Long-term reevaluation of spatially explicit models as a means for adaptive wildlife management

We evaluated a 20-yr-old spatially explicit model (SEM) that predicted the spatial expansion of reintroduced Persian fallow deer in northern Israel. Using the current distribution of the deer and based on multi-model inference we assessed the accuracy of the SEM's prediction and what other factors affected the population's current distribution. If the SEM's projection was still valid, the leading model in the multi-model inference would include only the SEM's projection as an explanatory variable with a good fit. Different leading models would reveal key variables overlooked when the SEM was constructed or changes in the landscape unforeseen at the time, thus assisting adaptive management and decision-making. We assessed deer presence from camera trap encounter counts analyzed using N-mixture models. Models included various combinations of seven predictors: the 20-yr predictions of an SEM developed during the initial phases of the reintroduction, three key landscape characteristics on which the SEM was originally based but updated to reflect current conditions, distance from the release site, elevation, and the distribution of gray wolves (a predator that was absent from the area when the SEM was developed). Competing models were ranked by Akaike information criterion (AIC). Wolf distribution was the key predictor explaining the current deer distribution, appearing in all three leading models (∆AIC < 2.0) and carrying 71% of the AIC weight (coefficient = -14.86 ± 5.6 [mean ± SE]). Of these three models, the SEM 20-yr prediction appeared in two, but explained only a fraction of the variance (coefficient = 0.001 ± 0.08). The contribution of all other predictors was negligible. While the SEM failed to accurately predict the 20-yr deer distribution, the divergence between its projection and reality pointed to the probable cause (wolves) of this discrepancy. The inclusion of the SEM prediction in the leading models indicates that had the wolves not spread to the study area, the predictions would still have merit suggesting that long-term SEMs can potentially be robust. Long-term reevaluation of SEMs can be beneficial even if model projections fail, as the process can uncover the specific factors driving this failure, supporting adaptive management procedures.