Abstract
During my PhD, I contribute to the development of indirect demographic inference methods. The aim is to provide the tools necessary to evaluate recent and local population history (mainly density and dispersal in the last dozens generations) using spatial and genomic data. This would inform conservation and management strategies while being less costly than approaches such as mark-recapture. To achieve this goal, we use a simulation-based approach. Our method links the spatialised simulator GSpace (Virgoulay et al., 2021), the summary statistics calculating library GSumStat and the summary likelihood statistical framework implemented in the infusion R package (Rousset et al., 2017). GSpace implements both coalescent and forward time models of isolation by distance. GSumStat allows efficient computation of a wide variety of commonly used summary statistics. Infusion uses an iterative approach to infer joint distributions of likelihood of multiple parameters, by leveraging tools such as random forest machine-learning method and multivariate Gaussian mixture models.