Based on sCARS Hyperspectral Modeling of Soil Organic Matter Content in Huaibei Plain
In order to determine the best inversion model of soil organic matter (SOM) of sand and Shajiang black soil in the Huaibei Plain,the best characteristic wavelength screening method was explored to improve the prediction accuracy of the model. The original spectrum was used to perform inverse-log reflectance(Log(1/R)),standard normal variable(SNV),continuum removal(CR),and first-order derivative reflectance(FDR) processing,and the stability competitive adaptive reweighted sampling(sCARS)was used to screen the charac-teristic variables,and the results obtained by the competitive adaptive reweighted sampling(CARS),correlation coefficient method(|r|≥0.47)and significance level method(p≤0.01)were compared and analyzed,and a partial least squares regression(PLSR)model of SOM content was established. And compare the accuracy differences. The results show that:(1)In the whole band,the SOM content is negatively correlated with the original spec-trum,positively correlated with the Log(1/R)spectrum,and significantly enhanced with the SNV spectrum. CR and FDR spectra showed different degrees of positive and negative correlation with SOM content.(2)Compared with the full band,the CARS and sCARS algorithms can effectively remove the spectral redundancy informa-tion,and the number of characteristic bands screened accounts for only 1％～5％of the full band. After screening,the accuracy of the model was higher,and the relative percent deviation(RPD)was greater than 1.8.(3)Com-pared with the CARS algorithm,the sCARS algorithm has better stability and accuracy. The screened characteris-tic bands are mainly distributed in the 800～850,1850～1900,2050～2500nm regions.(4)The Log(1/R)-sCARS model has the best accuracy,the determination coefficients(R2)of the modeling set and the prediction set are in-creased by 0.08 and 0.28,respectively,and the RPD value is 3.05,which is excellent for the prediction of SOM content.
soil organic matterShajiang black soilspectral transformationssCARS screeningpartial least squares model
NETL
NSTL
万方数据
