Numeric Modelling and Ensemble Discussion

[Dawlish]

Hi AM,

Thankyou for that link,i have Bookmarked it.

Interesting to see quite a large divergence from around the weekend,and some going for cold next week.(London).

It illustrates why the sypnotics showing in the models keep changing after about T120hrs.

I don't, personally, take any notice of ensembles. The charts will always change T+120+ and, in fact, to 13th March, despite the variability implied by the ensembles, that High pressure over the UK has been constantly there, for a week.

I mean, if, for instance, someone said that "beyond the reliable timeframe" whatever that competely nebulous phrase means, the gfs is unpredictable, why then does the same person talk ensembles? If the chance of the control run coming off is low, the chances of any ensemble run coming off is similarly low. In these cases, the control run is no better that any single ensemble run. Most of the time, that is true of T+200+ and I often use the phrase "white noise". When ensemble divergeance, or possibility is quoted further away than T+200+, I just smile and shake my head. Control/ensemble/time of day run means very little and forecasting accuracy, which si what I am really interested in, is demonstrably far too low to trust. If it wasn't, the Meo Office would produce a reliable 1/2 week ahead forecast and we'd plan our activities by it. They can't. hence the lack of use of any run as a predictor at T+200+. Hence the almost complete unreliability of ensembles.

Ensembles are mentioned so much, but, at the time that 12/13/14 High was established on the charts (for me), several days ago, the ensemble runs have, as pretty much always at that distance, been all over the place. The fact is mentioned about every 10th post, often with an implication that there are times when ensembles are useful. I know they can show a degree of convergeance a particular time in the future, but I think it is worth questioning the relevance of that convergeance and its usefulness in forecasting.

I'd be interested in hearing other views on this.

Paul

Edited November 7, 2006 by Dawlish

[Joneseye]

Ensembles for me show how much uncertainty there is and at what point the general uncertainty sets in. By referring back to the run you ccan pinpoint what is going on at that time and extrapolate where the colder and warmer ensembles members might go from there. That allows you to anticipate features to look for in order to arrive at the proposed destination.

In addition, if the control run is out on its own in terms of mild or cold, it is an outlier, or rather becomes one if subsequent runs do not start to follow its lead or show gathering support.

Its a statistical tool Paul, something you ought to love - you can see where the odds are of an evolution occuring by what support it has!

Well put - exactly why I look at the ensembles as well.

[Dawlish]

Interesting run once again. This time a greenland block is temporarily established and pressure remains generally high but it topples, only a day of strikingly cold temperatures. Some cold and some very warm conditions shown on this run.

Ensembles for me show how much uncertainty there is and at what point the general uncertainty sets in. By referring back to the run you ccan pinpoint what is going on at that time and extrapolate where the colder and warmer ensembles members might go from there. That allows you to anticipate features to look for in order to arrive at the proposed destination.

In addition, if the control run is out on its own in terms of mild or cold, it is an outlier, or rather becomes one if subsequent runs do not start to follow its lead or show gathering support.

Its a statistical tool Paul, something you ought to love - you can see where the odds are of an evolution occuring by what support it has!

I understand that completely and I have been interested in the ensembles before, for exactly that reason, but there's a logical inconsistency in there though. If the whole set of runs at, say, T+144, or even more distant proves to be pretty hopeless in 6+ days time - and a lot of TWS's excellent tracking shows this - then the ensembles were not actually providing a great deal of useful data at all, no matter when the convergence took place.

A good measure of this would be to forecast, using the times when the divergeance appears to start. Forecasts over a reasonably long period of time, using the control run, at that point of divergence, would show whether this point of divergence was of any practical use, or whether it is another of those "tools" that would be just as good left in the shed as being used!

Can it actually be used for forecasting? If it can't, it is there as a discussion tool and no more. I don't think it is useful, therefore I have decided not to use it. If anyone has any data on it's usefulness, I'd be really interested in seeing it.

In the end, there are a whole load of these "tools" out there, which are often referred to, with the implication that their use is obvious. The statistical lack of accuracy, in forecasting more than 7 days ahead, would suggest they may be of nothing like the usefulness that is often simply accepted, without a great deal of question.

Paul

PS The 06z takes the cold further East around the 12/13th again and the rest of the run is much milder.

Edited November 7, 2006 by Dawlish

[Big Bear]

I think the fact you think the ensembles can't be used for forecasting is proposterous. They work ever so well in conjunction with the charts up to +3 days out.

[Nick F]

Can it actually be used for forecasting? If it can't, it is there as a discussion tool and no more. I don't think it is useful, therefore I have decided not to use it. If anyone has any data on it's usefulness, I'd be really interested in seeing it.

In the end, there are a whole load of these "tools" out there, which are often referred to, with the implication that their use is obvious. The statistical lack of accuracy, in forecasting more than 7 days ahead, would suggest they may be of nothing like the usefulness that is often simply accepted, without a great deal of question.

Paul

Paul, your dismissiveness of the practical use of ensembles is a bit ill-informed I think. Ensembles produced by forecast models can lead to good forecasting skill for short and mid-range weather forecasts by using the ensemble mean as the best forecast - and this is what professional meteorological agencies use to base their forecasts on to predict say beyond a week. Also the spread of the forecast the ensembles additionally provides estimates of forecast certainty.

The ensemble mean is shown in red on the ensemble chart below from the 06z run, and may show an averaged 'trend' of different variables such as 2m temps, 850 temps, ppn etc:

Here's a a good summary of why emsemble forecasts are used to improve the skill of meduim range forecasts:

The traditional method of making a weather forecast is to take the best model available and run it until it loses it's skill due to the growth of small errors in the initial conditions. Skill is typically lost after 6 days or so, depending on the season. An alternate method that produces forecasts with skill up to 15 days after the initial forecast uses what is called "ensemble forecasting". Instead of using just one model run, many runs with slightly different initial conditions are made. An average, or "ensemble mean", of the different forecasts is created. This ensemble mean will likely have more skill because it averages over the many possible initial states and essentially smoothes the chaotic nature of climate. In addition, it is now possible to forecast probabilities of different conditions because of the large ensemble of forecasts available. A more detailed description of ensemble forecasting and CDC's forecast products follow.

http://www.cdc.noaa.gov/spotlight/12012001/

Edited November 7, 2006 by Nick F

[Dawlish]

I think the fact you think the ensembles can't be used for forecasting is proposterous. They work ever so well in conjunction with the charts up to +3 days out.

Sorry, BB, you've misunderstood me. I was talking, throughout my post, of forecasts 6/7+ days out, not up to +3.

I'm sure you are right that they are useful up to +3 days out and there, because the accuracy of the forecasts, to 3 days out, is good. Do you think it is preposterous that the forecast accuracy of 1 week+ forecasts, shows that the use of any current method is not, statistically, particulary useful? I always judge the usefulness, or otherwise, of forecasting techniques on their accuracy, not on their potential, or on the fact that people's anecdotal evidence, or belief, is testimony to accuracy.

I realise that this is a statement which will horrify people who feel that what people are doing on here, with LR gfs charts, is forecasting and getting results. In general, we don't; we hazard reasonable guesses and the better people qualify them all the time. "nailed on" and certainties are not terms best applied to reading the runes of the gfs. Probability forecasts would be so much better.

It has always been my premise that any forecast above, say, 7 days, is not accurate enough to trust, using any methods presently available. They are just not consistent and if the use is not consistent, what actually is the use, as one can't know which forecast is likely to be more accurate than another?

The thread title is "General Model Discussion" and that's exactly what we do on here; discuss, postlate, try. In the poorest sense, the use of ensembles is just as good...which actually means just as bad....as any other method currently being used at 1 week+. If there was an accurate method, the Met Office would be using it with demonstrable accuracy percentages. They aren't. Thus; a viable method does not exist - yet.

