基于多元统计方法的冬小麦叶面积指数光谱估测

Hyperspectral prediction on the leaf area index of winter wheat using multivariate statistical methods

叶面积指数(LAI)是评价作物生长状况的指标之一,利用光谱技术实现冬小麦LAI的快速、准确监测具有重要的意义。本文以连续两年的氮素运筹试验为基础,通过测定各生育时期的冠层光谱和LAI,并利用多元统计分析方法(偏最小二乘法,PLS;逐步多元线性回归,SMLR)提取氮素运筹条件下LAI特征波段和构建LAI估测模型。结果表明,光谱波段765、775、1060 nm进入到LAI的预测模型中,结合PLS中VIP参数和B-系数证实,以上波段与冬小麦LAI具有重要的关系;基于PLS-SMLR方法构建的预测模型R^2=0.699,RMSE=1.447,RE=0.275,经验证模型仍然具有较好的表现(R^2=0.689,RMSE=1.323,RE=0.285)。表明利用PLS-SMLR提取特征波段、建模的方法是可行的,可为作物LAI的快速诊断监测提供一定的理论依据。

Leaf area index （LAI） is one of the most important indices for evaluating crop’s growth. The rapid, realtime and nondestructive technology of hyperspectrum is widely applied on monitoring LAI. In this study, the effect of nitrogen addition level on LAI and the canopy spectral reflectance of winter wheat during 2012-2014 were determined. The sensitive wavelengths were determined and LAI monitoring models were constructed by using multivariate statistical analysis methods （partial least square, PLS; stepwise multiple liner regression, SMLR）. The results showed that the characteristic bands of 765, 775 and 1060 nm which were input into LAI spectrum monitoring model had an important relationship with LAI of winter wheat. This relation was confirmed by using the parameter of the variable importance for projection （VIP） and Bcoefficient. Moreover, the R2, RMSE and RE of the predictive LAI model were 0.699, 1.447 and 0.275, respectively, which were determined following the method of PLSSMLR. The validated model also had good prediction with R^2=0.689, RMSE=1.323, RE=0.285. It was concluded that the multivariate methods had potential applications on extracting the important wavelengths of LAI and constructing the predictive models. This study provides a basis for rapidly assessing the situation of LAI of winter wheat.