近红外光谱法测定土壤全氮和碱解氮含量

Prediction models of total and available soil nitrogen based on near-infrared spectroscopy

为探寻采用近红外光谱技术在野外快速测定土壤全氮和碱解氮含量的方法,采集土壤光谱信号,结合偏最小二乘法和主成分分析法,分别建立土壤全氮和碱解氮含量测定的定标模型。结果表明,采用PLS方法建模时,土壤全氮和碱解氮含量测定定标模型的精度较高。为提高模型的预测精度,采用多元散射校正、标准归一化、基线校正、卷积平滑和小波变换5种方法对光谱信号进行预处理,当用小波变换法对光谱信号进行去噪处理,并与PLS方法结合时,模型的预测精度最高,土壤全氮样品校正模型的相关系数为0.838 5,均方根误差为0.153 1,对应验证模型的相关系数为0.754 9,均方根误差为0.184 2,校正集和验证集土壤全氮含量预测值(y)与实测值(x)之间的关系模型分别为y=0.685 8x+0.198 0和y=0.621 4x+0.237 9;土壤碱解氮样品校正模型的相关系数为0.866 5,均方根误差为0.007 7,对应验证模型的相关系数为0.796 1,均方根误差为0.009 4,校正集和验证集土壤碱解氮含量预测值(y)与实测值(x)之间的关系模型分别为y=0.749 8x+0.019 4和y=0.700 7x+0.023 3。综合分析结果表明,应用近红外光谱技术对土壤全氮和碱解氮含量进行定量预测是可行的,且应用小波变换方法对光谱冗余信息进行预处理后,再与偏最小二乘法相结合可有效地提高模型的精度。

To explore the feasibility of measuring total and available soil nitrogen by using near-infrared spectroscopy in the field, the calibration models were respectively established on the basis of soil spectrum signals, as well as partial least squares method（PLS） and principal components analysis（PCA）. The results showed that the models of total and available soil nitrogen established by using PLS approach were more accurate. To further improve the precision of the models, five different pretreatment methods were adopted to process the spectrum signals, including multiplicative scatter correction, standard normalization, baseline correction, convolution, smoothing, and wavelet transformation. The highest precision model was derived from wavelet denoising combined with PLS. The correlation coefficient（R） and the root mean square error（RMSE） of the calibration model for total soil nitrogen were 0.838 5 and 0.153 1, respectively. The correlation coefficient and the root mean square error of the corresponding verification model were 0.754 9 and 0.184 2, respectively. The relationship models between the predicted and measured values of total soil nitrogen for the calibration data set and the verification data set were： y=0.685 8x＋0.198 0 and y=0.621 4x＋0.237 9, where x is the measured total soil nitrogen value, y is the predicted value of total soil nitrogen. In the calibration model of available soil nitrogen, R and RMSE were 0.866 5 and 0.007 7, respectively, and the corresponding values for the verification model were 0.796 1 and 0.009 4, respectively. The relationship models between the predicted and measured values of available soil nitrogen for the calibration data set and the verification data set were y=0.749 8x＋0.019 4 and y=0.700 7x＋0.023 3,where x is the measured available soil nitrogen value, y is the predicted value of available soil nitrogen. Therefore, it is feasible to apply NIR spectroscopy technology in quantitative determination of total and available soil nitrogen, and the wavelet transformation preprocessing method in combination with PLS can effectively improve the accuracy of the prediction models.