Soil-depth mapping at farm scale using dual-polarimetric SAR data

Soils in the southeastern pampas of Argentina commonly present a petrocalcic horizon which has great variability in depth at field scale. Developing conventional soil cartography in such conditions is laborious, time consuming and expensive. The energy backscattered and the degree of polarization of C-Band SAR data from Sentinel 1 (ESA) depends on soil moisture content, roughness and crop biomass. Hence, these data are potentially useful for providing auxiliary information to determine spatial patterns for soil depth. The aim was to propose and validate a technique to detect spatial-temporal changes of dual polarimetric SAR data, to determine spatial pattern of soil depth at field scale. Two farm-fields were selected for this study. In each field, a 50 m regular grid was used to sample soil depth. Random Forest regression was applied to determine the polarization and degree of dispersion with high and null predictive capacity of soil depth. Two-dimensional scatterplot between backscattering VV (σ_VV^o) and VH (σ_VH^o) in crop and fallow conditions were developed to characterize differences between polarizations with high and null predictive capacity of soil depth. Scatterplots with low backscattering values (in linear values) in both polarizations and with high amount of ground biomass had higher predictive capacity for soil depth. On the contrary, scatterplots with high and low values of σ_VV^o and σ_VH^o, respectively, had null predictive capacity. Results suggest that predictive capacity of soil depth from dual-polarimetric SAR data increases when backscattering is low, specifically when soil water content is a crucial factor for the change in the backscatter magnitude.