Estimation of subpixel vegetation density of natural regions using satellite multispectral imagery

A procedure is presented for estimating the subpixel fractional canopy density of natural or undisturbed semivegetated regions on a pixel-by-pixel basis using one satellite multispectral image and a physical modeling approach. The method involves applying a model of the bulk, nondimensional plant geometry combined with a simple model of canopy reflectance and transmittance to the red and near-infrared reflectance space of the atmospherically corrected satellite image. Shadow effects are parameterized assuming Poisson-distributed and geometrically similar plant canopies. The method is applied to the estimation of fractional cover and leaf area index, using Landsat thematic mapper imagery, of two physiologically different plant communities. The first is the Landes Forest, a coniferous region in south central France, during the June 1986 HAPEX-Mobilhy Experiment. The second is the semiarid Walnut Gulch basin of southeast Arizona that contains predominantly shrubs and grasses, during the June 1990 MONSOON Experiment. The procedure offers a physically based alternative to empirical vegetation indices for estimating regionally variable canopy densities of natural, homogeneous systems with little or no ground truth.