This is more statistical than S-Plus, but this group contains some
superb talent, so I'll forge ahead:
I need to analyze collections of gridded fields generated by
numerical models of the atmosphere. These collections are run as an
ensemble, where the initial conditions for each run vary slightly.
Due to sensitive dependence, the individual runs sometimes diverge in
groups or sets. There are 50 individual runs per case. The output of
each run consists of gridded parameter fields.
I want to apply cluster analysis to these runs to identify those that
are "similar" by some measure. I tried this once in the past with
gridded fields of the height of the 500 mb surface, but didn't have
mush success: while I could identify fields that looked similar,
every similarity/dissimilarity metric I could think of, and every
method available within S-Plus, failed to generate what I expected to see.
I'm revisiting the issue and have fund little in the literature that
addresses this particular problem, though my search has not been
nearly exhaustive. None of the texts I have on cluster analysis
appear to address this particular problem. It may be that the GIS
community has already tackled this problem, but I have no ready
access to the GIS literature.
Thus, I am open to suggestions:where should I look next? And if
anyone has suggestions about what I should try next, feel free to make them.
Kim Elmore
Univ. of Okla./National Severe Storms Laboratory
|