http://www.telegraph.co.uk/science/science-news/10738343/Forecasters-find-new-formula-for-long-range-weather.html

The news released to the press conceals as much as it reveals.

What does a "62%" success rate in winter forecasts actually mean? Rainfall? Temps? Frequency of storms? And is 62% really worth having, even if the MetO hope to increase this to 80%? That still leaves 2 chances in 5 that you as commercial operator might forward-buy gas, lay in stocks of salt etc and be left quite embarrased by the result.

Anyone able to fill in the details?

---

Somewhat more info at :

http://www.reportingclimatescience.com/news-stories/article/uk-met-office-announces-forecast-breakthrough.html

and

http://www.metoffice.gov.uk/news/releases/archive/2014/long-range-advances

 

 

---

This is the GloSea5 model; was presented last Spring.

http://adsabs.harvard.edu/abs/2013EGUGA..15.7649M 

According to Mr Fergusson it was used for past winter long range products. If so, not very accurate on a month by month basis when looking at the contingency planners product. Having said that, the weather we experienced over the winter was at the extreme end of anything that might be envisaged for the UK - a baptism by fire for any new forecasting product!

The presentation PDF.

http://www.ecmwf.int/newsevents/meetings/workshops/2013/Polar_prediction/papers_for_proceedings/SmithD.pdf 

It is reputedly very sensitive to seeing SSW events and is why the Met Office forecasts were playing around with the idea of a cold end to January - the model correctly forecast the warming but it did not propagate down - the rest of the winter is history.

I've not read the paper but would like to. TBH I've been astonished by the press coverage. As long ago as 2002 the UCL / Benfield Hazard NAO forecasts were suggesting an impressive and comparable correlation. See: 

http://www.ucl.ac.uk/mssl/climate-extremes/documents/NAO_Winter_Forecast_200203.pdf

As far as I'm aware these forecasts were discontinued. I can't remember the reason why but presumably it related to accuracy and their perceived value? 

The new MetO results sound as though they offer a similar level of accuracy and it's interesting to see they're focusing again on the NAO. Interesting because I thought they'd already established a decent correlation using May North Atlantic SSTs. See:

http://www.rmets.org/sites/default/files/pdf/presentation/20100317-folland.pdf

Predictability arises from re-emergence of previous May SST, pattern from under the summer thermocline. Can give cautious 6 month lead NAO forecasts -sign correct 2 out of 3 times.

TBH the more I think about this new announcement in the context of what has gone before the more confused I become. I need to read the paper (is it publically available?) because I may have got hold of the wrong end of the stick, but I'm not clear about the value add here.

---

The paper has this comment under the illustrations on page 11.

Figure 2: potential drivers of the NAO. (a) Atlantic tri-pole SST pattern and (b) mean sea level pressure response. Source: Rodwell et al 1999. (c)-(h) Observed composites of DJF mean sea level pressure associated with (c) ENSO (d) the 11 year solar cycle (e) major volcanic eruptions (f) the Quasi-Biennial Oscillation (QBO) (g) Atlantic multi-decadal variability (AMV) and (h) Pacific decadal variability (PDV). Units are standard deviations of annual (c-f) and decadal (g-h) timescales. Source: Smith et al 2012.

 

 

Utilising the NAO as a long range forecasting tool is foolhardy at best and will only lead to more ridicule if the MetO decide that this is to be one of the main tools they use for issuing LRF. Personally I think they should stick with the short to medium term, this is one area where they are head and shoulders above evryone else.

To echo DEW's question, I'm also baffled as to what a "62% success rate" implies, and not only as to whether it refers to temps or precipitaion or type of precipitation or whatever. 

For a start, how is "success" defined? How many categories are there into which each winter is divided? Two (below and above average)? In which case a 62/38 chance of success over a mere 20 years of hindcasting is no more significant, trustworthy or useful than a 50/50 one. Perhaps with, say, 5 categories (greatly above average, above average, average, below average and greatly below average) there might be a degree of significance, but no more than one of "this is slightly interesting; more work needed". 

A mere twenty seasons sounds a hopelessly inadequate testing period in any case. We have reconstructed, archived charts going back to the 19th century. Why not see how it would perform with those data inputted? You can't use the excuse of missing data for years in the distant past, as the whole point of an ensemble method (which this is) is to be able to tweak what data you do have in order to compensate.

Or have they committed the cardinal sins of cherrypicking years or, worst of all, devising the system on the basis of trying to make it fit recent results?

---

This was one of the things which surprised me. If you check out the UCL/BH paper you'll see their hindcasting apparently went back a lot further and appears to have achieved a similar correlation. The MetO developments were the lead story in The Times earlier this week (April 2nd) and the whole epsiode has left me perplexed. 

---

It does sound like either a non-story, an old story or a very old story. 

Why cobble together a press release claiming long-range accuracy (though failing to demonstrate it) at this particular moment? Perhaps partly to try and boost the credibility of long-range forecasts at a time when the IPCC has issued its own long-range forecast to great fanfare and accompanied by unequivocal demands that it be taken as gospel?

And perhaps partly to get a bit of old-fashioned publicity for the Met Office at a time when a long-range forecast is dominating the news? This seems to be a common reason for press officers to rehash old stories in such a fashion.

---

It is an old story. If you look at the presentation links in my first post you will see it goes back to Aril 2013.

Yes. In that case The Times and its editors have a few questions to answer. Perhaps the hack responsible had not realized this was a story from April 2013 rather thanApril 2014?

---

The blog on the Met Office web site is dated 2nd April 2014, so it is a new story.

http://www.metoffice.gov.uk/news/releases/archive/2014/long-range-advances

---

It looks as though the reference that nouska mentions is a presentation fron last year but that it has only just been submitted for peer review and publication in Geological Research Letters this week with little or no change. Plus also a met office press release which the press have just picked up on

 

---

Thre are two parts to using the NAO for winter forecasting.

First  you have to forecast the seasonal NAO(DJF).This usually will be a sign and anomaly figure.METO do make this forecast for each winter but I have yet to see any stats on accuracy of this forecast.It may be that just getting the sign right is enough to count as a correct forecast.

Secondly even if the forecast sign and value are OK then you have to look at how NAO values relate to the very  important winter temperature value.

I've just looked at the NAO for the winters 1873/74 to 2012/13.In these 140 years the NAO signs + or - and the expected  CET temperature anomalies ie warm or cold were correct in 72.5% of years (NAO + years 70% correct,NAO - years 75% corrrect.

Looking at values as well as sign METO are on record as saying that the correlation  beween actual NAO values and accomanying CET is 0.60 ie the NAO values account for c 40% of the CET variation.As forecast NAO values are  never going to be 100% so correaltions between firecast NAO and actual CET are likely be below 0.60.

Doies anyone have a link to buy/download the paper?

Thanks

 

Roger, Meteo France has a nicely illustrated Met Office power point PDF outlining the findings - the main article is paywalled.

http://www.meteo.fr/cic/meetings/2013/s2d/pres/tuam_7.pdf

 

Thanks Nouska.

I spoke with one of the authors at UCL a few years ago and the reason these forecasts were stopped was actually down to funding or lack of, because of the recession.

They did continue with their hurricane forecasts, I believe.

---

Interesting because as I read the MetO are saying there's commercial value with this level of accuracy which is apparently similar to what UCL we're claiming to have achieved. If that's the case you'd have thought UCL would have had no problems getting funding. The whole episode last week seems very odd to me.

---