Proportional estimation of land cover characteristics from satellite data

The land surface over much of the Earth is typically a complex mosaic of different types of vegetation. Traditional approaches that classify vegetation into discrete categories do not allow a description of such mixtures and gradients in vegetation types. Consequently, climate models and other types of global change models that require information on the global distribution of land cover types incorporate these idealized and somewhat unrealistic descriptions of the land surface. Analysis of satellite data potentially can lead to more accurate and realistic descriptions of the land surface as continuous variables of vegetation characteristics. Such variables would describe, for example, proportional mixtures of woody, non-woody, and no vegetation: deciduous and evergreen vegetation; and broadleaf and needleleaf woody vegetation. This paper explores two approaches to derive such continuous variables. With the first approach, multiresolution data from the NOAA AVHRR and Landsat Multispectral Scanner System (MSS) are used to estimate percentage cover of closed canopy broadleaf evergreen forest in central Africa based on empirical relationships. In the second approach, a linear mixture model is applied using AVHRR Pathfinder Land data to obtain a global map of proportional mixtures of woody vegetation (trees and shrubs), non-woody vegetation (herbaceous vegetation), and bare ground