The Serengeti is one of the most visited protected areas in the world. This ecosystem is particularly famous for the annual migration of around 1.8 million wildebeest, zebra and gazelles, making us wonder about a time when more wildlife could roam free across vast landscapes. But how do we know how many animals are there?
Despite the importance of knowing how our estimates differ from the true abundance and which type of monitoring errors should be minimized first, these are challenging questions to answer in the “real world”. These would require considerably complicated experiments to be conducted, especially given the potential financial and time costs involved. This is when simulation modelling comes in handy! Using a “virtual monitor” approach, it is possible to simulate “true” scenarios of wildlife abundance (taking into account, for example, animal distribution, population structure and herd size) and then simulate each step in the monitoring process (e.g. defining number of transects and time between photos) and potential errors (such as miscounting juveniles hidden behind adults and missing animals at further distances) to investigate how each of these factors may affect the final estimates (more specifically, their biases and variability).
Using this approach and the Serengeti as a case study, we (Ana Nuno, Nils Bunnefeld and EJ Milner-Gulland) just published a paper in the Journal of Applied Ecology which shows how simulation models can be used to improve monitoring under uncertainty. We investigated which components of monitoring should be prioritized to increase survey accuracy (difference between the estimates and the “truth”) and precision (uncertainty in the estimates) and showed how different monitoring budgets could affect the quality of the estimates obtained. Our findings indicate that aerial photo sampling does not seem to greatly underestimate the wildebeest population size but the level of their aggregation within the landscape is the most important factor explaining variation in the precision of our estimates (i.e. how much replicated estimates vary). For species with a more random distribution in the landscape, such as impala, underestimation was the main issue, which should be minimized especially by providing specialized training and calibration by observer.
In face of limited budgets and other conservation priorities, we must make sure that the resources for monitoring are well spent and allocated according to well defined priorities. In our paper, we give specific recommendations on how to improve monitoring of savannah ungulates by targeting the most important sources of errors or accounting for these effects when producing wildlife estimates. More importantly, we remind that wildlife monitoring starts in the office, when virtual monitoring can help us make better decisions. To implement robust management interventions, we should account for multiple types and sources of uncertainty (such as observation uncertainty) and then make decisions that consider their consequences and potential trade-offs.