基于多视角反射光谱的冬小麦冠层叶片氮素营养监测研究

通过2个蛋白质含量不同的冬小麦品种和4个氮素水平的试验，研究传感器在垂直冬小麦垄平面上，冠层不同观测角度的反射光谱与叶片氮素营养的关系，改进小麦冠层氮素光谱诊断的理论与方法。结果表明，在本试验选择的7种植被指数光谱特征参量中，2个小麦品种均表现为比值植被指数（RVI670，890]）与冠层叶片氮素含量（CLNC）相关性最高；不同视角的RVI与冠层叶片氮素含量关系中，0°、30°和90°的相关性最高；利用0°、30°和90°的RVI与CLNC进行模型拟合，其模型的决定系数是0°﹥30°﹥90°；在建立的0°模型中，京411小麦模型RMSE为0.2915，预测准确率为90.2%，中优9507模型RMSE为0.3827，预测准确率为87.2%。本研究证明改变反射光谱观测角度，能够提高冠层叶片氮素含量的光谱预测精度，不同小麦品种其光谱特征与冠层叶片氮素含量关系不同，在应用中要根据不同的小麦品种建立相应的模型。

Prediction Canopy Leaf Nitrogen Content of Winter Wheat Based on Reflectance Spectra in Different Directions

In order to improve the method of nitrogen spectral diagnosis, we conducted a experiment with two winter wheat (Triticum aestivum L.) cultivars Jing 411(low protein content) and Zhongyou 9507(high protein content) under nitrogen application levels of 0, 100, 200, and 400 kg N ha^-1. The wheat canopy spectral reflectance over 350–2 500 nm and canopy leaf nitrogen contents (CLNC) were measured at tillering, jointing, heading, milking, and mature stages. Seven vegetation indices (VIs) with view angles of 0°, 30°, 60°, 90°, 120°, 150°, and 180° to the vertical line to wheat row were compared with corresponding CLNC. The results indicated that CLNC had most significant correlations with RVI670,890] among seven spectrum vegetation indices, especially in the view angles of 0°, 30°, and 90°, and the predicting models for CLNC were further established on the basis of the correlations. The coefficient of determination (R²) of three models were sequenced as: 0°＞30°＞90°. Root mean square error (RMSE) and R2 between measured and estimated CLNC were employed to test the model reliability and predicting accuracy. RMSE and R² were different in different view angles and cultivars, the model based on the view angle of 0°had RSME of 0.2915 and R² of 0.902 for Jing 411, and RSME of 0.3824 and R² of 0.872 for Zhongyou 9507. There were lower RMSE and higher R² in the model established based on the view angle of 0°than in that of other angles. The correlation was different between canopy spectral reflectance and CLNC in different cultivars, indicating that different predicting models should be selected based on different cultivars.