基于Sentinel-1A与Landsat 8数据的北黑高速沿线地表土壤水分遥感反演方法研究  

Soil water content retrieval based on Sentinel-1A and Landsat 8 image for Bei'an-Heihe Expressway

作　　者：曾旭婧[1] 邢艳秋[1] 单炜[2] 张毅[3] 王长青[1] ZENG Xujing XING Yanqlu SHAN Wei ZHANG Yi WANG Changqing(Center for Forest Operations and Environment, Northeast Forestry University, Harbin 150040, China School of Civil Engineering, Northeast Forestry University, Harbin 150040, China College of Resources, Sichuan Agricultural University, Chengdu 611130, China)

机构地区：[1]东北林业大学森林作业与环境研究中心,哈尔滨150040 [2]东北林业大学土木工程学院,哈尔滨150040 [3]四川农业大学资源学院,成都611130

出　　处：《中国生态农业学报》2017年第1期118-126,共9页Chinese Journal of Eco-Agriculture

摘　　要：土壤含水量是影响水文和气候变化的基本参数,研究土壤含水量分布,对气候变化、水资源分布、农作物估产等有着重要的现实意义和科学价值。本文以2015年6月21日的Sentinel-1A(哨兵1号)双极化合成孔径雷达影像为基础,结合同时段辅助光学影像Landsat 8,对北安—黑河高速沿线地区不同植被覆盖程度下复杂地表土壤含水量进行反演研究,探讨不同极化组合方式在不同土地利用方式下的土壤水分含量反演结果。结果表明:VH极化及VH与辅助变量NDVI(Normalized Difference Vegetation Index)组合反演精度分别为52.1%和53.6%,整体效果并不理想。VV极化(VV Polarization)图像和双极化VV/VH(VH Polarization)组合在裸露和低植被地区反演更具有优势,其精度分别为75.4%和59.5%,而在高植被覆盖度地区并不适用。VH极化反演结果中耕地土壤含水量比实际值低9.37%,VV极化在低植被区域土壤含水量比实际值低10.45%,在灌木及耕地地区VV/VH反演结果精度比单极化及其组合反演结果低,最高精度模型的反演是VV结合NDVI。VV与辅助变量NDVI结合能综合反映复杂地表环境下土壤含水量,其精度达84%,标准误差RMSE为2.07,比VV极化反演精度提高8.8%,RMSE比VV极化降低2.704。VV与辅助变量NDVI组合方式在中等植被覆盖地区土壤含水量反演更具有优势,并能够更好地发挥哨兵1号C波段合成孔径雷达在土壤水分研究中的潜力与有效性。Soil water content is one of the basic parameters that affect hydrological variability and climate change. It has important practical significance and scientific value for climate change, water resources and estimation of crop yield to study the distribution of soil water content. To probe new ways of soil water content retrieval in complex vegetation coverage area,this study analyzed soil water contents of complex surfaces with various degrees of vegetation cover along Bei＇an-Heihe Expressway using images from Sentinel-1A dual-polarization Synthetic Aperture Radar （SAR） for 21 June 2015. Also Landsat 8 images were integrated as assisted optical image for the same satellite transit time. Then the results of the inversion of soil water content under different land use types and polarization combinations were discussed. Backscattering coefficients of different polarization modes were extracted using the water cloud model. Support vector regression algorithm was used to estimate surface soil water content based on the soil inversion parameters. The applicability of different polarizations in the retrieval of soil water content on complex surface was also discussed. The results showed that VH polarization retrieval accuracy was 52.11%, while combined VII polarization with normalized difference vegetation index （NDIV） retrieval accuracy was only 53.6%. This was not satisfactory for the vegetation zone. VV polarization and dual polarization ratio of VV/VH images were very sensitive to bare land and low vegetation cover land, for which retrieval accuracies were respectively 75.4% and 59.5%. These methods were, however, not applicable in areas with moderate or high vegetation cover. The results of VH polarization inversion for arable lands soil water content was 9.37% lower than the measured value. Also the inversion value of VV polarization for areas with low bush was 10.45% lower than the measured value. The inversion results for dual polarization ratio of VV/VH in shrub and arable lands were not as good as t

关 键 词：Sentinel-1A 土壤含水量 后向散射系数 极化方式 支持向量回归 

分 类 号：TP79[自动化与计算机技术—检测技术与自动化装置] S152.7[自动化与计算机技术—控制科学与工程]