邓 超1，苏 南1，潘晓婷2，马丽丽2，林 薇1

　　(1.水利部南京水利水文自动化研究所，江苏 南京 210012;

　　2.扬州市生态环境局，江苏 扬州 225000)

　　摘 要：土壤含水量作为地表的重要参量之一，对地球能量循环、水循环、碳循环及生态环境都有十分重要的意义。以南京市金川河流域为研究区，融合哨兵2号L2A数据和Landsat 8遥感数据2种数据源，分别采用偏最小二乘法(PLSR)、最小二乘-支持向量机(LS-SVM)、反向传播神经网络(BPNN)和随机森林(RF)等4种建模方法，建立遥感数据与土壤含水量之间的关系，并进行模型的验证与评价。结果表明：1)土壤含水量与哨兵2号和Landsat 8各波段反射率均呈负相关关系，和海岸带监测波段(波长为430～450 nm)和近红外波段(波长为2100～2300 nm)相关性最佳;2)融合后的遥感数据相较于单一遥感数据源，预测土壤含水量的能力更佳，最优模型R2达0.996，均方根误差仅为0.003g/g;3)4种建模方法中，建模效果从好到差依次为PLSR，RF ，LS-SVM，BPNN。融合哨兵2号L2A和Landsat 8数据，结合PLSR建模方法可进行土壤含水量的精准反演，相较于现有研究反演精度大大提升，对研究该地区地表与地下水循环和生态环境治理有一定参考价值。

　　关键词：土壤含水量;多源遥感数据;融合;反演;PLSR

　　Inverse estimation of soil moisture based on multi-source remote sensing data fusion

　　DENG Chao1，SU Nan1, PAN Xiaoting2,MA Lili2, LIN Wei1

　　(1.Nanjing Research Institute of Hydrology and Water Conservation Automation, Ministry of Water Resources, Nanjing 210012, China;

　　2.Bureau of Ecology and Environment of Yangzhou, Yangzhou 225000,China)

　　Abstract： As one of the important parameters, soil moisture is of great significance to affect the earth’s energy cycle, water cycle, carbon cycle and ecological environment. Taking the Jinchuan River Basin in Nanjing as the case study, the remote sensing data form Sentinel 2 L2A and Landsat 8 are gathered and fused. The relationships between remote sensing data and soil water content are established respectively by using four modeling method including Partial Least Squares Regression(PLSR), Least Square Support Vector Machine (LS-SVM), Back Propagation Neural Network (BPNN) and Random Forest (RF). And moreover, the relationships are verified and evaluated. The results show that:1) There is a negative correlation between soil water content and the reflectance of Sentinel 2 and Landsat 8,but it is best correlated with coastal zone monitoring band(wavelength 430~450 nm) and near infrared band (wavelength 2100~2300 nm); 2) Compared with the single remote sensing data source, the fusion data has a better ability to predict soil water content, and R2 of the optimal model is 0.996, and the RMSE is only 0.003g/g; 3) Among the four methods, the performance from good to bad is PLSR, RF, LS-SVM, BPNN. Combined with PLSR method, precise inversion estimation of soil moisture content can be carried out, and the inversion accuracy is greatly improved. It can provide more information for the study of surface and groundwater circulation and ecological environment management in this area.

　　Key words：soil moisture;multi-source remote sensing data;fusion;inverse estimation;PLSR