The Role of Attribute Selection in GIS Representations of the Biophysical Environment

This article questions the arbitrary selection of input attributes for the definition of landform classes and other fiat objects that are used to represent the biophysical environment in geographic information science. It suggests that attribute selection influences the characterization of both geographic and attribute space in these applications. Hence digital elevation model-based fuzzy c-means landform classification relies on sensible selection of terrain attributes to generate fuzzy landform classes (memberships) with biophysical meanings. A case study employed several sets of sensitivity tests and evaluated how selections of terrain attributes may affect the outputs of fuzzy c-means landform classifications. The results showed an average classification difference of 37 percent when different numbers of attributes are used and 18 percent when similar terrain attributes are swapped. Effects of attribute selection also show obvious dependence on spatial resolution and number of classes. These results indicate that the current approach of selecting terrain attributes (not only for landform classifications but also for other applications) according to tacit expert knowledge needs to be better justified. Because the fuzzy c-means classification method is essentially data-driven, the adoption of an exploratory approach as a part of this method is crucial. Such an approach may help to identify membership distributions (and corresponding classifications) that summarize the correspondence between landforms and specific biophysical patterns.