[GRASS-user] Interpolatint climate varibles with RST

[Jose Gomez-Dans]

Hi,
I have run some tests to interpolate climate variables using v.vol.rst
and v.surf.rst (trying to decide which one to use, really). I have
read the paper by Helena Mitasova and others, and we have decided to
search for tension and smoothing parametes by carrying  out a cross
validation analysis (as mentioned in the paper, we vary the tension
and smoothing values and work out the crossvalidation. You then choose
the tension and smoothing where the RMSE is minimal). The results we
get for our area (550x330 km, around 100 meteo stations) are
reasonable with the default parameters (the xvar RMSE is similar to
that obtained by kriging, slightly better in the case of RST).
However, the RMSE drops for large values of tension (>100) and low
values of tension (1,2...). I have found a paper by Chiara Sboarina
presented at the GRASS users 2002 conference that quotes tension
parameters of 1 and 2. This is quite different from the tension value
of Helena, around 45 and 70 (depending on whether 2D or 3D RST is
used). Our results imply are more in line with those of Chiara, but
they somehow look suspect :)

Does anyone have any experience in choosing these parameters that can share?

Cheers,
Jose