Population synchrony studies in animal populations have been central in ecology as a tool to understand drivers of ecological systems. It is a common approach to use capture-recapture (CR) methodologies, which provide information on sampling error and population size, and assume multinomial sampling. Several studies have highlighted the consequences of not addressing the variation in the sampling error in their respective state-space approaches, in which the population processes are often assumed to be first-order autoregressive (AR (1)). To understand the potential impacts of different models for sampling error, we assessed the performance of two estimation methods for the coefficients of AR (2) population cycles, using simulated CR data with two trap sessions. The baseline method was to fit the AR (2) model to the log-transformed observed counts. The second method consisted of estimating individual probabilities for the different capture histories using a multinomial likelihood in R-INLA, and fit the AR (2) model to the population size log-transformed estimates from the Horvitz-Thompson estimator. We subsequently applied both methods on an 18-year dataset of Gray-sided Voles, collected in Arctic Norway, along a 200 km extension from coastal to inland tundra.
Results/Conclusions
Our results show that method performance is highly dependent on the variation of capture probabilities of the different individuals, as well as the structure of the underlying AR process. Interestingly, the two methods performed similarly in most simulation settings, although the second one, which incorporated capture history information, consistently obtained better estimates than the baseline method. The small detected difference in both methods was influenced by the under-sampling of low capturability individuals, as well as bias of the Horvitz-Thompson estimator for low capture probabilities. This translated in similar estimates from both methods to the real dataset, showing an inverse gradient in both AR coefficients, from the coast to inland. We highlight the need to perform simulations to assess adequacy and limitation of models prior to ecological experiments, as available data might not be enough to obtain reasonable estimates, which will necessarily have an impact on conclusions.