Estimating Vegetation Water Content By Combining Multi-Spectral And Radar Imageries

Vegetation is an important part of the biosphere of the hydrological cycle and provides the necessary organic matter for the survival of other organisms.Water is the main component of vegetation and one of the main factors controlling plant photosynthesis,respiration and biomass.The dynamic changes of vegetation water content(VWC)reflect many basic biophysical processes of the plant,such as vegetation temperature self-regulation,leaf transpiration,and cell expansion,which are closely related to global carbon,water,and energy cycle processes.The photosynthesis of green plants provides the energy for the ecosystem and determines the morphological structure of an ecosystem.Studying the remote sensing estimation method of VWC is of great significance for understanding hydrological cycle and maintaining ecosystem balance.The first part of this paper evaluates and improves the remote sensing estimation method of grassland VWC.Firstly,the measured values of grassland VWC in Inner Mongolia in 2017 and 2018 were used to evaluate the uncertainty of grassland VWC based on the linear model and exponential function model of normalized difference vegetation index(NDVI),and then the regression coefficient of the original model was improved;Secondly,the relationship between VWC and NDVI is reconstructed according to the type of grassland,and the estimation error(RMSE and ubRMSE)of VWC is reduced from 0.078kg/m~2 to 0.047kg/m~2.The results of this paper will help to improve the understanding of grassland VWC remote sensing estimation methods and the attribution of uncertainty.The second part of this paper analyzes and evaluates the VWC remote sensing estimation method of corn.In this paper,combined with Sentinel-1 and Sentinel-2 data,the C-band VV and VH polarization radar backscatter coefficient,NDVI,normalized differential infrared index(NDII),water stress index(MSI)and global water index(GVMI)are obtained.In the multi-spectral vegetation index,NDVI has the highest estimation accuracy(R~2=0.76,RMSE=1.20kg/m~2,ubRMSE=1.15kg/m~2,Bias=-0.33kg/m~2).In radar remote sensing,the VV and VH polarization backscatter coefficients have little difference in the estimation accuracy of VWC.The overall performance is that the error in the early stage of vegetation growth is large,and the error is reduced in the late stage of vegetation growth.When VWC is greater than 4kg/m~2,the multi-spectral vegetation index will be saturated,and cannot be used to estimate VWC.When VWC at the range of 4kg/m~2 to 6kg/m~2,the backscatter coefficients of VV and VH polarization still have a good correlation with VWC(R~2=0.46 and R~2=0.36,respectively).According to the characteristics of multi-spectral vegetation index and backscatter coefficient,this paper proposes a VWC estimation method combining NDVI and backscatter coefficient(VV or VH polarization),which is expected to solve the problem of premature saturation of multi-spectral vegetation index and the influence of soil background scattering before radar.The results show that the estimation accuracy of NDVI+VV and NDVI+VH are close.In the VWC estimation of NDVI+VV,R~2=0.81,RMSE=0.97kg/m~2,ubRMSE=0.97kg/m~2,Bias?0kg/m~2;In the VWC estimation of NDVI+VH,R~2=0.80,RMSE=0.98kg/m~2,ubRMSE=0.98kg/m~2,Bias?0kg/m~2.Meanwhile,the statistical results of VWC segmentation accuracy show that:NDVI is recommended to estimate VWC at the range of 0kg/m~2 to4kg/m~2;NDVI+VH is recommended to estimate VWC at the range of 4kg/m~2 to6kg/m~2;NDVI+VV is recommended when that is greater than 6kg/m~2.In order to analyze the VWC estimation error characteristics of the combined multi-spectral vegetation index and radar backscatter coefficient,the spectral reflectance and radar backscatter coefficient under different soil moisture,different soil surface roughness and vegetation conditions were simulated using the PROSAIL model and water cloud model(WCM),The error analysis of the optical method and the radar method in this paper are carried out respectively,and the conclusions obtained are consistent with the conclusions in this paper,explaining the reason why the combination of NDVI and backscattering coefficient improves the estimation accuracy,and proves that the conclusions of this paper are reliable.The analysis results also indicate that although the combined optical and radar remote sensing methods have improved the VWC estimation accuracy as a whole,it is recommended to use multi-spectral vegetation index or backscatter coefficient alone to estimate VWC for corn of a specific growth period.