基于叶片反射光谱估测水稻氮营养指数

【目的】基于叶片反射光谱建立快速、无损监测水稻氮营养指数（nitrogen nutrition index,NNI）的估算模型。【方法】2018—2019年开展2个水稻品种（徽两优898和Y两优900）及5个氮肥梯度（施氮量为0、75、150、225和300 kg·hm^-2,分别记为N₀、N₁、N₂、N₃、N₄）的田间小区试验,测定关键生育期不同叶位叶片反射光谱和植株NNI,构建多种光谱指数的水稻NNI监测模型。【结果】单叶及叶位组合的敏感波段均分布在540 nm的绿光波长处,其与近红外波段构成的窄波段比值指数SR（R₉₀₀,R₅₄₀）可较好反演水稻NNI。但不同叶位叶片窄波段比值指数与水稻NNI的预测精度表现不同,顶3叶（L₃）预测精度最好（R²=0.731,RMSE =0.130,RE=11.6%）,顶2叶（L₂）次之（R²=0.707,RMSE =0.136,RE =12.2%）,顶1叶（L₁）最差（R²=0.443,RMSE =0.187,RE =14.7%）;顶2叶和顶3叶组合平均光谱（L₂₃）的预测精度优于单叶水平和其他叶位组合（R²=0.740,RMSE =0.128,RE =11.5%）。再将窄波段比值指数SR（R₉₀₀,R₅₄₀）近红外与绿光区域分别重采样50 nm和10 nm,所构建的宽波段比值指数SRAR_{（900±50）},AR_{（540±10）}]模型精度较SR（R₉₀₀,R₅₄₀）未明显降低,且在L₂₃水平下2个模型的模型精度和预测精度基本一致（R²=0.740,RMSE =0.128,RE =11.5%）。水稻NNI小于1时与产量呈线性的正相关关系（P<0.05）,大于1时产量趋于平稳。【结论】L₂和L₃叶片反射光谱为监测水稻NNI的敏感叶位,其中叶位组合L₂₃可提高模型预测精度。基于叶片反射光谱构建的多种波段比值指数（SR（R₉₀₀,R₅₄₀）和SRAR_{（900±50）},AR_{（540±10）}]）可快速估测水稻NNI,从而为不同传感器对水稻氮营养指数估测监测研究提供了理论依据。

Determination of Suitable Band Width for Estimating Rice Nitrogen Nutrition Index Based on Leaf Reflectance Spectra

【Objective】The research aimed to analyze the relationship between rice (Oryza sativa L.) nitrogen nutrition index (NNI) and leaf spectral reflectance characteristics of leaf on different positions, so as to provide an effective method for nondestructive and timely evaluation of NNI in rice.【Method】Field experiments were conducted with different N application rates and rice cultivars across two growing seasons during 2018-2019, and the leaf hyperspectral reflectance of 350-2 500 nm of leaf on different positions and the plant NNI were measured during key fertility growth stages to construct a variety of spectral index model for rice NNI monitoring.【Result】The results indicated that green band (540 nm) at leaf level was the sensitive band for estimating NNI, and narrow band ratio index SR (R₉₀₀, R₅₄₀) composed of near infrared band and green band could be used to retrieve NNI of rice. However, the prediction accuracy of narrow band ratio index and rice NNI of leaf on different positions were different. In terms of prediction accuracy, the best single leaf position was the third leaf (L₃) from the top (R²=0.731, RMSE=0.130, RE=11.6%), the second leaf (L₂) from the top followed (R ²=0.707, RMSE=0.136, RE=12.2%), and the top one (L₁) was the worst (R ²=0.443, RMSE=0.187, RE=14.7%). The averaged spectra of L₂ and L₃ (L₂₃) was the optimum leaf spectra combination, which contributed to improving the predictability to NNI(R ²=0.740, RMSE=0.128, RE=11.5%). The samples were resampled at 50 nm and 10 nm in the near infrared region (900 nm) and green region (540 nm) respectively, and the accuracy of the wide band ratio index SR (AR_(900±50), AR_(540±10)) was not significantly lower than that of SR (R₉₀₀, R₅₄₀). The model accuracy and prediction accuracy of the two models were basically the same at L₂₃. When the NNI of rice was less than 1, there was a significant positive linear correlation with the yield, and then it tended to be stable. 【Conclusion】The results showed that the reflectance spectra of L₂ and L₃ leaves were sensitive for monitoring NNI for rice, and L₂₃ could improve the prediction accuracy of the model. Multiple band ratio indices SR (R₉₀₀, R₅₄₀) and SR (AR_(900±50), AR_(540±10)) based on leaf reflectance spectra could be used to rapidly estimate rice NNI, which provided a theoretical basis for monitoring rice NNI with various sensors.