Data Assimilation Model For Satellite Remote Sensing-Based Monitoring Of Soil Moisture Content In The Hetao Irrigation Area

Soil moisture serves as a vital factor in the interconnectivity of soil-vegetationatmosphere,acting as a carrier for matter and energy circulation in the soil system.It has a direct impact on plant and crop growth,yield,and the stability of the regional ecological environment.Data assimilation provides a feasible way to achieve high-precision monitoring of continuous spatiotemporal changes in regional soil moisture.This study focuses on the Shahaotou canal in the Jiefangzha irrigation area of the Hetao Irrigation District,using GF-1 satellite remote sensing imagery as the data source.We established a quantitative relationship between remote sensing spectral reflectance and soil moisture content,constructed a soil moisture content inversion model at different depths,and selected the optimal inversion model as the observation operator.We then constructed the HYDRUS-1D soil moisture content simulation model,analyzed and calibrated the sensitivity of the model parameters,and used the calibrated HYDRUS-1D model as the model operator.Based on the data assimilation algorithms of ensemble Kalman filter and untraced Kalman filter,we constructed corresponding data assimilation models to explore the feasibility and performance of different data assimilation methods in this study.Our aim was to obtain more accurate spatiotemporal information of regional soil water content,thereby improving the monitoring precision and accuracy of soil moisture content in the Hetao irrigated area.Furthermore,we aimed to provide technical support for drought monitoring and warning,as well as water resources management and utilization in the irrigated area.The main research contents and conclusions are as follows:(1)An inversion model of soil water content in the river-loop irrigation area based on GF-1 satellite remote sensing was constructed.The optimal combination of independent variables for spectral indices at different depths was determined by the full subset screening method.Three algorithms,including multiple linear regression,BP neural network,and support vector machine,were used to construct soil moisture content inversion models at various depths.The accuracy of these different models was compared,and the optimal inversion models were selected to invert soil moisture content at different growth stages of irrigated areas.The results demonstrate that the optimal independent variable combination based on full subset screening exhibits high sensitivity to soil moisture content.The inversion accuracy and stability of the model after screening were significantly improved.Furthermore,we found that the BP neural network model had the highest accuracy among the three algorithms,with the modeling set R2adj of each depth mostly higher than 0.5,and the validation set R2adj:ranging from 0.417 to 0.633.Therefore,the BP neural network model,after screening,is the best model for retrieving farmland soil moisture.(2)A soil moisture content simulation model based on HYDRUS-1D was constructed.The model was established by setting corresponding initial conditions,boundary conditions,and discrete divisions.The sensitivity analysis of the main parameters in the model was carried out using the graph comparison method,with meteorological data and groundwater level data as input.The simulated parameters were then calibrated according to the measured soil moisture content data.The calibrated model was used to simulate soil water dynamics in irrigated areas.Results showed that the soil moisture content simulation results of the HYDRUS-1D model were sensitive to the saturated water content and shape parameter α.During the simulation period,the simulated soil moisture content fluctuated greatly at the surface depth of 0～20cm,and was relatively stable at the depth of 20～40cm and 40～60cm.In terms of simulation accuracy,there were some differences in the simulation effect of the model at different depths,with the best simulation effect observed at a depth of 20～40cm.However,the simulated values of soil water content were generally consistent with the measured values,reflecting changes in soil water content at different depths over time in the irrigated area.(3)Based on the best inversion model obtained from satellite remote sensing for soil moisture content in Hetao irrigation area as the observation operator,and the calibrated Hydraus-1D model as the model operator,this study aimed to construct different soil moisture content assimilation models using various assimilation algorithms.The model accuracy of different data assimilation algorithms was compared to determine the most effective one.The results demonstrated that the different data assimilation algorithms could considerably enhance the simulation accuracy of the numerical model of soil moisture content.Furthermore,the assimilation effect of the EnKF was found to be superior to that of the UKF.Accordingly,the accuracy of different soil water content monitoring models was found to be as follows:EnKF assimilation model>UKF filter assimilation model>remote sensing inversion model>Hydrus-1D simulation model.The results of soil moisture content at different depths,based on the EnKF assimilation algorithm,indicated that the 20～40cm depth yielded the best results,followed by the 40～60cm depth and 0～20cm depth.The RMSE,MAE,and RE at a 20～40cm depth were 0.007,0.006,and 0.025,respectively,while NER was 0.773.