GFS doing something different with the low on Monday/Tuesday... very slidey. Its now showing what the GEM was showing a few days ago! 

And that’s been the case for me, the last 10 days and the next 10 day, 7-9 degrees in the day and 2-4 degrees at night, with rain symbols for most days, depressing. 

The MetO raw has upped the ante here - it's now firmly in the mild category, offering up 10/10/10/9/11/9/9 for the next week (the 91-20 average here is 8.1 at this time of year). Given the way we've had winds with an easterly component for quite a while, and will do for a while yet, it really is quite dreary to see. 

It remains the case that the models are most reluctant to bring anything below -2 at 850 down here for the next 10 days - even the GEM, which magics up an upper cold pool in its 12z op run, manages to steer it effortlessly away from the SE. 

About the only good thing that can be said from a coldie point of view is that the chance of deep cold, the rainbow I keep mentioning, remains at around a 1-in-3 chance in 10 days+ time [as per GEFS for London via MC]. I'd really hope to see that edging closer, rather than just going "poof" as it did last time around.

We really have to give up as all scrooge horrible models kept saying NO to us despite we got favourable set up pattern.  If we were in the past, newspapers weather page say yes, ceerfax and teletext say yes, older BBC forecasters on TV say yes with charts showing LP moving east simple.  12z ensembles very zzzz, need zonality to return to bring some sunshine and cold nights with northerly after departing Lows, or bring the last year early Spring weather.  

https://www.wetterzentrale.de/en/topkarten.php?map=1&model=aifs&var=1&run=12&time=270&lid=OP&h=0&mv=0&tr=6#mapref 

I’m not bothered with the AI offerings to be fair. They’ve randomly wafted around more than Trump’s mullet in hot air.

I think Greenland HP and LP over Scandinavia is better than what we having now.  It need to break this current pattern that been stalement since the last cold spell.

Spot on. They're pattern-matching our previous climate, not our current one, and as I've said before it'd be more useful if they also assimilated previous runs' T+24, T+48 etc charts rather than just using ECM reanalysis data.

No different to any other model then.

As Darren eluded to, without definitive data or any in-depth understanding of how the AI models were trained, I’d say they work from scratch every run and have no bias apart from the original training data which I think is 1990-2020 or thereabouts. They therefore err on the side of what we might of expected 15yrs ago and not the new reality of 2026.

I’d be very interested if anyone has references to white papers or any publication dealing with the new AI modalities.

True. Too much random chaos beyond say the 144hr mark. 

I still recall that it was the GFS AI which first picked up on the high about a month ago that 'jarred' the pattern that we are still locked into today. 

You’ve explained this over and over. Maybe it’ll sink in eventually. 

I'm sorry but I do remember this forum from  20 years ago and back then it was all about the era of the modern winter, oh how the climates not like it used to be blah blah blah.  But then we had a run of very cold winter spells 08/09- 09/10 10/11. 

But once again we find ourselves saying the same old same old like you have above, and nsrobins post about using data that is "15" out of date as it doesn't represent our climate now. Do we not learn? 

I repeat what I say - the "AI" models are trained on old data and are pattern matching our old climate. 

https://www.ecmwf.int/en/about/media-centre/aifs-blog/2025/introducing-anemoi-training-ready-version-era5 

("Using a time period spanning 1979 to 2023")

https://www.ecmwf.int/en/newsletter/178/news/aifs-new-ecmwf-forecasting-system 

It remains a fact that snowy midwinter easterlies were common in the 80s. They were still frequent in the 90s, then we had the 2-week long easterly in 2005, which Philip Eden speculated might be the last proper easterly spell. I commented a couple of weeks ago, when the "AI" models were producing epic cold and snowy runs day after day, that they were pattern-matching 2005. They showed the classic pattern of the Siberian High ridging westwards over Scandinavia, then the "ping pong" of highs between Scandinavia and Greenland which kept the deep cold air topped up. This was a very familiar pattern and it's why the east and even the SE saw lots of snow back in the day - here, as I've said many a time, a fall of an inch or two would be nothing to write home about, whereas in the 80s and 90s falls of 6 inches were commonplace... or more. And even back in the 80s, comparisons were made with the classic spells of the 40s, 50s, 60s and 70s, which saw even more snow! I can't comment on those, though, as I wasn't alive to experience them.

