Hyperspectral Vegetation Indices and Their Relationships with Agricultural Crop Characteristics

The objective of this paper is to determine spectral quantities of each crop. The best of these two-band indices were further tested to see if soil adjustment or nonlinbands that are best suited for characterizing agricultural crop biophysical variables. The data for this study comes ear fitting could improve their predictive accuracy. The best of the narrow band NDVI models explained 64% to from ground-level hyperspectral reflectance measurements of cotton, potato, soybeans, corn, and sunflower. 88% variability in different crop biophysical variables. A strong relationship with crop characteristics is located in Reflectance was measured in 490 discrete narrow bands between 350 and 1,050 nm. Observed crop characterisspecific narrow bands in the longer wavelength portion of the red (650 nm to 700 nm), with secondary clusters tics included wet biomass, leaf area index, plant height, and (for cotton only) yield. Three types of hyperspectral in the shorter wavelength portion of green (500 nm to 550 nm), in one particular section of the near-infrared predictors were tested: optimum multiple narrow band reflectance (OMNBR), narrow band normalized differ(900 nm to 940 nm), and in the moisture sensitive nearinfrared (centered at 982 nm). This study recommends a ence vegetation index (NDVI) involving all possible twoband combinations of 490 channels, and the soil-adjusted 12 narrow band sensor, in the 350 nm to 1,050 nm range of the spectrum, for optimum estimation of agricultural vegetation indices. A critical problem with OMNBR models was that of “over fitting” (i.e., using more spectral crop biophysical information. Elsevier Science Inc., channels than experimental samples to obtain a highly 2000 maximum R2 value). This problem was addressed by comparing the R2 values of crop variables with the R2 values computed for random data of a large sample size. The combinations of two to four narrow bands in INTRODUCTION, RATIONALE, AND OMNBR models explained most (64% to 92%) of the BACKGROUND variability in crop biophysical variables. The second part The spectral data from the current generation of earthof the paper describes a rigorous search procedure to orbiting satellites carrying broad band sensors such as identify the best narrow band NDVI predictors of crop Landsat Thematic Mapper (TM), Le Systéme pour l’obsbiophysical variables. Special narrow band lambda (k1) ervation de la terre (SPOT) high resolution visible versus lambda (k2) plots of R2 values illustrate the most (HRV), and the Indian Remote Sensing (IRS) Linear effective wavelength combinations (k1 and k2) and bandImaging Self-Scanning have limitations in providing acwidths (Dk1 and Dk2) for predicting the biophysical curate estimates of biophysical characteristics of agricultural crops (Fassnacht et al., 1997; Thenkabail et al. * Center for Earth Observation (CEO), Yale University 1995; Wiegand et al. 1991; Wiegand and Richardson, † International Center for Agricultural Research in the Dry Areas 1990), natural vegetation (Friedl et al., 1994), and in (ICARDA), Aleppo, Syria quantifying other terrestrial ecosystem characteristics, Address correspondence to P. S. Thenkabail, Center for Earth such as soil characteristics; stress due to weeds, water, Observation (CEO), Department of Geology and Geophysics, Kline Geology laboratory, P.O. Box 208109, 210 Whitney Avenue, Yale Univerand nitrogen deficiencies or excess; crop phenology; and sity, New Haven, CT 06520-8109, USA. E-mail: prasad.thenkabail@ fallow, forest, and agricultural interactions (Moran et al., yale.edu Received 19 February 1999; revised 15 June 1999. 1994; Running, 1989). These limitations have motivated