Crop discrimination is the basis for vegetation mapping; one of the first steps to crop monitoring and mapping efforts. More specifically, this is used to; characterize, model, classify and map crops, species composition, crop type, biophysical & biochemical properties, disease and stress, nutrient, moisture, crop productivity etc. These changes affect crop reflectance which such that the reflected spectra has differences. Hyperspectral sensors, a new development offers to solve the crude spectral categorization; narrow contiguous bands (1-10nm) sensitive to subtle differences in spectral behavior to attain a higher accuracy. Despite the many studies and comparisons on crop discrimination using hyperspectral imagery for crop discrimination, few studies have been done in Africa, hence this study. Additionally, a selection of bands is needed to solve dimensionality as well as provide optimal data for discrimination. This study offers a comparative study of the performance of hyperspectral (Hyperion) and multispectral (Landsat ETM+ and EO-1 ALI to determine crop discrimination. Crop discrimination was determined using Stepwise Discriminant Analysis, Principal Component Analysis and a correlation study between Hyperion bands to determine redundant bands. From stepwise discriminant analysis, a subset of wavebands is selected to discriminate crops with their variability scores of 61%, 48 and 45% for Hyperion, ALI and Landsat respectively. Principal component analysis generated principal components for wavebands with most lying the 1200-1600nm region. Correlation analysis produces lambda vs lambda plots to all from which bands redundant bands are selected. Classification accuracy is done using Discriminant analysis to using a selection of bands that generate 95% accuracy for Hyperion, 87% for ALI and 85% for ETM+.

show more