General Notes On GFS Checks

[johnholmes]

hi all

Yep, the GFS checks for T+168 are back. Read the 34th check at T+168 GFS checks.

As soon as I get time I will do an overall summary on, what is now, 11 months of checks on this run, that is 12Z for T+168 hours.

Overall, it cannot be described as anything but impressive. From T+144 the overall pattern is usually correct. By that I mean the isobar flow, main pressure centres, 500mb flow, and 'surface' flow over the UK, is pretty accurate.

Those who have read my posts from last October forgive this next couple of paragraphs.

For those of you who have joined Net Weather over the past few months just a few comments.

GFS is NOT infallible. It is just about the best computer prediction there is, about up to the Met Office out to T+96. Beyond that, its way out in the lead. Okay, it does get it wrong, but rarely within the T+120 hours time frame.

Again I hear comments like, so what, should not a computer be able to go way beyond that and be even more accurate? In theory yes, but the real world of Meteorology is not like that. Improvements at the time scale T+120 and beyond will happen, but only slowly. Remember that 5 years ago to be able to predict the general weather pattern was well nigh impossible beyond about T+96, ten years ago it was the same beyond T+72. So it is improving, but be patient, give the Meteorologists and computer experts time. In my view we should, in most of your lifetimes, not mine(!), be able to be as accurate at T+368 as we now are at T+144. Beyond that Im not sure.

Anwway, enough of my rant.

Enjoy the GFS thread, and enjoy all that Net Weather has for us all.

regards

John

[The Enforcer]

Let's hope this puts an end to the nonsense of people posting charts well beyond T+144 and then saying "look - we're about to have another 1962/3".

[TG312274]

Well anything beyond 5-6 days is still very difficult to forecast of course. Last Sunday on Countryfile they were talking about snow across the Scottish mountains this weekend. They've now gone very quiet on that.

[Portland Paul]

Well anything beyond 5-6 days is still very difficult to forecast of course. Last Sunday on Countryfile they were talking about snow across the Scottish mountains this weekend. They've now gone very quiet on that.

<{POST_SNAPBACK}>

I had my doubts all the time about the snow showers on Sunday gone!

[Roger J Smith]

Could see some snow on highest peaks of Scotland Saturday, coldest air sweeps across the region around 09-15z then there is gradual warming at most elevations Sunday. I don't think there will be any snow below 3000 ft though.

On this subject of accuracy of models, I have been taking similar notes over North America for the Canadian global model, which is probably about on a par with the GFS. I have roughly the same perceptions, that it is generally accurate on large scale patterns 96 hours and to some extent 120 hours in advance. There are a few exceptions, though. For example, a very strong east coast storm developed rapidly from the Gulf of Mexico on Dec 24, 2000 and by the 26th it was a 958 mb low near Halifax, NS. Before the 72h forecast there was little hint of this system. All of the North American models had a poor 96h forecast track for Katrina, placing it near Panama City FL at the time when it was actually approaching New Orleans.

I was actively forecasting in the mid 1970s and the models at that stage were fairly reliable to perhaps 36h so things certainly do show a steady improvement.

The factor that might place a ceiling on model accuracy beyond 144h is that the energy producing weather systems may be (according to research I am doing) partly supplied by interactions between the earth's magnetic field, atmosphere and the solar system magnetic field. If that is so, you can understand that no matter how sophisticated the computer models become and how much data we acquire to initialize them, they will have no further capability than they do today to anticipate the peaks of energy that might become available ten or fifteen days ahead of time. There is also, within this, the second question of whether or not such interactions would all be predictable, or whether some of them are random (Sun emits flares at day 8 of forecast with effects on earth's weather from day 10 onward, for example). I have not really made that much progress on that question, but from a theoretical point of view, there is no guarantee that super-computers will necessarily be able to produce accurate long-range forecasts more than about ten days in advance, unless we understand these processes and can incorporate them into the models.

