Statistical Estimation Of Backscattering Coefficients In X-Band Over Bare Agricultural Soils

Empirical, semi empirical or physical modeling approaches have been proposed to simulate the microwave signal that can be observed during periods of bare soil. The aim of this study is to evaluate the performance of a statistical approach (Random Forest algorithm) to estimate the backscattering coefficients in X-band. Regular satellite images are acquired by TerraSAR-X, over an agricultural region located in southwestern France. The analyses take advantage of a ground dataset covering a wide range of soil conditions: smooth to rough surfaces $(\mathrm{h}_{rms}0.5 -7.9$ cm) monitored in dry to saturated conditions (2.4-35.3%). Once trained and validated on half the dataset, the statistical algorithms shows better performance than previous approaches based on semi-empirical or physical models, with correlations above 0.83 and errors below 1.07 dB. The sensitivity of the statistical algorithm keeps consistent with others models, with a higher importance of top soil moisture than surface roughness or texture variables.