[GRASSLIST:1351] Re: r.surf.idw2 time

Eric G . Miller egm2 at jps.net
Wed Jan 10 22:08:00 EST 2001


On Wed, Jan 10, 2001 at 07:26:42PM -0500, cbsled at ncia.net wrote:
> I am seeing a processing rate of about 8000 interpolated output points
> per hour.
> 
> Running r.surf.idw2 with 4 as number of reference points, much less
> than the default of 12.  However, if this makes any difference, the
> interpolation ratio is high, about 12.5/1.
> 
> The data being interpolated is USGS 1:24-25000 DEM data.

I'm wondering why you are interpolating a USGS DEM.  They don't have any
holes.  Are you wanting to interpolate at the margins?  If so, you may
be more interested in using some combination of r.patch/r.resamp.rst
and/or possibly r.mfilter/r.mapcalc.  r.surf.idw2 is better when you're
trying to make a surface from a "sparse matrix" of known points.  
  
> The system is running GRASS 5.0 b10 compiled on Redhat 6.2, on a AMD
> K6-2 400.
> 
> I didn't think the math for this function was very complex, being a
> type of weighted average, as I understand it, without any iterative
> components.
> 
> Am I wrong?
> 
> Is this performance normal?

In the manual page for r.surf.idw2 you'll see:

NOTES

    The amount of memory used by this program is related to the number
    of non-zero data values in the input map layer. If the input raster
    map layer is very dense (i.e., contains many non-zero data points),
    the program may not be able to get all the memory it needs from the
    system.  The time required to execute increases with the number of
    input data points. 

Since a DEM will only have zero (NULL) data at the margins, successful
processing will take a long, long time (if it ever completes).

> I need to know because I intend to process over 100 files of this
> type, each with about 24,000,000 output points.  At this rate the task
> will take over three *years* on this system.  It would take ~12 days
> to process a single file just to determine if the function is working
> correctly.

-- 
Eric G. Miller <egm2 at jps.net>




More information about the grass-user mailing list