基于PLSR的土壤有机碳预测模型中建模组与验证组最优数量关系  

Optimal Quantity Relationship Between Modeling Group and Prediction Group in Soil Organic Carbon Prediction Model Based on PLSR

作　　者：丁建军[1] 章盛 孙超[1] 米铁[2] DING Jianjun;ZHANG Sheng;SUN Chao;MI Tie(School of Physics and Information Engineering;School of Chemistry and Environmental Engineering,Jianghan University,Wuhan 430056,Hubei,China)

机构地区：[1]江汉大学物理与信息工程学院,湖北武汉430056 [2]江汉大学化学与环境工程学院,湖北武汉430056

出　　处：《江汉大学学报（自然科学版）》2018年第5期404-408,共5页Journal of Jianghan University：Natural Science Edition

基　　金：国家科技支撑计划资助项目(2013BAD11B02);教育局市属高校产学研项目(CXY201803);湖北省高等学校优秀中青年科技创新团队计划资助项目(T201420)

摘　　要：基于土壤有机碳含量现场快速测定技术研究,提出了利用可见-近红外光谱技术对土壤样本进行可见-近红外反射光谱分析的方法,选取400~1 100 nm波段光谱经S-G平滑加一阶微分滤波预处理后,利用偏最小二乘回归分析(PLSR)建立土壤有机碳预测模型。结果显示,当建模组样本数与验证组样本数之比为52∶53(约为1∶1)时,决定系数R^2=0. 98,均方根校正标准偏差RMSEC=0. 25。这说明将建模组样本与验证组样本的数量关系比设定为1∶1是建立基于PLSR的土壤有机碳预测模型的最优条件。Based on rapid determination technology research of soil organic carbon content in scene, an analysis method of visible and near infrared reflectance spectroscopy for soil samples was proposed. The 400 - 1100 nm band spectrum was pretreated by S-G smoothing and with first order differential filtering, and the prediction model of soil organic carbon was established by partial least squares regression analysis （PLSR）. The results showed when the ratio of the sample number of the modeling group and the sample number of the predicted group was 52 ： 53 （about 1 ： 1） , the determining coefficient was R^2 = 0.98 , and the standard deviation of the root mean square error of calibration was RMSEC = 0. 25. These results indicate when the ratio of the sample number in the modeling group and in the prediction group is set to 1 ： 1, it is the best condition to establish the prediction model based on PLSR for soil organic carbon.

关 键 词：土壤有机碳 可见-近红外光谱技术 偏最小二乘回归 

分 类 号：S153.6[农业科学—土壤学]