小麦叶层氮含量估测的最佳高光谱参数研究

Research of the Optimum Hyperspectral Vegetation Indices on Monitoring the Nitrogen Content in Wheat Leaves

【目的】作物体内氮素状况是评价长势和预测产量的重要指标。小麦植株氮素营养的快速监测和无损诊断对于精确氮素管理具有重要作用。本文旨在通过对高光谱信息的精细分析和信息提取,探索建立小麦叶片氮含量(LNC,leaf nitrogen content)估算的最佳波段、光谱参数及监测模型。【方法】利用连续4年的系统观测资料,采用精细采样法,详细分析350～2500nm波段范围内原始光谱反射率及其一阶导数光谱的任意两两波段组合而成的主要高光谱指数与小麦冠层叶片氮含量的定量关系。【结果】发现小麦叶片氮含量的最佳波段为位于红边的690、691、700和711nm以及近红外波段的1350nm;基于归一化光谱指数NDSI(R1350,R700)和NDSI(FD700,FD690)、比值光谱指数RSI(R700,R1350)和RSI(FD691,FD711)、土壤调节光谱指数SASI(R1350,R700)(L=0.09)和SASI(FD700,FD690)(L=-0.01)构建氮含量监测模型,决定系数(R2)分别为0.851和0.857、0.842和0.893、0.860和0.866。利用独立试验资料对模型检验的结果显示,模型测试的精度(R2)均大于0.758,RRMSE均小于0.266,尤其是高光谱参数RSI(FD691,FD711)和SASI(FD700,FD690)表现最好。【结论】总体上,利用精细采样法确定最佳波段,构建植被指数和氮含量监测模型,可显著提高模型的精确度和可靠性,从而为快速无损诊断小麦叶层的氮素状况提供新的波段选择和技术途径。

[ Objective ] Crop nitrogen status is an important index for evaluating the growth status and forecasting grain yield. Non-destructive and fast monitoring of leaf nitrogen status can play a significant role in precision nitrogen management in wheat production. The primary objective of this study was to explore the optimum wavebands, vegetation indices and quantitative models for estimating leaf nitrogen content （LNC） in wheat by precise analysis on canopy hyperspectral information in relation to leaf nitrogen status. [Method] On the basis of detailed data from 4-year field experiments under varied nitrogen rates and wheat eultivars, a systematic analysis was undertaken on quantitative relationships of LNCs to major hyperspectral indices composed of any two wavebands with original reflectance and its derivative within the full spectral range of 350-2 500 nm. [Result] The results showed that the optimum wavebands for LNC were 690 nm, 691 nm, 700 nm and 711 nm in the red edge range, and 1350 nm in the near-infrared range. The derived key hyperspectral parameters were NDSI （R1350, R700） and NDSI （FD700, FD690） as Normalized Difference Spectral Index, RSI （R700, R1350 ） and RSI （FD691, FD711） as Ratio Spectral Index, and SAVI （R1350, R700）（L=0.09） and SAVI （FD700 , FD690）（L=0.01）as Soil Adjusted Spectral Index. The LNC monitoring models developed from these spectral indices gave the coefficients of determination （R^2） no less than 0.842. Testing of the derived equations with independent experiment data produced R^2 values over 0.758 and RRMSE lower than 0.266, with best performance from the model based on RSI （FD691, FD711） and SAVI （FD700, FD690）（L=0.01）. [Conclusion] Overall, the method of discovering the optimum bands and constructing the spectra indices and building the model based on the systematic precise analysis methods can enhance the precision and reliability of the N monitoring model, and provide new waveband choice and technical approach for non-destructive and fast monitoring of the canopy nitrogen status in wheat.