On Mar 6, 7:36*pm, "Eskimo Will" wrote:
"Martin Rowley" wrote in message

...

On 06/03/2013 18:23, John Hall wrote:
In article ,
* Martin Rowley writes:

... I understand that the UK Met suite of models is catching the EC
output up. EC has been regarded as the best output for some time,
though I'm not sure why that should be.

Well the graph contained within the linked to document appears to show
that it is.

... What I'm not sure of is *why* the EC is consistently better .. does
anyone know?

Isn't it all down to a better analysis?

Will
--

It is an interesting link for keeping up to date on what UKMO are up
to.

But, have I missed something or what?
Verification of what?
I assume output for each grid point against outcome as our erstwhile
Dawlish poster calls it.
But what output? Cloud cover? Precipitation?

I am continuall y trying to convince neighbours and Jo Public that the
Met O forecasts are not rubbish.
But their perception is coloured by the errors in occurence and timing
that are encumbent with the local forecasts and the postcode
forecasts.
Particularly of cloud and rain.

Len
Wembury

"Martin Rowley" wrote:
... What I'm not sure of is *why* the EC is consistently better ..
does anyone know?

[and]
On Mar 6, 7:36 pm, "Eskimo Will" wrote:
Isn't it all down to a better analysis?

.... that would seem to be the 'core' difference between centres: after
all the physics, mathematical handling/manipulation, computational
'gizmos' etc., must be well-known through peer-reviewed literature and
exchange working of personnel.

But /why/ the differing analyses needs more work. If one centre has got
a better handle on the initial state that should really be fed through
to other workers; given that all analyses are based on a previous model
iteration (often run several times to provide a nominal 'base'), then
it's almost a self-fulfilling process; better analysis - better
forecast - better 'background' for next cycle - better assimilation of
new data - better forecast etc.

[and]
Len Wood wrote:
But, have I missed something or what?
Verification of what?
I assume output for each grid point against outcome as our erstwhile
Dawlish poster calls it.
But what output? Cloud cover? Precipitation?
I am continuall y trying to convince neighbours and Jo Public that the
Met O forecasts are not rubbish.
But their perception is coloured by the errors in occurence and timing
that are encumbent with the local forecasts and the postcode
forecasts.
Particularly of cloud and rain.

.... I'm pretty sure that such verification is based on differences based
on contour/isobaric analyses - but others may know better. What I /do/
know (from occasionally seeing the output) is that there is a *vast*
amount of explicit/deterministic & probabilistic data now output from
the EC factory at Shinfield which is valued highly by all NMSs
throughout Europe and elsewhere.

I agree with your comment about day-to-day / local specifics; today was
a good case in point. My wife had me pottering in the garden yesterday
(she can't do it herself) putting in a few plants 'because they said it
would rain today [Wednesday]': no rain - back to the watering can.

Martin.


--
West Moors / East Dorset
Lat: 50deg 49.25'N, Long: 01deg 53.05'W
Height (amsl): 17 m (56 feet)
COL category: C1 overall

On Wed, 06 Mar 2013 21:23:29 +0000
Martin Rowley wrote:

But /why/ the differing analyses needs more work. If one centre has
got a better handle on the initial state that should really be fed
through to other workers; given that all analyses are based on a
previous model iteration (often run several times to provide a
nominal 'base'), then it's almost a self-fulfilling process; better
analysis - better forecast - better 'background' for next cycle -
better assimilation of new data - better forecast etc.

Used to be that the ECMWF model didn't run until nearly 12 hours after
data time, about 9 hours after Met Office, so it would obviously have
more data. Don't know whether that's changed.

--
Graham P Davis, Bracknell, Berks.
Carlos Seixas, Sonata nº 1: http://www.youtube.com/watch?v=YXox7vonfEg
And for something completely different, Cumberland Gap:
http://www.youtube.com/watch?v=AsU-LTwx8Co

On 07/03/2013 08:14, Graham P Davis wrote:
On Wed, 06 Mar 2013 21:23:29 +0000
Martin Rowley wrote:

But /why/ the differing analyses needs more work. If one centre has
got a better handle on the initial state that should really be fed
through to other workers; given that all analyses are based on a
previous model iteration (often run several times to provide a
nominal 'base'), then it's almost a self-fulfilling process; better
analysis - better forecast - better 'background' for next cycle -
better assimilation of new data - better forecast etc.

Used to be that the ECMWF model didn't run until nearly 12 hours after
data time, about 9 hours after Met Office, so it would obviously have
more data. Don't know whether that's changed.

.... some information he-

http://www.ecmwf.int/products/foreca...ry_system.html

It looks as if what we would call 'cut-off' times have been reduced (I
can access EC OP products around T+6 for example, which must include
post-processing time), but that not all available observations are used
in /that/ run .... they are, however, used eventually to drive the
background for the *following* run via the update/background 4DVAR
analysis routines. Very clever!

However, although I'm not up to speed on current MetO procedures, I'm
pretty sure that they too follow a similar process, so that can't be the
whole story.

Martin.


--
West Moors / East Dorset
Lat: 50deg 49.25'N, Long: 01deg 53.05'W
Height (amsl): 17 m (56 feet)
COL category: C1 overall

On Wednesday, 6 March 2013 19:52:02 UTC, Len Wood wrote:

Verification of what?

This is what I find most poor about most verifications: stuff like 500mb root mean square error is hardly of use to the person on the ground who might want to try and understand which forecasting model is best.

Would be great to see how well Lamb weather types or Grosswetterlagen can be forecasted in advance in different models - e.g. Scandinavian block is forecast on average 5.5 days ahead by ECMWF but only 4.5 days ahead by the GFS. This would, I guess, require a human input - but something I notice that Bruce Messer tried his hand at on his blog a few months ago.

It would also be interesting to contrast the raw model T+0 output with the observations assimilated into the model to compare these differences...

Richard