Substituting space for time: Empirical evaluation of spatial replication as a surrogate for temporal replication in occupancy modelling

Abstract

AbstractOccupancy models that account for detection probability are important analytical tools in conservation monitoring. Traditionally, occupancy models relied on detection/non‐detection data generated from temporal replicates for estimating detectability. Due to logistical challenges and financial costs involved, many large‐scale field studies instead use spatial replication as a surrogate. The efficacy of the two approaches and their statistical validity has generally sought support from simulation‐based inferences rather than empirical data.Using the sloth bearMelursus ursinusas an example, we compared estimates of occupancy and detection probabilities obtained from temporal and spatial sampling designs. We carried out temporally replicated camera trap surveys and spatially replicated sign surveys across a 754‐km2area around Bhadra Tiger Reserve in the Western Ghats of India.We sampled along forest/coffee plantation roads in 58 grid cells of 13 km2each, treating these cells as independent sites. We used the standard single‐season model for the camera trap survey data, and the single‐season correlated detections model (with Markovian dependence) for the sign survey data, and incorporated ecological covariates that likely influenced occupancy and detection probabilities.Occupancy estimates from the two surveys and corresponding modelling approaches were similar [ = 0.58 (0.03) for camera trap surveys; = 0.56 (0.03) for sign surveys]. In both cases, the influence of covariates corroborated oura prioripredictions. Site‐level estimates of occupancy from the two methods were highly correlated (r = .78). We generated a combined estimate of sloth bear occupancy in the region as an inverse‐variance weighted average of the two estimates [ = 0.57 (0.02)].Synthesis and applications. Studies that aim to evaluate occupancy models should account for spatial variation in occupancy/detection probabilities, particularly when making inferences on species–habitat relationships. We show that spatial replication can serve as a good surrogate for temporal replication in occupancy studies, which may be useful for distribution assessments of species when field resources are limited or logistical challenges preclude traditional survey approaches that yield temporally replicated data. Our results therefore provide a basis for efficient targeting of funds and field resources, particularly for practitioners involved in monitoring species at large landscape scales.