In article
,
Dawlish writes:
snip
The ECM verifies, most of the time, as the best
forecasting model at 5 days, but even with the ECM, accuracy in
forecasting temperatures has varied between 0.91 and 0.72 over this
last month, with the mean accuracy being 0.847. Granted, this is
higher than the other 3 models that they evaluate (gfs in second place
with 0.793), but we are talking about the model being wrong about 15%
of the time even and only, at 5 days.

That raises the question of how wrong does it have to be to be flagged
as wrong? If we don't know that, then those statistics don't tell us
very much. To complicate matters, there are at least two factors to be
considered when determining whether the model is "correct": the size of
the discrepancy between forecast and actual temperature, and the
proportion of the area being forecast for over which that discrepancy
occurs. And are we talking about surface temperatures, which are the
ones of prime concern to the layman, or - as I suspect - 850mb
temperatures?
--
John Hall
"If a man will begin with certainties, he shall end in doubts;
but if he will be content to begin with doubts,
he shall end in certainties." Francis Bacon (1561-1626)