基于可见/近红外漫反射光谱的土壤有机质含量估算方法研究

Rapid detection method of soil organic matter contents using visible/near infrared diffuse reflectance spectral data

为研究不同土壤颗粒粒径对可见/近红外光谱分析技术在土壤有机质含量快速检测应用中的影响,获取粒径为0.169-2 mm和〈0.169 mm的2种土壤样本（各53个）的可见/近红外光谱（325-1 075 nm）,分别建立各自的主成分-反向传播神经网络（PCA-BPNN）、最小二乘-支持向量机（LS-SVM）和偏最小二乘法（PLS）土壤有机质含量检测模型.结果表明：当土壤粒径为0.169-2 mm时,所建立模型的土壤有机质含量预测相关系数r均在0.84以上,且预测均方根误差（RMSEP）都在0.20以下;而当土壤粒径〈0.169 mm时,所建立模型的预测相关系数r均不超过0.71,而RMSEP都在0.23以上;对于相同粒径的土壤,PLS模型对土壤有机质含量的预测效果优于LS-SVM和PCA-BPNN模型.说明不同土壤颗粒粒径会显著影响可见/近红外光谱对于土壤有机质含量的预测结果.

The effect of different soil particle sizes on the visible/near infrared spectra to detect soil organic matter contents was analyzed.The visible/near infrared spectra of soil samples with the particle sizes of 0.169-2 mm and 〈0.169 mm were measured at the range of 325-1 075 nm.Then the principal component analysis-back propagation neural network（PCA-BPNN）,least squares-support vector machine（LS-SVM） and partial least squares（PLS） models were established to detect the soil organic matter contents.When the soil particle size range was from 0.169 to 2 mm,the correlation coefficients（r） of prediction of all three models were above 0.84 and root mean square errors of prediction（RMSEP） were below 0.20.When the soil particle size range was smaller than 0.169 mm,the r values of the models were below 0.71 and RMSEPs were above 0.23.When either the soil particle size range was from 0.169 to 2 mm or smaller than 0.169 mm,PLS models obtained better results than LS-SVM and PCA-BPNN models.The overall results show that the difference of soil particle size can significantly affect the prediction results of visible/near infrared spectra to detect the soil organic matter contents.