基于Sentinel-1数据的黑河中游土壤水分反演

基于Sentinel-1合成孔径雷达 (SAR) 数据及相同时段的中分辨率成像光谱仪（MODIS）和Landsat 8两种归一化植被指数（NDVI），构建变化检测模型以估算黑河中游的高分辨率土壤水分，并探讨模型中具体参数设置对估算精度的影响。结果表明：①在对后向散射系数时间序列的差值 ( \mathrm{\Delta } \sigma ) 和植被指数 ( \mathrm{V} \mathrm{I} ) 进行线性建模过程中，MODIS NDVI和Landsat 8 NDVI这两种植被产品所构建的模型在 \mathrm{\Delta } \sigma - \mathrm{V} \mathrm{I} 空间中所选取的采样点比例分别为2%和4%时，各自取得最优精度; ②以土壤水分反演为目标，使用Landsat 8 NDVI构建的变化检测模型略优于使用MODIS NDVI构建的变化检测模型，两种模型的均方根误差RMSE分别为0.040 m³/m³和0.044 m³/m³，相关系数R分别为0.86和0.83; ③对于变化检测方法的关键参数，若使用低分辨率的SMAP/Sentinel-1 L2_SM_SP土壤水分数据分别代替站点观测的土壤水分初始值和缩放因子 (即两个连续时相土壤水分变化的最大值 \mathrm{\Delta } M s_{max} ) 这两个参数，则土壤水分RMSE将分别增加0.01 m³/m³和0.04 m³/m³。即土壤水分缩放因子这一参数的误差对反演结果的影响大于土壤水分初始值误差对反演结果的影响，故采用高精度的缩放因子进行变化检测估算。研究结论对于利用新兴的Sentinel-1 SAR数据，通过变化检测算法准确获取高分辨率土壤水分信息具有实际参考价值。

Sentinel-1 based Soil Moisture Estimation in Middle Reaches of Heihe River Basin

In this study, a change detection model, constructed using the Sentinel-1 Synthetic Aperture Radar (SAR) data and the simultaneous Normalized Difference Vegetation Index (NDVI) products from the Moderate Resolution Imaging Spectroradiometer (MODIS) and Landsat 8 sensors, is applied to estimate soil moisture in middle reaches of the Heihe River Basin, and the effects of two key parameters on retrieval accuracy are comprehensively investigated. The results show that: (1) when constructing the empirical relationship between backscattering coefficient difference ( \mathrm{\Delta } \sigma ) and Vegetation Index (VI) required by change detection model, the optimal sampling ratios in the ( \mathrm{\Delta } \sigma - V I ) space are approximately 2% and 4% for MODIS NDVI and Landsat 8 NDVI, respectively; (2) the Landsat 8 NDVI-based change detection model slightly outperforms the MODIS NDVI-based model in soil moisture retrieval accuracy, with Root Mean Square Error(RMSE) of 0.040 m³/m³ and 0.044 m³/m³respectively; (3) for the key parameters of the change detection method, replacing the ground-based initial soil moisture and scaling factor (maximum soil moisture difference between two adjacent dates \mathrm{\Delta } M s_{max} ) by the low-resolution SMAP/Sentinel-1 L2_SM_SP data will increase the RMSE by 0.01 m³/m³ and 0.04 m³/m³ respectively. Comparing to the parameter of initial soil moisture, the error in soil moisture scaling factor will lead to more significant degradation in the performance of the change detection method, thus it is recommended to use the high precision scaling factor for soil moisture estimation. This study confirms the promising potential of Sentinel-1 data for retrieving high-resolution soil moisture via change detection method and provides practical insight into its application.