> ----------
> Van: Willy Bas Loos
> Verzonden: woensdag 11 april 2001 17:49
> Aan: 'Edzer J. Pebesma'; ', [EMAIL PROTECTED]'
> CC: Dirk Zoetebier
> Onderwerp: RE: indicator kriging
>
> Hi,
>
> I've tried what you suggested (set order = 1), and it works. I don't get
> any more strange answers.
> Nonetheless: Can someone tell me what causes this so-called order relation
> violation? I like knowing what I'm doing. Also, the added command seems to
> slow down the process, and I need a considerable amount of
> interpolations....
>
> background info:
> My (large) data set contains no other values than 1 and 0.
> All data points are at least 1000 map units apart. It does not seem to be
> ill conditioned kriging, as mentioned in section C.1.2 of the manual.
The radius that I use is rather large: 30 000 map units. The strange effects
do not occur when I use 8 data points (radius +/- 3000 map units) insted of
a radius command (!).
> To my understanding, in ordinary kriging there is no calculation of trends
> of any kind, like there is in universal kriging. What happens is (again,
> this is only my idea), that ordinary kriging predicts a value by a
> weighted average. How can this result in values that are just a little
> below zero and a bit above 1???
>
> Willy-Bas Loos
>
>
>
> ----------
> Van: Edzer J. Pebesma[SMTP:[EMAIL PROTECTED]]
> Verzonden: vrijdag 30 maart 2001 8:42
> Aan: Willy Bas Loos
> CC: Dirk Zoetebier; '[EMAIL PROTECTED]'
> Onderwerp: Re: indicator kriging
>
> Willy Bas Loos wrote:
> >
> > Hi,
> >
> > I'm using ordinary block kriging on a 0/1 dataset (indicator kriging).
> > Gstat predicts some negative values, an a few greater than 1.
>
> You need to inform gstat that it's dealing with indicators. You can
> define an indicator variable with:
>
> data(something): ... , I=0.5, ... ; # use 0.5 as indicator cutoff value
>
> Be aware that anything below 0.5 now becomes 1, and above 0.5 becomes 0
> (see manual). Now gstat knows it's kriging indicators. Order relation
> violation correction (correcting estimated probabilities outside [0,1]
> etc).
> can be controlled by
>
> set order = 1; # or some other value; again, see manual
> --
> Edzer
>
>