[R-sig-Geo] Daily precipitation kriging

[Helen Greatrex]

Hi Ivan,

I think I have an answer to your problem. Maybe you're getting the 
errors because you're inputting data at more than one time step into the 
kriging process?

Also there might be a slightly more elegant way of approaching the 
situation.

Normally, you would krige at one point in time -(i.e. the structure of 
gold ore doesn't change over time) However, unless you have an extremely 
dense raingauge network, you're not going to obtain a reasonable 
variogram doing this for daily rainfall. Also it's reasonable to assume 
that rainfall on different days has the same spatial structure over a 
short enough time period (i.e. all the days in January). So you can pool 
the data together and make a 'January' variogram and krige using that.

At Reading, we do this through using a climatological variogram. 
Basically, you 'normalise' each day's rainfall by dividing through by 
the total daily variance to create the variogram, then at the end of the 
kriging process you remultiply each day's estimate to give you final 
estimates. The method is explained more briefly in Teo & Grimes, 
"Stochastic Modelling of rainfall from satellite data", J. Hydrology 
346, 2007. We also have another publication on 'Geostatistical analysis 
of rainfall' which has just been accepted for publication in a special 
edition of Geographical Analysis on ‘Geostatistical Methods in 
Geography', which should explain the process in detail.

I've created a set of self contained R scripts to do the entire process 
(plus an extensive user manual which should explain things further). 
I'll send it to you as a separate e-mail to stop large attachments 
getting sent to everyone on the list.

I'm aware this e-mail hasn't answered your specific coding problems, but 
I hope it helps!
Best wishes
Helen







-- 
-----------------------------
Helen Greatrex (PhD Student)
Room 1U09
Department of Meteorology
University of Reading
Earley Gate
PO Box 243
READING
RG6 6BB