The way that one researches such a question is to produce long-range forecasts from a theoretical model, beyond the time horizon, where accuracy or lack thereof would give an indication of the validity of the assumptions being made. I am doing that on a regular basis, which is partly why I have expanded my work day from the previous 12 hours to a more intensive 14-16 by looking in detail at the UK and western Europe. I also look in less detail at other regions. Part of the research is to compare timing functions in different regions, to see how postulated cause and effect actually plays out on a regional basis, not just in the original area of interest, North America.

To give one example of this, I will be checking weather maps for strong low pressure near the UK on various dates that I have identified as likely candidates for storm development. Once I get enough info, I will use this to make long-range forecasts that would provide a basis for "how to" extend the models beyond the 10-day time frame where they normally devolve into a sort of random noise at present. All of the above assumes that various other meteorological principles would remain in play into this longer time frame, so that it's a blend between the external cause and effect and the more conventional energy cycles of atmosphere, SST and all the factors that go into the working parts of the GFS type models already.

To give one example of a strong low affecting the UK, likely to occur beyond the usual time horizon of the models, mark down October 23-24.

[johnholmes]

interesting post Roger, perhaps we should ecgange pm's on what we are each doing.

regards

John

[johnholmes]

hi Roger

tks for pm will reply as soon as i get time to read it closely.

In the meantime why 23-24 October, how did you get that?

regards

John

[Big Bear]

Has anyone noticed the possible link up of the Greenland & Siberian HP in about +8 days time??

What would happen if this came off now, and what would happen if it occured mid-winter??

Thanks

BB

[Adi F]

In the meantime why 23-24 October, how did you get that?

<{POST_SNAPBACK}>

Good Question John, thats 24 days or 576 hours out.... which models go that far out? i take it they are not for public use or are experimental

[Arrows]

If you look at the +360 and the +384 setup you see a deep low to the nw moving south with a MB 970. The low in the +360 to the north eventually merges with the low from the west and pushes westwards which is different from whats been happening recently where everything has moved NE. The high thats protected most of the Uk from the high winds has 1020mb and 1025mb. When this low comes aross the high is only 1005mb and 1010mb with -15 over most of the Uk. Not strong enough to keep the low away. The low from +372 and +384 intensifies and tightens as it moves into the fragile front already over the UK. Take a look and tell me it doesn't look good. I'm thinking about asking met-check there opinion.

Edited September 29, 2005 by theredarrows

[The PIT]

If you look at the +360 and the +384 setup you see a deep low to the nw moving south with a MB 970. The low in the +360 to the north eventually merges with the low from the west and pushes westwards which is different from whats been happening recently where everything has moved NE. The high thats protected most of the Uk from the high winds has 1020mb and 1025mb. When this low comes aross the high is only 1005mb and 1010mb with -15 over most of the Uk. Not strong enough to keep the low away. The low from +372 and +384 intensifies and tightens as it moves into the fragile front already over the UK. Take a look and tell me it doesn't look good. I'm thinking about asking met-check there opinion.

<{POST_SNAPBACK}>

FI stuff and will more than likely be gone from the next run. GFS has modified the cool spell this weekend (downgrading) but has overall stuck with it. I'd only use T96+ for trend settings and only then if more than one run kept the same trend. Lately the GFS has been really switching around a lot more than it used too I feel.

[johnholmes]

not so Pit, at least in my opinion with the 12Z run, its been remarkably consistent, which is what I was trying to highlight, evenwith the fairly unusual state of the atmosphere in the Gulf of Mexico over the past 6-8 weeks.

Yes, out to T+96 and its rare it ever is wrong in the basics, the flow over the Uk and its source, even the isobar is often predicted within 50-60 miles of where it ends up. To me that is astonishing. Even out to T+120 its unusual for it to be far wrong, and I try to be as objective as I can.

regards

John

[Arrows]

where can you view the t charts that you keep mentioning? I can't find them on this site unless I'm being blind. Help me out here people.

Edited September 29, 2005 by theredarrows

[johnholmes]

http://www.wetterzentrale.de/topkarten/tkavneur.html

enjoy

John

[Arrows]

thanks. The low at that time seems to be developing a hook and as it stands nothing is standing in its way.

