A linear model to predict with a multi-spectral radiometer the amount of nitrogen in winter wheat

The objective was to develop an optimal vegetation index (VIopt) to predict with a multi-spectral radiometer nitrogen in wheat crop (kg[N] ha-1). Optimality means that nitrogen in the crop can be measured accurately in the field during the growing season. It also means that the measurements are stable under changing light conditions and vibrations of the measurement platform. Different fields, on which various nitrogen application rates and seeding densities were applied in experimental plots, were measured optically during the growing season. These measurements were performed over three years. Optical measurements on eight dates were related to calibration measurements of nitrogen in the crop (kg[N] ha-1) as measured in the laboratory. By making combinations of the wavelength bands, and whether or not the soil factor was taken into account, numerous vegetation indices (VIs) were examined for their accuracy in predicting nitrogen in wheat. The effect of changing light conditions in the field and vibrations of the measurement platform on the VIs were determined based on tests in the field. VIopt ((1+L)*(R2NIR+1)/(Rred+L) with L=0.45), the optimal vegetation index found, was best in predicting nitrogen in grain crop. The root mean squared error (RMSE), determined by means of cross-validation, was 16.7 kg[N] ha-1. The RMSE was significantly lower compared to other frequently used VIs such as NDVI, RVI, DVI, and SAVI. The L-value can change between 0.16 and 1.6 without deteriorating the RMSE of prediction. Besides being the best predictor for nitrogen, VIopt had the advantage of being a stable vegetation index under circumstances of changing light conditions and platform vibrations. In addition, VIopt also had a simple structure of physically meaningful bands