Study on Line Imaging Spectroscopy as a Tool for Nitrogen Diagnostics in Precision Farming

Within nitrogen diagnostics hyperspectral imaging has a great potential overcoming many of the problems connected with derived estimates based on canopy reflectance measurements with a low spatial resolution. With this technology it is possible to collect spectral information from thousands of canopy sub areas (pixels) with a resolution of only a few mm. Hence it is now possible to obtain pure leaf estimates of the reflected light provided that the pixels have been pre-classified. However prior to this study it had not been shown that a high spatial resolution combined with a pixel filtration can actually improve nitrogen diagnostic in winter wheat. A hyperspectral line imaging system VTTVIS with a resolution of 1.8 mm (commercialised by Specim Ltd. Finland as Imspector V7) was able to predict the leaf chlorophyll concentration 30% and the leaf nitrogen concentration 38% more accurate than a hyperspectral non-imaging system with a lower resolution of 3.3*10 mm using the spectral range from 438-756 nm in winter wheat at 16 nitrogen levels provided no other plant stress was present. This improved performance using the VTTVIS was due to a two-band classification extracting relevant green leaf pixels using upper and lower thresholds for both R550 and the Normalised Difference between Green and Red (NDGR) [R550R670]/[R550+R670] prior to estimating the mean reflectance spectra used in the prediction models. The optimal limits modelling leaf chlorophyll concentration and leaf nitrogen concentration were found separately. Through a detailed theoretical description and based on ray tracing results of the dispersion Prism-Grating-Prism unit in the equipment, significant image distortion within both the spectral and spatial dimension of the hyperspectral image were proved. Further pattern noise in form of e.g. fixed pattern noise (FPN), lens vignetting, slit width variations, wavelength dependent PhotoResponse NonUniformity (PRNU), fringing patterns, and narrow band quantum efficiency response variation were shown to introduce significant bias in the sensitivity response of the VTTVIS. All known authors using PGP based systems have not described procedures for handling these systematic noise sources. Thus it is likely their published results are confounded with significant system bias. Fast calibration procedures under field conditions using the solar spectrum were developed and tested for spectral calibration with accuracy of ± 0.5 nm and for minimising system pattern noise. However due to limitations of the equipment, it was not possible to develop a comprehensive procedure for completely correcting these errors.