Stand basal area extension over forestland by different classification algorithms applied to Landsat ETM+ imagery

摘要: Conventional ground-sampling-based inventories are able to give a proper statistical assessment of the extent, condition and productivity of forest ecosystems. However, given the usual sampling intensities, they usually cannot directly support an effective estimation and visualization at the local level of the spatial distribution of the measured attributes. Instead, mapping forest variables and associated characteristics are fundamental for forest management and planning, and they also represent an essential basic information source for many modelling tasks. A spatial modelling experimentation based on the integration of remotely sensed images, sample field measurements and GIS technologies targeted to produce forest stand basal area maps is here presented. Testing was carried out in Central Italy (Mediterranean biogeographical region), where more than 300 geocoded sampling field plots based on a single-stage systematic cluster design had been implemented to routinely inventory wood and non-wood forest attributes. Relying on these ground data, two non-parametric classification methodologies (k-NN and fuzzy classifier) were applied to Landsat 7 ETM+ images to spatialize and map stand basal area. Different configurations of the classifiers have been tested, with the distinctive experimental purpose of assessing the effects by spectral distances computed in increasingly more complex ways.