Inverse Modeling Study Of Salinity And Content Of Heavy Metals In Soda Saline Based On Visible-Near Infrared

Soil is an important part of the natural ecosystem and an important material basis for human survival and agricultural production.With the rapid socio-economic development,human activities,such as industrial emissions,mining,smelting,hazardous solid waste disposal,etc.,lead to various pollutants such as heavy metals entering the soil through atmospheric deposition,sewage irrigation,etc.The persistent heavy metals in the soil can be absorbed by plant tissues and enter the biosphere,and continuously enrich in the soil causing soil salinization and soil heavy metal pollution,both of which are the causes of global Both are major causes of global desertification and soil degradation,yet China has very limited arable land and food security is particularly important.The traditional method to obtain soil condition is field soil survey and analysis,which is not only time-consuming and laborious,but also can only achieve small area survey with poor representativeness,and it is difficult to achieve large area real-time dynamic monitoring.Therefore,how to quickly and accurately invert the heavy metal content of saline soils over a large area is an important research topic to ensure food security.In this paper,a sensitive waveband selection mechanism for the inversion of salinity and heavy metal content in saline soils is proposed,and a quantitative inversion model of salinity,heavy metal manganese(Mn),cobalt(Co)and iron(Fe)content and visible-near infrared spectral data of saline soils is developed.Then,three spectral indices,namely,ratio(RI),difference(DI)and normalized(NDI),were constructed based on the pre-processed spectral data,and the model training samples were determined by the correlation analysis between the spectral indices and the salinity and heavy metal contents.The partial least squares algorithm,radial basis neural network algorithm and random forest algorithm were used to model and invert the salinity and heavy metal contents of saline land;finally,the sensitive band combinations with significant correlation between the spectral indices and the contents of salinity,manganese,cobalt and iron were determined by the accuracy analysis of the gradient cyclic modeling of correlation coefficients,and the optimal inversion model of salinity and heavy metal contents of saline land was established,and the following conclusions were mainly drawn.1)The raw spectral data were preprocessed using five preprocessing methods:envelope removal,S-G smoothing,multiple scattering correction,SG smoothing followed by multiple scattering correction(SG+MSC)and multiple scattering correction followed by SG smoothing(MSC+SG).The results show that the raw spectral preprocessing algorithm based on multiple scattering correction has the best effect on spectral quality improvement in heavy metal content inversion applications,and the raw spectral preprocessing algorithm based on S-G smoothing has the best effect on spectral quality improvement in salt content inversion applications.2)Using the data processed by five spectral pretreatment methods,three spectral indices were firstly constructed as RI,DI and NDI,and then the Spearman rank correlation coefficients between heavy metal content and salinity and the spectral indices were analyzed separately,and finally the sensitivity of the spectral indices with significant correlation to salinity,Mn,Co and Fe content were determined by the accuracy analysis of the correlation coefficient equal gradient cycle modeling wave combinations.From the statistical results of sensitive band combinations selected by the optimal selection principle,it can be seen that the ratio index and normalized index have the largest contribution to the sensitive band combinations in both the heavy metal content inversion application and the salinity inversion application,which are the main data characteristics of the modeled inversion samples.3)The inversion models of salinity,Mn,Co and Fe contents of saline soils were developed based on partial least squares regression,random forest algorithm and radial basis neural network algorithm,respectively,using the sensitive waveband combinations selected by the optimal selection principle as data sources.The results show that the random forest algorithm has the highest modeling accuracy for Mn,Co and Fe content inversion applications,with R2 of 0.75,0.92 and 0.91 for Mn,Co and Fe content inversions,respectively,and the average relative accuracy of 90.83%,92.45%and 93.90%,respectively.The radial basis neural network had the highest modeling accuracy in the salinity inversion application with an R2 of 0.98 and an average relative accuracy of 99.56%for the salinity inversion of salinity.This study provides an effective method for the accurate and rapid analysis of salinity and heavy metal content in saline soils,which is of great practical significance for the management of the "double pollution" of soil salinization and heavy metal pollution.