[GRASSLIST:7758] cross validation on v.srf.rst. Does it work?

[Jaime Carrera]

Hi all,

I'm trying to use the cross validation option on both
v.srf.rst and v.vol.rst. However I'm getting some
strange results for the slovakia rainfall data set. 
1) I ran v.srf.rst as follows: 
GRASS 6.0.0 (slovakia3d):~ > v.surf.rst input=rain3d
zcolumn=int_1 cvdev=cv1 -v mask=mask at PERMANENT

Which wrote a table with difference values on
Postgres.

2) Then I created another table without the gauging
station #122 (which I called rain3d_2) and
interpolated using the same parameters as above:
GRASS 6.0.0 (slovakia3d):~ > v.surf.rst input=rain3d_2
zcolumn=rain mask=mask at PERMANENT elev=vsurf1_wo122

AFAIK cross validation removes one point at a time,
interpolates using the remainder points and computes
the difference between the interpolated value at a
certain location (i.e. station 122) and the real value
(record at station 122).  Accordingly if I query the
raster map vsurf1_wo122 at the coordinates of station
122 and substract the interpolated value to the real
value I should get the value that was written to the
cross validation table for station 122, so I query the
map:
GRASS 6.0.0 (slovakia3d):~ > r.what input=vsurf1_wo122
east_north=4416904,5414029
4416904|5414029||1057.9642333984

The real value at station 122 is 798 mm; according to
these values the difference is: 1057.96-798 = 259.96.
The value that was written by the v.srf.rst module on
the cross-validation table is -5.31.

Can someone tell me what's going on here?

Cheers,

Jaime



__________________________________________________
Correo Yahoo!
Espacio para todos tus mensajes, antivirus y antispam ¡gratis! 
Regístrate ya - http://correo.espanol.yahoo.com/