Just to head off one branch of possible criticism on that; if anyone thinks that is a post that says "don't bother trying", then they just don't know my writing! Always bother. Always experiment. Don't be put off by any criticism, but accept the current weaknesses of what you do. There is a means out there of forecasting accurately at T+200+. It just hasn't been found, monitored, evaluated and exploited.......yet......unless Piers Corbyn really has cracked it. I do wish he'd make his research public.

Paul

[Dawlish]

Paul, your dismissiveness of the practical use of ensembles is a bit ill-informed I think.

Hi Nick,

"Ill informed" wouldn't be right. I am well informed. I understand ensembles, control runs, mean ensembles, very well. I understand the mechanics of numerical forecasting well until it gets to the maths(!) and I have a good knowledge of the gfs ensembles and their production. It is just that I question their use in informing accurate forecasting - especially in the medium range. It isn't that I'm ill informed - it is my knowledge of ensembles that allows me to sensibly question - it is that I've decided that they are not useful enough to forecast at T+200+ with accuracy. Your link is interesting, but without saying how accurate that mean ensemble is, at a projected forecast time, what is the real point of it, apart from simply being interesting?

Sometimes we have these ideas that things, such as ensembles, are useful in forecasting, but are they of use in forecasting, or are they of use in justifying the forecast that is being made?

Paul

Is he still using sunspots to forecast the weather with the bookies????

He now runs a company selling the information to corporations instead. It is a private company, so it does not file public accounts, but I read that his annual income was up to £5m a few years ago.

He certainly gained intial publicity doing just what you say, Kippure (Solar activity, rather than just sunspots, I think) and was on the TV several times, maybe 5-10 years ago. I don't know whether he still does the same!

Paul

Edited November 7, 2006 by Dawlish

[Thundery wintry showers]

I don't think ensembles in themselves are useless: they give a good idea of the amount of agreement on a particular outcome. They can't be taken as gospel as even if there is 100% agreement the assessment can change suddenly over 24 hours, but strong agreement usually means more chance of the outcome coming off.

Seeing the current pattern develop reminds me a lot of the Novembers of 1999, 2001 and 2004; I think this November could well turn out to be like those (which makes my 7.7C CET projection look good for the moment). The northerly incursion around 18th-20th that I hinted at in my LRF at the beginning of November could realistically come to fruition from this setup, though I don't expect any potentially snow-producing polar incursions before then. The high is too dominant and we can expect alternation between mild cloudy and cold bright spells, with dry weather persisting away from the north and west of Scotland.

[Nick F]

Sometimes we have these ideas that things, such as ensembles, are useful in forecasting, but are they of use in forecasting, or are they of use in justifying the forecast that is being made?

Paul

Well unless you can come up with a more successfull way of forecasting the mid-range Paul, your statistical spots won't suffice I'm afraid, then using the ensemble means from model output, and also other methods in conjuction is the best way currently of perhaps getting some reasonable skill in forecasting for the medium term.

Would ECMWF bother running more ensembles than the other models (GFS, UKMO) if they thought using that it ensembles were a waste of time for improving forecasting accuracy?

Edited November 7, 2006 by Nick F

[phil nw.]

Dawlish Today, 10:53 Post #174

Cumulonimbus

Group: Members

Posts: 1,925

Joined: 24-November 05

From: Guess!

Member No.: 4,727

QUOTE(phil n.warks. @ 7 Nov 2006, 10:03 AM)

Hi AM,

Thankyou for that link,i have Bookmarked it.

Interesting to see quite a large divergence from around the weekend,and some going for cold next week.(London).

It illustrates why the sypnotics showing in the models keep changing after about T120hrs.

I don't, personally, take any notice of ensembles. The charts will always change T+120+ and, in fact, to 13th March, despite the variability implied by the ensembles, that High pressure over the UK has been constantly there, for a week......

Hi Dawlish,

To save space i have just quoted part of your post.

I usually look at the Ensembles when the models chop and change so much as they have been doing lately.

Differences are some times observed earlier than T120hrs. and this signals caution when interpreting the sypnotics.

There are times,especially,in settled conditions,where there is little difference way beyond T120 hrs. amongst the members

so you would likely see a lot of model agreement further out.

