[R-sig-Geo] some problems with krigeST

[Edzer Pebesma]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1



On 10/09/2013 05:23 AM, 胡艺 wrote:
> Dear all,
> 
> I'm a newer to spatio-temporal analysis though I am quite familiar
> with spatial one. When coming to spatio-temoral prediction using
> krigeST, I have some queries as follows:
> 
> 1. it seems that the krigeST can only deal with data with full
> layout format (e.g., STFDF class), however our data in practice,
> which is often the case, come with sparse format (e.g., STSDF
> class) because of different No. of samples in each time point or
> missing values. Under this circumstance, I mean only with STSDF
> class, what I can do if I still want to make spatio-temporal
> prediction? just sort out a  STFDF format from the available data?
> or else
> 

please look at the development version of gstat at
https://r-forge.r-project.org/R/?group_id=378
and in particular demo(localKrigeST) -- we addressed this issue on a
hackaton three weeks ago, and you're welcome to test ;-)

> 2. I happened to see a replying email by Edzer to someone which
> said krigeST was designed to exploit separaility of spatio-temporal
> covariance models. If this is the case, what dose other types of
> spatio-temporal variogram (e.g., productSum, sumMetric, and
> simpleSumMetric) mean? Do them predict with a jiont spatio-temporal
> covariance? if so, why one needs to specify purely spatial and
> temporal variogram if the jiont spatio-temporal covariance is
> specified?

This was true for the first incarnation of krigeST; now it supports
more ST variogram model classes, many of which are non-separable. Look
at (the examples of) ?vgmST to see how they work.

> 
> 3. In terms of specification of arguments in vgmST, it seems to be
> too subjective to specify arguments for varigram of space and time,
> for example, time = vgm(1, "Exp", 1.5 * 3600 * 24)). How could I
> know range, sill, or nugget for the varigram?

This is a modelling / inference question. I will not go any further
here than saying that this is not trivial. Some hints at what you can
do with gstat are found in the package vignette("st"). Please don't
complain about the lack of options, but write working and readable
code. If you would contribute it, we will include it and acknowledge you!

- -- 
Edzer Pebesma
Institute for Geoinformatics (ifgi), University of Münster
Heisenbergstraße 2, 48149 Münster, Germany. Phone: +49 251
83 33081 http://ifgi.uni-muenster.de GPG key ID 0xAD3A77F1
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.11 (GNU/Linux)
Comment: Using GnuPG with Thunderbird - http://www.enigmail.net/

iQEcBAEBAgAGBQJSVPuBAAoJEM1OCHCtOnfxzycH/38pUdCuUY19ijIzgDZuRawk
oLLWEPjlQ8ZiXUjyy6ezUhymL5Lo8n+eVzaoGslm4S7xGesgO9AsfadoiwMRVpX7
yMDRZcrMJ6ZhSOjQv1gCwZsXZnMUkEe75rkxsEjFd2uJDmYXmmUaTQ8EwpXF3L7s
yQSwJ5AQPjB0sIpbl9Q+iiqplNDGLt/dPrnPPVGlC5PpcIcgZIU+Tb2KuIIryzu/
7aW/jdG18aM3z0Ofj7sok6h6UmyoZXASiOTM+Jdc411/OQHRPgYZV+Eo4lQTU+3y
sZS6yVvGw/F4qG2fhkAdGlZg/n6OL3T6EpH2dBkIsm5gY+jToQAGlRBK+gFgkUg=
=gwcm
-----END PGP SIGNATURE-----