Résumé
Model-based analysis of neutral genetic data allows to indirectly estimate demographic and historical parameters such as population sizes, migration rates or divergence times because those parameters shape the repartition of genetic variability within and between populations over time. In numerous species, dispersal is spatially-limited (individuals preferentially find geographically close mates) and individuals may be spread over a continuous habitat rather than aggregated into discrete panmictic populations. However, inference methods accounting for localized dispersal still bear a number of limitations in terms of biological complexity of the underlying spatial models and of type of information brought by the analyses (e.g. which parameters can be estimated, as well as their biological interpretation). In this study, we used a new recent simulation-based inference method coupled with a spatial genetic data simulator to infer local demographic parameters of population in a continuous habitat. Our results show that we can estimate with good precision more parameters than with previously available methods, notably by independently inferring population density, dispersal rate and the shape of the dispersal distribution. In contrast to competing studies, we reach these results without assuming that the total population size or the habitat size is known. Instead, these results are possible because the simulations do not require coalescent approximations (such as assuming large population size and small migration rate), and because simulation-based methods can exploit summary statistics for which no simple analytical expectation is known. These results highlight the power of simulation-based inference in population genetics and pave the way for new demogenetic inferences under more realistic spatial population genetic models.