Estimating occupancy using spatially and temporally replicated snow surveys

Occupancy modelling is increasingly used to monitor changes in the spatial distribution of rare and threatened species. Occupancy methods have traditionally relied upon temporally replicated surveys to estimate detection probability. Recently, occupancy models with spatial replication have been used to estimate detection probabilities over large geographical areas that are difficult to survey repeatedly. We developed occupancy models that combine spatially and temporally replicated data and applied them to snow-tracking surveys of six species, including wolverine Gulo gulo and Canadian lynx Lynx canadensis. We surveyed thirty-nine 100-km² cells and used 1-km trail segments within cells as spatial replicates. We surveyed 56% of the cells once and 44% of the cells between 2 and 14 times, resulting in a total of 872km surveyed. We compared four occupancy models that incorporated spatial correlation in detection probability and hierarchically estimated occupancy at two spatial scales: cell occupancy and segment presence. We detected strong serial correlation in probability of detection for all species. Our models with serial correlation had higher occupancy estimates with larger confidence intervals than models assuming segments were independent and exchangeable. Spatial and temporal replicates have identical power to detect decreases in occupancy when survey segments are independent, but spatial correlation in detection probability can reduce the power of spatial replicates. The effects of spatial correlation are more pronounced when detection probability is low. Application of temporal replicates to spatial replicated surveys increases the precision of occupancy estimates, but sampling design trade-offs between number of sites and spatial versus temporal replicates need to balance levels of spatial correlation in detection probability with costs to visit sites.