Accuracy and precision of nitrogen estimation can be improved by hyperspectral remote sensing that leads effective management of nitrogen application in precision agriculture. The objectives of this study was to identify nitrogen (N) sensitive spectral wavelengths by evaluated different approaches. Two years study was conducted during 2011 and 2012 at Northwest A & F University, China, to determine the relationship between leaf hyperspectral reflectance (350-1075 nm) and leaf N contents of corn (Zea mays L.) temporally under five nitrogen rates (0, 60, 120, 180, and 240 kg/ha pure nitrogen) were measured at five key developmental stages. The fitting of linear, non-linear regression models and their diagnosis accuracy for determining nitrogen nutrition were compared among the single (R) and dual (R1+R2) spectral reflectance, spectral ratio (SR), NDVI, GNDVI, SAVI, first order differential transform based on area, position and their derivative SVIs and REIP (Red Edge Inflection Point). We chose 2-3 high coefficient of determination (R2) with high F value models that best suited to each growth stage and tested their accuracy among self-species and cross-species examination analysis based on RMSE (Root mean square error) and RE (Relative error). The results showed that Y = -2.2954Ln(x) + 0.2925 was the best regression prediction model for real time estimation of leaf N contents at 10-12 leaf stage followed by Y = -0.187 - 7.932x – 3.452x2, Y = 4.070-2.304x-52.177x2, Y = 2.390-0.793lnx and Y = 5.129e-2.317x at silking stage, tasseling stage, late dent stage and 6-8 leaf stage respectively. GNDVI overall showed the highest R2 both at 10-12 leaf stage (0.88) and silking stage (0.80). The results presented that the universal practicability of these models may be good in real time estimation of leaf N contents under wide range of corn species by using hyperspectral remote sensing in real time.
