Quantification of subpixel cover fractions using principal component analysis and a linear programming method: Application to the coastal zone of Roscoff (France)

In this study, principal component analysis. and the linear programming method "affine algorithm" were jointly used to provide an unmixing method and to calculate fractions bounds of main seaweeds and intertidal components from mixed pixels. The method was tested using CASI airborne imagery data taken over the Roscoff coastal zone (Brittany, France). The contribution of the number of spectral bands used to identify many component types was also examined. Between four and six components can be distinguished depending on the spectral richness of the original image. The accuracy of fraction cover types was estimated by comparing calculated proportions with ground data. Whatever the number of bands and ground categories, the coefficient of determination R-2 was higher than 0.62. Much more information and more accurate results were obtained with 13 spectral bands (R-2 from 0.81 to 0.96). (C) Elsevier Science Inc., 1998.