## [ai-geostats] Uncertainty of Conditional Stochastic Simulated results

Expand Messages
• Dear List: I got a question about the uncertainty associated with the Conditional Stochastic Simulated results, the question is: I have soil C data for a
Message 1 of 2 , Oct 10, 2004
Dear List:

I got a question about the uncertainty associated with the Conditional
Stochastic Simulated results, the question is:

I have soil C data for a sample area, within that area, there are woodland
and grassland. The variance of C data in woodland is higher than that in the
grassland. Now I want to use Conditional Stochastic Simulation to simulate a
bunch of images of soil C on the area. I will use one single variogram model
to do the simulation. The questions is: Do the estimated C data more
variable under woodland than those under grassland (simulated soil C data in
woodland have a larger variance than those in grassland)? or a single
variogram model means uniform or random variance of simulated data?

Thank you very much

Feng Liu
Texas A&M University
• Hi Feng, You are in the ideal situation where it would make sense to separate the two populations and do the variography and simulation separately for each
Message 2 of 2 , Oct 10, 2004
Hi Feng,

You are in the ideal situation where it would make sense
to separate the two populations and do the variography
and simulation separately for each stratum, since clearly
the pattern of variability of soil carbon is expected to be
different under these two landcovers (I am sure that besides the
variance, the range and nugget effect must be pretty different)
and also the average C concentration should vary a lot .
In most situations it is difficult to consider different populations since:
1. we might not have enough data within each stratum to compute a
reliable variogram,
2. the boundaries of the different strata might be fuzzy,
3. the multi strata approach creates discontinuities in the map
that are not physically realistic.
I don't know your data but I am pretty sure that none of these limitations
apply.

Now to answer your question, the use of a single semivariogram implies
an assumption of stationarity, hence the fact that the
variance/covariance does not depend on the location within the study area,
which is clearly not your case. Even wen using a single variogram,
the variability in your simulated map will be influenced to some
extent by your data (conditioning observations). Hence, I suspect
that more variability will appear on woodland anyways.
By analogy, the same happens when you use an isotropic semivariogram
in simulation/estimation, while the data display strong anisotropy.
The simulated/estimated map will display some anisotropy, which is
always reassuring.

Cheers,

Pierre
<><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><>

Dr. Pierre Goovaerts
President of PGeostat, LLC
Chief Scientist with Biomedware Inc.
710 Ridgemont Lane
Ann Arbor, Michigan, 48103-1535, U.S.A.

E-mail: goovaert@...
Phone: (734) 668-9900
Fax: (734) 668-7788
http://alumni.engin.umich.edu/~goovaert/

<><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><>

On Sun, 10 Oct 2004, Feng Liu wrote:

> Dear List:
>
> I got a question about the uncertainty associated with the Conditional
> Stochastic Simulated results, the question is:
>
> I have soil C data for a sample area, within that area, there are woodland
> and grassland. The variance of C data in woodland is higher than that in the
> grassland. Now I want to use Conditional Stochastic Simulation to simulate a
> bunch of images of soil C on the area. I will use one single variogram model
> to do the simulation. The questions is: Do the estimated C data more
> variable under woodland than those under grassland (simulated soil C data in
> woodland have a larger variance than those in grassland)? or a single
> variogram model means uniform or random variance of simulated data?
>
> Thank you very much
>
> Feng Liu