Its usefull to observe how many members favour a particular scenario going forward,ie temps.inorder to get an idea of how probable a certain trend is likely to be,say in a few days,for example.

I note other posters have already responded to your POV on this quite thoroughly already so i don`t think i need add anymore .

[Dawlish]

Well unless you can come up with a more successfull way of forecasting the mid-range Paul, your statistical spots won't suffice I'm afraid, then using the ensemble means from model output, and also other methods in conjuction is the best way currently of perhaps getting some reasonable skill in forecasting for the medium term.

Would ECMWF bother running more ensembles than the other models (GFS, UKMO) if they thought using that it ensembles were a waste of time for improving forecasting accuracy?

I know and I understand. It wouldn't help the current accuracy by issuing the challenge to me! I wish forecasting accuracy was higher, but it isn't. My spots have nothing to do with this, of course, they are only for events, as I feel there is no accurate means of forecasting mid-range on a regular basis. They aren't statistical though; instead they involve time-delayed coincidence. The fact that ensembles are issued doen't imply they will improve forecasting accuracy, Decisions are made by gfs as to what information to release from the models. There may be a demand, but is it for ensembles because they increase forecasting accuracy? The proof of the pudding is in the forecasting accuracy....... which is low.

Paul

Edited November 7, 2006 by Dawlish

[The Enforcer]

I'd be interested in hearing other views on this.

I find ensembles a useful tool, but mainly to demonstrate how unreliable the latter two thirds of a run are, usually as a response to someone bigging up cold or mild on the control run more than a week away. The only real variability on the ensembles is what I term the 'divergence point', usually between 5 and 10 days, after which the ensembles normally scatter to show a whole host of scenarios. Occasionally they will reconvene later in the run, but as they will be coming from a variety of angles, this is likely to be a coincidence and is rarely reproduced on subsequent runs. On the other hand, if the ensembles were to show a divergence point out to, say, T+300, then hyping up the control run's projection at T+300 would be justified if it were in line with the majority and poo-pooing it would be reasonable if it was a clear outlier.

[Dawlish]

Dawlish Today, 10:53 Post #174

Cumulonimbus

Group: Members

Posts: 1,925

Joined: 24-November 05

From: Guess!

Member No.: 4,727

QUOTE(phil n.warks. @ 7 Nov 2006, 10:03 AM)

Hi AM,

Thankyou for that link,i have Bookmarked it.

Interesting to see quite a large divergence from around the weekend,and some going for cold next week.(London).

It illustrates why the sypnotics showing in the models keep changing after about T120hrs.

I don't, personally, take any notice of ensembles. The charts will always change T+120+ and, in fact, to 13th March, despite the variability implied by the ensembles, that High pressure over the UK has been constantly there, for a week......

Hi Dawlish,

To save space i have just quoted part of your post.

I usually look at the Ensembles when the models chop and change so much as they have been doing lately.

Differences are some times observed earlier than T120hrs. and this signals caution when interpreting the sypnotics.

There are times,especially,in settled conditions,where there is little difference way beyond T120 hrs. amongst the members

so you would likely see a lot of model agreement further out.

Its usefull to observe how many members favour a particular scenario going forward,ie temps.inorder to get an idea of how probable a certain trend is likely to be,say in a few days,for example.

I note other posters have already responded to your POV on this quite thoroughly already so i don`t think i need add anymore .

I understand Phil, but with respect, the number of members favouring a particular run makes no difference to the probability of its outcome, unless that outcome showed sinificant success over time, based on that number of member runs supporting it. "caution when interpreting the synoptics" only means so much.

I agree that forecasting in settled conditions is much easier than in changing conditions ie forecasting the continuation of a blocking high is easier than forecasting any other synoptic situation.

Good debate I appreciate the views. I hope you see that accuracy is how I judge usefulness and low accuracy = low usefulness.

Paul

Edited November 7, 2006 by Dawlish

[TEITS]

I have to agree with Dawlish about the ensembles.

