基于遥感数据同化的土壤含盐量估算方法

为探究同化遥感数据对监测区域尺度土壤含盐量时空信息的适用性，以河套灌区沙壕渠灌域为研究区，以高分一号卫星影像为数据源，通过灰度关联法筛选光谱指数，采用岭回归法构建不同深度的土壤含盐量反演模型，使用集合卡尔曼滤波同化算法将遥感数据应用于HYDRUS-1D模型中，开展区域尺度不同深度土壤含盐量的同化研究。结果表明，基于不同深度土壤含盐量的岭回归法模型，其R²均在0.64以上，RE为0.14～0.22,反演精度较高，得到的反演值较为准确；在单点尺度上，与模拟值、反演值相比，同化值更接近实测值，其EFF为0.84～0.93,NER为0.61～0.73,均为正数，且RMSE降低到0.006%～0.011%,提高了HYDRUS-1D模型模拟精度；在区域尺度上，不同深度同化值的r均为0.94以上，NER为0.61以上，优于模拟值和反演值，且同化精度随着深度的增加而降低。本文基于遥感数据和HYDRUS-1D模型的集合卡尔曼滤波同化研究，提高了土壤含盐量的模拟精度，对提高监测区域尺度土壤含盐量时空信息的精度具有一定的参考价值。

Estimation Method of Soil Salinity Based on Remote Sensing Data Assimilation

Soil salinization seriously restricts sustainable agricultural development, and it is a main environmental problem in arid and semiarid regions. Therefore, the method of assimilating remote sensing data is used to monitor spatial and temporal information of soil salinity in a regional scale, which is of great significance to management of soil salinization. The feasibility of soil salinity estimation to assimilate HYDRUS-1D model and remote sensing data was explored by using ensemble Kalman filter. The study area was located in Shahaoqu Irrigation District of Hetao Irrigation District. The remote sensing data was obtained by GF-1 satellite. Spectral indexes were screened by gray correlation method, and inversion models of soil salinity at different depths were constructed by ridge regression models. Then remote sensing data was applied to HYDRUS-1D model by using ensemble Kalman filter to carry out assimilation study of soil salinity of different depths in a regional scale. The main conclusions were as follows: based on ridge regression models of soil salinity at different depths, R2 were above 0.64 and RE were 0.14~0.22. Inversion accuracies were relatively good and inversion values were relatively accurate. In a single point scale, compared with inversion values and simulation values, assimilation values were closer to measured values. EFF of assimilation values were 0.84~0.93 and their NER were 0.61~0.73. They were all positive values. And their RMSE were reduced to 0.006%~0.011%. These results showed the scheme of data assimilation improved simulation accuracies of HYDRUS-1D model. In a regional scale, r of assimilation values were above 0.94 and their NER were above 0.61. And they were better than r and NER of inversion values and simulation values. Meanwhile, with increase of depth, the accuracy of assimilation was decreased. The results indicated that data assimilation greatly improved simulation accuracies of soil salinity at different depths by using ensemble Kalman filter. The research result can provide certain reference value for improving monitoring accuracy of soil salinity in a regional scale.