Using radar remote sensing to predict topsoil properties at farm field-scale

Soil cartography available at farm field-scale is a key factor to implement crop-site management practices. The aim was to determine the backscattering of SAR data with high predictive capacity for soil properties and to validate an approach to predict the spatial pattern of six soil properties in farm fields conditions. Fifty one C-Band SAR images data (5.36 GHz) were obtained from Sentinel-1 (ESA) platform. Two farm-fields located in the Pampas of Argentina were used in this study. Apparent soil electrical conductivity (ECa) was previously measured in both fields using Veris 3100 soil sensor. A regular grid soil sampling scheme was carried out for each field. An hydric balance was estimated for crop and fallow seasons. Random forest regression was applied for: (i) to determine when radar SAR polarization had the highest predictive capacity; and (ii) to predict clay, silt and sand content, soil depth and ECa. In general, our results suggest that the backscattering during the crop-summer season with water stress condition had higher predictive capacity of soil properties. Soil depth was the best predicted using C-band SAR data followed by clay and sand contents. These properties are strongly associated with the dynamic of soil water storage. ECa was better predicted at 0-0.3 m than 0-0.9 m for both fields. Soil properties were in general better predicted for field 1 which could be related to an effect of crop residues on C-band SAR data predictive capacity. We found that this capacity depends on the interaction among spatial dynamic of soil coverage, soil water storage capacity and temporal dynamic of water availability.