Research On Soil Moisture Inversion Based On GF

Soil moisture refers to the amount of water content in the soil,and it is an important component of soil water cycle and ecosystem.Soil moisture not only plays an important indicative role in agricultural production,ecological environment,and water resource management,but also affects the temperature changes and precipitation formation through the evaporation of soil water.Studies have shown that soil moisture exhibits significant spatial and temporal variations due to factors such as soil texture,temperature,topography,and geographical environment.Therefore,high spatiotemporal resolution monitoring of soil moisture is of great significance.Traditional soil moisture monitoring methods usually use point-based ground observation or manual sampling methods,which can obtain high-precision soil moisture information but have poor spatiotemporal continuity and require a lot of manpower,material and financial resources.Although the existing soil moisture products obtained based on remote sensing technology have advantages such as global coverage and temporal continuity,their spatiotemporal resolution is often low,which is not enough to characterize the short-term and fine spatial scale changes of soil moisture trends.As China’s first geostationary orbit remote sensing satellite,the GF4 satellite has the advantages of high temporal resolution and relatively high spatial resolution,and has been successfully applied in forest fire monitoring,flood disaster monitoring,water body extraction,etc.,but there are few reports on its application in soil moisture inversion.Therefore,in order to improve the spatiotemporal resolution of soil moisture,this paper takes Guiyang as the study area,uses GF4 optical remote sensing images as the main data source,and collaborates with temperature,precipitation,elevation and Sentinel-1 microwave data to carry out soil moisture inversion research based on GF4.The main work and conclusions of this paper are as follows:（1）To address the shortcoming that the normalized vegetation index is prone to saturation in medium-height vegetation cover areas,a linear model for soil moisture inversion in Guiyang City（hereinafter referred to as MLR＿W）was developed using multiple linear regression method with GF4 red-band reflectance(R_red),near-infrared-band reflectance(R_nir)and wide-range dynamic vegetation index（WRDVI）as explanatory variables and actual soil moisture measurements as dependent variables,and compared with the traditional The modeling accuracy and soil moisture prediction accuracy of the MLR＿W model were compared with those of the traditional exponential model and the multiple linear regression model with R_red,R_red and NDVI as explanatory variables.The results show that the MLR＿W model has higher model fitting accuracy with mean absolute error（MAE）and root mean square error（RMSE）of 5.05%and 6.38%,respectively,and the prediction accuracy is only lower than that of the MPDI exponential model（MAE and RMSE of 5.17%and 6.29%,respectively）,but because the MLR＿W model does not need to determine soil lines,it is easier to implement in practice than the SM＿M is easier to implement.（2）To address the problem of possible complex nonlinear relationships between the influencing factors and soil moisture,the RF and GAM models were constructed using Random Forest（RF）and Generalized Additive Model（GAM）,respectively R_red,R_red with NDVI,Albedo,Elevation,Precipitation and Temperature as the influencing factors.The RF and GAM soil moisture inversion models were constructed using Random Forest（RF）and Generalized Additive Model（GAM）,respectively.The experimental results showed that the ranking of the influence factors on soil moisture diagnosed by the RF and GAM models were basically the same;however,compared with the RF model,the GAM achieved a higher accuracy of soil moisture prediction,with MAE and RMSE of 4.99%and 6.27%,respectively,which were 0.29%and 0.08%lower than those of the RF model.（3）Based on the proposed GAM model,the linear/nonlinear relationships between each influencing factor and soil moisture and the interaction synergy between different influencing factors were revealed,and the results showed that:R_red,R_nirand soil moisture have linear relationships,while NDVI,Albedo,Precipitation and Temperature have complex nonlinear relationships with soil moisture;R_red-R_nir,R_red-Precipitation,R_red-Temperature,R_nir-Precipitation,R_nir-Temperature and Precipitation-Temperature six cross terms significantly affect soil moisture at the level.（4）To address the defects that optical remote sensing is easily affected by clouds and vegetation and the traditional BP neural network model is easily trapped in local optimum,we construct a synergistic optical remote sensing and microwave remote sensing with the input parameters extracted from GF4 optical remote sensing,the incidence angle,the radar backscatter coefficient of VV polarization mode（VV_σ）and the radar backscatter coefficient of VH polarization mode（VH_σ）in collaboration with the incidence angle of Sentinel-1 radar data.The inversion model of soil moisture based on GA-BP neural network optimized by genetic algorithm was constructed.The experimental results showed that compared with the Support Vector Machine（SVM）model and the traditional BP neural network model,GA-BP achieved better soil moisture prediction accuracy,with R²=0.866,MAE and RMSE of3.73%and 4.72%,respectively,between the predicted soil moisture values and the reference values,and MAE decreased by 2.22%and 2.56%,and the RMSE decreased by 3.17%and3.23%,respectively.Compared with the GAM model constructed based on optical remote sensing,the accuracy of the GA-BP neural network model was also significantly improved,and the MAE and RMSE of soil moisture prediction values were reduced by 1.26%and1.55%,respectively.