Narrow-Band and Derivative-Based Vegetation Indices for Hyperspectral Data

Hyperspectral remote sensing imagery was collected over a soybean field in central Illinois in mid-June 2001before canopy closure. Estimates of percent vegetation cover were generated through the processing of RGB (red, green, blue)digital images collected on the ground with an automated crop mapping system. A comparative study was completed to testthe ability of broad-band, narrow-band, and derivative-based vegetation indices to predict percent soybean cover at levelsless than 70%. Though remote sensing imagery is commonly analyzed using reference data collected at random points overa scene of interest, the analysis of the hyperspectral imagery in this research was performed on a pixel-by-pixel basis overthe field area covered by the automated crop mapping system. Narrow-band and derivative-based indices utilizing the finerspectral detail of hyperspectral data performed better than the older broad-band indices developed for use with multispectraldata. Specifically, second-derivative indices measuring the curvature in the green region (514-556 nm), longer wavelengthred region (640-694 nm), and short wavelength NIR (712-778 nm) performed well. Narrow-band indices, based on thestandard ratio index equations, which used values from the blue (472-490 nm) and green (514-550 nm) regions, alsoperformed well for many of the datasets. The performance of all indices was shown to suffer over areas of brighter soilbackground, and the use of ratio-based narrow-band indices that did not incorporate NIR reflectance values performed bestin this case.