基于UV-Vis-SWIR漫反射光谱的土壤多属性同时快速测定

摘要

Rapid and simultaneous determination of 7 soil properties i. e. ,soil organic matter,total nitrogen,available nitrogen,a-vailable phosphorus,available potassium,slowly available potassium and pH was investigated byultraviolet-visible-short-wave-near-infrared diffuse reflectance spectroscopy combined with chemometrics. In total,81soil samples were collected. Then,50samples were used as calibration set,16 samples were used as validation setand the rest 15 samples were the test set. For each soil property,by in-vestigating variouspre-processing,variable selection methods and their combinations,the optimal PLSR models were obtained. The predictive performance of the obtained optimal models were estimated by the independent test set. Predictions of soil organic matter and total nitrogen were excellent with a determination coefficient of 0. 94 and 0. 97,respectively. Approximate quantitative predic-tions were obtained for available nitrogen and pH with a determination coefficient of 0. 76and 0. 66,respectively. For available potas-sium,prediction with a determination coefficientof 0. 52was acceptable. Predictions of available phosphorus and slowly available po-tassium were not satisfactory,with a determination coefficientsmaller than 0. 6. UV-Vis-SWIRDRS is proved to be a hopeful predic-tion method of soil properties.%采用紫外-可见-短波近红外漫反射光谱( UV-Vis-SWIRDRS)结合化学计量学对土壤中有机质( SOM)、总氮( TN)、碱解氮( AN)、速效钾( AK)、缓效钾( SAK)、有效磷( AP)及pH共7种属性建立了快速同时低成本的定量分析方法。土壤样本共81个,其中50个分为校正集,另16个为验证集,剩余15个为测试集。对每种土壤属性,考察了多种光谱预处理,变量选择以及它们的组合方法处理土壤光谱后建立的PLSR模型预测效果,选择出了最佳的光谱数据处理方法及其对应的最佳PLSR预测模型,并以独立的测试集样本对最佳模型预测性能进行客观评价。对于有机质和总氮,测试集决定系数(R2t)分别为0．94和0．97,模型可用于准确定量分析；对碱解氮和pH,R2 t分别为0．76和0．66,模型定量预测结果良好；对于有效磷,速效钾及缓效钾,R2 t小于0．6,但模型的预测效果和以往报道中的预测效果相差不大。研究表明,利用UV-Vis-SWIR DRS对土壤多属性进行快速同时低成本定量分析是可行的,有望用于土壤信息检测。