Hyperspectral versus multispectral crop-productivity modeling and type discrimination for the HyspIRI mission

Publication Type:
Journal Article
Remote Sensing of Environment, 2013, 139 pp. 291 - 305
Issue Date:
Filename Description Size
Thumbnail2012006901OK.pdf1.78 MB
Adobe PDF
Full metadata record
Precise monitoring of agricultural crop biomass and yield quantities is critical for crop production management and prediction. The goal of this study was to compare hyperspectral narrowband (HNB) versus multispectral broadband (MBB) reflectance data in studying irrigated cropland characteristics of five leading world crops (cotton, wheat, maize, rice, and alfalfa) with the objectives of: 1. Modeling crop productivity, and 2. Discriminating crop types. HNB data were obtained from Hyperion hyperspectral imager and field ASD spectroradiometer, and MBB data were obtained from five broadband sensors: Landsat-7 Enhanced Thematic Mapper Plus (ETM+), Advanced Land Imager (ALI), Indian Remote Sensing (IRS), IKONOS, and QuickBird. A large collection of field spectral and biophysical variables were gathered for the 5 crops in Central Asia throughout the growing seasons of 2006 and 2007. Overall, the HNB and hyperspectral vegetation index (HVI) crop biophysical models explained about 25% greater variability when compared with corresponding MBB models. Typically, 3 to 7 HNBs, in multiple linear regression models of a given crop variable, explained more than 93% of variability in crop models. The evaluation of λ1 (400-2500nm) versus λ2 (400-2500nm) plots of various crop biophysical variables showed that the best two-band normalized difference HVIs involved HNBs centered at: (i) 742nm and 1175nm (HVI742-1175), (ii) 1296nm and 1054nm (HVI1296-1054), (iii) 1225nm and 697nm (HVI1225-697), and (iv) 702nm and 1104nm (HVI702-1104). Among the most frequently occurring HNBs in various crop biophysical models, 74% were located in the 1051-2331nm spectral range, followed by 10% in the moisture sensitive 970nm, 6% in the red and red-edge (630-752nm), and the remaining 10% distributed between blue (400-500nm), green (501-600nm), and NIR (760-900nm).Discriminant models, used for discriminating 3 or 4 or 5 crop types, showed significantly higher accuracies when using HNBs (>. 90%) over MBBs data (varied between 45 and 84%).Finally, the study highlighted 29 HNBs of Hyperion that are optimal in the study of agricultural crops and potentially significant to the upcoming NASA HyspIRI mission. Determining optimal and redundant bands for a given application will help overcoming the Hughes' phenomenon (or curse of high dimensionality of data). © 2013 Elsevier Inc.
Please use this identifier to cite or link to this item: