Dear Pierre,
It looks as if your covariance matrix is not positive-definite. The sill of
the cross-variogram should not exceed the square root of the products of
the sills of the variograms (i.e., 0.35 should not exceed sqrt(0.2*0.5)).
Violating this condition may cause very strange results.
Gerard
At 05:59 PM 5/23/2001 -0400, you wrote:
>Hi all,
>
>I am trying to simulate a locally gaussian field using gstat. The command
>file looks like this:
>
>#
># Strates: 0=Sud 1=Nord 2=Voie navigable
>#
>data(HG0): 'Hg89Export.txt', x=4, y=3, v=5, log, max=12, min=3, radius=3000,
>b=[0] ;
>data(HG1): 'Hg89Export.txt', x=4, y=3, v=5, log, max=12, min=3, radius=3000,
>b=[0] ;
>data(HG2): 'Hg89Export.txt', x=4, y=3, v=5, log, max=12, min=3, radius=3000,
>b=[0] ;
>data(): 'GrilleLSF.txt', x=1, y=2, s=3 ;
>variogram(HG0): .2 Exp(3266) ;
>variogram(HG1): .5 Exp(3266) ;
>variogram(HG2): .35 Exp(3266) ;
>method: gs;
>set output='pr89sim.out';
>set nsim=50;
>
>The prediction file contains 1531 positions located on a square grid
>covering a lake divided in three strata (North shore, South shore and center
>channel). The output has the following strange characteristic: The
>variability of predicted values at a given prediction location depends on
>its order in the file. In the original GrilleLSF.txt file, the prediction
>locations go from west to east, south to north. The standard deviation of
>predicted values, depicted here as the size of the circles, looks like this:
>
>Notice the sudden increase in variability around x=544000 which, given the
>conditioning data, is rather unexpected. When the order of the file is
>reversed, everything else kept the same, the variability of simulated values
>becomes :
>
>Now the most variable section is in the west, up to x=551000. It is as if
>the first, maybe 500, points in the data() file were not treated the same as
>the rest. What might cause this?
>
>I hope the figures make it througt the list server.
>
>Thanks in advance.
>
>Pierre Gagnon
>Centre Saint-Laurent
>105 McGill, Montr�al
>(Qu�bec) H2Y 2E7
>514 496 1456
>http://www.qc.ec.gc.ca/csl/ <http://www.qc.ec.gc.ca/csl/>
>
>
Gerard B.M. Heuvelink
Institute for Biodiversity and Ecosystem Dynamics
Science Faculty
Universiteit van Amsterdam
Nieuwe Achtergracht 166
1018 WV Amsterdam
The Netherlands
tel +31 20 5257448 (secretary +31 20 5257451)
fax +31 20 5257431
email [EMAIL PROTECTED]
http://www.frw.uva.nl/soil/Welcome.html
