Soil Salinization Inversion And Risk Assessment Based On Fractional-order Differentiation And Machine Learning

Soil salinization,as one of the forms of soil degradation,hinders the planning of agricultural land use and agricultural production and life,and slows down the urbanization process to a certain extent.To fully understand the development mechanism of regional soil salinization and to develop a set of rapid and effective quantitative monitoring and risk assessment techniques for regional soil salinization is the way to meet the above challenges,and it is also one of the hot issues of remote sensing research on soil salinization monitoring.Therefore,this study takes soil physicochemical data and optical remote sensing data as the main data sources,and tries to combine regional soil salinity remote sensing monitoring research with the advantages of fractional order differencing and machine learning applications to improve the accuracy and efficiency of salinity monitoring,and at the same time,develops a spatio-temporal soil salinity risk assessment method based on a comprehensive scoring method while combining the advantages of RS and GIS technologies.The specific findings are as follows.(1)In this study,the Sentinel-2 MSI data were used as the data source to develop a linear model for soil salinity prediction using spectral indices with a model correlation of 0.676.(2)Using HJ-1A/HIS hyperspectral data as the data source and combining the fractional order differentiation and machine learning research methods,the results show that the highest model accuracy is achieved under the 1.2 order differentiation transform.(3)The risk of soil salinization is different in the wet and dry seasons.The risk of salinization is relatively high in the wet season,while the high risk areas in the wet season turn to medium risk in the dry season as the season changes from the wet season to the dry season(May to August),and moderate risk areas(wet season)to low risk areas and no risk areas(dry season).In this study,the spatial distribution of soil salinization and the driving causes of salinization risk were considered comprehensively,and a more comprehensive analysis of soil salinization was conducted.In terms of theory,this study introduces fractional order differentiation and machine learning to enrich the research methods of salinity monitoring and analysis,and provides some theoretical basis to promote the research process of salinity monitoring in arid and semi-arid regions.In terms of application,this study constructs a model for salinity inversion analysis through algorithmic combination,which improves the accuracy of salinity monitoring and breaks through the bottleneck of low accuracy and incomprehensive monitoring parameters of single technology in saline soil monitoring.The potential driving factors of soil salinization in the study area were explored with full consideration of climate and environment,providing theoretical support for salinization in Ebinur Lake Wetland National Nature Reserve and a priori reference for monitoring and management of other salinized areas.