s.surf.tps-memory

[Helena Mitasova]

> We have GRASS 4.1 and, with the purpose of producing a DEM of our region, 
> we are going to use s.surf.tps. The input files are point files. GRASS4.1
> is on 2 Sun Sparc stations 10 and 20 with respectively 32 and 64 Mb of RAM.
> We would be interested to know if GRASS loads all the points in the RAM 
> before starting the interpolation . 

Yes it does, it has to do it to create the quadtree for segmentation.

> In this case we wonder if there is an
> ideal number of points of the input files that GRASS is able to process
> before swapping on the HD (with an exponential increase in the processing 
> time).
> This of course will influence the size of the areas into which we are
> going to split the region to create the DEM.

I did not have problems myself as I was interpolating data sets with
up to 150000 points, but I know that there were people working with much larger
data sets, they can probably share their experience.
However, before going into spliting the region I suggest to increase dmin to
at least the grid cell size or maybe a little bit more (default is half
the grid cell size), so that the points which are not necessary get removed
before the quadtree is built - it often happens that more than 50% of points
is removed without any significant loss of detail and the computation
is much faster.

If you have troubles with memory because your output DEM is too big for
your workstation, please contact me for a library version of s.surf.tps 
(unless you have downloaded it already some time ago from moon, it is not 
there any more as it is being replaced by a floating point version and we
don't support it any more).

I hope it helps,

Helena Mitasova

Geographic Modeling and Simulations Laboratory
Dept. of Geography
University of Illinois at Urbana-Champaign
helena at gis.uiuc.edu