As we all know +144 is about the limit of the reliability of the models although this can vary between +72 to +204 depending on the current synoptics. So what I personally do is just look at the ensembles upto +144 because like Dawlish correctly says if the control run is showing wildly different synoptics in FI on each subsequent runs then as a forecast tool the ensembles are useless because these to would show huge variations. The problem again is the Chaos theory which is why these supercomputers struggle predicting beyond 7days. One of the reasons for these ensembles is to actually take into account the Chaos theory by changing the starting points and then to see if the same pattern emerges. Unfortunately IMO all these ensembles do is cause greater confusion if you want to know what is going to happen in FI because you only need to look at the SLP ensembles for +276 for example to see the wildly different synoptics.

Like I say alot depends on the current synoptics as to where FI begins so sometimes I only look at the control/ensembles upto +72 and other times it can be +204. I will admit to go looking for E,lys at +384 come winter time :lol:

[The Enforcer]

The ensembles show 4 runs going for much colder weather from around the 12th or 13th.

To illustrate the point I was making above, your assessment concerns a period after the divergence point on the 06z ensembles (5 days). There may well be 4 cold outcomes, but what about the other 6? Balance of probabilities tends to dictate that the 6 mild runs are more likely to be closer to the truth.

[SMU]

Afternoon All- My breif Lunchtime Post-

Paul-

Its often been described that you rub people up the wrong way with your slight 'bais' towards milder weather, same way I have a love of Cold weather -

I dont wish to be critical because I doubt your responses today have been in any way shape or form critical or dismissive to other people..- However-

I do think you have a problem here-

It would seem that yours & indeed all others quest in the model output discussion is to achieve that 'accuracy' beyond 5 even 7 days-

I do note 'yours' as a quest as much as Say the forecasting people on here because you are here most days 'challenging' around the accuracy point of view-

What is Slightly frustrating I guess especially from a forecasters point of view is that

we know 7 days is a vast period of time in weather & equally modelling- so what is it you expect to achieve in terms of detail on a forecast at day 7 ( 168)??

If your expecting a Country wide forecast with finer detail painted around temps, Pressure, Windflow & precipitation then your still going to be 'challenging' these sorts of forecasts in 5-10 years time-

For every poster that highlights a certain 'predictive tool' used for forecasting the mid term your general rebuttle is that it is not a statistically proved method & that we should dismiss it or ignore it-

However there is NO method ot tool that can give us accuracy rate that your expecting to achieve-

The methods that forecasters adopt to produce a forecast in this timeframe is with general reference to the Ensembles-

HOWEVER the ensembles that are used by the general population on here - being the WZ/W.Online 850 Mb charts are probably NOT the ideal ones to generate a forecast because there are the greatest of errors in this forecasting depth-

Available at the NCEP modelling Suite is-

The Ensemble mean which is BEST used at the 500 Mb level - This then shows the likely 500 Mb Hemespheric flow patterns-

A forecaster would then use the other data availble like SST's, Current Jet flow, Time of year to generate an expected forecast-

I understand you wish to challenge & understand- however If you challenge everything, then dismiss everything you are left with nothing except a GFS run, An ECM Run & some other low res models-

You cant sit on the fence then critique other methods-- it doesnt make sense.....

S

Steve

Edited November 7, 2006 by Steve Murr

[nick sussex]

I can understand the criticism of the ensembles as by and large theres always quite alot of scatter later in the timeframe however I think they can be useful when looking at general trends, normally when scatter shows up within 7 days this shows the borderline situation as in the current output with colder air to the east of the uk and only small changes either shifting it completely away or bringing it further west.

The use of superensembles which are generated using several different models and have better results looks the way forward but, as a forecasting tool in terms of probability the normal ensembles are still important and give the forecaster an idea of how likely any given situation is within the more reliable timeframe.

