Vegetation indices for the detection and classification of leaf nitrogen deficiency in maize

Nitrogen is an important nutrient with respect to crop growth, development and yield. Hence, site specific and optimal nitrogen fertilization requires knowledge of spatial nitrogen distribution and deficiencies in the field. Optical methods to determine leaf nitrogen concentration (LNC) have advantages over laboratory methods because of lower costs and faster performance. Together with e.g. unmanned aerial vehicles (UAV), optical methods can also be used to acquire LNC information with high spatial resolution. The main goal of this research was therefore to determine the most suitable vegetation indices for the detection and classification of nitrogen differences and deficiencies in maize (Zea mays L.). Hyperspectral images from 450 nm to 998 nm of fully expanded maize leaves from four different nitrogen treatments (0.72–2.88 g N/plant) were acquired under controlled light conditions and the corresponding LNC were determined. Then optimal wavelength-pairs for two predefined vegetation index formulas, the normalized difference spectral index (NDSI) and the ratio spectral index (RSI), were identified. Finally, the performances of the identified vegetation indices and selected vegetation indices from the literature to predict and classify LNC were assessed by means of a simulated pattern map that reflects spatially varying LNC classes. It was found that a wavelength from the red edge region (718 nm) was the most significant for LNC (r = 0.92). The vegetation index formulas considered (NDSI and RSI) showed the best performances when wavelength-pairs from the red-edge and NIR region (722 nm, 950 nm) were combined. Both vegetation indices showed a strong relationship with LNC (R²=0.90 for NDSI, R²=0.86 for RSI) and performed best at predicting LNC classes and their distribution in the simulated pattern map (accuracy=91.7 %, kappa=0.87).