Fuzzy k-means classification of topo-climatic data as an aid to forest mapping in the Greater Yellowstone Area, USA

Previous attempts to quantify topographic controls on vegetation have often been frustrated by issues concerning the number of variables of interest and the tendency of classification methods to create discrete classes though species have overlapping property sets (niches). Methods of fuzzy k-means classification have been used to address class overlap in ecological and geographical data but previously their usefulness has been limited when data sets are large or include artefacts that may occur through the derivation of topo-climatic attributes from gridded digital elevation models. This paper presents ways to overcome these limitations using GIS, spatial sampling methods, fuzzy k-means classification, and statistical modelling of the derived stream topology. Using data from a ca. 3600 km(2) forested site in the Greater Yellowstone Area, we demonstrate the creation of meaningful, spatially coherent topo-climatic classes through a fuzzy k-means classification of topo-climatic data derived from 100 m gridded digital elevation models (DEMs); these classes were successfully extrapolated to adjacent areas covering a total of ca. 10 000 km(2). Independently derived land cover data and middle infrared corrected Landsat TM derived estimates of Normalised Difference Vegetation Index (M-NDVI) at 575 independently sampled sites were used to evaluate the topo-climatic classes and test their extrapolation to the larger area. Relations between topo-climatic classes and land cover were tested by chi (2) analysis which demonstrated strong associations between topo-climatic class and 11 of the 15 cover classes. Relations between M_NDVI and topo-climatic classes proved to be stronger than relations between M_NDVI and the independent cover classes. We conclude that the fuzzy k-means procedure yields sensible and stable topo-climatic classes that can be used for the rapid mapping of large areas. The value of these methods for quantifying topographic controls on biodiversity and the strength of their relations with computed NDVI values warrant further investigation.