Felly sgrifennodd Martin Rowley :
Following various threads recently relating to Meteograms, 'Hauptlauf'
and the applicability of ensembles many days out, I have added the
following to the FAQ pages relating to models:-

http://homepage.ntlworld.com/booty.w...NS_diagram.htm

Thanks Martin, very informative.

I assume that the "Hauptlauf" is the same as you call the "control run"
under the entry for "Ensemble" in the FAQ. So therefore, the Hauptlauf
uses the data as it is known at present to form the model, and the other
ensemble members use the same data, but with small deliberate alterations
(errors) introduced. Right?

Whilst understanding that this is a valuable and reasonable way to do things,
I would (following on from my PhD research in multivariate data analysis!)
suggest an alternative approach. I would not introduce any deliberate
errors, but would use different sets of variables; maybe even different
subsets of variables from the "Hauptlauf". One thing that came out strongly
from my research is that the best model is rarely formed by using ALL the
available variables, but rather certain subsets (if well chosen) would
perform much better. This is known as the "Principle of Parsimony", yet
tends to be misunderstood and often ignored.

Before anyone asks: no I wasn't trying to predict the weather! Rather,
the origin of olive oil samples! But the same theory should apply.

Adrian
--
Adrian Shaw ais@
Adran Cyfrifiadureg, Prifysgol Cymru, aber.
Aberystwyth, Ceredigion, Cymru ac.
http://users.aber.ac.uk/ais uk