Abstract

A new method for calibrated maximum likelihood (ML) classification of forest types has been developed for the Swedish CORINE land cover mapping project. The method corrects for the tendency of the ML-algorithm to over-represent dominant classes and to under-represent less frequent ones. The correction procedure iteratively adjusts the prior weights until class frequency in the output corresponds to objective (field-inventoried) estimates of the frequency for each class. National forest inventory data measured from a five-year period is used both for the estimation of relative frequency and to derive spectral signatures for each class. The method was implemented operationally in an automated production line which enabled rapid production of a country-wide forest type map from Landsat TM satellite data. This paper describes the details of the method and demonstrates results from operational use

Published in

Publisher: Swedish National Board of Forestry

Conference

ForestSAT 2005

Authors' information