基于RADARSAT-2全极化数据反演周期性地表土壤湿度

Use of fully polarimetric RADARSAT-2 data to retrieve soil moisture of periodic surfaces

土壤湿度是农作物估产和旱情监测的关键参量。目前常用的土壤湿度反演模型都建立在随机粗糙地表条件下,对周期性垄行结构的土壤并不适用。基于RADARSAT-2全极化数据和野外实测数据,分析了交叉极化(vh)后向散射系数对垄行方位角无明显响应;同极化(hh,vv)对方位角的响应为余弦函数,但在方位角为90o±2o位置易出现偏离曲线的异常高亮度值。通过雷达影像上采样点的实测值和Oh模型推算的理论值,拟合得到周期性地表和随机地表之间的误差函数,进而对同极化影像进行纠正。纠正后的同极化比(p)去除了方位角和异常值的影响,交叉极化比(q)受到异常值的影响。通过Oh模型中的p和vh对研究区的地表参数进行反演,17个检验点的验证结果表明,预测的土壤湿度平均相对误差为11.13%,标准差为0.0256cm3/cm3;预测的均方根高度平均相对误差为13%,标准差为0.1315cm。结果与随机粗糙地表土壤湿度和均方根高度的反演精度相当,证明了该模型的有效性。

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.