[GRASS-user] Tuning segmented processing in v.surf.rst

[Pierre Roudier]

Hi,

I've been working with v.surf.rst on large datasets, and I can
appreciate its scalability when using segmented processing.

However, the wiki page [0] is not very clear on that topic, so I ran a
few (very informal) tests for different values of segmax and npmin on
a subregion of my study area (449*450=202050 cells):

(1) augmentation of npmin values, segmax constant

segmax npmin time (real)
20 50 1m27.178s
20 100 3m12.316s
20 150 7m23.835s
20 200 14m18.463s

(2) augmentation of segmax values, npmin constant, but have to keep
npmin > segmax

segmax npmin time (real)
10 100 4m10.401s
15 100 3m37.057s
50 100 1m23.217s
100 150 1m42.070s
200 250 1m34.099s
500 600 3m54.364s

(3) for reference, with the other v.surf.* methods:

v.surf.bspline 4m04.990s
v.surf.idw 0m58.686s

So, at the end, I am a bit puzzled about the choices I should take to
interpolate the full dataset. I'm very keen on using v.surf.rst, as it
improved the computing time a lot on the whole scene (say 20 minutes
vs 1 day for v.surf.idw), but I just don't feel comfortable choosing
random parameters.

Any hint on how to tune those parameters? That would make a great add
to the wiki page IMHO.

Pierre

[0] http://grass.osgeo.org/wiki/RST_Spline_Surfaces

-- 
Scientist
Landcare Research, New Zealand