Things were already changing 20 years ago, as Philip Eden noted, and of course since then we've not had any of those classic midwinter easterlies - we've had some northerly outbreaks, we've had "blink and you'll miss it" ridges zipping across Scandinavia and swiftly collapsing, but one thing we've not had is that prolonged, repeating "ping pong" easterlies of old. There are all sorts of other changes too, for example you could (rarely) get sleet or snow, briefly, as cold fronts cleared eastwards from Atlantic lows back in the 80s - as there was that much more cold air around.

It's clear as day that the pattern matchers, trained as they are on old data, have been using the examples of easterlies in their database in this current spell. After all, in 2005 we had a similar setup: a prolonged jet far to the south, a persistent block to the NE, and we did indeed get weeks of cold and snow (here at least) out of it. This time? Not even close, and as per the rainbow analogy the pattern matchers eventually twigged and removed the cold air and stellar synoptics. 

FWIW the ECM "AI" model uses the specially-prepared ERA5T dataset - which is a subset of their ERA5 reanalysis data specially formatted for training purposes. It covers plenty of fields but crucially doesn't include any historical model runs, just starting conditions. I would argue that having the model input from previous runs is essential - as it could then learn *why* things went wrong, rather than just how a similar - but not the same - situation evolved in the past.

EDIT: Two other points worth mentioning. One, the climate has warmed by 1.5C down here between 61-90 and 91-20 and yes, that has a marked effect on our weather patterns. And two, as those who've been around on here for 25 years will remember, pattern matching was once very much in vogue - the NOAA composite charts were much posted back in the day during winter. And you know what? More often than not they didn't reflect what eventually happened. The best that can be said about this spell so far is that we know in the 80s and 90s we'd have been much more likely to have seen cold and snow from it.

We are getting close to to a SSW by about the 13th February.  Cold Spring? Hope not.

As I mentioned yesterday, the ECM output has been very bullish about showing colder air reaching the UK in its runs over the last couple of days.

As for -8°C t850s, from a local perspective, we manage that here on Tuesday on the 00z GFS op run and then from Friday for a week or so. It is one of the colder ensemble options but not without support. In reality it probably means slightly colder rain.

FAX - LP mid-Atlantic circulating troughs and fronts around Britain at least to Wed, occasionally pepped up by a deeper trough as on Tue. HP 1026 mb Scandi and LP 993mb Italy having little effect on British weather

GFS Op - As above but the LP eventually sinks further south and tries to bring in an E/NE-ly Tue 10th/Wed 11th, never really getting to the S; then after a bit of uncertainty we're back to LP Atlantic and HP continent with Britain in the sandwich getting strong S-lies Sun 15th.

ECM - (only available to T+240, Thu 5th at time of posting,) Like GFS but that LP sinks south sooner (Fri 6th) and starts up E-lies then, but not that cold.

AIFS - maxima 8C to Wed 11th, dipping but possible recovery shortly after, rain on most days, winds mostly southerly with occasional E or N (as on 11th). Edinburgh, maxima around 5C but with dip as for London, rain most days, wind E-ly

GEFS - mean temp close to norm, supported by ens members until about Wed 11th, then ens spread develops markedly in both directions, rain on many days, more persistent and heavy after 11th esp in SW. In Scotland, once the spread develops there's a distinct cluster of colder members dragging down the mean, further N there's less pptn but more continuous and more likely to be snow (say, 60%)

(Brief ,but I'm away from home and have to go to breakfast)

Thanks for clarifying Darren. I thought I’d acknowledge this as the previous post mentioned me by name.