Water And Salt Information Diagnostic Model Of Salinized Soil Based On Hyperspectral Remote Sensing

The problem of salinization will cause soil degradation,crop yield reduction,ecological damage,and severely restrict agricultural development and sustainable land use in this area.Hyperspectral remote sensing technology can effectively reflect ground feature information.Therefore,hyperspectral remote sensing can be used to monitor soil water and salt information to provide a basis for the rational development of saline soil.This paper takes the Jiefangzha irrigation area in Hetao irrigation area as the research area,collects soil samples with different salinization degrees,and analyzes and obtains soil moisture content,total salt and eight main water-soluble base ions(K⁺?Ca²⁺?Na⁺?Mg²⁺?Cl^-?SO₄^2-?HCO₃^-?CO₃^2-)content and other parameters.Combined field and laboratory soil hyperspectral data(ground hyperspectral data)and Gaofen 5 hyperspectral data.Perform fractional order differentiation processing on these data and extract spectral parameters(band reflectance and spectral indices).Then use different variable screening methods and extreme learning machine models to establish soil moisture content,total salt,and eight major ions based on ground hyperspectral data and satellite hyperspectral data,and comprehensively evaluate the model based on different evaluation indicators.The main conclusions of this paper are as follows:(1)Diagnostic models of salinized soil moisture content based on fractional differential spectroscopy was constructed.The spectrum of salinized soil is affected by SMC and SSC,but the effect of SSC can be weakened by the selection of effective bands.In some cases,FOD(e.g.0.5 and 0.75 orders)can have better estimation than the integer order derivatives(i.e.,1st and 2nd orders)and the original reflection spectrum.The ELM model based on the0.75 order derivative spectrum and the random frog algorithm provides an optimal model prediction.Our study confirmed the potential of in-situ VIS-NIR spectroscopy for SMC estimation.The good applicability of the random frog algorithm in soil spectroscopy was further confirmed in this study.Variable selection algorithms(i.e.,VIP,CARS,and RFA)are capable of selecting useful spectral variables,simplifying the model,and improving the prediction accuracy.In general,RFA has the best prediction but a long calculation time.An alternative approach is the application of the VIP method due to its less required time to process the algorithm with a relatively high precision.(2)Diagnostic models of soil salinity information based on fractional differential spectral indices was constructed.The results indicated that the correlation between the contents of soil salt and soluble ions and FOD spectra increased as the derivative order increased.The optimal FOD spectrum of soil salt content(SSC),K⁺,Na⁺and SO₄^2-were the0.9-order derivative spectrum while the optimal FOD spectrum of Ca²⁺,Mg²⁺,CO₃^2-,HCO₃^-and Cl^-were 1.65,0.85,1.90,0.80 and 1.55 order derivative spectrum,respectively.The newly proposed three-band index(TBI4)showed a better correlation with data(e.g.,TBI4 of Mg²⁺had the best effect,and its R² was 0.92).In most cases,the IRIV-ELM models outperformed MCUVE-ELM and BOSS-ELM models in estimating the contents of soil salt and soluble ions than.Difference existed among the optimal prediction accuracy of the contents of soil salt and different ions:the prediction accuracy of SSC,Ca²⁺,Na⁺,Mg²⁺,Cl^-and SO₄^2-was very high while that of CO₃^2-was high and K⁺and HCO₃^2-was relatively lower.However,these models have reached and even exceeded the quantitative prediction level.Therefore,we concluded that the contents of soil salt and soluble ions could be estimated using FOD algorithm and the optimal combination of spectral indices.(3)The model for remote sensing diagnosis of soil water and salt information based on the fractional differential spectral indeices of the Gaofen-5 satellite was constructed.The low-order FOD could process GF-5 AHSI more effectively than high-order FOD,changing the image contrast,thereby increasing some features of the ground objects.0.75-order FOD could effectively improve the correlation between indices and SMC as well as SSC.Among them,PI had the best effects on SMC and SSC,and its correlation coefficient(|r|)reached0.85 and 0.70,respectively.The SMC and SSC prediction models based on the 0.75-order spectrum had achieved good prediction results,R_v² reached 0.83 and 0.74,RPD was 2.24 and1.76,and RPIQ were 1.95 and 2.06,respectively.Therefore,these two models could effectively predict the regional distribution of SMC and SSC,and thus can provide a basis for scientific irrigation and soil ecological restoration.(4)The soil salinity diagnosis model based on remote sensing data fusion of Sentinel-2and Gaofen-5 satellites was constructed.The spectral distortion of the fusion image obtained by using three fusion algorithms such as GSA,Hy Sure and CNMF is small.The fusion image based on GSA fusion algorithm has the best spectral fidelity and spatial fidelity.Combining the GSA fusion image with the 0.75-order fractional differential algorithm can obtain a prediction model with higher accuracy,and its RPIQ can reach 2.54.The distribution map of soil salt content based on GSA fusion images has a more refined distribution.Therefore,the fusion of image data solves the problem of low spatial resolution of hyperspectral satellites to a certain extent.Data with high spectral resolution and high spatial resolution characteristics provide a guarantee for the establishment of an effective soil salt model.Use the built model to draw the distribution map of soil salinity in the area,which provides a basis for soil salinization management and soil ecological restoration.