Soil Salinity Changes And Its Hyperspectral Response Of Soil Profiles In The Western Lakeside Oasis In Bosten Lake

Profile soils with frequent energy migration and material conversion,could track changes in soil salinity information.Therefore,the monitoring of soil salinity should pay attention to the change of salt in soil profile,which was essential for salt damage prevention,saline soil utilization and sustainable agricultural development.The spectral reflectance of soil was an apparent comprehensive reflection of the soil nature,which could effectively monitor the dynamic changes of soil salinity in a short period of time.As the lifeline of the spectral database,near-Earth spectral measurement data was the premise of established quantitative parameter statistics and mechanism estimation models,and it was also the basis for the joint application of“air-ground”with aerospace and aerial remote sensing data.How to extracted effective salt characteristic response information from soil hyperspectral data was the key to monitored soil salinity quickly and accurately.This study selected the western lakeside oasis of Bosten Lake as the research target area,and explored the seasonal variation,profile change and spatial distribution characteristics of soil salinity through field soil sample collectioned and laboratory salt determinationed,characterized the difference in soil salinity reflectance characteristics of different salinity profiles in dry and wet seasons by spectroscopy analysis,salt characteristic response spectrum index of single-band and index of two-band combination were constructed by algorithm of mathematical amplification.Finally,hyperspectral estimation model of soil salinity based on partial least squares regression(PLSR)and support vector machine regression(SVR)were established,and the model accuracy was verified by independent samples.The main conclusions were as follows:(1)The average soil salinity in the dry and wet seasons of the study area mainly concentrated in the surface layer,and gradually decreased with the increase of soil depth.Average salt content of 0?10 cm layer in autumn was 2.66 g·kg^-1,and it increaseed to 4.27 g·kg^-1 in spring,which was 1.95 and 2.32 times of the salt content of 50?100 cm layer,and the soil salinity of each layer in the spring profile was higher than that in the autumn,especially in the 10?20cm layer,and the salt accumulation rate was 80.81%.The soil salt profile types were mainly divided into three types:surface aggregated profile,oscillating profile and equably distribution profile.The SSC of each layer of the soil profile was mainly controlled by characteristic ions:Cl^-,Na⁺+K⁺,SO₄^2-.In topsoil,the salt composition were mainly sulfate-chloride type and sulfate type,and the salinization grade were mainly mild and moderate salinization.From the edge of the internal farmland and the lower reaches of the Kaidu River to the northern and southern parts of the lakeside wetland,the contents of salinity,SO₄^2-,C^l-,and Na⁺+K⁺gradually increased,and the spatial distribution of each characteristic ion was more consistent in autumn,but the difference was greater in spring.(2)Soil spectral curves of different salt profile types in the dry and wet season were basically the same in morphology and variation trend.The continuum removal spectra had distinct and different depth absorption valleys at 415 nm,510 nm,690 nm,910 nm,1140 nm,1400 nm,1800 nm,1910 nm,2200 nm,and 2330 nm.The average reflectance of spring soil increased with salt content was significantly higher than that of autumn,while the average coefficient of variation was reversed,especially at590nm,800nm,1810nm and 2150nm.Distribution of the soil average spectral curve in each layer of surface aggregated profile was the most obvious,as the depth of the soil layer increased,the reflectivity decreased.Soil average spectral curve in each layer of oscillating profile was centered and irregularly arranged.And soil average spectral curve in each layer of the equably distribution profile was the most concentrated.(3)Among the 17 kinds of mathematically transform spectra,the correlation between spectral reflectance and soil salinity was significantly increased by first-order differential spectrum,square root first-order differential spectrum,logarithmic first-order differential spectrum and reciprocal first-order differential spectrum.The best sensitive band were located at 1230 nm,1230 nm,1083 nm and 2307 nm,respectively,and the maximum correlation coefficient R was up to 0.63,(p<0.01).At0.01 significant level,the DSI index characteristic response spectrum region was:X:1760?1800nm,Y:1720?1810nm,the correlation coefficient R² highest value DSI(1780,1765)reached 0.54;RSI index characteristic response spectrum region was:X:1760?1810nm and Y:1720?1790nm range,the R² highest value RSI(1785,1756)was 0.48;NDSI index characteristic response spectrum region was mainly located at:X:1760?1810nm and Y:1720?1790nm range,X:In the range of 1960?2000nm and Y:1230?1330nm,the R² highest value NDSI(1780,1742)was 0.57.(4)Compared with the linear modeling method of PLSR,the nonlinear soil salt content prediction model established by SVR method had higher precision.Based on the logarithmic first-order differential spectral significant feature band and the grid search(GS)method for parameter optimization,the constructed salt SVR model had good precision and strong stability.The R_V²of model was 0.85,RMSEV was 0.59g·kg^-1,and RPD was 2.49.Based on the significant characteristic band combination of normalized exponential spectrum and the optimization of GS method parameters,the constructed salt SVR model had the highest accuracy and the strongest stability,the prediction accuracy index R_V² was as high as 0.86,RMSEV was 0.55 g·kg^-1,and RPD was 2.61.