Résumé
Oilseed rape (OSR) genes can escape fields in space via pollen and seeds and in time via volunteers resulting from seeds lost before or during oilseed rape harvests. Previous simulation studies and field observations showed that co-existence at the landscape level of contrasting OSR varieties such as genetically modified (GM) and non-GM varieties require costly measures that are difficult to implement, such as isolation distances between OSR fields and stringent volunteer control in all fields and road margins. In the present study, two local strategies, non-GM buffer zones aroundGMfields and discarding the harvest of boundary plants of non-GM fields, were tested in a simulation study using the GENESYS model. This model, which was modified here to improve the prediction of volunteer reproduction and pollen dispersal, quantifies the effects of cropping system on gene flow in landscapes over the years. The evaluation of the improved model version with independent field observations showed that cross-pollination of fieldswas accurately predicted up to 50m and systematically underestimated above that distance, though fields were still ranked correctly. A corrective factor should therefore be applied to predictions in case of low regional OSR and GM proportions (which result into large distances between fields). To study the effects of the two local strategies, simulations were carried out in four contrasted case studies, based on different landscapes, regional OSR and GM proportions, cropping systems and OSR varieties. In these situations, harvest discarding was needless when volunteers were well controlled because the GM impurity in non- GM harvests was well below the EU labelling thresholds; it was useless when OSR contained male-sterile plants or when the landscape was infested by volunteers because large parts of the fields had to be discarded to respect impurity thresholds, which is financially unacceptable. Buffer zones were useful if the non-GM varieties comprised male-sterile plants or if the non-GM fields were small, but only if volunteers were well controlled and harvest impurity was already close to the labelling threshold. Buffer zones appeared considerably more efficient in reducing harvest impurity than harvest discarding because they not only increase the distance between GM and non-GM fields but also diminish the proportion of GM pollen in the total pollen cloud over landscapes. Harvest discarding could possibly be interesting for isolated OSR crops surrounded by bare ground or small crops, an effect that was not tested here. The present study showed that efforts should be dedicated to characterizing and modelling of gene flowat the landscape level, aswell as the collection of datasets for the evaluation of these models in many landscapes and configurations. Local measures tested here are insufficient for ensuring harvest purity. Therefore, future work should focus on finding additional measures are necessary to manage OSR volunteers which constitute the major risk factor for OSR harvest purity.