Hyperspectral vegetation indices and their relationships with agriculturalcrop characteristics

The objective of this paper is to determine spectral bands that are best su ited for characterizing agricultural crop biophysical variables. The data f or this study comes from ground-level hyperspectral reflectance measurement s of cotton, potato, soybeans, corn, and sunflower. Reflectance was measure d in 490 discrete narrow bands between 350 and 1,050 nm. Observed crop char acteristics included wet biomass, leaf area index, plant height, and (for c otton only) yield. Three types of hyperspectral predictors were tested: opt imum multiple narrow band reflectance (OMNBR), narrow band normalized diffe rence vegetation index (NDVI) involving all possible two-band combinations of 490 channels, and the soil-adjusted vegetation indices. A critical probl em with OMNBR models was that of "over fitting" (i.e., using more spectral channels than experimental samples to obtain a highly maximum R-2 value). T his problem was addressed by comparing the R-2 values of crop variables wit h the R-2 values computed for random data of a large sample size. The combi nations of two to four narrow bands in OMNBR models explained most (64% to 92%) of the variability in crop biophysical variables. The second part of t he paper describes a rigorous search procedure to identify the best narrow band NDVI predictors of crop biophysical variables. Special narrow band lam bda (lambda(1)) versus lambda (lambda(2)) plots of R-2 values illustrate th e most effective wavelength combinations (lambda(1) and lambda(2)) and band widths (Delta lambda(1) and Delta lambda(2)) for predicting the biophysical quantities of each crop. The best of these two-band indices were further t ested to see if soil adjustment or nonlinear fitting could improve their pr edictive accuracy. The best of the narrow band NDVI models explained 64% to 88% variability in different crop biophysical variables. A strong relation ship with crop characteristics is located in specific narrow bands in the l onger wavelength portion of the red (650 to 700 nm), with secondary cluster s in the shorter wavelength portion of green (500 nm to 550 nm), in one par ticular section of the near-infrared (900 nm to 940 nm), and in the moistur e sensitive near-infrared (centered at 982 nm). This study recommends a 12 narrow band sensor, in the 350 nm to 1,050 nm range of the spectrum, for op timum estimation of agricultural crop biophysical information. (C) Elsevier Science Inc., 2000.