The inaccuracy in the forecast for today has led to alot of criticism of
the Met Office on this newsgroup. However I feel that it highlights an
important issue in weather forecasting; the chaotic nature of the
atmosphere.

The atmosphere is a chaotic system and as such predictions of its future
state at a given time are always prone to uncertainty. The extent of
that uncertainty depends on the uncertainty in your initial conditions
which you feed into your numerical model.

Forecasts relating to areas of rapid development are very sensitive to
the initial conditions. A slight change in the initial state will lead
to large differences in the forecasts. The storm today is good example
of such a sensitive region, as was the 1987 storm.

The best way to improve forecasts in these situations is either to 1)
improve your estimate of the inital state of the atmosphere at that time
-this can be done through an improved data assimilation system or by
targetting extra observations to be taken in this region. or 2) to take
account of the uncertainty through using an ensemble forecast.

In 1987, the storm was not forecast to develop as much as it did, todays
forecast overestimated the development (and also produced a slight error
in the track which unfortunately led to a big difference in the weather
experienced on the ground). However, an ensemble of forecasts would have
revealed the large uncertainty associated with these forecasts. Extra
observations would have helped reduced the spread (uncertainty) of the
ensemble and given us a more accurate forecast.

The point I am trying to make (in a very long winded way) is that a
single forecast will not yield as much information as an ensemble in
these sensitive conditions. People saying "the GFS model did better than
MO" are missing the point. They are both to all intents and purposes
very similar models giving in effect individual ensemble members. The
fact that they are different merely represents the high uncertainty of
the situation, not that one is "better" than the other.

The best way to have forecast this storm more accurately would have been
to take more observations from the sensitive region. This is an exciting
new area of research. People interested in this should look at the
THORPEX program:

http://www.mmm.ucar.edu/uswrp/programs/thorpex.html

Hopefully, within the next 10 years the results of this should improve
our forecasts dranatically for situations like these...

"Steven Glean" wrote in message
...
The inaccuracy in the forecast for today has led to alot of criticism of
the Met Office on this newsgroup. However I feel that it highlights an
important issue in weather forecasting; the chaotic nature of the
atmosphere.

The atmosphere is a chaotic system and as such predictions of its future
state at a given time are always prone to uncertainty. The extent of
that uncertainty depends on the uncertainty in your initial conditions
which you feed into your numerical model.

Forecasts relating to areas of rapid development are very sensitive to
the initial conditions. A slight change in the initial state will lead
to large differences in the forecasts. The storm today is good example
of such a sensitive region, as was the 1987 storm.

The best way to improve forecasts in these situations is either to 1)
improve your estimate of the inital state of the atmosphere at that time
-this can be done through an improved data assimilation system or by
targetting extra observations to be taken in this region. or 2) to take
account of the uncertainty through using an ensemble forecast.

In 1987, the storm was not forecast to develop as much as it did, todays
forecast overestimated the development (and also produced a slight error
in the track which unfortunately led to a big difference in the weather
experienced on the ground). However, an ensemble of forecasts would have
revealed the large uncertainty associated with these forecasts. Extra
observations would have helped reduced the spread (uncertainty) of the
ensemble and given us a more accurate forecast.

The point I am trying to make (in a very long winded way) is that a
single forecast will not yield as much information as an ensemble in
these sensitive conditions. People saying "the GFS model did better than
MO" are missing the point. They are both to all intents and purposes
very similar models giving in effect individual ensemble members. The
fact that they are different merely represents the high uncertainty of
the situation, not that one is "better" than the other.

The best way to have forecast this storm more accurately would have been
to take more observations from the sensitive region. This is an exciting
new area of research. People interested in this should look at the
THORPEX program:

http://www.mmm.ucar.edu/uswrp/programs/thorpex.html

Hopefully, within the next 10 years the results of this should improve
our forecasts dranatically for situations like these...

What a great post Steven

Completely echos my views. Agree 100%

Regards
Paul

Hmm.... in my opinion the UKMO rely far too much on 'Super computers'
instead of getting back to old nitty gritty of proper forecasters\weather
ships\observations\practical stuff etc.
As an enthusiast I certainly woudn't get on their back for this one, but
short term detailed forecasts are getting worse by the year, forgetting
about today's storm. This is because the aim now is to improve on long range
forecasts, Global Warming programs.
Of course the chaotic nature of the atmosphere plays a part, 1+1 never = 2
in this game, but super computers and Global Warming programs are not the
answer.
Didn't they pull out Weather ships prior to the '87 Storm?


--------------------------------------------------------
"Steven Glean" wrote in message
...
The inaccuracy in the forecast for today has led to alot of criticism of
the Met Office on this newsgroup. However I feel that it highlights an
important issue in weather forecasting; the chaotic nature of the
atmosphere.