The current uncertainty is down to what happens over greenland with indecision about what the NAO and AO will do and so it might be a few days yet before any definite trend is picked up. :lol:

Edited November 7, 2006 by nick sussex

[weathernewbie]

I know and I understand. It wouldn't help the current accuracy by issuing the challenge to me! I wish forecasting accuracy was higher, but it isn't. My spots have nothing to do with this, of course, they are only for events, as I feel there is no accurate means of forecasting mid-range on a regular basis. They aren't statistical though; instead they involve time-delayed coincidence. The fact that ensembles are issued doen't imply they will improve forecasting accuracy, Decisions are made by gfs as to what information to release from the models. There may be a demand, but is it for ensembles because they increase forecasting accuracy? The proof of the pudding is in the forecasting accuracy....... which is low.

Paul

I dont understand the point about Ensembles. Without seeming to wade in half way through. The fact forecasting accuracy is low, has nothing to do with any single tool used to forecast weather.

(To which Ensembles are but one element of an array of "tools" available)

The difficulty in forecasting is to do with a numerous string of factors, related to our geographical location. An nothing to do with what tools are used, or how individaully they rate as forecasting tools for the mid term.

[tugmistress]

Rather than clog up the general model discussion, i thought i would open this one and move all the relevant posts here.

[Dawlish]

Thanks for the input. I'm happy to be damned with the faintest of praise! I don't mind, mate honest, it's a good way of making a point; just steer away from the personal, if you would. :lol: ). What I said really does make complete sense, Steve and you are looking at this from the point of view of an amateur forecaster (as we all are) defending amateur forecasters and thus missing the point completely about accuracy. I understand very well how you and others are pull together all sorts of strands to produce a forecast. I also understand your frustrations at 7+days out being inaacurate. I understand all too well that amateur forecasters may see what I say as a challenge. i'm one of them and the impossibility of being consistently accurate frustrates me to hell too.

However, it does not change the fact that accuracy is the thing that is lacking - in ALL amateur and professional forecasting at 7+ days. If I didn't say that, what do people really think? Do they think that the amateur forecasts are in some way "good", or "great"? Well, yes; when judged on the content of the meteorology. I've said many times that all the LRFs and medium term ones are that. I have great respect for the learning that is displayed, but all too often that "good forecast", or "great forecast" has an element of outcome in the praise. Unfortunately, much of the outcome is judged, on here, by the amount of the forecast devoted to predicting when it will get cold. I'm pretty sure that drives the majority of winter forecasts on netweather. Let's see if I can get the cold bits right, rather than let's assess what this possible evidence is showing me and see if I can come up with a forecast, as a result.

Forecasts can only be good, great, or whatever, if they stack-up, over time, with a statistical accuracy of 75% or higher. That's what I'm looking for in 7 day+ forcasting and that's what I'm looking for in LRF. No-one is achieving that with consistency, therefore the tools that are being used are not good enough, or the forecasting is not good enough.

It's not sitting on the fence, quite the opposite, it is stating a truth so we can judge these forecasts in the correct light. If I rub you up the wrong way (odd term to use in this excellent debate) maybe it's because you don't agree with me, but the only way to challenge back is to produce the evidence that ensembles, or any other "tool" is used to produce forecast accuracy, or accept that what is being done in amateur forecasting, produces results which are simply not accurate, over any period of time and against any measure. I know that would be hard to accept, but the evidence for the veracity of my statement is overwhelming. I wouldn't make it otherwise.

Paul

PS Off for 2 days to do "great work" in North Somerset. Hope it doesn't snow!

Edited November 7, 2006 by Dawlish

[Shiny_Bottom_1]

An interesting topic this.

A few points, whilst I am thinking:

1) A forecast is a prediction, and not a set-in-stone testiment

2) The ensembles reperesent the outcome of 10 (or is it 11) members. One being raw data, the rest being modified.

3) The ensembles are used for 'possible' movement of synoptics (possible equating to prediction).

4) Because the ensembles represent the predictions, it is neither correct from T+0 to T=384...the difference being that the further down the line, the lower the accuracy.

So....

Point 4 really is just mathematics. At point T+0, you have 10 values, with 10 predictions for (lets say) T+24.

