基于近红外光谱及 BP 神经网络分析法磁预测森林土壤有机碳含量

摘要

为快速测定森林土壤的有机碳含量，从取自小兴安岭带岭林业局东方红林场的120个土壤样品中采集350～2500 nm的土壤近红外光谱数据，对光谱做一定的预处理后，运用主成分分析法压缩提取前8个主成分，结合BP神经网络非线性方法建立土壤有机碳含量的预测模型并进行验证。结果表明，验证集的相关系数为0.78002，均方根误差为0.5002，预测集的相关系数为0.84941，均方根误差为0.4538。应用近红外光谱技术及BP神经网络非线性方法建模可以有效地预测土壤的有机碳含量，为野外大面积快速测定森林土壤碳含量提供了技术依据。%To rapidly determine forest SOC content , the spectra of 120 soils samples from Dongfanghong forest farm of Dailing Forestry Bureau located in the northeast Lesser Khingan Mountains were scanned with a vis -NIR spectrometer in the 350 to 2 500 nm after some pretreats , and the first eight principal components were compressed and gained by principal component analysis ( PCA ) . With combination of BP neural network nonlinear method , the prediction model of SOC content were established and validated .The result showed that the correlation coeffi-cient (R) and root mean square error (RMSE) of validation and test set were 0.780 02, 0.849 41 and 0.500 2, 0.453 8 respectively .In this sense , NIR technology and BP neural network nonlinear method could be a good tool in the prediction of SOC content , and could provide a feasibility to determine forest SOC content in the field widely and quickly .