[GRASSLIST:1088] Public domain D8 flow accumulation?

[David Finlayson]

I have been working with rasters on the order of
1000x1000 to 10000x10000 cells (high resolution
watersheds of small areas and low resolution
watersheds of continental mountain ranges). 
r.watershed apparently does not scale well to these
size rasters.  I plotted its progress through a
calculation on the Himalayas (USGS Hydro1k data) and
came up with a formula to estimate how long it would
take to complete (after 3.5 days it was at 37%):

%complete = 0.43 (cpu minutes) ^ 0.52

For the Himalayas (1792 X 3051 cells), this resulted
in 100% completion after 25 days of cpu time!  There
must be a better way to do this in GRASS or does
anyone know of a public domain D8 flowaccumulation
algorithm that can be used instead of r.watershed? 
This would be particularly useful here because the
USGS Hydro1k data comes with flow direction grids.

Note: The machine was a K6-3D with 192Mb of ram
(several hundreds more in swap).  r.watershed was
running the ram module (not seg) and there was no
obvious thrashing so I assume this is a computation
issue not an I/O problem.

=====
--
David Finlayson
david_finlayson at yahoo.com
University of Washington 
Box 351310 
Seattle, WA   98195 - 1310

__________________________________________________
Do You Yahoo!?
Thousands of Stores.  Millions of Products.  All in one Place.
http://shopping.yahoo.com/