[Gstat-info] conditional sequential simulation, and value honoring

[G. Allegri]

Thanks Jon,
the two models I've tried are:

  model psill range
1   Nug  0.40     0
2   Sph  0.15   750

and

  model psill range
1   Nug  0.00     0
2   Sph  0.40    10
3   Nug  0.00     0
4   Sph  0.15   750

In the second one I've setted a shperical structure with nugget 0 with
very short range (10 meters). The differences between the results
obtained from both the models are very small (0.1-1 dS/m).
As you said, maybe the deviations are due to the 0.4 nugget (respect
to a a-priori variance of 0.99, being a normal-transformed variable),
having a strong effect even on the short distances within a cell (50
meters).



2008/4/8, Jon Olav Skoien <j.skoien at geo.uu.nl>:
> Hi Giovanni,
>
>  The result of a conditional simulation depends very much on the variogram
> you use as input. Short correlation length or large nugget give larger
> deviations from observations also at short distances. Large deviations can
> also occur close to some observations even if you have high correlation, due
> to the random component in the conditional simulation.
>
>  A more detailed answer to your question is difficult unless you provide
> more information about the observations and the variogram used in your
> simulations.
>
>  Cheers,
>  Jon
>
>  G. Allegri wrote:
>
> >
> > Hi.
> > I've done some work with krige() to run conditional simulations on my
> > datas. It's a simulation grid with 6458 nodes on 50x50 meters cells,
> > and 198 observed values (sparsed all over the grid). I run the
> > simulation with nsim=500, nmax=100 and range=700, choosed on the base
> > my variogram model range.
> > At the end of the 500 simulations of done the mean for every cell, and
> > then I've overlayed the values in the 198 locations. If I compare the
> > observed values and the values retrieved from the cells, I obtain very
> > different values. I.e. on a scale of 0-70 dS/m, if the observed value
> > is 52, I may obtain a simulated value of 15 in the cell it falls in,
> > even if it's the only observed value ina range of 200-300 meters.
> > Shouldn't it weigh much more in the simulated value? I would expect
> > something that honor more the true value...
> > Am I wrong in expecting this from the conditional simulation algorithm?
> >
> > Thanks,
> > Giovanni
> > _______________________________________________
> > Gstat-info mailing list
> > Gstat-info at geo.uu.nl
> > http://mailman.geo.uu.nl/mailman/listinfo/gstat-info
> >
> >
>