Quantitative Determination of Bandpasses for Producing Vegetation Indices from Recombined NEON Hyperspectral Imagery

Abstract:

Hyperspectral imaging systems can be used to produce spectral reflectance curves giving rich information about composition, relative abundances of materials, mixes and combinations. However, as each spectral return from these systems is a vector with several hundred elements, they can be very difficult to process and analyze, and problemeatic to compare within, across, and between datasets over time and space. Vegetation indices (e.g. NDVI, ARVI, EVI, et al) attempt to combine spectral features in to single-value scores. When derived from calibrated and atmospherically compensated reflectance data, these indices can be quantitatively compared. Historically, these indices have been calculated from multispectral sensor data. These sensors have a handful (4 to 16 or so) of bandbasses ranging from 20 nm to 200 nm FWHM covering specific spectral regions for a variety of reasons, including both intended applications and system limitations. Hyperspectral sensors, however, cover the spectrum with many, many narrow (5 to 10 nm) bandpasses. This allows for analyses using the full, detailed spectral curve, or combination of the bands in to regions by averaging or in to composites using transforms or other techniques. This raises the question of exactly which bands should be used and combined in what manner for ideally deriving well-known vegetation indices typically made from multispectral data. In this study we use derivatives and other curve and signal analysis techniques to analyze vegetation reflectance spectra to quantitatively define optimal bandpasses for several vegetation indices and combine the 5 nm hypserspectral bandpasses of the NEON Imaging Spectrometer to synthesize them.