This option allows you to speed up the clustering process during linkage clustering by reducing the number of calculations that the program performs when determining which clusters to merge. Recall that in linkage clustering dissimilarity values are calculated for each possible pair of members in the two sets of cluster elements being evaluated. This process can be time consuming, especially for raster data sets.
In the "Advanced" page of the Constrained Clustering dialog, you have the opportunity to enter values that will define the subsampling process. Define what fraction of the locations to sample and the minimum number of samples (i.e., this overrides the fraction chosen if taking a fraction leaves too small a sample).
See also: