基于Sentinel-1/2的土壤盐分含量反演研究

本试验利用Gram-Schmidt（GS）变换将Sentinel-1A雷达影像与Sentinel-2A多光谱影像进行融合，并分析雷达影像、多光谱影像及融合影像各波段与吉林省白城市表层土壤含盐量的相关性，建立研究区土壤含盐量的反演模型，对研究区土壤含盐量进行制图。研究结果表明：Sentinel-1A 的VH、VV波段后向散射系数与研究区土壤含盐量均呈显著正相关，可用作土壤盐碱化监测的遥感数据源；合适的数学变换可以提升Sentinel-1A、Sentinel-2A及融合影像与土壤含盐量的相关性，其中，Sentinel-1A的VV波段与Sentinel-2A第5 波段融合后，其二次方变换与土壤含盐量的相关系数达到0.820；引入合适的盐分指数可以有效改善Sentinel-2A及融合影像与土壤含盐量的相关性，其中，融合影像的盐分指数(D₂D₄)/D₃与土壤含盐量相关系数达到0.889；利用融合影像及盐分指数(D₂D₄)/D₃建立的研究区土壤含盐量反演模型Y=86.260X-66.206X²-5.312，模型决定系数达到0.791，均方根误差为1.884 g·kg^-1，表明将Sentinel-1A雷达影像与Sentinel-2A多光谱影像进行融合来提升土壤含盐量反演精度的方法切实可行。

Inversion of soil salt content based on Sentinel-1/2

In this experiment, Sentinel-1A radar image and Sentinel-2A multispectral image were fused by Gram Schmidt (GS) transform, and the correlation between radar image, multispectral image and each band of fusion image and surface soil salt content in Baicheng City, Jilin Province was analyzed. The inversion model of soil salt content in the study area was established to map the soil salt content in the study area. The results showed that there was a significant positive correlation between the VH and VV band backscattering coefficients of Sentinel-1A and the soil salt content in the study area, which could be used as a remote sensing data source for land salinization monitoring. Appropriate mathematical transformation improved the correlation between Sentinel-1A, Sentinel-2A and fusion images and soil salt content. After the fusion of VV band of Sentinel-1A and the fifth band of Sentinel-2A, the correlation coefficient between quadratic transformation and soil salt content reached 0.820. The introduction of appropriate salinity index effectively improved the correlation between Sentinel-2A and fusion image and soil salt content. Among them, the correlation coefficient between salinity index (D₂D₄)/D₃ of fusion image and soil salt content was 0.889. The inversion model of soil salt content in the study area established by using the fusion image and salt index (D₂D₄)/D₃ is Y=86.260X-66.206X²-5.312. The model determination coefficient was 0.791, and the root mean square error was 1.884 g·kg^-1, indicating that the method of combining Sentinel-1A radar image and Sentinel-2A multispectral image to improve the accuracy of soil salt content inversion was feasible.