Estimating population impacts via dynamic occupancy analysis of Before-After Control-Impact studies

Background/Question/Methods

Estimating environmental impacts on populations is one of the main goals of wildlife monitoring programs, which are often conducted in conjunction with management actions or following natural disturbances. In this study we investigate the statistical power of dynamic occupancy models to detect changes in local survival and colonization from detection/non-detection data, while accounting for imperfect detection probability, in a Before-After Control-Impact (BACI) framework. We simulated impacts on local survival and/or detection probabilities, and asked questions related to: (1) costs and benefits of different analysis models, (2) confounding changes in detection with changes in local survival, (3) sampling design tradeoffs, and (4) species with low vs. high rates of turnover. 

Results/Conclusions

Estimating seasonal effects on local survival and colonization had little effect on power to detect changes in local survival. Estimating a parameter that accounted for pre-treatment differences in local survival between Control and Impact sites decreased power by 50%, but it was critical to include when such differences existed. When the experimental treatment had a negative impact on species detectability, but analysis assumed constant detection, the Type I error rates were inflated (0.20-0.33). There was low power (<0.5) to detect a 50% decrease in local survival for all combinations of sites (50, 100), seasons sampled (8, 12), and visits per site per season (4, 6). Unbalanced designs performed worse than balanced designs with the exception of the case of treatments being implemented in different seasons at different sites. Surveying more seasons After Impact resulted in modest power gains, but at least three seasons Before Impact were required.  Occupancy studies conducted in a BACI design offer the opportunity to detect environmental impacts without the costs of intensive studies. Given the low power to detect small changes (20%) in local survival, these studies should be used when researchers are confident that major treatment impacts will occur or large sample sizes are obtainable.