Epsilon Archive for Student Projects

Abstract

Camera trapping has become crucial in wildlife research, enabling detailed observations of elusive and nocturnal species with limited human interference. The use of occupancy modelling to analyse camera trap data is also rapidly increasing, aiding in the assessment of habitat selection, species distribution, and multi-species dynamics while considering imperfect detection. However, the design of camera trap studies, typically involving large grids with limited cameras, often results in sparse data and low detection probabilities which challenges model accuracy. These low detection probabilities can hamper model performance and resulting inferences, however to what extent this is true for multi-state occupancy models is still unknown. In this study, I simulated datasets with 60 and 200 sites, with various detection probabilities between 0.001 and 0.9 and their impact on multi-state occupancy models to establish thresholds required for model reliability. The results revealed that detection probabilities must exceed 0.1 for model convergence, and that values above 0.2 minimized bias and eliminated convergence issues. Additionally, more than 60 sites are required in multi-state occupancy models for finding habitat relationships, which is more than found for single-state occupancy models. The case study on moose and roe deer revealed that while aggregation of survey length increases detection probabilities it also increases uncertainty in estimates. Models with daily detection histories failed to converge, while weekly aggregation improved fit for some parameters, however there still had large uncertainty. Simplifying state complexity enhanced model performance despite low detection probabilities but could not fully mitigate the effects of the sparse data. I conclude that when detection probabilities are too low, complex occupancy models are difficult to fit. Strategies like deploying multiple cameras, targeted camera placement or using bait can increase detection rates but may introduce biases. However, the gained model performance from higher detection probabilities might outweigh these biases. Ensuring higher detection probabilities and adequate data collection is essential for reliable model outcomes. This study highlights key thresholds and considerations for improving multi-state occupancy models using camera trap data, aiding in the design of more effective wildlife research studies.