Sensitivity of texture of high resolution images of forest to biophysical and acquisition parameters

This article presents a quantitive analysis of the sensitivity of textural information of high resolution remote sensing images of a forest plantation (Les Landes, France) to a number of biophysical parameters: crown diameter, distance between trees and rows, tree positioning, leaf area index (LAI), and tree height. Influence of acquisition parameters (spatial resolution, spectral domain and viewing, and illumination configurations) is also investigated. The work is realized with the discrete anisotropic radiative transfer model (DART) simulated images with which we quantify texture with variograms. Results point out the complex dependency of variogram characteristics (range, sill, amplitude of oscillations) on biophysical and acquisition parameters. Neglect of spatial variations of the reflectance of canopy elements, as in most geometric-optical models, can lead to important errors. This stresses the interest of accurate radiative transfer models, such as DART. Although tree crown diameter is the most influential biophysic parameter, its influence may be totally masked by acquisition parameters. Finally, theoretical results were tested against high resolution airborne data (1.67 m resolution). Although encouraging results were obtained, this work both confirms the difficulty of extracting reliable texture information from real remote sensing data, and stresses the usefulness of radiative transfer models for studying the texture of high resolution satellite images. (C) Elsevier Science Inc., 1998.