Statistical Methods for the Detection and Space-Time Monitoring of DNA Markers in the Pollen Cloud

The analysis of pollen grains finds applications in fields as diverse as allergology, paleoecology, apiculture and forensics. In contrast with morphological identification methods that require the visual inspection of individual pollen grains, recently-developed genetic approaches have the potential to increase both the scale and resolution of pollen analyses. In the first part of this dissertation, I describe efficient experimental designs to determine the prevalence of a genetic marker in an aggregate pollen sample from the results of DNA amplification by polymerase chain reaction (PCR). The method is based on the theory of limited dilution assays and takes into account potential sources of assay failure such as DNA degradation and PCR inhibition. In the following parts, I show how the genetic composition of air-sampled and bee-sampled pollen can be used to infer spatial characteristics of the floral landscape. Through individual-based simulations of the foraging behavior of honey bees, I obtain theoretical relationships between the genetic differentiation of pollen loads collected at a beehive and the spatial genetic structure of the plant populations visited by foragers. At a larger scale, I present a hierarchical Bayesian model that describes the distribution and spread of common ragweed in France by integrating annual pollen counts from aerobiological stations and presence data from field observations. As the capacity for pollen sampling and analysis increases, these models could be expanded to describe in more detail the biological and physical processes affecting pollen production and transport, and thus provide better predictions for ecological applications such as the control of invasive species