便捷式近红外光谱仪在土壤养分中的预测研究

Study on Prediction of Soil Nutrients by Portable Near Infrared Spectrometer

采集位于云南省昆明、安宁、弥勒3个地区的350份土壤样品,利用便携式近红外光谱仪进行光谱的扫描并构建全氮、全钾、全磷和有机质4项养分的近红外预测模型。结果表明,在950~1 650 nm,不同地区的土壤样品光谱的轮廓较为接近;全氮、全磷、有机质的最佳预处理方法为一阶导数,全钾的最佳预处理方法为标准正态变量变换(SNV),光谱数据经过预处理后可提高模型的预测能力,并降低模型的复杂度;在土壤养分的PLS预测模型中,全氮、全钾、全磷和有机质的决定系数(R^(2))分别为0.789 9、0.910 8、0.947 0和0.833 6,RPD值分别为2.108、2.903、3.938和2.238,模型的拟合效果和预测能力均较好,基于便携式近红外光谱分析技术能实现对土壤养分含量的预测。

350 soil samples were collected from Kunming,Anning and Mile in Yunnan Province,and the spectra were scanned by portable near infrared spectrometer,the near infrared prediction model of total nitrogen,total potassium,total phosphorus and organic matter in soil were established.The results showed that the spectral profiles of soil samples from different regions were close in wavelength bands of 950-1650 nm.The best pretreatment method of total nitrogen,total phosphorus and organic matter was the first derivative,and the best pretreatment method of total potassium was SNV.The preprocessed spectra could improve the prediction ability and reduce the complexity of the model.In the PLS prediction model of soil nutrient,the determination coefficients(R^(2))of total nitrogen,total potassium,total phosphorus and organic matter were 0.7899,0.9108,0.9470 and 0.8336 respectively,and the RPD values were 2.108,2.903,3.938 and 2.238 respectively.The fitting effect and prediction ability of the model were good.The soil nutrient content could be predicted by portable near infrared spectroscopy.