Research On Method Of Satellite Remote Sensing Monitoring Soil Salinization Based On Data Assimilation

Soil salinization is an important reason for restricting agricultural production in Hetao Irrigation Area.Accurate acquisition of soil salinity information is of great significance for soil salinization control and agricultural irrigation.Satellite remote sensing can monitor soil salinity in large area,accurately and in real time.Using means of data assimilation to introduce satellite remote sensing data into soil salinity transport model,which can more accurately monitor the spatial and temporal dynamics of soil salinity at a regional scale.Therefore,the study object was soil salinity at different depths in the Shahaoqu Irrigation Area.The remote sensing images of GF-1 satellite were uesd to obtain spectral reflectances and spectral indices were calculated.Correlation analysis and gray correlation method were used to screen and construct combinations of spectral indices.Soil salinity inversion models were established to obtain multi-temporal remote sensing inversion values of soil salinity at different depths.The HYDRUS-1D model was used to simulate transport of soil moisture and soil salinity at different depths,and simulation results were calibrated and verified.Assimilation algorithms of En KF and UKF were used to introduce remote sensing inversion values of soil salinity into the HYDRUS-1D model.The data assimilation schemes of soil salinity at different depths were constructed and sensitivity analysis were carried out.Model accuracies were compared and analyzed,and the best data assimilation scheme was determineded to monitor soil salinity in regional area.The main conclusions of this study were as follows:(1)Based on the remote sensing images of GF-1 satellite,a time serise inversion models of soil salinity were constructed at different depths.The inversion models based on gray correlation method had higher accuracy than the inversion models based on correlation analysis.Spectral indices were screened based on gray correlation method,and the inversion models were constructed by using PLSR models,SR models and RR models.It was found that RR models had the highest accuracy,followed by PLSR models,and the worst inversion accuracy was SR models.And inversion accuracies of models decreased with increase of depth.Therefore,RR inversion models based on gray correlation method were used as the observation operator of the data assimilation scheme of soil salinity.(2)The HYDRUS-1D model based on the simulation of dynamic transport of soil moisture and soil salinity was constructed.At three depths,the RMSEs of soil moisture and soil salinity simulation were less than 0.03,and good simulation results were obtained.Measured values of soil moisture and soil salinity were closed to simulated values.And accuracies of soil moisture and soil salinity simulation results decreased with increase of depth,and the simulation effect was best at depth of 0-20 cm.At the same time,soil moisture and soil salinity at depth of 0-20 cm fluctuated greatly due to external conditions such as irrigation,rainfall and evaporation,while fluctuations of soil moisture and soil salinity at depth of 20-40 cm and 40-60 cm were relatively small.Therefore,it was feasible to use the HYDRUS-1D model to transport of soil moisture and soil salinity.(3)Data assimilation schemes of En KF and UKF based on soil salinity were constructed at different depths.According to results of sensitivity analysis,it was found that the optimal ensemble numbers of En KF data assimilation scheme were 50,and its assimilation accuracies decreased with increase of observation error and model error,while assimilation accuraces of UKF data assimilation scheme also decreased with increase of observation error and model error.At a single point scale,En KF assimilation values were the highest accuracy,followed by UKF assimilation values,the third was inversion values,and the worst accuracy was simulation values.And their accuracies decreased with increase of depth.At a regional scale,accuracies of En KF assimilation values and UKF assimilation values were higher.Their r were above 0.8 and NER were above 0.5,which were better than accuracies of simulation values and inversion values.And accuracies of En KF assimilation values were higher than that of UKF assimilation values.At three depths,the best assimilation effect was depth of 0-20 cm,followed by depth of 20-40 cm,and the worst assimilation effect was depth of 40-60 cm.Therefore,En KF data assimilation scheme was used to monitor soil salinity at different depths at a regional scale,which can more accurately obtain the temporal and spatial information of soil salinity at different depths at a regional scale.