At T+24, when it becomes T+0, the outcome for the next T+24 (i.e. T+48 when at original value T+0) is going to be different again. Now keep multiplying (using time and number of members)...as time moves, so do the predictions.

Basically, the ensembles are not consistants, as nothing within the graph is actually 'real time'.

So really, this is when points 1, 2 and 3 come in. Predictions.

Look at it another possible way. Take the lottery for example.

You want to do the lottery over a 16 week period (16 days in terms of the ensembles).

You have 49 balls to choose from. You have a chance in T+1week of approximately 13mil to win the lottery.

Now how about winning T+1week and T+2 weeks? Are the chances the same? Definately not, they are much less.

The same principle is applied to the ensembles.

The only way I would foresee having an accurate model, is if you were using values that were linear. The weather unfortunately is not linear. It does not move in a set pattern.

So how do you foresee the future in terms of weather? You 'predict' it, or in other words, forecast it. The same principle applies to all methods of forecasting a non-linear set of values.

Therefore, I see the ensembles of use for forecasting. The above explains that the further out you go, the less accurate it is 'likely' to be. But that doesnt mean that an outcome cant be forecast further out.

The models arent random selections of outcomes, but linked to so many other factors (sea temps, past weather patterns and many other influences).

In other words, they are as accurate as anything else that is available (subject to any private forecasting tools that we do not see of course).

Can there be a tool that can predict the future of our weather accurately? Well, understanding the mathematics behind it....No!

But should we really ignore them because they dont show what we want? Should we ignore the models because they are less accurate the further out they go? No!

Why not? Because the models and ensembles are purely and simply forecasting tools...predictions...exactly what a forecast is!

[Jack Wales]

I'm fairly new to the MB and there are undoubtedly many on here who are more knowledgeable than me but I would like to 'chip in' on this one if it's ok.

The ensembles will never give you any degree of accuracy as they are positively and negatively weighted to varying degrees, so there is bound to be scatter in FI. It's like driving your car in a straight line and then doing it again but turning the steering wheel slightly, after 10 meters the car will be in pretty much the same position as before after 100 meters your position will be considerably different and it's the same with the ensembles

The object of the ensembles is firstly to be indicative of the accuracy of the control run. If there is a wide scatter from the start, it means that confidence in that particular run is low, alternatively if there is a wide correlation in the ensembles the chances of the control run synoptics coming off are more likely.

Secondly the ensembles do help identify trends. If you look in FI during the summer and you note that one of the ensembles is out on its own predicting 18C at 850hpa. You would normally dismiss this as an outlier and rightly so, however the following day two members are taking the same route and the next day three members etc. How many times have we seen patterns like this evolve over the summer for heat and for winter cold.

If anything there is nothing wrong with the ensembles, it's us as human beings who need to improve our ability to read them both mathematically and visually.

As for predicting our weather with 75% accuracy, it depends how strictly you set your targets and measure results. I could predict the next 10 winter CET’s to be between 2.5C and 5.5C and be confident of meeting the 75% success rate! Without wanting to sound flippant I'm just emphasising how such a target would be difficult to measure or quantify.

The beauty of meteorology is that it's an evolving science. Whereas Maths and Physics are right/wrong, black/white etc, meteorology will never be that. We will continue to improve in our quest for better accuracy and to do this we need ensembles, LRF's, alternative theories etc. It's easy to dismiss something because it doesn't sound right at the time (i.e the world is round etc) but we need to keep our minds open.

Apologies if I've waffled on, I think I need a drink now!

Regards

Adrian

[johnholmes]

welcome Jack and what a very good, logical and well argued post.

Keep em coming and enjoy Net Weather.

John

[SMU]

welcome Jack and what a very good, logical and well argued post.

Keep em coming and enjoy Net Weather.

John

Tonights 12 Z ensemble mean-

This is what should be analized.... ( Pinched from Raleigh on Eastern US)

http://www.easternuswx.com/bb/index.php?ac...st&id=61307

S

[Jack Wales]

Thanks for the welcome, I look forward to future discussions across the MB.