help - s.surf.tps "segmentation fault" error

[Kenn Gardels]

I have been giving s.surf.tps a real run for its money the last couple
of weeks, and have had no problems like you describe (I'm working on a
DECStation).  Couple of observations that might help generally, and
could possibly do strange memory things on the RS6000 (based on other
AIX problems I've heard about):

1 - you may want to add a slight (1 or less) smoothing factor, at
least to see if that fixes your overshoots.

2 - try to prevent extrapolation of z values by adding a grid of
constant values (probably 0) in a donut around your real data.  I know
this sounds kludgy, but it seems to make the algorithm behave better.

3 - don't use a region definition to either limit the processed area
or reduce extrapolation - it creates unpredictable edge effects.  If
you need to reduce your total number of sample points, mask the site
list in s.menu, save it in a new site file, and use that.

4 - try reducing dmin down to the minimum or nearly so.  You may be
throwing away so many points that what's left is not representative of
the terrain, and that makes the algorithm unstable.  Just guessing on
this one...

5 - find an AIX wizard.

Good luck!!!

-kenn
> 
> 
> I am getting segmentation faults (core dumped) under some conditions,
> running s.surf.tps, but not under others.  In general, if I
> use the default settings for tension, and npmin < 200, things run ok.
> But my map ends up with effects between "segments", and 
> with "overshoots" due to extreme topography.  From the manual,
> I gather I should then increase both tension and npmin. But if I
> increase the tension, or use npmin > 200 (not thoroughly
> tested) I get a crash, immediately after the map has been "read",
> just before the any "percent done" value shows up.
> 
> Computer: RS6000 model 340.
> 
> Grass:    4.0 with (I think) all the updates.
> 
> Map:      -digitized using v.digit, topography of a province in
>            Thailand, with extreme topography changes: flat valleys 
>            between very steep mountains.
>           -converted to a site list, using v.out.point | s.in.ascii
>           -about 240,000 points.
>           -testing s.surf.tps on smaller regions of 2000 - 5000 points.
> 
> Results:
>  set segmax=35 npmin=200, others default:       runs ok, poor surface
>  set segmax=35 npmin=250 tension=80:            crash
>  set segmax=25 npmin=200 tension=80:            crash
>  set segmax=35 npmin=300, others default:       crash
>  set segmax=35 npmin=250, others default:       crash
> 
> where each crash is a "segmentation fault (core dumped)", when the
> "percent done:" message shows up, but no percent is done yet.
> 
> Can anyone help?
> 
> Tim.
> 
>  -------------------------------------------------------------
>   Tim Martin                   *
>   Spatial Information Systems  *   These opinions are my own:
>   University of Alberta        *      My employer has none!
>   martin at cs.ualberta.ca        *
>  -------------------------------------------------------------
> 
>