Abstract:
Apart from surface roughness and soil moisture, the azimuth angle of a crop row is also a contributing factor of the backscattering coefficients of periodic bare soil surfaces. The commonly used models for estimating soil moisture could be applied to randomly rough surfaces, but not periodic surfaces. Based on the fully polarimetric RADARSAT-2 and field sampling data, this paper analyzes the response of backscattering coefficients to the azimuth angles of the crop rows. The like-polarized (hh or vv) backscattering is very sensitive to azimuth angles, assuming trigonometric function. Abnormal high value appears at the position around 90o. Thus, adjustment to the like-polarized backscattering coefficients is needed in order to remove the influence from azimuth angles. The cross-polarized (vh) backscattering shows a random distribution, reacting insensitively to the changes of azimuth angles. Assuming that the backscattering coefficient from periodic surfaces is the result of a random function (related to soil moisture and root mean square height) results in adding a cosine function (related to azimuth angles). Thus, through the difference between backscattering coefficients measured by SAR and calculated by an Oh model, the fitting error functions were acquired and could be regarded as the difference between periodic surfaces and random rough surfaces, and then the like-polarized images could be corrected. The correlation coefficients between the corrected backscatter coefficients and soil moisture are 0.626 and 0.775 respectively in hh and vv polarization modes, which are significantly improved compared with the results before correction. The scatter of the corrected co-polarized ratio p was randomly distributed with no abnormal value around 90o. This proves that co-polarized ratio p could remove the effect of azimuth angles and abnormal value. While the cross-polarized radio q could remove the effect of azimuth angles to some extent, it was affected by the abnormal value around 90o. Finally, the soil moisture and root mean square height of the study area can be estimated by solving two equations (vh and p). This study selected 17 sampling points as checkpoints. The correlation coefficient between estimated soil moisture and measured soil moisture was up to 0.88, with the average relative error of 11.13% and the standard deviation of 0.0256 cm3/cm3. The correlation coefficient between estimated root mean square height and measured root mean square height was 0.76, with the average relative error of 13 % and the standard deviation of 0.1315 cm. There was no significant difference in accuracy between the samples with azimuth angles of 90o and the other samples. The inversion accuracy of the corrected Oh model for periodic surfaces is very close to the models of randomly rough surfaces. The modified model is reliable and applicable for periodic surfaces.