Digital Mapping of Soil Surface Texture Using Soil Legacy Data and Fully Polarimetric Radarsat-2 Images.

Precision farming and conservation, through the optimal use of fertilizers and water resource management, enables farmers to increase crop yields and quality while ensuring environmental protection. The inability to obtain accurate spatial variability of soil characteristics rapidly and inexpensively constitutes one of the biggest limitations of precision farming. The digital mapping of soil surface texture (DMSST) should be at the basis of the decision-making process in precision farming because this primary soil property is implied in establishing appropriate yield potential and assessing environmental risks (soil erosion, nutrient leaching and greenhouse gas production).

The two critical aspects of the DMSST are the compositional nature (sand, silt and clay content) of texture data and the often nonlinear relationships between soil and environmental variables. In order to investigate these relationships, after an isometric log-ratio transformation of soil texture data a set of linear methods (ordinary kriging, cokriging, regression kriging) and the novel nonlinear technic (the ε-insensitive support vector regression - SVR) have been tested for DMSST. The covariates used were multiple polarizations and polarimetric parameters extracted from H/A/a, Freeman and Durden, and Touzi decompositions of RADARSAT-2 data.

The results showed that 1) using RADARSAT-2 covariates highly improves digital soil map accuracy and 2) the SVR provided the best performance predictions compared to other interpolation technics. Its kernel-based transformation of the nonlinear approximation and its robustness make it a valid alternative to linear interpolation technics for DMSST.