Further development of camera-based estimates of abundance for unmarked animals

Abstract

Estimating abundance is a cornerstone of wildlife management, conservation, and research, and a variety of approaches have been developed to meet this need. Camera traps have been used to estimate abundance for decades, with earlier deployments using a mark–recapture framework and requiring uniquely identifiable individuals. Recently, there has been a surge in methods using camera trap data that do not require individual identification, including instantaneous sampling (IS) and space-to-event (STE) methods. The IS and STE methods involve randomly deploying cameras to simultaneously take photos at fixed intervals and have the advantages of not requiring estimates of animal movement rate and detection probability and are generally robust to variation in animal movement patterns and space use in contrast to nonrandom camera deployments. We describe three alternative estimators of animal density based on IS and STE methods and demonstrate how each are derived. We examined the performance of IS, STE, and the three alternatives using simulated data for 90 different scenarios representing five population densities, three different grouping behaviors, three assumptions of animal movements and habitat selection, and two levels of sampling effort (i.e., 50 and 130 cameras). We evaluated the accuracy of abundance estimates and assessed bias in four alternative approaches for estimating the SE and CIs. We found all estimators were unbiased when there was no habitat selection or when habitat selection was uncorrelated with viewshed area. However, IS was the only estimator that remained unbiased when habitat selection was related to viewshed area. Nonparametric bootstrapping across cameras was the least biased approach for estimating SEs, and unbiased CIs were obtained via a semi-parametric approach. Across our scenarios, we found that when densities were >2 individuals/km2 and when >50 photos with animals were observed, a CV in the abundance estimate <20% was generally achieved with the IS method. Our results emphasize the importance of random placement of cameras, as we show that nonrandom deployments such as placing cameras along roads or game trails, without understanding and accounting for the potential habitat selection, can significantly bias abundance estimates.