[GRASSLIST:752] Re: Spatial Clustering or Network Problem?

[Michael Barton]

Hi Trevor. 

Why don't you convert the vector grid to a raster grid, at the resolution of
your vector grid. Then use r.clump. My understanding is that it combines
contiguous groups of cells with similar values into 'clumps'.

This is not the same as clustering, of course, but may be what you want.

The other idea is to cluster grid centroids (x,y,val) using something like
K-means, and doing it iteratively increasing the number of groups. Then look
at the clustering results to identify the optimal number of groups and that
clustering solution.

Michael
__________________________________________
Michael Barton, Professor of Anthropology
School of Human Evolution & Social Change
Center for Social Dynamics & Complexity
Arizona State University

phone: 480-965-6213
fax: 480-965-7671
www: http://www.public.asu.edu/~cmbarton



> From: Trevor Wiens <twiens at interbaun.com>
> Date: Mon, 17 Apr 2006 23:50:58 -0600
> To: GRASSLIST <grasslist at baylor.edu>, STATSGRASS <statsgrass at grass.itc.it>
> Subject: [GRASSLIST:732] Spatial Clustering or Network Problem?
> 
> I have an interesting problem and after spending some time looking
> through the clustering documentation in R, I'm not sure that is the
> right approach.
> 
> I have a region for which a series of bird surveys were conducted
> within a vector defined grid. The coverage is not complete and I want to
> assess how best to clump the grid squares based on a survey
> completeness. Thus some grid squares will have a had no survey and
> others will be partial and others will be complete. I want to group them
> irregularly so that I end up with groups of squares that have had n
> number of completed surveys conducted.
> 
> When I first thought of this problem I thought of R and using some
> clustering mechanism, but it would appear to me that these are all
> based on the idea of grouping data based on similarity of value and
> specifying at the outset how many groups you will end up with at the
> end. My problem is the opposite in that I want to value for each group
> to be as close to each other as possible, but I don't know how many
> groups I will end up with.
> 
> Then I started wondering about turning the centroids of these squares
> into a network using v.delaunay and using a network approach
> to the problem, although I'm not sure how I might do this.
> 
> Any suggestions on how to approach this problem would be appreciated.
> 
> Thanks
> 
> T
> -- 
> Trevor Wiens 
> twiens at interbaun.com
> 
> The significant problems that we face cannot be solved at the same
> level of thinking we were at when we created them.
> (Albert Einstein)