Comparison Of Salt Content Estimation Models Of Surface Salinized Soil Based On Hyper-spectral Data In The Lower Reaches Of Kaidu River Basin

Soil salt content was an important index to reflecte the soil salinization degree andstatus in tillage soil,observation of soil salt content played an important foundationrole in the research of monitoring and prediction for soil salinity. the characteristics ofsoil salt and spectral as well as the correlation of soil reflextivity to soil salt contentwere analyzed, which was based on fields investigation,soil sampled,salt contenttested and soil hyper-spectral data measured.the estimated model of surface soil saltcontent in study area was built by using Multiple linear regression,Partial leastsquares （PLSR） and BP neural network,after model comparison,the better estimationmodels of all kinds of soil salinization in study area were determined.The result showed that:（1） The statistical characteristics of the surface saline alkali soil salt content in thelower reaches of Kaidu River Basin,which was taken on as the object of theresearch,was analyzed;the types of soil salinity were classified as chloride-sulfate,sulfate, sulfate-chloride by calculated the milligram equivalent of Cl^-to SO₄^2-instandard of soil salinity classsiction in Xinjiang irrigation. thought the soil salinizationclassification was counted, the type of sulfate was54.09%, chloride-sulfate was42.11%, the type of sulfate-chloride was only3.80%.（2）There was stronger multicollinearity among the spectral data,which made itdifficult to establish stable regression models.the result showed that the sensitivity andcorrelation of soil reflextivity to soil salt content were improved by differentialtransformation of the first derivative and second derivative about the hyper-spectraldata.（3）Diagnostic spectrums sensitive between the soil salt content and thehyper-spectral data was mainly loaded around350¹330nm,1430¹830nm and1830²400nm.the iagnostic spectrums sensitive of the chloride-sulfate salinizationwas mainly loaded around350⁹50nm,1100¹330nm and1330nm¹850nm;theiagnostic spectrums sensitive of Sulfate salinization was mainly loaded around350¹300nm,1430¹750nm and1750nm²400nm;the iagnostic spectrums sensitiveof Sulfate-chloride salinization was mainly loaded around350¹330nm,1430¹830nm and1830nm²400nm.（4）The Multiple linear regressions analysis,Partial least squares regression and BPneural network were used to build the soil salt estimation models by spectral data andsoil salt content. through model comparised,for the type chloride-sulfate:the first derivative spectral reflectance data was used to build PLSR soil salt contentestimation model for estimating soil salt content was better,R²=0.91,RMSE=0.33,F=39.57;for the type of sulfate:the first derivative spectral reflectance data was usedto build the multiple linear regression model for estimating the soil salt content wasbetter,R²=0.851,RMSE=0.31,F=56.75;For the type of sulfate-chloride:the firstderivative spectral reflectance was used to build the Multivariate linear regressionmodel was better for estimating soil salt content,R²=0.90,RMSE=0.22, F=95.96.