A preliminary analysis of topsoil organic carbon contents and stocks spatial distribution in a region of France (Region Centre)

We present a preliminary assessment of Soil of Organic Carbon (SOC) contents and stocks in a region of France. We use legacy data on soil profiles having either SOC content (3297 profiles) or SOC stock (620 profiles) measurements. We estimated the contents and stocks for a standard layer corresponding to 0-30 cm. Ancillary variables used in the model are the DEM (50-m resolution) derivatives, Land Cover (2006 Corine database), various polygon soil class maps and airborne gammay-ray spectrometry data. The prediction models used Boosting Regression Trees (BRT), a methodology combining regression trees-a binary split regression for predictors-and boosting, an adaptive learning machine methodology. For SOC content the main contributing factors are Land-Cover, Beven index, U, K, and various DEM derivatives. The importance of gamma-ray data may be linked to SOC through non-direct relations with other soil parameters such as soil depth, clay content, parent material nature, degree of weathering and illuviation, etc. Using more detailed soil maps improves the quality of the prediction. When predicting SOC stocks the gamma-ray data (especially K) have a higher relative contribution to the model.