应用近红外光谱预测水稻叶片氮含量

Estimating leaf nitrogen content with near infrared reflectance spectroscopy in rice

以水稻(Oryza sativa)新鲜叶片和干叶粉末两种状态的样品为研究对象,基于近红外光谱(NIRS)技术,应用偏最小二乘法(PLS)、主成分回归(PCR)和逐步多元回归(SMLR),建立并评价了水稻叶片氮含量(NC)近红外光谱模型。结果表明,基于PLS建立的模型表现最好,鲜叶氮含量近红外光谱校正模型校正决定系数RC2为0.940,校正标准误差RMSEC为0.226;干叶粉末氮含量的近红外光谱校正模型RC2为0.977,RMSEC为0.136。模型的内部交叉验证分析表明,预测鲜叶氮含量内部验证决定系数RCV2为0.866,内部验证标准误差RMSECV为0.243;预测干叶粉末氮含量RCV2为0.900,RMSECV为0.202。模型的外部验证分析表明,预测水稻鲜叶氮含量的外部验证决定系数RV2大于0.800,外部验证标准误差RMSEP小于0.500,预测干叶粉末氮含量的RV2为0.944,RMSEP为0.142。说明,近红外光谱分析技术与化学分析方法一致性较好,且基于干叶粉末建立的近红外光谱预测模型的准确性和精确度较新鲜叶片高。

Aim Our primary objective was to establish an effective method of near infrared reflectance spectroscopy （NIRS） for estimating leaf nitrogen content in rice,which would help with nitrogen diagnosis and dressing fertilization in rice production. Methods Using the techniques of partial least square （PLS）,principal component regression （PCR） and stepwise multiple linear regression （SMLR）,we established four NIRS-based models for estimating nitrogen content （NC） in fresh leaf and leaf powder of rice cultivars under varied nitrogen application rates. Important findings The coefficient of determination （RC2） and root mean square error for calibration （RMSEC） of NC models with fresh leaf were 0.940 and 0.226,respectively,whereas the RC2 and RMSEC of NC models with leaf powder were 0.977 and 0.136,respectively. We tested the accuracy of models with independent experiment datasets by the determination coefficient （RCV2） and root mean square error of cross-validation （RMSECV）,and the determination coefficient （RV2） and root mean square error of external validation （RMSEP）. With fresh leaf,the RCV2 and RMSECV of NC models were 0.866 and 0.243,respectively,while the RV2 was 0.800 and RMSEP was 0.500. With leaf powder,the RCV2 and RMSECV of NC models were 0.900 and 0.202,respectively,whereas the RV2 and RMSEP were 0.944 and 0.142,respectively. Overall,the performance of the models with leaf powder is better than that with fresh leaf in rice.