"Len Wood" wrote ...

I always thought Martin, that ensembles at least in part take care
of
uncertainties in inputs, caused by for example butterflies and
elephants.

.... I agree that up to (roughly) 6 days, the 'perturbation' method
should cope with extracurricular inputs; beyond that period, slight
variance of initial conditions can have dramatic downstream (in both
time and space) effect that, as we've seen, even 'tightly clustered'
plumes (or other types of ensemble output) can be significantly in
error.

What they do not help with as regards NWP, are the errors involved
in
finite differencing of the differential equations, and
parameterisations of all sorts of feedbacks at various scales, not
least the subgrid scales.

.... indeed, and I have a healthy skepticism of such beyond that period
(~6 days); I've seen too many 'certain' outcomes go awry to give too
much credence to any one particular run.

*If* successive runs (over, say, three days / open to debate) indicate
a similar output, then I might consider giving a %age probability for
6 to 9 days of 70%, and at 10 days, perhaps 60% (but I'd need to be
certain that the outcome was climatologically & dynamically sensible)
.... but even then, I seen several examples of 'rock solid' ensemble
output for the 6 - 10 day period that don't verify to be sucked into
any higher probs.

Beyond 10 days ... well, it's all good knockabout stuff I suppose :-)

Martin.



--
Martin Rowley
West Moors, East Dorset (UK): 17m (56ft) amsl
Lat: 50.82N Long: 01.88W
NGR: SU 082 023