2364[ai-geostats] Re: Multivariate kriging using terrain fabric

Expand Messages
• Feb 1, 2006
• 0 Attachment
My apologies for having taken so long to reply to everyone. It's been a busy week. I do appreciate the terminolgy warnings as regards UK vs. KT; I have a bad habit of being a little loose with terms outside of my field.

> Maybe I am being thick but how do you have a
> relationship between two variables but say there is
> no correlation? If you have high with high and low
> with low, you should find a direct correlation
> between the two variables.

An excellent question! Obviously, I wasn't too clear when I stated my problem. Basically, when I said they share the same trend/fabric, I meant that they share the same anisotropy. Because we have continous coverage of the secondary variable (bathymetry), I can (hypothetically anyway, in practice this may be more difficult) derive an anisotropy ellipse for each point I'm interested in estimating. The magnetics (my primary variable) will follow the anisotropy pattern of the bathymetry, as they're both produced by the same process.

Is there any method that allows the variogram model of the primary variable to vary spatially as a function of a secondary variable?

Assuming I can derive an anisotropy ellipse from the bathymetry that can be related to the variogram model of the magnetics, is it even remotely theoretically sound to allow the variogram model to vary spatially as a function of some other variable?

I apologize again if I'm not making much sense. It's 6:00am, and I haven't slept in a while, to say the least, so I'm sure I sound a little deranged!

Thanks,
-Joe

---- Original message ----
>Date: Mon, 30 Jan 2006 00:00:47 +0000 (GMT)
>From: Isobel Clark <drisobelclark@...>
>Subject: Multivariate kriging using terrain fabric
>To: Joe Kington <jdkington21@...>
>
> Joe
>
> Maybe I am being thick but how do you have a
> relationship between two variables but say there is
> no correlation? If you have high with high and low
> with low, you should find a direct correlation
> between the two variables.
>
> If you are saying the trends are similar but the
> residuals are uncorrelated, then I understand. Are
> you using co-located or non-co-located cross
> variogram on the residuals? Or on the original data?
>
> First stage UK (estimating the trend) fits local
> trends, which can be linear and approximate a
> complex larger surface in a ‘piece-wise‘
> fashion. Similarly with the modelling stage of KED.
>
> Maybe I can help if you can straighten my geriatric
> brain out ;-)
> Isobel