Trom

This is exactly the conclusion I formed some time ago. There's no doubt that they have an effect but the correlations between all these measures or at least current understanding of them seems to be limited. I very much admire Tamara et al who are willing to share there knowledge as our understanding of the interplays between them develops. They are often the target of some posters when things don't go to plan but I think we need to accept that our understanding and hence usefulness of these measures In terms of forecasting rather than hind-casting will improve over time. pioneering is how I see this area right now and I have great respect to those who are willing to share their musings publicly.

With regards to verification stats are they for the global output or more geo specific. The only reason I ask is that if they are for the model output as a whole I'd imagine the UK with its complicated global position would have lower correlations. Is this a fair comment to make? Edit: Sorry ignore I see this has been answered later in the thread.

Is the GFS the most progressive post it's update? It was the first to hold the Atlantic back.

Poor old GFS getting a bashing. Don't forget it was the first model that attempted to hold back the Atlantic when Meto and ECM were being far more progressive. Noticeable as the GFS has the reputation (deserved or not) for being the most likely to default to zonal. Does it still have this post upgrade? It certainly has a tendency to sniff something out then underplay it relative to ECM before moving towards the Euro output. I read somewhere that it's slightly inferior stats were due to the initialisation data and that it had some years ago been run with ECM initialisation data which improved it's verification stats. Also thank you to those that have championed more volatility information being posted with mean charts. I notice quite a few posts of means along with ensemble charts which personally I've found to improve my ability to interpret those mean charts without having to delve around to find volatility. I'm guessing if the trends continue this place will be a lot busier and all the behavioural dissonances will rear their heads. Anyway so much more positive than a few days back.

"But as shown in this draft paper Seasonal forecasting of the polar stratosphere and its coupling with the troposphere (Seviour et al - with Scaife again) the Northern hemisphere zonal wind forecast made in November for Dec-Feb has a correlation of only 0.16 with the ERA-Interim reanalysis - https://github.com/wseviour/glosea5-paper/blob/master/paper.md" 0.16 wow that's almost random.

This goes back to the point I've made a few times on here, that mean without some kind of dispersion measure is an odd way to describe a data set. I know of no other field of study other than this website that mentions one without the other. It's just bizarre to describe distributions without a volatility measure. But of course if runs tightly cluster around a mean it tells us that there is a trend. By their very definition mean charts wont be able to sniff out a model change (as the positive and negative deviations will cancel out to some extent), but when the standard deviation of ensembles increases it tells you that there is greater uncertainty and hence the mean becomes less significant relative to volatility as the important measure. Of course when the standard deviation of runs is low relative to the mean then it becomes useful. I'm not having a go at any individual posters but really think that if you are going to post a mean then you need to post the ensembles so give it some kind of context, so that forum users can glean whether they are looking a significant trend or just a sign of uncertainty - both useful to know when interpreting the models. Observing frequencies of particular outcomes can also be useful to determine implied probabilities in the forecast data so get looking for clusters.

Wow I found that really cool. We went from mildish and hammering rain to sleet and snow blizzard in about 10 minutes and to see the dew points plummet like that isn't an every day occurrence. In fact I can't remember a snow to rain event quite as extreme as that in my lifetime. I agree I've seen deeper snow but it's still 2 inches when I was wondering if I'd see a flake. You are not that far from me. Always a funny thing with snow though small differences can lead to different precipitation rates. Must of been in the sweet spot for this one. Having said that my family is living in the garden in a caravan whilst our house is refurbished and the up coming weather is going to be tough for us. Can't deal with ice days when your water supply is a relatively poorly insulated hose!

Thank you - you don't see that very often in these Isles. I guessed it had to look a bit like this from the charts but wow that really is a dead Atlantic stream. Like someone flipped a switch. Not really any energy heading North or South from the US let alone West.

Would anyone happen to have a chart of the jet stream for roughly the same time. Sorry I don't have access right now.

Excellent post - the obsession with mean runs (which I haven't noticed in prior years) in FI seems very odd to me. As an Economist looking at data distributions and describing them by mean without a dispersion measure seems pointless. Even with a dispersion measure you would still want to see skew. From my perspective all these charts give you is a very broad brush trend of all the ensembles. When you look at the dispersion patterns in FI they are so vast. Median would be a better measure given how most people use them. There are dispersion measures using standard deviation which give a good idea of how well the op is supported relative to ensemble mean. Nick F posted one back on page 168. I'd like to see much more of this data being posted on the model thread. The charts he posted showed exactly where the models had the greatest variability in the northern hemisphere. We have a lot of data produce by the models but this measure goes under reported on the forum. Essentially it's telling us where the models are solid and where they are uncertain. Nick F if you could post more (or let me know where they can be found) that would be great.

The map below shows the current distribution of Buzzards in the UK. Are you expecting an easterly to shift them west across the country? Sorry for the tease

Ok I've got that now. So right shows actual standard deviation of operating run from ensemble mean. The left shows the standard deviation divided by long run mean standard deviation so we can get an idea of whether the right hand diagram is particularly unusual. or close to the normal spread. Thanks for your time.

Hello Nick can you give me a little info on how to interpret that. What mean is being used to calculate these measures is it the current ensemble mean or some kind of long run average? The chart title infers the former in the first diagram. Isn't it just showing where the 500hpa heights are the most uncertain per the ensemble data and where the models show more confidence? I only ask as the second model shows some areas +/- 25 standard deviations from the mean which would give a huge range of potential outcomes which has me questioning whether I'm interpreting it correctly. (Also as an aside is there median data available as this is less likely to be skewed by outliers).

With regards to your second point Matt if you had factored that in it would be classic behavioural bias of the gambler's fallacy. I lecture a bit of behavioural finance and find the model thread to be rife with it. For those of you who have a little interest in this area wikipedia has a great summary of the cognitive biases know to date https://en.wikipedia.org/wiki/List_of_cognitive_biases Kind of model related so I hope the mods will let it slide by

Some digs at the meto tonight on the model pages. Reality is they do a great job but the info we get is terrible given it is tax payer funded. We just get the statements about the long range models but don't see the output. I get that they need to attract external revenue but so does ECMWF and even they give us more output. Personally to have this wealth of knowledge domestically and tax payer funded should mean we get to see a lot more of it in my opinion. The loss of the BBC contract might mean the meto come under greater pressure to release more of their data. It's great that Ian F posts but it's not really a proper substitute for actually seeing the data. With the loss of the BBC they move further towards being an academic institution. As a lecturer myself the mandate I was told was "publish or die" and I think this resonates in academic circles. Anyway this thread is where you are meant to be able to moan - so there this one's been bugging me for some time.

The mode is a way of describing a distribution of data in this case ensemble runs. mode is the most frequently occurring observation. Unlike means and medians data can have more than one mode. Bimodal simply means two modes in other words their are two observations occurring with the same frequency. In the context of Ian's post it would mean there are two distinct clusters in the ensembles.

There is a slight irony in posting "predicting the weather 10 days ahead made a fool of us..." before predicting cold and ice by the end of autumn in the very next sentence. Nonetheless synoptically we remain in a much better place than in previous years so I do see your point. The vortex continues it's fragmented movement and looks a long way from being cohesive. Can't see any signals for cold and ice for the whole country as yet but the potential is certainly there. Just nice to have a season that is thus far free from raging zonality. Think back to the last two years with the compact vortex over Greenland and the high pressure constantly to the south and east of the UK and the difference is stark. I do agree at this stage of autumn/winter to focus on the specifics of each run is a little pointless but to focus on the long run broad patterns is encouraging if colds what you want. Given this is my house at present and my family is living in the garden in a caravan the law of the sod says it's bound to be cold.

Well Costa I'm not sure I totally agree with your comment "Unless your up in the highlands or Pennines, I can't see -4 to -6 uppers cutting the mustard!". It seems to me that the volatility of the weather has increased over my life time and we seem to see more extremes. A lot seems to be due to sea temperatures and their effect on global pressure rises and falls which seem to have thrown a lot of theories up in the air. It's important to remember that this year has been very unusual what with a massive el-nino, a massive SSW break up at the strat level right now and don't forget hurricanes in the atlantic in January! A -4 to -6 in a NE who knows? Unlikely to actually happen but I would not totally disregard those synoptics. Nor would I bet my house on them.

Yes but a huge upgrade - 10 uppers were no where near the UK on previous runs. in fact -5 was only glancing certain areas and Scotland on previous runs. Nice if it holds come tomorrow mornings output.

I know it's small consolation but I have to visit Wisconsin regularly in Winter (my head office is there) and it often gets down to these kind of crazy temperatures. It's actually not much fun as you can't spend any significant time outside. It is cool that the Mississippi freezes in the town where I visit (you can drive out cut a hole and go ice fishing). But in pretty much writes off outdoors as a place to be. The weather for frolicking and building snowmen it is not. Still the models show the strat dominating the trop forcings. Some signs of change going forward but nothing that indicates persistent cold at the moment. As a few posters have mentioned the synoptics actually look good at face value until you realise there's no cold near by to tap into.

Yes agreed it can be faster. I don't remember the confidence interval that was used for that range of values. I remember at the time being struck by the size of the range rather than its maximum and minimum values. Was 2009 an outlier or is this recurring theme? We certainly have had interesting background variables in the trop this season with an el nino event coupled with warm Indian ocean waters, the Kara heights, a series of rossby waves hitting the strat and a hurricane in the Atlantic in January. Now we get to see how a warming will interact with these variables. A lot water to flow under the bridge before we know where we stand there but it should be interesting to how the pieces of the jigsaw fall.

I seem to remember that when I was doing back ground reading on SSW last year the academic view was between two weeks to two months for the effect to really be felt in the trop. So certainly some variability there if my memory is correct.

He mentions "extensive" rather than "stronger" does this mean he's talking about area rather than extremes of temp?

Himalayas I believe (at the current time)

Feb is rare? So much to play for especially given the models inability to cope with the synoptics.