Occupancy-based methods are commonly used to model the distribution, habitat use, and relative abundance of species. In particular, occupancy-based methods are often recommended for monitoring species, allowing researchers to track population trends using detection-non-detection data alone. Occupancy models, however, have proven difficult to apply to rare, highly mobile species. Species' movements outside of sampling areas may lead to violation of the geographic closure assumption of occupancy models and overestimated occupancy probability. Low detection probability may further inflate occupancy probability estimates. We developed a novel continuous-time, multi-scale occupancy model to simultaneously account for closure assumption violations and low detection probability. We used a simulation study to test our model relative to a discrete-time multi-scale model, and we conducted a power analysis to assess the ability of an instantaneous occupancy parameter to detect trends in abundance, relative to standard occupancy alone. The continuous-time model was competitive with the discrete-time model and was generally computationally faster than and outperformed the discrete-time model when detection probability was low. The instantaneous occupancy parameter outperformed occupancy in terms of power to detect trends when we used an implicit (i.e., estimating occupancy independently in each primary occasion) dynamic occupancy model, but performed no better when we used an explicit (i.e., estimating colonization and extinction) dynamic occupancy model. We applied both discrete-time and continuous-time multi-scale occupancy models to a case study of data collected on wolverines (Gulo gulo) in Washington, USA. We found improved precision in estimates with the continuous-time model and that asymptotic occupancy of wolverines was high, but short-term use of any given area was low. Our multi-scale, continuous-time occupancy model can be used to detect trends in abundance of rare, highly mobile species, regardless of how occupancy dynamics are modeled. Furthermore, our model can allow for more efficient data collection and analysis than traditional discrete-time or spatially multi-scale approaches, as our model uses all available detections and requires only one detector per sampling unit by substituting time for space.