Glossary for Geospatial Science

clustering:� A process in which multiple, registered, overlapping, rasters are reduced to a single raster, which is referred to as a cluster map.� The rasters used in clustering represent analytical data (such as images and elevations).� Different clustering methods (such as K-means, Fuzzy C-means, and Isoclass) use varying logic to accomplish this mapping and dimensional reduction.� In general, all of the methods use the concept of testing multiple values for each cell against all other cell values to determine which subpopulation each cell is most similar to and should be grouped with.� All clustering methods first make a preliminary set of groups by testing all cells or a sample of them.� They then proceed by various methods to test individual cells and dynamically redefine the clusters until every cell is satisfactorily assigned a cluster membership in a single cluster, which is recorded at that cell�s position in the cluster map.� The maximum number of clusters to form and the number of refinement iterations is usually controlled during setup.� (See also: labeling, unsupervised classification)