The atmosphere is a chaotic system and as such predictions of its future
state at a given time are always prone to uncertainty. The extent of
that uncertainty depends on the uncertainty in your initial conditions
which you feed into your numerical model.

Forecasts relating to areas of rapid development are very sensitive to
the initial conditions. A slight change in the initial state will lead
to large differences in the forecasts. The storm today is good example
of such a sensitive region, as was the 1987 storm.

The best way to improve forecasts in these situations is either to 1)
improve your estimate of the inital state of the atmosphere at that time
-this can be done through an improved data assimilation system or by
targetting extra observations to be taken in this region. or 2) to take
account of the uncertainty through using an ensemble forecast.

In 1987, the storm was not forecast to develop as much as it did, todays
forecast overestimated the development (and also produced a slight error
in the track which unfortunately led to a big difference in the weather
experienced on the ground). However, an ensemble of forecasts would have
revealed the large uncertainty associated with these forecasts. Extra
observations would have helped reduced the spread (uncertainty) of the
ensemble and given us a more accurate forecast.

The point I am trying to make (in a very long winded way) is that a
single forecast will not yield as much information as an ensemble in
these sensitive conditions. People saying "the GFS model did better than
MO" are missing the point. They are both to all intents and purposes
very similar models giving in effect individual ensemble members. The
fact that they are different merely represents the high uncertainty of
the situation, not that one is "better" than the other.

The best way to have forecast this storm more accurately would have been
to take more observations from the sensitive region. This is an exciting
new area of research. People interested in this should look at the
THORPEX program:

http://www.mmm.ucar.edu/uswrp/programs/thorpex.html

Hopefully, within the next 10 years the results of this should improve
our forecasts dranatically for situations like these...

Brilliant and best post on the subject in my opinion, Stephen. I hope more
people read it whatever their standpoint.

Dave

In uk.sci.weather on Mon, 12 Jan 2004 at 16:46:43, Steven Glean wrote :
The inaccuracy in the forecast for today has led to alot of criticism of
the Met Office on this newsgroup. However I feel that it highlights an
important issue in weather forecasting; the chaotic nature of the
atmosphere.

Snip

I think the real issue is the loss of public confidence that ensues when
dire warnings are not borne out.
--
Paul Hyett, Cheltenham

Email to pahyett[AT]activist[DOT]demon[DOT]co[DOT]uk

...The best way to improve forecasts in these situations is either to 1)
improve your estimate of the inital state of the atmosphere at that time
-this can be done through an improved data assimilation system or by
targetting extra observations to be taken in this region. or 2) to take
account of the uncertainty through using an ensemble forecast...
Admirable sentiments overall-
caveats-
recent work has suggested that the important factors in errors out to 7-10 days may not be chaos but
ongoing deficits in the models .ie there's still lots of room for imrovements.
Ensemble forecasts do not always pick up outlying ,severe events.What do you invest yr
computational/financial resources in-
improving model resolution,ensemble forecasting,and/or data aquisition and assimilation?

--
regards,
david
(add 17 to waghorne to reply)

Waghorn wrote:
...The best way to improve forecasts in these situations is either to 1)
improve your estimate of the inital state of the atmosphere at that time
-this can be done through an improved data assimilation system or by
targetting extra observations to be taken in this region. or 2) to take
account of the uncertainty through using an ensemble forecast...
Admirable sentiments overall-
caveats-
recent work has suggested that the important factors in errors out to 7-10 days may not be chaos but
ongoing deficits in the models .ie there's still lots of room for imrovements.
Ensemble forecasts do not always pick up outlying ,severe events.What do you invest yr
computational/financial resources in-
improving model resolution,ensemble forecasting,and/or data aquisition and assimilation?


-yes, I agree, obviously the models are not perfect. Ideally, an
ensmeble forecast should take into account ALL of the uncertainty
involved in the forecast ie. model error, initial condition error and
data assimilation error. At the moment most ensemble forecasts
concentrate on initial condition error as that is judged to be a
significant source of error. I don't doubt that model error may be just
as significant (if not more so in some cases). However, taking account
of this model error properely is not an easy task.

If you have taken into account all of the sources of error mentioned
above then all outliers or severe events would be picked up. But, I
suppose the downside of taking into account all of this uncertainty is
that your forecast will just show a very large range of possibilities
and you will be left none-the-wiser as to what is actually going to happen!

IMHO, the best improvements in short-medium range forecasts would be
made by investing in a better observation network that would include the
possibility of targetting specific sensitive regions (using aircraft,
dropsondes, etc.) each day to reduce the analysis error. That and an
improved data assimilation system would (I believe) really make a
difference to situations like those experienced over the last few days.

I suppose in the end, where you invest your resources depends on what
timescale you are interested in forecasting...

I'd love to discuss this subject further if anyone has any opinions or
comments...