[R-sig-Geo] Doubts about kriging (2 part)

[GEMA FERNANDEZ-AVILES CALDERON]

Hi to everybody:


I would like to thank your suggestions, especially from the following
people.
 

-          David: very interesting suggestion. We will think about this.

-          Ashton: We have not more monitoring stations. 

-          Ruben: Box-Cox problems apart, Why do you choose to elaborate
an index with the observed values using PCA and then kriging the index? 

Don't you think that is better to krige every observed variable and then
elaborate the index? Myers (1983) suggests that this last option is
better in terms of MSE. However, in the literature everybody use the
first procedure.

Thanks in advance.

 Best regards,
Gema.







-----Mensaje original-----
De: David S. Bieri [mailto:dsb at vt.edu] 
Enviado el: jueves, 02 de octubre de 2008 23:09
Para: GEMA FERNANDEZ-AVILES CALDERON
Asunto: RE: [R-sig-Geo] Doubts about kriging

Hi Gema,

Just a quick general observation: Kriging/interpolation techniques for
CAPs
(criteria air pollutants) is problematic due to the way in which
atmospheric
influences distort the source-receptor distribution [ie how pollution
travels from source of emission (point sources? mobile source?) to where
it
is measured]. Without some sort of atmospheric model, you would need at
least a source-receptor matrix to get meaningful results 

[see eg. http://www.epa.gov/AMD/ModelDevelopment/index.html]

Kriging works well for things like precipitation, temperatures etc.
because
the spatial dispersion process that underlies them is very different
from
air pollutants.

Hope this helps,
David



-----Original Message-----
From: r-sig-geo-bounces at stat.math.ethz.ch
[mailto:r-sig-geo-bounces at stat.math.ethz.ch] On Behalf Of GEMA
FERNANDEZ-AVILES CALDERON
Sent: Thursday, October 02, 2008 5:25 AM
To: r-sig-geo at stat.math.ethz.ch
Subject: [R-sig-Geo] Doubts about kriging

Dear list,
 
I have two problem and I would like to know you opinion.
 
1.  I am kriging 6 environmental variables, SO2, NOx, O3, CO, NO, PM10,
to
built and enviromental index in R. I have standardized the observed
values
and I am implementing in geoR  the function likfit to obtain the
parameter
lambda (Box-Cox transformation) to see if the standarized variables are
normal. But I have error using the likfit function. 
It could be that the error arise because the standarized variables
contain
negative values?
 
2. I have values of the 6 variables at 25 locations but I want to use a
linear combination of the six variables to deal with only a variable.
Finally, I want to interpolate this only variable in several sites where
I
have not observed data. 
What do you think is better? 
- To interpolate each one of the six variables and then elaborate the
linear
combination.
- Or to elaborate a linear combination with the observed values and
finally
krige the linear combination.
 
Thanks in advance,
Gema Fernandez-Avilis
 
Universidad de Castilla-La Mancha.
(Spain)


	[[alternative HTML version deleted]]