Call For Forecasts -- Predict 20 March Before 4 March

[Glacier Point]

Thanks for all the hard work Roger.

From my perspective, the forecast parameters: positive Arctic Oscillation; and, Global Wind Oscillation moving through a phase 2-3 evolution looks likely verifying so the correlation map is coming out at a very satisfactory value. The shift from the first forecast issued to the second was based upon total and relative angular momentum being lower and again this looks like the correct base assumption. The pressure pattern around the UK is not so good and a dropping off of the North Atlantic Oscillation looks the main culprit here although as noted, the broad arrangement across the grid was good.

With regard to the correlation vs pressure pattern, I think the grid was set sufficiently large to warrant an anomaly correlation scoring assessment. However, perhaps for next time we could increase the grid square to say 20N - 80N and 280W - 80E so as to equalise out the potential local variations and bring correlation and pressure anomaly ?

[PB22]

Thanks for all your work on this Roger, was certainly an interesting exercise and was fun watching the countdown to D-day and how things developed.

[Roger J Smith]

GP, do you know if they have ever tried to double the range on those model maps you were submitting? I mean, what if you took pressure anomalies and doubled them? The maps you showed were fairly accurate as to upper level flow, and I think they would have scored even higher with a larger pressure differential from highest to lowest. Is there a web location where we can view these maps? I would like to try using them as background flow for my own research data.

Here are some other non-statistical comments on validation.

The UKMO real time analysis shows a weak frontal wave near the Irish Sea coast of n.w. England and the German site wetterzentrale shows an occluded front northeast through Scotland. These are remnants of a somewhat stronger feature that gave 5-8 mms of rain in parts of Ireland on the 19th. Of our contest maps, the only ones that showed anything like these features include:

RJS has a frontal system in roughly the correct location

STF has a frontal system too far east by about half a day

PB shows a warm front in the general vicinity of these features

some other maps imply but do not show a frontal trough

the NW ensemble has a trough through the Irish Sea.

PIT mentioned in his verbal, no-map forecast, a front exiting Scotland on the 20th and improving weather behind it.

Now as to the ridge of high pressure running WSW-ENE through France, southeast England, Belgium, Holland and Germany with a centre (1035 mbs) in western Germany, the closest map depictions of this would seem to be:

MB seems to come the closest, with his centre too far west but a fairly accurate ridge axis.

ICE had a reasonably close ridge, centered too far west, and extending too far north in the eastern third of the grid

PB had a good intensity estimate and was just a touch too far south

RJS had a ridge parallel and 3-5 deg south of the actual

BLU had a ridge somewhat too far north and weaker than reality near the North Sea.

the GP maps had a ridge across the southern third of the validation grid

A few other housekeeping notes. I gave a hat tip to the PIT above, but BFTP had a verbal forecast that resembled my map forecast and suffers from the same intensity problems, while Jethro went for a cool easterly which matched conditions earlier in the week, so she wasn't too far off on her month-ahead stab in the dark either, really.

As for MB's 12-hour map timing error, of course that is the entrant's problem in any given case, when the call for forecasts specifies a map time, but in any case, I checked the validations and they are almost identical, if marginally higher for 00z reality.

In closing, I would say that any future map workshop/challenge should (as GP suggests) involve a larger validation grid of at least 70N to 40N and 30W to 30E, a different base map would be required to get all those grid points onto the base map. Thanks to Coast for supplying the base map for this exercise though. I would also suggest that another person step up to the plate with a date for another round of maps. I'm willing to do the validations in the same way except that I would concentrate on the pressure differentials more than correlation. Another rule change that I would suggest would be to cut off entries 21 days in advance to prevent even the mere suggestion of comparing the 16d GFS at day 17. Not that I think it did anyone any good, the results are pretty well scattered across the time frame of the entries.

[Stationary Front]

Thank you Roger for all your efforts over the last few weeks. For the record, my map was derived from looking at the last 25 March 20th's and comimg up with a sort of average chart from those previous 25.

[MurcieBoy]

Yes, many thanks from me too Roger for all your efforts.

My forecast was made using my usual method and variances were expected. As stated before, I shall be looking at the variances from these trials to see what other information I need to take into account to formulate the definitive DNA for a particular day.

[Thundery wintry showers]

Yes, thanks a lot for the work- it must have been a huge amount of effort!

I was pleasantly surprised by how close I got, as my prediction was a pure "stab in the dark" and was also very specific re. the positioning of lows and highs, which is always risky in a correlation-based analysis. As Roger mentioned, some of the earlier GFS runs threw up a pattern that bore a strong resemblance to my prediction, but as we got nearer the time everything got revised much further south.

Of course, if I'd got it right, it would've been more through luck than judgement!

[Coast]

Great fun and I have sort of proved something for myself in this particular instance.

With very little knowledge in comparison to many on here I thought that rather than try and guess, I'd take the Wetterzentrale archive since 1948 for the 20th March and see how many charts were broadly similar. After a dozen or so goes backward and forward, I found quite a few that shared similar pressure and positions which I then superimposed on each other to get a crude 'average'- this was my chart!

Hardly scientific and we have all talked about pattern matching hit and misses before. But I wanted to just see if rather than taking a wild stab, I could get close with a sort of average - it seems not!

Oh well, back to the drawing board and maybe I'll try something different next time.

[Iceberg]

Yes my thanks as well, my entry was blind luck with a little bit of historic analysis.

At the end of the day this rather re-inforces my view about most LRF's. When talking about general themes a month or so ahead, it's possible with little weather knowledge but some common sense to produce something that is roughly accurate a fair amount of the time.

We can talk about solar, moon, JTO correlations but are they really any better than the above. ?

[johnholmes]

Re the accuracy of any individual chart

I’ve outlined what seem to me to be the important aspects, where the surface lows and highs are, and the zone of strong winds.

Perhaps someone could come up with a marking guide?

The key to the ability to predict surface features 20-30 days ahead with sufficient accuracy is, perhaps similar to my suggestion in a previous post? That is to specify several locations around the UK with the predict for wind direction and strength, temperature and general weather?

On this occasion, as those using pattern matching was that the chart turned out to be a decent match with what has happened about the same time on average over a number of years. The real break through would be to predict a pattern that was not suggested by pattern matching and for it to be right. Do that 4 out of 5 times and you may be on to something.

Probably one of the most eagerly watched forecast outputs is from Stewart (GP). No slight intended Stewart but his chart is a fair bit out in locating the surface features so suggesting that many used the idea of pattern matching or memories of how mid March can sometimes be.

An interesting exercise though and again many thanks to Roger for all the work. I do believe that pressure checking is perhaps the way forward rather than correlation. All models individually and compared to one another from T+xyz out tend to zero in on the actual chart. Thus the spread from a long way out, usually, becomes very much less obvious as the time counts down. Hence why ensemble forecasting is seen, at the moment, by the major weather centres, as the current way forward in producing forecasts for customers.

Edited March 21, 2011 by johnholmes