Integration of spatial analysis and fuzzy classification for the Estimation of forest Parameters in Mediterranean Areas

Satellite images are often insufficient to provide reliable estimates of forest parameters in complex Mediterranean areas, where the spectral response of vegetation is influenced by several interacting factors. In the current work spatial analysis and Geostatistics are used to produce additional information which can supplement that derived from remotely sensed data. In particular, an approach based on kriging was applied in a study area in Tuscany (Central Italy) to estimate forest composition and structure, and the same was done with a fuzzy classification procedure using bitemporal Landsat-TM images. Since the two methods produced per-pixel estimates of error variance, their outputs could be integrated by means of an optimal merging methodology. The results of the experiments, evaluated by statistical comparison to independent ground references, showed that both methods provided good estimates of forest composition and structure at stand level. Furthermore, the information derived from the two sources was partly nonredundant and could be efficiently merged to improve the estimation accuracy for both forest parameters.