Analysis of Different Hyperspectral Variables for Diagnosing Leaf Nitrogen Accumulation in Wheat

Hyperspectral remote sensing is a rapid non-destructive method for diagnosing nitrogen status in wheat crops. In this study, a quantitative correlation was associated with following parameters: leaf nitrogen accumulation (LNA), raw hyperspectral reflectance, first-order differential hyperspectra, and hyperspectral characteristics of wheat. In this study, integrated linear regression of LNA was obtained with raw hyperspectral reflectance (measurement wavelength = 790.4 nm). Furthermore, an exponential regression of LNA was obtained with first-order differential hyperspectra (measurement wavelength = 831.7 nm). Coefficients (R(2)) were 0.813 and 0.847; root mean squared errors (RMSE) were 2.02 g·m(−2) and 1.72 g·m(−2); and relative errors (RE) were 25.97% and 20.85%, respectively. Both the techniques were considered as optimal in the diagnoses of wheat LNA. Nevertheless, the better one was the new normalized variable (SD(r) − SD(b))/(SD(r) + SD(b)), which was based on vegetation indices of R(2) = 0.935, RMSE = 0.98, and RE = 11.25%. In addition, (SD(r) − SD(b))/(SD(r) + SD(b)) was reliable in the application of a different cultivar or even wheat grown elsewhere. This indicated a superior fit and better performance for (SD(r) − SD(b))/(SD(r) + SD(b)). For diagnosing LNA in wheat, the newly normalized variable (SD(r) − SD(b))/(SD(r) + SD(b)) was more effective than the previously reported data of raw hyperspectral reflectance, first-order differential hyperspectra, and red-edge parameters.