Response Of Temporal And Spatial Changes Of Soil Salinity To Water-saving Transformation In Shen-Wu-irrigated Area

Taking the implementation site of water-saving reconstruction project and the inter-city transfer pilot of water rights union of the Yellow River Trunk in Inner Mongolia--Shen-Wu-irrigated area of Hetao Irrigation District of Inner Mongolia as the research area,and taking the reduction of yellow water diversion amount of irrigation area and the reduction of canal system leakage after the completion of water-saving reconstruction project as the research background,the spatial distribution of soil salinity and the change of local land use type combined with remote sensing technology and spatial interpolation technology,Soil salinization and land use types in typical irrigation areas of Hetao irrigation area were investigated,as well as the spatio-temporal distribution characteristics and change rules of salinized soil standard grades under different land use types.Quantitative inversion and precision evaluation of spatio-temporal distribution of soil salinization in irrigated land were carried out.The following conclusions are drawn from the research:(1)Before the water-saving reform,the mean range of interannual soil salt content in Shen-Wu-irrigated area was 2.97～3.48g/kg,and after the water-saving reform was1.36～3.01g/kg,which decreased by 32.25 % year on year.After the reduction of the amount of yellow water diversion and the influence of the canal lining on the leakage of the canal system,the soil salt content after the water-saving transformation has an obvious decreasing trend compared with that before the transformation.(2)Before the water-saving reform,the average salt content of soil in the ploughing layer before spring sowing was between 3.38 and 3.85g/kg,the growth period was between 2.73 and 2.85g/kg,and the autumn harvest was between 3.09 and 3.42g/kg.After the water-saving reform,the average salt content of topsoil in the same period was between 1.64～2.91g/kg,1.31～2.92g/kg and 0.84～2.63g/kg.Compared with the same period before the water-saving transformation,the water saving transformation decreased by 51.47 %,51.96 % and 72.82 %,respectively.The coefficient of variation of soil salinity in surface layer,surface layer and deep layer before water-saving transformation was basically moderate variation before spring sowing,growth period and autumn harvest,while it was strong variation after water-saving transformation.Therefore,water-saving transformation had a great impact on soil salinity in irrigation field,and the spatial-temporal distribution of soil salinity in irrigation field became more complicated.(3)Before and after the water-saving reform,the cultivated land area of irrigation area increased by 149.93km2,the grassland area decreased from 611.87km2 in 2016 to480.36km2 in 21,and the wasteland area decreased by 22.83km2.After the completion of the water-saving renovation project,the salinization degree of soil in irrigation area was improved,and so was the soil environment of salt wasteland(4)The overall change trend of soil salinization standard grade of each land use type in irrigation area was from moderate to mild soil salinization to non-salinization.After the completion of the water-saving renovation project,due to the accumulation of salt in the low-lying parts of the local topography of the irrigation area,some small natural waters dried up,more severe salinization and salinization in the irrigation area appeared.(5)19 spectral indices,vegetation indices,salt indices and soil salt content commonly used were selected and calculated.After analyzing their correlation,9 indices with correlation greater than 0.55 were selected to construct soil salt inversion models,and the coefficient of determination of the multiple linear regression model was 0.518.The fitting accuracy of the multiple linear regression is not enough to support the accuracy required by the research.Therefore,the regression models of Adaboost,BP neural network,gradient lifting tree,KNN,random forest and decision tree are respectively used to construct soil salinity inversion models,compare the accuracy of multiple machine learning models,and use the data of 70 % in April 2021 for calibration.30 % of the data is used as verification set.Among various machine learning models,the random forest regression model has a higher degree of fitting,and the coefficient of determination of its verification set is up to 0.86.