Performance of Hierarchical Abundance Models on Simulated Bat Capture Data

Abstract

The ability to accurately estimate abundance is crucial to ecologists, conservationists, and managers to provide insight on species status, population trends, and viability. Acoustic detection and occupancy modeling can provide an understanding of resource use for bats, but these methods do not estimate how many bats are in an area, or how these numbers change over time. In North America, there is a heightened need to estimate bat abundance and trends in response to white-nose syndrome (WNS) and other threats to bat populations. We assessed the performance of the N-mixture model for repeated count data and the general multinomial-Poisson model for removal sampling to estimate bat abundance from simulated mist-net capture data. We evaluated performance under varying numbers of sites and visits, detection probabilities (P), and population sizes. We simulated four scenarios with a total of 85 combinations of parameter values each containing 1,000 replications. We used the UNMARKED package in R to fit the N-mixture and removal models. We calculated relative bias (RB), mean absolute error (MAE), and mean absolute percent error (MA%E) from model estimates to evaluate model performance. Relative bias, MAE, and MA%E decreased as p and bat abundance increased for all models. The removal model outperformed the N-mixture model in all scenarios except when P = 0.05. The N-mixture model had low RB, MAE, and MA%E when bat abundance was ≥ 70 and P > 0.5, but in other scenarios, errors were large. The mean of estimates from the removal model were unbiased and RB, MAE, and MA%E were very low for most scenarios. Use of the removal model with data from repeated mist-net surveys may allow resource managers and conservationists to better quantify how resource management and landscape composition affect bat species abundance and overall populations.