夏玉米叶片全氮含量高光谱遥感估算模型研究

Total nitrogen content estimation models of summer maize leaves using hyperspectral remote sensing

在5种不同施氮量和2种夏玉米品种处理下,分别在玉米拔节期、大喇叭口期、抽雄期、吐丝期、乳熟期测定了玉米冠层高光谱反射率及其对应叶片的全氮含量。选取了470、550、620和720 nm 4个代表性光谱波段,分品种对叶片全氮含量与原始光谱反射率、光谱反射率一阶微分以及部分高光谱特征参数(基于光谱位置、面积、植被指数的特征参量)分别进行线性回归和非线性回归拟合。在每个生育时期,选择决定系数和F值最高的模型3个,并分别用第二年测定的光谱和全氮含量数据分别对两个品种进行均方根差和相对误差的验证,选择均方根差和相对误差较小的拟合模型。结果表明,在拔节期、大喇叭口期、抽雄期、吐丝期和乳熟期,玉米叶片全氮含量最佳拟合光谱参量分别为R720、DR720、SDb、DR550和DR550,玉米叶片全氮含量最佳高光谱遥感估算模型依次为Y=5.129e-2.317x、Y=3.421-10.010x-477802.331x3、Y=4.070-2.304x-52.177x2、Y=-0.468-0.528lnx和Y=-2.390-0.793lnx。

Based on the treatments of five nitrogen fertilizer application amounts and two cultivars of summer maize, crop canopy spectral reflectance and total nitrogen content of maize leaves were measured at the jointing stage, huge bellbottom stage, tasseling stage, silking stage and milk stage. The canopy spectral reflectance in 470, 550, 620 and 720 nm wavelength of hyperspectral remote sensing were chosen to establish liner and nonlinear regression relationship between leaf total nitrogen content and canopy spectral parameters for each cuhivar, which includes original spectral reflectance, first order differential transform, and part of hyperspectral characteristic parameters （ i. e. spectrum position, area, characteristic parameters of vegetation index）. Three models with high coefficients and F values of each cultivar at each growth stage were chosen to verify root mean square error and relative error with the second year data of spectral reflectance and total nitrogen content of two cultivars separately. The smallest root mean square error and relative error models were taken as the best models. The results show that： at the jointing stage, huge bellbottom stage, tasseling stage, silking stage and milk stage of maize, spectrum parameter for the best fitting regression relationship with leaf total nitrogen content was R720, DR720, SDb, DR550 and DR550, respectively. And the best model to estimate total nitrogen content of maize leaf based on above best spectrum parameter of hyperspectral remote sensing in five growth stages was Y = 5. 129e-2.317x, y = 3. 421 - 10. OlOx - 477802.331x3, Y =4.070 -2.304x -52.177x2, Y = -0.468 -0.5281nx and Y = -2.390 -0.7931nx, respectively.