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

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 VH 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 the inversion results for single polarization. For the various combinations, the inversion with the highest precision model was that for the combination of VV with NDVI. In summary, the combination of VV and auxiliary variable NDVI comprehensively reflected soil water content in complex surface environments. The goodness of fit (R²) of VV polarization combined with NDVI was 84% and the calculated root mean squared error was 2.07. In comparison with VV polarization, the retrieval accuracy improved by 8.8% and the calculated root mean square error decreased by 2.704. The combination of VV polarization with NDIV had more advantages for the inversion of soil water content for the regions with middle vegetation cover. The application of combined VV polarization with NDIV increased the potential and effectiveness of Sentinel-1A c-band synthetic aperture radar in areal study of soil water content.