Out-of-sight or out-of-site? Forays and detection probabilities in single-season occupancy models

Occupancy models have become popular in wildlife survey analyses because they account for the frequent failure to detect individuals of targeted species. Those individuals sometimes move outside sampling sites, i.e. exhibit temporary emigration. In such cases, occupancy models may become difficult to interpret or even misleading either at the species or the individual level, because they confound presence at the site, availability for detection given presence, and actual detection by the observer. We quantified the probabilities of these three components with spruce grouse (Falcipennis canadensis) in southern Quebec, Canada. We conducted call-response surveys of 24 grouse monitored by radio-telemetry. We defined sites empirically as circular areas of 83 m radius centered on the observer, corresponding to the maximum detection distance obtained. Based on telemetry locations, grouse were present at the site during 42 % of the surveys. Six stationary grouse were present during surveys, but were never detected. Thus, only individuals that moved in the presence of the observer (89 %) were considered available for detection. Individuals available for detection were detected in 51 % of the cases. We simulated detection histories and built single-season occupancy models, based on the empirical relationship between detection probability and the distance measured between observers and grouse. When temporary emigration was ignored, site occupancy was {psi} = 0.89, and the associated probability of detection was p = 0.23. When instances of temporary emigration were dropped, estimates were {psi} = 0.88 and p = 0.41. Using only grouse available for detection, estimates were {psi} = 0.87 and p = 0.42. Disentangling the components of detection probabilities had little impact on occupancy estimates, but showed a major effect of temporary emigration on estimated detection probabilities.