Cartographie de la biomasse forestiere a l'aide des donnees d'inventaire forestier et des images TM de Landsat (French text).

摘要

Biomass estimation is important for sustainable forest management, monitoring of global change and forest productivity models, and the need to map forest biomass is increasing rapidly since the Montreal and Kyoto protocols. Several methods for estimating biomass by remote sensing have been developed, but their comparative advantages have not been evaluated for areas in Canada. This study compares four methods of mapping forest biomass on a pilot region located in Western Newfoundland using Landsat-5 image: (i) the development of direct radiometric relationships between Landsat TM reflectance or spectral indices and biomass values measured on forest inventory plots, (ii) the k-NN method used in the Finnish National Forest Inventory, (iii) the application of biomass tables on an unsupervised classification of the TM image by land cover, and (iv) a method that combines the use of image classification, biomass tables and the forest stand maps. A last method was used as a reference map for the validation and was produced by the application of biomass table to the forest stand maps. The results of each method were evaluated using independent validation plots and a comparison with a baseline biomass map. The methods combining the classification process, biomass tables and forest stand maps gave the best overall estimations, followed by the k-NN method. The method that applied biomass tables on the classified TM image was the one that showed the least interesting results, as well as the baseline map. The direct relationships method was also discarded for implementation purposes because it presented important disadvantages.