[R-sig-Geo] Question about temporal Kriging
hzambran.newsgroups at gmail.com
Thu Oct 2 10:55:22 CEST 2008
I'll really appreciate any ideas/comments about the following situation:
I need to carry out some spatial interpolations of daily precipitation
on several catchments, using point measurements (rain gauges).
The precipitation regime is quite different between the northern part
of the catchments and the southern one, and among seasons (summer,
In principle, I guess that one (not optimal) way of doing this is to
select one variogram for each day and use it for carrying out the
interpolations on that day. However, my problem is that I need to do
that for 30 years (~11000 days), an it would be a very time-consuming
I've heard that Bayesian Maximum Entropy (so far, I don't have
knowledge on this field) is appropriate for tackling this situation,
but I haven't found an R package for doing BME interpolations.
At the other hand, I was thinking on probably fit a spatio-temporal
regression model for interpolating the daily precipitation at
different times, and then analyse the residuals with OK or UK, using
the time as a third dimension in gstat.
However, regarding to the last approach I have some questions:
-) What should be the minimum number of points in each time step for
obtaining a meaningful 3D semivariogram ?.
-) What should be the time scale used as third dimension ?. The xy
coordinates are in UTM (~6000000, ~400000). If the time coordinate
varies within [0, ~11000], will the analysis suffer of numerical
problems due to differences in magnitudes among the 3 dimensions ?.
Could you suggest any other practical way of tackling this problem ?.
Thanks in advance,
Mauricio Zambrano B.
Dept. of Civil and Env. Engineering
University of Trento, Italy
Linux user #454569 -- Ubuntu user #17469
"Predicting the future is hard,
especially if it hasn't happened yet"
More information about the R-sig-Geo