Date:         Fri, 29 Sep 2006 11:15:23 -0700
Reply-To:     oumar ka <[log in to unmask]>
Sender:       Biological Conservation and GIS <[log in to unmask]>
From:         oumar ka <[log in to unmask]>
Subject:      fuzzy k-means and raw DEM
Content-Type: text/plain; charset=ascii

Hi all,

I have a region where soil survey maps have showns distinct sub-areas based on soil types, geology, and also vegetation. This can also somehow be observed in a visual interpretation of the hillshade map, which suggest a relationship with elevation.

I am trying to do something which (because of my lack of knowledge in the field) I don't know if it is feasible and/or sensible at all. 

1. I would like to apply a moving window based voariogram algorithm to calculate varogram ranges and map them coutinously through the whole area covered by the DEM.
2. Then I would use this range map to classify the DEM into distinct areas following a fuzzy k-means (preferably) or a k-means approach.
3. Later I would fit a separate variogram model to each class in 2) and derive the range in that class.
4. Finally I would use each range calculated in 3) as "the search radius" to calculate 2 DEM derivatives (elevation percentile and local relief) in a plan to delineate landform units using an algorithm that uses these 2 derivatives as part of its input variables. This I believe would optimize the calculation of these 2 parameters since the search radius is estimated differently in the original algorithm.

I was wondering of course if it is possible at all and sensible to use fuzzy k-means to segment raw DEM data in such a way, and  if yes how could this be done? If sensible, is it then possible to use variogram ranges (derived as mentioned above) as kind of soft information to fuzzy k-means (or k-means) a DEM, or this could be simply done on the DEM directly.

Thank you very much for giving me your thoughts on this. I am not very familiar with this subject, so plz excuse my ignorance if there's something that's not clear in my message. 

Oumar

**********************************************************
CONSGIS LISTSERV
To unsubscribe send to [log in to unmask] this email message
SIGNOFF CONSGIS
Questions?  Contact Pete August, [log in to unmask]
**********************************************************