Résumé
The structural organization of a forest canopy is an important descriptor that may provide spatial information on above-ground biomass and carbon stock estimate. We test here the potential of a powerful method of canopy texture analysis from very-high remotely sensed images, such as commercial IKONOS or freely available Google EarthTM images, for studying the wet tropical forests of the Western Ghats of India. The multivariate ordination of Fourier spectra allowed us to classify 1557 125-m square canopy images with respect to canopy grain, i.e. a combination of mean size and frequency of tree crowns per sampling window. Topography variations proved to influence image texture and to induce a “local” effect, arising from the presence of pronounced geomorphologic features that bias the ordination of relatively few windows, as well as a more “global” effect related to an increase or decrease of trees shade size according to the underlying terrain and having little but significant influence over windows ordination. A partitioned standardization, based on a hillshade model computed from a Digital Elevation Model was used to mitigate the effect of topography. Using forest parameters based on 7 1-ha ground plots, we shown that IKONOS-derived canopy textural index enable us to discriminate forested scenes according to the dominance of specific ranges of spatial patterns in the image texture and that this ordination highly and reasonably correlate with structural parameters of modeled forested scenes (e.g. R² = 0.87 for mean crown diameter, R² = 0.95 for mean tree density) and ground plots (e.g. R² = 0.54 for mean tree diameter, R² = 0.63 for stand basal area and above ground biomass estimate), respectively. Google Earth-derived results were consistent regarding the reference commercial image, although the poor model calibration led sometimes to un-realistic prediction values.