基于NIR-PLS的土壤碳含量预测模型研究

以东北小兴安岭林区带岭林业局东方红林场的土壤为研究对象,对120个土壤样品近红外光谱做去噪、Savitzky-Golay平滑和多元散射校正预处理,利用偏最小二乘（PLS）法建立关于土壤碳含量和吸光度之间的定量分析模型,并进行模型校、验证及部分预测集样品碳含量预测.结果表明：主成分数为4时,模型最优.校正模型的决定系数R2和均方根误差（RMSE）分别为0.784和5.752;验证模型的决定系数R2和均方根误差（RMSE）分别为0.621和7.521,预测集样品的实测值和预测值的决定系数R2达到0.735,均方根误差RMSE为7.202,预测标准差SEP为10.356.应用近红外技术可以实现对小兴安岭次生林土壤碳含量的有效预测,为大面积快速测定土壤碳含量提供理论依据与技术支撑,进而为林分土壤碳循环的相关研究提供新的思路.

Predicting Soil Carbon Content Based on the Near Infrared Spectroscopy and Partial Least Squares

The forest soils were collected from Dongfanghong forest farm in Dailing Forestry Bureau of northeast Xiaoxing＇an Mountains. The near infrared spectrums of 120 soil samples were preprocessed with the methods of denoising, Savitzky-Golay, and multiplicative scatter correction （MSC）. Then, a quantitative analysis model was established based on the carbon content and the ab- sorbance by using partial least squares （PLS） method. The model was calibrated and validated, and then used to predict the carbon content of some samples of the prediction set. Results showed that when the principal component number was 4, the model was option- al. The determination coefficient （R2） and root mean square error （RMSE） were 0. 784 and 5. 752, respectively, for the calibra- tion model; The corresponding values were 0. 621 and 7. 521 , respectively, for the verification model; After prediction set were pre- dicted, the determination coefficient between measured and predicted values was 0. 735 with root mean square error and standard error of prediction of 7. 202 and 10. 356. The research showed that the application of near infrared technology can achieve effective prediction of secondary forest soil carbon content of Xiaoxing＇an Mountains and provide theoretical basis and technical support for determining soil carbon content widely and quickly and then provide a new train of thought for the relative research in soil carbon cycle of forest stand.