Avoiding biased estimates: measuring observer avoidance bias in distance sampling estimates of animal populations

Accurate monitoring of animal populations provides crucial information to assess anthropogenic impacts and the success of mitigation efforts. Although distance sampling is the most common method for estimating population densities in mammals and birds, these and other species often violate assumptions that animals do not respond to human observers, causing biased estimates. We present an agent-based model for estimating this bias and evaluating its impact on population density estimates. Animals in our 2D agent-based model follow programmed rules. Models parameterised with either: no human avoidance, or detection of the observer by animals and fleeing behaviours, and are compared to determine the extent of bias that avoidance behaviour can introduce into population density estimates. We test this method with empirical data on responsive movement by Diana monkeys and lesser spot-nosed monkeys from hunted regions of the Gola Forest in Liberia and Sierra Leone. We find that our empirically observed i) distance at which monkeys detected observers and ii) flight initiation distances caused up to a 16% underestimate in population density. We also parameterise the model with theoretical avoidance behaviour to demonstrate the potential scale of the problem, finding up to a 93% density underestimate in extreme - but biologically plausible - cases. By making informed decisions regarding the parameterization of the model, our approach has the potential to complement existing distance sampling techniques, leading to more accurate estimates of species density. This, in turn, can aid in the effective allocation of conservation resources, making our model and its future improvements a valuable tool for researchers in the field.