Data Mining-Based Agricultural Typologies to Support Public Decision-Making: Investigating Methodological Issues on the Case Study of Pesticide Reliance of Oilseed Rape Crop.

Typologies are based on similarities between individuals. They have been identified as relevant tools for agricultural policy-making. Numerous examples based on the statistical analysis of a dataset can be found in the literature. They are often applied to farming systems and combine factor analysis (FA) and hierarchical clustering (HC). Despite these numerous applications, implementation to a new case study raises methodological issues. These issues include (i) the theoretical framework of the typology, (ii) the type of variables to use as input of FA (i.e. continuous, categorical or mixed), (iii) the number of factor to use as input of HC. As pesticide use is a major concern in France and oilseed rape as a crop is a major pesticide consumer, we investigated these issues, while building a typology of current cropping systems for oilseed rape.

We based our analysis on a survey from public statistic, gathering information about 2000 oilseed rape plots in France. Attributes describing crop management plan and crop sequence were selected. They were declined into both continuous and categorical variables, to perform FA on only continuous, only categorical, and mixed variables. For each FA, various numbers of factors has been tested as input of HC. Classifications performed were compared on the basis of the number of plot changing of cluster. A set of criterions (including pesticides reliance but not only) were calculated to compare the inter-cluster variability and inter-region variability of performances. Finally, we focussed on some regions to identify lever for action to reduce pesticide reliance.

We first identified variables combinations that structure types and are resilient to all methodological alternatives tested. Doing so, we distinguished between types presenting contrasted performances in view of the considered criterions. This approach made it possible to quantifiy possibilities to reduce pesticide use, but could be adapt to other issues.