[The PIT]

not so Pit, at least in my opinion with the 12Z run, its been remarkably consistent, which is what I was trying to highlight, evenwith the fairly unusual state of the atmosphere in the Gulf of Mexico over the past 6-8 weeks.

Yes, out to T+96 and its rare it ever is wrong in the basics, the flow over the Uk and its source, even the isobar is often predicted within 50-60 miles of where it ends up. To me that is astonishing. Even out to T+120 its unusual for it to be far wrong, and I try to be as objective as I can.

regards

John

<{POST_SNAPBACK}>

I've been looking at both runs and the shift can be quite large so I'm probably compairing the two.

I've followed some of comparrisons and sometimes I disagree with your comments as preasure systems have been displaced enougth to alter the actual the actual weather experianced on ground level. What will be interesting is how far in future will models be able to predict.

[johnholmes]

Like I said Pit I try to be as accurate and unbiased as I can when doing the checks. I've been doing them now for almost 12 months, so unless I have been making mistakes regularly I still feel my comments are valid.

If you have precise examples of my 'errors' please pm me and we can discuss any apparent discrepancy.

regards

John

[The PIT]

Like I said Pit I try to be as accurate and unbiased as I can when doing the checks. I've been doing them now for almost 12 months, so unless I have been making mistakes regularly I still feel my comments are valid.

If you have precise examples of my 'errors' please pm me and we can discuss any apparent discrepancy.

regards

John

<{POST_SNAPBACK}>

Will do in a friendly manner.

[Roger J Smith]

As far as charts that go +360 and +384 being accurate, I seem to recall a discussion on this site when I first joined a few weeks ago about a big cold outbreak being advertised for Sep 29-30 over the UK. Well, some of the elements of that verified and some didn't, I would have to observe. I think the maps I recall seeing showed a weak low crossing the North Sea with a massive NW flow following behind it. In reality, what's likely to verify at forecast time is a strong low north of the North Sea with a brief shot of colder air reaching mainly Scotland.

This illustrates my earlier-stated principle that computer models will eventually run out of steam as unforeseen but major energy developments overwhelm the calculations of system evolution. This should surprise nobody, really -- it is a matter of common knowledge in meteorology that weather systems have life cycles of about 5-8 days and for any computer model, no matter how reliable through 96-120 hours, to try to compute what today's frontal wave south of Tokyo will look like on Day 15 when it is (after two or three major transformations and a picked up tropical system or two) cruising towards Ireland, is just beyond their capability. It may forever remain there because as I was mentioning earlier, there could very well be external factors in the complex cause-and-effect of meteorology, factors such as geomagnetic energy, interactions between our atmosphere, magnetic field and solar-system magnetic field, that have to be understood in order to go that far forward with a forecast.

I'll put this in much simpler terms. Suppose you're on a motorway in good visibility and you can see miles ahead of you. You can then make realistic predictions of where you will be five, ten, maybe fifteen minutes from now. But then try to imagine where you'll be in an hour, or two hours, if your journey then takes you into a large city. Even if you stay on the same motorway, your calculations, while subject to less error, are still more and more theoretical and prone to error.

Or, just use the widely accepted logic of the NHC in hurricane track forecasting, itself one of the less accurate of the many things that weather models do on a regular basis. The 24-hour error may be 50 miles, the 5-day error may be 300 miles, and the 10-day error in the rare cases when a tropical system can be predicted for ten days, is virtually half the width of the Atlantic Ocean. Since the average long wave in the atmosphere is about the width of the Atlantic Ocean, these longer-range model errors start to become quasi-random.

