Improving occupancy estimation when sampling disturbances and animal movements violate the closure assumption

Background/Question/Methods

Site occupancy models that account for species imperfect detection are increasingly utilized in ecological research and wildlife monitoring.  Occupancy models require replicate surveys to estimate detection probability over a time period where the occupancy status at sampled sites is assumed closed (i.e., no changes in occupancy).  Unlike mark-recapture models, few studies have examined how violations of closure can bias occupancy estimates. Our study design allowed us to differentiate among two processes that violate the closure assumption during a survey season: 1) repeated destructive sampling events that result in either short- or long-term site avoidance by the target species and 2) sampling occurring over a time period during which non-random movements of the target species result in variable occupancy status.  We used dynamic occupancy models to quantify the potential bias in occupancy estimation associated with these processes for a terrestrial salamander system.  We sampled 126 transects repeatedly within a single field season using invasive sampling techniques and constructed 17 candidate models that portrayed different sources of variation in occupancy and detection probabilities. Finally, we conducted simulations to evaluate how violations of closure can bias occupancy estimates when local extinction occurs within a field season.

Results/Conclusions

Model ranking indicated that both repeated sampling and non-random movement of organisms were evident in our system.  We observed a chronic decrease in occupancy associated with cumulative sampling and non-random salamander movements over the entire field season.  We also observed a strong but temporary disturbance effect on salamander detection probability associated with repeated sampling within a 24-hr. period.  Our simulation study revealed general sensitivity of estimates from single-season occupancy models to violations of closure, with the strength and direction of bias varying between scenarios. Bias was minimal when extinction probability or the number of sample occasions was relatively low.  Our research highlights the importance of addressing closure in occupancy studies and we provide multiple solutions, using both design- and model-based frameworks, for minimizing bias associated with non-random changes in occupancy and repeated sampling disturbances.