Robust retrieval of soil moisture at field scale across wide-ranging SAR incidence angles for soybean, wheat, forage, oat and grass

Abstract

Surface soil moisture is estimated by inverting physical scattering models for low- crops using L-band airborne SAR data over the incidence angle range from 30 to 50°. The forward simulation is accurate with 1.7, 1.8, and 2.3 dB rmse for grass, wheat (including oat and forage), and soybean respectively assessed over 38 fields and during the full growth period for soybean. Dominant scattering mechanisms revealed by the model are found consistent across the incidence angle range: surface (VV) and double bounce (HH) for grass, double bounce for both polarization of wheat, and surface (young plant) and volume (mature plant) for both polarization of soybean. The model fidelity is robust across incidence angles between 30 and 50°, benefiting from rigorously simulating the three scattering mechanisms and successfully simulating the effect of vegetation and roughness.The soil moisture estimates are accurate to unbiased rmse of 0.041, 0.059 and 0.060 m3/m3 and correlation of 0.71, 0.83, and 0.69 for grass, wheat, and soybean fields, respectively over the soil moisture dynamic range up to 0.5 m3/m3, and soybean's full growth cycle. Vegetation water content and surface roughness are retrieved simultaneously to the soil moisture solutions, thus constraining the inversion of soil moisture. The retrieval performance is fairly uniform across the incidence angle, partly because of the reliable forward modeling across the incidence angle and also thanks to the reliable retrieval strategy. The robustness against the incidence angle supports frequent global mapping.The use of both HH and VV offers superior soil moisture retrievals than single-channel VV or HH input perfoms, by 0.034 m3/m3 unbiased rmse for grass and by 0.01 m3/m3 for wheat and soybean. The VV-based retrieval performed better than HH for grass and soybean, where surface scattering was important, but either choice resulted in similar retrieval accuracy for double-bounce dominanted wheat. No need for quad-pol and multi-angular observations would allow the retrieval approach to be applicable for spaceborne global mapping.