[GRASS-user] r.neighbors, wide filtering

[Michael Barton]

On Dec 5, 2008, at 1:08 AM, <grass-user-request at lists.osgeo.org> wrote:

> Message: 7
> Date: Thu, 04 Dec 2008 18:37:56 +0000
> From: John Stevenson <john.stevenson at manchester.ac.uk>
> Subject: [GRASS-user] r.neighbors, wide filtering
> To: GRASS user list <grass-user at lists.osgeo.org>
> Message-ID: <49382384.6010206 at manchester.ac.uk>
> Content-Type: text/plain; charset=ISO-8859-1; format=flowed
>
> Hi,
>
> For my research, I am testing a mesh denoising algorithm on  
> topographic
> data.  It smooths the surfaces much like using r.neighbors
> method=average or r.neighbors method=median, and, depending on
> settings,  gives similar results to r.neighbors when size <5.  The
> advantage of the algorithm is that it has some ability to preserve
> features.  In cases where more significant smoothing is necessary
> (r.neighbors size > 5) it is much better at preserving minimum and
> maximum elevations etc as it converges on a stable solution for the
> smoothed landscape.
>
> My question relates to understanding in which cases is it  
> necessarily to
> smooth to such an extent?  From what I have seen e.g. taking speckle  
> out
> of SRTM DEMs, smoothing by such extremes removes a lot of useful
> information and results in unrealistic surfaces.
>
> Has anyone come across a situation/dataset or type of analysis where
> they need to smooth with r.neighbors (method=average/median, size >  
> 5)?
> I would be interested to know, and to see if this algorithm would be  
> useful.

John,

In fact, we were using r.neighbors median smoothing with size=7. We  
have created a recursive script to model surface erosion and  
deposition on large landscapes over long time frames (centuries). We  
have been using r.terraflow or r.flow to model water accumulation as  
part of the script. However, both of these create rapidly growing  
spikes and pits due to approximations in the flow calcuations. We  
needed a 7x7 median smoother to take these out each cycle.

We have just switched to the new r.watershed (up to 80x faster than  
the old version in some tests). This is now considerably faster than  
r.terraflow AND much more accurate. It does not produce the same  
spikes and pits. Over short time frames (decades) no smoothing is  
needed; over large time frames (centuries) we hope to get by with much  
smaller smoothing windows (we are testing 3x3 now).

Michael

____________________
C. Michael Barton, Professor of Anthropology
Director of Graduate Studies
School of Human Evolution & Social Change
Center for Social Dynamics & Complexity
Arizona State University

Phone: 480-965-6262
Fax: 480-965-7671
www: <www.public.asu.edu/~cmbarton>