In the context of agrosystem management in order to reduce the use of pesticides, we aim to understand population dynamics at the landscape scale of the codling moth (Cydia pomonella), an insect pest in apple orchards. To analyse statistically the genetic markers data (22 microsatellites) obtained from a thousand georeferenced individuals (sampled in 51 different orchards), we developed a spatially explicit metapopulation model. The objective is to estimate population sizes and migration rates as functions of landscape factors (spatial heterogeneities). First the structure of homogeneous genetic groups are analysed from genetic marker data, according to numerical Bayesian approach, with a non-supervised clustering method (STRUCTURE). Second the estimated sub-populations allelic frequencies are used as inputs of a mechanistic-statistical model to estimate dispersion in the heterogeneous landscape. The model associates a reaction-diffusion model for the mechanistic description of insect movements with a population genetic model for the statistical modelling of marker data.