Inversion of bare soil moisture by the least squares support vector machine approach combined with SPM

The least squares support vector machine (LS-SVM) techniques are applied to the inversion of soil moisture. The backscattering properties of bare soil under different radar parameters are numerically simulated by using small perturbation method (SPM). After data sensitivity analysis, with the L-band radar frequency(1.4 GHz) and dual angle of incidence(40°/50°) selected, designed a variety of inversion scheme, herein the single polarization, dual polarization and co-polarization ratio of the backscattering coefficient are selected as the microwave signal sample information. Through appropriate training, the least squares support vector regression techniques are adopted to estimate soil moisture under different inversion schemes. The inversion results demonstrate high accuracy when multiple incident angles and the ratio of co-polarization backscattering coefficients are used as the microwave signal sample information. Comparison with the results of the artificial neural network (ANN) proved the validity and the anti-noise ability of the presented method, thus providing a new approach for the real-time retrieval of soil moisture.