Study On Prediction Of Soil CEC In Middle Subtropical Based On Spectral Reflectance

The soil cation exchange capacity(CEC)is the basic physical and chemical properties of soil,which is the main reference of soil classification system for Argosols,Ferrisols and Ferralsol.Traditional CEC determination methods are time-consuming and laborious,however the spectral technology has the advantages of low cost,no pollution,convenient and fast.Therefore,it is of great significance to use the spectral technology to predict the soil CEC.In this paper,the target of research is that use the spectral reflectance to predict the soil CEC content,the research area is in Mid subtropical zone of South China,the research area includes transition zone of Argosols,Ferrisols and Ferralsol,in this area,we collected 501 soil samples and determined the content of CEC,organic matter content,pH value,mechanical composition and the spectral reflectance of each soil samples,we changed the spectral reflectance into 4 forms based in the pretreatment of spectral reflectance data and divided the soil samples into 4 types.Analyse the spectral feature of soil CEC and extract the characteristic band of soil CEC with 3 methods.Use the PLSR and BP neural network to build the direct and indirect forecasting model of soil CEC.The main conclusions are as follows:(1)Determined the spectral characteristic bands of soil CECRespectively use the correlation analysis,multivariate stepwise regression,and genetic algorithm based on the first derivative(R'),two order differential(R''),the logarithm of reciprocal(log(1/R))and the envelope removal of soil spectral reflectance data,obtained the characteristic spectral bands of soil CEC in a total of 154,including: 491-518 nm,767-781 nm,785-788 nm,1224-1238 nm,1413-1432 nm,1440-1445 nm,1890-1920 nm,1526nm,1994-2003 nm,2118-2130 nm,2225-2235 nm.(2)Established a direct prediction model of soil CEC with soil spectral reflectanceIn this paper,we ensured the influence factors of CEC content which included organic matter content,soil texture,soil pH and soil parent material by referring to previous research results.In order to obtain higher model accuracy,the soil sample were divided by spectral characteristics,influence factors of CEC content,soil parent material and the soil texture,on the basis of spectral transformation and sample classification,respectively using PLSR and BP neural network method to build the spectral prediction model of soil CEC with full band and feature band.(3)Established an indirect prediction model of soil CEC with soil spectral reflectanceIn order to investigate the spectral prediction of soil CEC more comprehensive,in this paper,we firstly established the prediction model of the spectrum with CEC influence factors(soil organic matter,clay and sand content),and then used the organic matter,clay,gravel and pH as input variables to establish CEC prediction model,finally achieved the purpose of using spectral reflectance to predict the soil CEC.(4)optimize the technological process of soil CEC spectral predictionContrast the prediction model between(2)and(3),concluded that using the whole band,the genetic algorithm based on the first derivative(R'),the sample classification type with spectral characteristic,the spectral reflectance data based on BP neural network method are the optimal technological process of spectral prediction of soil CEC,its R2 and RPD were respectively 0.636 and 1.78,the prediction accuracy compared with the original spectra(R2 and RPD respectively 0.371 and 1.43)have greatly improved.Offering new ideas to determine the soil CEC more fastly and efficiently for Argosols,Ferrisols,and Ferrallsols.