What I was saying above about timing theoretical events and using the one example of October 23-24, I could just elaborate briefly on that. My research is mainly in North America and I'm extending it to western Europe in more detail this year to try to build up the basis for global modelling of longer range forecasts. My research system uses a grid of timing lines in the atmosphere, so that predictable events that I believe are associated with weather events can be timed and therefore located on these timing lines. I am working around the idea that there are nine main timing lines around the northern and probably southern hemispheres. Four of these that I am more used to working with are off the west coast of North America, another in the lee of the Rockies, then another through the Great Lakes and mid-Atlantic states, and a fourth near the east coast of Newfoundland. The next one downstream from these is apparently through Ireland and southwest England (the timing lines tend to be somewhat curved in a NNW-SSE orientation). You could do the math and figure out where the others are, approximately. Therefore, what I am looking at is the timing of energy (low pressure systems or frontal waves) across these timing lines. There is also the question of latitude and intensity to consider, and that's part of the research as well. Without going into more detail, October 23-24 represents a strong energy peak where weather systems should be crossing these timing lines. Since I expect the long-term flow pattern to remain essentially zonal or fast WSW for much of October, I would be looking for this event to be a deep low somewhere near Ireland on the 23rd. I'm more at the observing than forecasting stage of this part of the research now. When the winter closes in, I'll list some other dates for possible major weather systems and some ideas about what to expect from them. I've said elsewhere that I'm looking for a mild, windy and stormy winter pattern that evolves into a cold blocking pattern in February, so these individual storm possibilities would be within that context. I wouldn't bet much on a white Christmas, there's a deep low indicated for Dec 26 near 60N 10W.

[The PIT]

Good post.

On todays modeling what would you call accurrate? A low predicted at T384 hrs being 200 miles out of position.

I'm told that the GFS doesn't have the same data set for each run which explains why it jumps around a bit and possibly comes up with complete nonesense at times.

[The Enforcer]

Interesting stuff RJS. Whilst I wouldn't expect you to put your neck on the line, it would be useful for you to compile a report in hindsight showing how your predictions compared with reality.

Obviously, the main limitation of your method is the reliance on no blocking occurring in the research period. On the basis of recent autumns/winters, I guess this is a reasonable assumption to make. That said, I expect that any assumption is based on another set of data.

In broader terms, the pattern you have outlined would not paint a pretty picture for the forthcoming UK winter. It would appear to be basically a re-run of last winter with mild air dominant until february, by which time northern lattitudes would be so mild that cold incursions would lack any real punch.

[Thundery wintry showers]

It would depend IMO on how far north the jet tracked during the early winter. In the mild zonal winter 1994/95, the jet was in a fairly 'normal' position, such that the Arctic remained cold and the polar incursions of Spring 1995 were quite potent. On the other hand, winter 2004/05 had the jet tracking exceptionally far north, so that the Arctic became exceptionally warm.

[johnholmes]

Good post.

On todays modeling what would you call accurrate? A low predicted at T384 hrs being 200 miles out of position.

I'm told that the GFS doesn't have the same data set for each run which explains why it jumps around a bit and possibly comes up with complete nonesense at times.

<{POST_SNAPBACK}>

00, 06, 12 and 18Z all have the same basic data but other information is also available at different times, ie satellite data, aircraft reports, byous etc, and is assimilated into the run, as I said at the beginning, at different times. This is why I alsways say, stick with one run, don't follow each change from 00 to 06 to 12 to 18 as you will become pretty lost pretty quickly on some days. There is much less variability if you stick with one run. By all means look at each one but for forecasting use just the one.

regards

John

[Adi F]

I have to second what John has said in his last post.

I have only been forecasting for a very short time and am by far the least experienced forecaster on the team but John has been training and guiding me through a real minefield of information. And the one thing he has always said to me is to always use the same run to do your forecasts from.

This proved to be a life saver on the 21st of September when I was trying to write the close up forecast for the storms at the end of that week. Every chart I looked at was giving me conflicting data but because I stuck to the one run I was able to put up a fairly accurate forecast.

John and I actually discussed the problem with the miss matched data and I explained that I chose one data set and stuck with that to do the forecast and he admitted he was struggling to tie things down.

Its like coffee find the product you like and stick with it, it will get you through any eventuality.

[Arrows]

ADI AND JOHN your right.

I find working with more than one confuses me. There's so many. If you use more than 1 you get mixed up with time of runs and whats what.