Evaluating camera trap data for studying spatiotemporal avoidance and predation between mammal species across a gradient of habitat degradation

Abstract

Ongoing climate change and anthropogenic disturbance negatively impact biodiversity. We must be able to capture, monitor and understand how ecosystem processes, such as, species interactions, are impacted so that we are better positioned to protect against further biodiversity loss. One inherent challenge is data collection. Camera traps enable us to remotely capture large volumes of data with minimal disturbance to behaviour, but current automated classification methods are unable to generalise well across locations. I investigate the ability of convolutional neural networks to generalise across a gradient of habitat degradation within a camera trap dataset collected in tropical forest. I found generalisability was poor, but was helped by using a detector-classifier combination. Methods are needed to detect interaction signals from the large volume of camera trap data. Here, I apply statistical methods to test for spatiotemporal avoidance across land-use and disturbance gradients, using the hypothesised avoidance of humans by bearded pigs as a case study. The results did not support the hypothesis, but highlighted the need to understand the data requirements to power the method. Using an agent-based model to simulate animal movement and generate a camera trap dataset, I test our ability to detect species interactions at varying population and camera trap densities, and interaction strengths. The results showed the difficulty in discerning the type of interaction, and in detecting avoidance behaviour across a range of parameters, despite the large volume of simulated data. The use of camera trap data for ecological analyses is a growing field, with the potential for transformative analysis, including in our understanding of species interactions. Reliable rapid processing of the images, as well as sensitive methods to detect interactions, are, however, still lacking, and further development is needed to better quantify species’ responses to anthropogenic disturbance in order to identify the species most impacted.