Estimation of organic matter content and pH value of regional-scale tobacco-growing soils using hyperspectral imaging
To achieve accurate and rapid estimation of organic matter content and pH value of tobacco-growing soils on a regional scale,296 soil samples were collected from the tobacco-growing areas in Sichuan Province,and the partial least squares regression(PLSR)estimation models for soil organic matter content and pH value were established using hyperspectral imaging.The results showed that:1)The accuracy of the established estimation models was enhanced after preprocessing the original spectra with most of the single and combined preprocessing methods.2)There were obvious differences in the accuracy of PLSR models established by different preprocessing methods and characteristic band screening methods.Based on the combined preprocessing method of Savitzky-Golay(SG)smoothing filtering and max-min scaling(MMS),the estimation model for the soil organic matter content established by using correlation analysis(CA)to screen extremely significant characteristic bands had the best accuracy;the coefficient of determination(R2)of the model validation set was 0.758,and the residual prediction deviation(RPD)was 2.03.Based on the combined preprocessing method of the first derivative(D1)and multivariate scattering correction(MSC),the estimation model of soil pH established by using principal component analysis(PCA)to screen characteristic bands had the best accuracy;the R2,root mean square error(RMSEV),and RPD of the model validation set were 0.724,0.59,and 1.90,respectively.Therefore,it is feasible to use hyperspectral imaging technology to estimate the organic matter content and pH value of tobacco-growing soils with high accuracy on a regional scale.The organic matter content can be estimated by the SG-MMS-CA(P＜0.01)-PLSR model,and the pH value can be estimated by the D1-MSC-PCA-PLSR model.
Tobacco-growing soilHyperspectral imagingpH valueOrganic matterCombined preprocessingPartial least squares regression
国家科技期刊平台
NSTL
万方数据
