Comparing estimates of population change from occupancy and mark–recapture models for a territorial species

Monitoring studies often use marked animals to estimate population abundance at small spatial scales. However, at smaller scales, occupancy sampling, which uses detection/nondetection data, may be useful where sites are approximately territories, and occupancy dynamics should be strongly correlated with population dynamics. Occupancy monitoring has advantages in that it is less expensive and invasive, and marked animals are not needed. Here, we used empirical data to determine whether and when change in occupancy is a good proxy for population change for a territorial species. As part of this overall goal, we also compared maximum‐likelihood estimates using a model‐averaging approach with a Bayesian MCMC approach. We used field data collected from 1993 to 2013 on three study areas for California spotted owls ( Strix occidentalis occidentalis ), a territorial species. Although correlations for trajectories of realized population change (Δ t ) between territory occupancy and Pradel models were moderate to high for Bayesian MCMC ‐based estimates and high for model‐averaged estimates, magnitudes of the trajectories were different with the Pradel model reporting greater magnitudes of change. For the two areas showing a decline, Δ t for the Pradel model was approximately 20–30% lower than for the occupancy model, and 25% higher in the area showing an increase. These differences can arise because the occupancy model is less sensitive, in that if two owls share a territory, the loss of one may be reflected in survival and, consequently in Δ t by the Pradel model, but because the territory remains occupied it is not reflected by the occupancy model. Bayesian MCMC ‐based and model‐averaged estimates of Δ t were in close agreement in pattern (correlation ≥0.74) and magnitude (relative differences of last Δ t were ≤5%) for both occupancy and mark–resight models. Results from the Pradel model may lead to conservation actions necessary to avoid high extinction or extirpation risk for small populations, while results from the territory occupancy model may result in status quo management. We found both Bayesian MCMC ‐based and model‐averaged estimates of Δ t robust approaches to evaluate population trends. However, we recommend the Bayesian MCMC approach for estimating risk (e.g., probability of declines) for retrospective analyses.