[R-sig-Geo] Time-extent/time-lag in spatiotemporal variograms

[piero campa]

Dear experts,

rather than a technical question, this is more about how to set up a model
for spatiotemporal kriging predictions.

With special regard to the fitting of a spatio-temporal model, whichever the
covariance model I want will use, I was wondering /how deep/ in time should
my `spacetime` ST*DF object be.

The temporal memory I am expecting to have should drive the cutoff distance
of the variogram in time, but how many temporal units of data should I
(better) have in case I want, say, N tlags in the variogram?

I know that ~30 pairs per lag are suggested for a stable variogram. Assuming
I have a pattern in space that permits such rule-of-thumb to be satisfied,
in case I have e.g. 3 different days with the same spatial pattern, I could
also compute my stable experimental spatiotemporal variogram with 2-3
temporal lags with no worries of not having enough pairs for meaningful
statistics (even at the border lags I would have a sufficient number of
observations).

Thanks for any guidance/reference,
Piero







--
View this message in context: http://r-sig-geo.2731867.n2.nabble.com/Time-extent-time-lag-in-spatiotemporal-variograms-tp7581911.html
Sent from the R-sig-geo mailing list archive at Nabble.com.