Spatial Variability And Spectral Estimation Of Salinity And Organic Matter In Salinized Soil

At present,due to the comprehensive effects of various factors,the phenomenon of soil salinization in Xinjiang is becoming more and more serious.Unreasonable development and utilization and disturbance of human activities may lead to lack of soil nutrients and decreased fertility.Therefore,the development of soil management green agriculture has become the current an important mission.As an emerging method for estimating soil composition,the visible-near-infrared spectroscopy has the advantages of high accuracy,convenience and rapid-fire,time saving,manpower and material saving,etc.It has been widely used in the acquisition of soil composition information,exploration and reconnaissance,digital mapping,fine farming and intelligent agricultural development.In order to clarify the spatial distribution characteristics of soil salinity and organic matter under the influence of human activities,and the spatial variability characteristics of soil salinity and organic matter at different depths,seek the optimal spatial interpolation method of soil organic matter under different disturbance degrees,establish the prediction model of saline soil with high accuracy and good stability.In this paper,the typical salinized soil in Fukang City,Xinjiang,was taken as the research object,and soil sample and spectral information were collected in the field.Based on GIS and geostatistics,descriptive statistics,correlation and variation characteristics of soil organic matter were analyzed.Spatial interpolation analysis of soil organic matter under different human activities was carried out by using ordinary Kriging method,inverse distance weight method,radial basis function method,local polynomial method.Based on field measured VIS-NIR spectral data,multiple linear regression(MLR),support vector machine(SVM)and random forest(RF)models were used to estimate the soil salinity of each area.The spatial distribution and spatial variability of soil salinity and organic matter under different disturbance levels and different depths in this area,as well as the response characteristics of surface soil salinity spectrum were analyzed.The optimum spatial interpolation method of soil organic matter content and the optimum spectral estimation method of soil salt content under different disturbance degrees were discussed.The research has a certain significance for the prevention and restoration of local salinization and the sustainable development of ecological environment.At the same time,it provides a theoretical basis for improving the soil fertility status and rational development and utilization of salinized land in this area.The main results are as follows:(1)Soil salinity increased with the increase of human disturbance activities,while organic matter content decreased gradually.The coefficient of variation of soil salinity and organic matter content varies from 15.45%to 60.19%in each disturbance degree,which belonged to medium variation.At the same time,the stronger the disturbance degree,the lower the correlation between soil salt and organic matter content,that is,area A>area B>area C,the correlation coefficients are:-0.614,-0.412,0.043;With the increase of human disturbance activities,soil salinity and organic matter were more and more strongly affected by random factors.The higher the spatial variability caused by spatial autocorrelation,the higher the base ratio of soil salinity and organic matter was 12.5%and 9.68%respectively in the unmanned disturbance area,which had a strong spatial correlation.(2)With the increase of depth,soil salinity increased first and then decreased.The maximum value appeared at the depth of 40~60 cm,and the content of organic matter was 42.38g/kg,while the content of organic matter decreased gradually,and the salinity and organic matter in each layer belonged to medium variation.There was a significant negative correlation between soil salinity and organic matter in the surface layer(0~20cm)and subsurface layer(20~40cm),and a significant positive correlation in the latter three layers.The ratio of block to base of soil salinity ranged from 1.23%to 23.42%,which had strong autocorrelation.The organic matter content of surface and subsurface layers was moderate autocorrelation,and the latter three layers had strong spatial correlation.(3)With the increase of disturbance degree,the content of soil organic matter decreased gradually,the spatial variability changed from weak to moderate,and the spatial correlation changed from strong to weak,that is,the stronger the disturbance degree was,the greater the soil organic matter was affected by random factors.At the same time,with the increase of human activities,the accuracy of four spatial interpolation methods for predicting organic matter was decreasing.The OK method was more suitable for the spatial prediction of organic matter in the undisturbed area with strong spatial structure,and its determinant coefficient R~2 was 0.625.The RBF method was more accurate for the spatial interpolation analysis of soil organic matter in the area of human disturbance and severe human disturbance.The prediction accuracy R~2 was 0.562 and 0.434,respectively.(4)Comprehensive spectral response and correlation analysis determined the significant characteristic bands of soil salinity,the area A was 492 nm,526 nm,532nm,the area B was 919 nm,2055 nm,650 nm,and the area C was 467 nm,876 nm,2197 nm.Regardless of the degree of disturbance,the fitting effect of the three models from high to low was RF>SVM>MLR.The soil salt content model of three different regions established by RF method had Rc~2 ranged from 0.750 to 0.896,Rv~2ranged from 0.735 to 0.831,and RPD ranged from 1.765 to 2.097,which was obviously better than the other two methods in modeling accuracy and could accurately predict soil salinity.Regardless of the modeling method,the prediction accuracy was the highest in the area A,with RPD ranging from 1.796 to 2.097,followed by the area B with RPD ranging from 1.713 to 2.034,and the salinity quantitative estimation accuracy in the area C with RPD ranging from 1.597 to 1.802,it was slightly lower.