基于地基微波辐射观测的土壤水分反演算法评估

Evaluation of Soil Moisture Retrieval Algorithms based on Ground-based Microwave Radiation Observation

当前常用的被动微波土壤水分反演算法有水平极化单通道算法、垂直极化单通道算法、双通道算法、微波极化差比值算法和扩展双通道算法,5种反演算法具有不同的差异,对这些反演算法进行系统的评估和分析将有助于反演算法的改进和星载高精度土壤水分产品的发布。为了避免直接采用卫星产品验证时的尺度匹配、空间异质性等问题,基于地基L波段微波辐射观测以及配套的土壤和植被参数测量数据,对这5种反演算法进行了实现、对比和分析,得出以下结论:①单通道算法具有最佳的反演性能,水平极化单通道算法反演结果具有最高的相关性(相关性系数R=0.83),垂直极化单通道算法反演结果具有最小的反演误差(均方根误差RMSE=0.028 m^3/m^3,偏差BIAS=-0.011 m 3/m^3),但单通道算法需要精确的植被含水量输入;②其余3种算法能脱离植被辅助数据的使用,性能略差但也能满足星载微波传感器的探测指标要求(小于等于0.04 m^3/m^3);其中,扩展双通道算法和微波极化差比值算法的土壤水分反演结果比双通道算法略差,但本例中扩展双通道算法在植被含水量反演方面更具优势。

The commonly used passive microwave soil moisture inversion algorithms include Single Channel Algorithm at H polarization(SCA-H),Single Channel Algorithm at V polarization(SCA-V),Dual-Channel Algorithm(DCA),Microwave Polarization Ratio Algorithm(MPRA)and Extended Dual Channel Algorithm(E-DCA).The five retrieval algorithms have different performance,systematic evaluation and analysis of these inversion algorithms will contribute to the improvement of the retrieval algorithm and the release of satellite soil moisture products.Verification of satellite product could bring some problems,such as scale matching and spatial heterogeneity.In order to avoid these issues,the above five soil moisture inversion algorithms are imple‐mented,compared and analyzed based on ground-based microwave radiometer observation and supporting soil and vegetation parameter measurement data.The results show:(1)SCA has the best inversion performance.SCA-H has the highest correlation(R=0.83),and SCA-V has the smallest inversion error(RMSE=0.028 m^ 3/m ^3,BIAS=-0.011 m^ 3/m ^3),but SCA needs the accurate vegetation water content as an input.(2)The oth‐er three algorithms can get rid of the use of vegetation-aided data,with slightly poor performance but also meet the satellite detection requirements(less than or equal to 0.04 m^ 3/m^ 3).Among them,E-DCA and MPRA are slightly worse than the DCA.However,E-DCA is more advantageous in the vegetation water content inversion in our study.