Using relational decision trees to model out-crossing rates in a multi-field setting

Abstract

Nearly three-quarters of the genetically modified maize (the insect resistant type MON 810, also called Bt maize) produced in the EU are cultivated in Spain, where the share of Bt maize cultivation in some regions (Catalonia) is very high (above 70%). In order to ensure coexistence with the production of conventional maize and satisfy the 0.9% EU threshold for adventitious presence of authorized genetically modified (GM) material in conventional (non-GM) maize crops, a set of preventive coexistence measures must be applied. These measures usually include the setup of large and fixed isolation distances, pollen barriers, flowering coincidence, crop rotation and other measures, which are very hard to fulfill in a multi-field setting. Basic empirical and modeling studies that explore the feasibility of coexistence between GM and non-GM crops focus on pair-based interactions between fields while multi-field studies build upon them, attempting to consider the complexity of gene flow under crop management practices.In this study, we use the methodology of relational data mining (which can take into account several coexistence measures at the same time) to predict gene flow from GM to non-GM maize fields under multi-field crop management practices at a local scale. The approach extends the pair-based assessments of out-crossing rate by considering all neighboring fields within the entire study area, along with the farming practices applied to them. The estimation of the out-crossing rates is performed by using a PostgreSQL relational database that is analyzed with the algorithm TILDE for building relational classification trees. In building the trees, TILDE explores the relations describing spatial aspects, maize flowering and crop management practices for the 400ha maize oriented production area Pla de Foixà in Catalonia, Spain, in the period 2004–2006.Our approach proposes a new methodology to predict the level of adventitious presence on a multi-field setting, where the influence of more than one GM field is considered at the same time. The structure of the obtained models can be used in the design of coexistence measures of the second generation, which should not be used individually but treated as synergetic coexistence measures, offering different alternatives to achieve a particular coexistence threshold (e.g., 0.9%, 0.45%, or 0.1%). The possibility to consider multiple measures simultaneously makes farmers more flexible in their management decisions as compared to the rigid use of isolation distance only, which is currently the most commonly recommended coexistence measure.