Research On Application Of Coupling Machine Learning And Microwave Data To Monitoring Soil Water And Salt In Oasis

With the rapid development of China's economy,a large number of land resources have been developed and utilized on a large scale,resulting in a gradual increase in the area of contaminated land,threatening the stability of the ecological environment and the safety of agricultural production.Therefore,there is an urgent need to carry out soil pollution prevention and control work.Soil moisture and soil salinity are the main factors affecting the soil ecological environment in arid areas,and timely and accurate monitoring of the distribution of soil water and salt is the key to mastering the soil conditions in arid areas.This study will take the Weiku Oasis in Aksu,Xinjiang as the research object,and use the Dobson dielectric model to analyze the dielectric properties of the soil.Using Sentinel-1A microwave data as remote sensing data,the WCM model and AIEM model are coupled to construct a soil moisture retrieval model for the Weiku oasis.The combination of microwave data and machine learning algorithms is used to construct a soil salinity inversion model in the Weiku oasis to monitor soil water and salinity in arid areas.The main conclusions of this research are as follows:(1)Using the Dobson dielectric model to simulate the influence of soil volumetric water content,soil salinity,temperature,frequency,and soil bulk density on the dielectric constant,the study found that the soil dielectric constant is affected by many factors,but the soil water and salt The content has the most significant influence on the dielectric constant.The real part of the dielectric constant is mainly controlled by the soil water content,and the real part of the dielectric constant is proportional to the volumetric water content.The imaginary part of the dielectric constant acts together with soil water and salt,and the imaginary part of the dielectric constant shows an exponential decrease trend with the increase of volumetric water content,which is proportional to the salt content of the soil.The simulated dielectric constant is fitted with the measured dielectric constant,and the fitting result is good,indicating that the Dobson dielectric model is suitable for this study area.(2)Combine the AIEM model with the WCM model.First,use the AIEM model to simulate the relationship between the backscattering coefficient and the soil moisture and roughness under the conditions of different root mean squares,correlation lengths,and water content,and find the relationship between the backscattering coefficient and the soil There is a logarithmic relationship between moisture and roughness.Secondly,the WCM model is used to eliminate the contribution of vegetation to the backscattering coefficient to obtain the radar backscattering coefficient of bare soil.The AIEM model and WCM model are used to construct a soil moisture inversion model,and the measured data is substituted into the soil moisture data for accuracy verification.The determination coefficient R2=0.906.The results show that this method can be used for soil moisture inversion in the study area.(3)Using the backscattering coefficients of the two polarization modes of Sentinel-1A,constructing 20 polarization combinations 2D index and soil moisture as characteristic variables,using Boruta function to screen the characteristic variables,and finding that the Boruta feature selection algorithm filters out the Characteristic variable with good correlation of soil salinity.Combine the selected characteristic variables with machine learning methods to establish a soil salinity model.Through the comparison of model evaluation indicators,it is found that the five models established for soil salinity have good inversion accuracy.Among them,the rpart model has the lowest accuracy and the Cubist model has the highest accuracy.Its verification set R2=0.822,RMSE=3.064,indicating that the use of radar backwards The combination of scattering coefficients and machine learning algorithms can achieve accurate inversion of surface soil salinity,and can effectively monitor the distribution of salinization in arid areas.According to the inversion results,in the study area,non-saline soil accounted for72.68%,mild saline soil accounted for 9.75%,moderate saline soil accounted for13.07%,severe saline soil accounted for 4.21%,and extremely moderate saline soil accounted for 0.27%,the soil salinity showed a gradual increase trend from the inside to the periphery of the oasis.In summary,this study mainly uses microwave remote sensing to couple multiple models to build a soil moisture retrieval model in the study area,and combines microwave radar data with machine learning algorithms to retrieve soil salinity in the Weiku oasis.The two models established in this study have achieved good inversion accuracy,and can monitor soil water and salt in arid areas,so as to carry out targeted soil ecological environmental protection work in the later period,so as to better protect the arid area ecology Stable environment and sustainable development of agriculture and economy will help revitalize rural areas and build a beautiful Xinjiang.