Arnaud Bensadoun and Hervé Monod
National Institute for Agricultural Research (INRA), Research Unit (UR), France
Frédérique Angevin
INRA, Unit for Research Support (UAR), France
David Makowski
INRA, Joint Research Unit (UMR), France
Antoine Messéan
INRA, UAR, France

In the European debate about GMOs, the coexistence between GM and non-GM crops is a major stake. The regulatory coexistence measures currently considered by Member States mostly rely on fixed separation distances at a national scale. Several spatially explicit modeling approaches have been studied to help determine these separation distances. However the formalism used in those models and the availability of relevant and independent data for calibration and validation make the uncertainty analysis of those models almost impossible. The study presented here aims at developing an alternative model-based approach with emphasis on uncertainty to better adapt coexistence rules to any specific situation. The research work focuses on the use of Bayesian methods to design a collection of statistical models at the scale of an agricultural landscape. Those models yield cross-pollination rates in non-GM fields and are flexible enough to adapt to the available in situ information. Thanks to the Bayesian approach, estimates are computed as distributions whose dispersion depends on the amount and quality of available data; the more abundant and accurate the data, the narrower the distribution. In addition to model construction, we propose a coherent approach to select the best model for a given situation. The selection does not only rely on goodness of fit but also on the quality of the resulting decision for a given threshold. Models are already compatible with the decision support tool of the EU project PRICE.

Key words: Bayesian methods, coexistence, decision support, gene flow, pollen dispersal.