Fuzzy k-means clustering of fields in an elementary catchment and extrapolation to a larger area

A practical problem in land management in southern France is the classification of cultivated fields according to their hydrological properties as a function of soil and topography. Lack of spatial contiguity and overlap in data space render conventional numerical methods for soil and landscape classification unreliable. Fuzzy k-means with extragrades have been used to create a spatially coherent classification of fields for a test area in Roujan (Herault, Languedoc). Several different classifications were tested for geographical coherency using the Geary contiguity coefficient. The stability of the classes was also tested using a less precise data set for characterizing the same fields. To test the usefulness of this approach, the best classification was used to allocate fields from a larger area adjacent to the test site to the established classes. The increase in extragrade sites is used as a measure of robustness, representativity and transferability of the classification. Our,results show that a fuzzy classification gives systematically better mappable and less error-sensitive clusters than a classical crisp classification. Furthermore, it provides more information, particularly concerning the representativity of the elementary catchment.