High-resolution landform classification using fuzzy k-means

Previous attempts to devise automated methods of landscape classification have been frustrated by computational issues related to the size of the data set and the fact that most automated classification methods create discrete classes while 'natural' interpreted landscape units often have overlapping property sets. Methods of fuzzy k-means have been used by other workers to overcome the problem of class overlap but their usefulness may be reduced when data sets are large and when the data include artefacts introduced by the derivation of landform attributes from gridded digital elevation models. This paper presents ways to overcome these limitations using spatial sampling methods, statistical modelling of the derived stream topology, and fuzzy k-means using the Distance metric. Using data from Alberta, Canada, and the French pre-Alps it is shown how these methods may easily create meaningful, spatially coherent landform classes from high resolution gridded DEMs. (C) 2000 Elsevier Science B.V. All rights reserved.