Estimation of forest attributes by integration of field sampling and remotely sensed data under Mediterranean environments

Abstract

The use of remotely sensed data taken from satellite platforms is increasing within forest inventory and monitoring programs. To better model the complex relationships between spectral signatures and forest attributes the use of flexible classification procedures is needed: in such a context the k-Nearest Neighbour (k-NN) is one of the most used and efficient non-parametric classifier. The aim of this study is to test different versions of the k-NN method to estimate stand volume, one of the most common forest attributes, in two different study sites in central Italy. Different methods for spectral distance computation are inter-compared and analyzed. Within the examined conditions the spectral distance modified by a multiregressive method achieves the best performance in terms of estimation accuracy, and therefore seems to be able to optimally use the information content of the available spectral bands.