Research On Soil Moisture Retrieval Of Crop Covering Area Based On Gaofen3 Radar Data

At present,the soil moisture inversion based on radar satellites is mainly Radarsat-2 and sentinel-1,there are few soil moisture inversions based on domestic GF3 data.Therefore,based on the GF3 radar image,Landsat8 optical image and measured parameters,combined with the PROSAIL,AIEM and water cloud models,taking wheat in the jointing stage of Luancheng County as the research object.The soil moisture estimation formulas of high and low incidence angles of GF3 data and soil moisture inversion model based on PSO-ELM improved by particle swarm optimization algorithm were constructed.There are three main aspects:?1?The soil direct backscatter coefficient of the GF3 co-polarized data was obtained based on the water cloud model.First,based on the PROSAIL model,measured parameters and Landsat8 images,the best index(Normalized Difference Water Index NDWI₁₆₄₀)was selected from the eight commonly used vegetation moisture indexes;Then,based on the AIEM model,measured surface parameters and GF3 original image,the A and B values in the water cloud model were obtained,and the vegetation canopy water content from landsat8 and the GF3 original image were used to remove the effect of wheat,and the GF3 soil directly Backscatter coefficient(VV_soil?HH_soil)was obtained.Finally,the correlation shows that the VV_soil?HH_soil are consistent with the soil direct backscattering coefficients of the AIEM model(VV_AIEM?HH_AIEM),where R² of the VV polarization is 0.8427 and R² of the HH polarization is 0.8216.?2?Soil moisture inversion of GF3 high and low incidence angle data based on AIEM model.The backscattering coefficient(VV_AIEM?HH_AIEM)of various factors in different ranges of various parameters are characterized through the AIEM model.After studying the correlation between various factors and the backscattering coefficient,the soil moisture estimation formula was developed.Combined with 49°and 29°soil direct backscattering coefficient images of GF3(VV_soil?HH_soil),the combined roughness sl characterizing the surface relief was got to obtain soil moisture.It can be known from the accuracy verification that both VV_soil and HH_soil can better invert soil moisture.Among them,the accuracy of VV_soil is higher,R²=0.5956,RMSE=0.0415m³m^-3,and the accuracy of HH_soil is lower,R²=0.4737,RMSE=0.0538m³m^-3.?3?Using optimized extreme learning machine algorithm?PSO-ELM?to obtain soil moisture.The global optimal value found by the particle swarm optimization algorithm is used to generate the input weight and offset of the extreme learning machine to make up for the randomness of the extreme learning machine.The error analysis of 100 iterations of VV and HH polarization shows the training effect of VV polarization is better than HH;meanwhile,the predicted conclusion of the measured soil moisture shows that VV_soil is superior to HH_soil,where R²=0.6436,RMSE=0.0456 m³m^-3.Therefore,the soil moisture information was finally obtained based on the VV_soil data,and according to the inversion results:the soil moisture inversion value obtained through the VV_soil has a good correlation with the actual measurement value,where R²=0.4556,RMSE=0.0568 m³m^-3.