Weather advancements

[lustyd]

Sorry, but you are showing your blinkered attitude. If what you say is correct then you, as a data science expert could set up a totally automated service put you money where your mouth is. Then give a thought to all the other forecasts and users of forecasts.

One of us is very blinkered but it's not me. I have been working on this stuff, and don't need to put my own money into anything. People pay me to work on these things (or did, I'm retiring this month at 45 as a result of my skillset). This entire thread was started by me posting a link to people like me showing that we can and do do better than people like you precisely because we're open minded.

 

[franksingleton]

These data sets already exist, they just aren't being used for this purpose.

You still have not explained how you would collect information such as starling murmurations, for example.

You've clearly not read anything I've written, or you're purposefully ignoring what I've written so I won't carry on responding to you. I've addressed this point multiple times on this thread and never once suggested it would be used for immediate forecasting.

And you are not explaining why you think that it is possible to predict weather for the next few days anywhere in the world with such a limited amount of data, regardless of whether you are using NWP or AI which, I do accept, is highly likely to be an important factor in prediction. A problem with trying to discuss AI/ML with you is that you continually move your feet without ever shifting your ground.

 

[lustyd]

A problem with trying to discuss AI/ML with you is that you continually move your feet without ever shifting your ground.

The problem talking to you is that you're unwilling to try to understand what I'm saying and you're wedded to your old methods. I've been quite consistently trying to help you to understand and you've demonstrated in almost every post that you're not even willing to consider modern data science methods as a possibility. You rule things out without even understanding what I was saying, based on old and outdated assumptions about data and processing.

You keep saying you're interested in the topic, so I can only assume you're oblivious to your closed mindedness.

 

[Marsali_1]

The problem talking to you is that you're unwilling to try to understand what I'm saying and you're wedded to your old methods. I've been quite consistently trying to help you to understand and you've demonstrated in almost every post that you're not even willing to consider modern data science methods as a possibility. You rule things out without even understanding what I was saying, based on old and outdated assumptions about data and processing.

You keep saying you're interested in the topic, so I can only assume you're oblivious to your closed mindedness.

I'm sorry to intervene in a post that was not specifically directed at me, but you are wrong. As an outside observer of the to and fro between you and Frank I can tell you that, from reading his replies to you, I can see that he has been trying to understand your point of view. The problem is that you have not been adequately explaining your point of view. You have said, several times, that you are an expert without demonstrating that expertise. You have contemptively dismissed legitimate counter-points by saying to the effect that it is "old school". You perceive someone else's legitimate reasons for a point of view as being archaic and inflexible.

I admit I don't have Frank's professional knowledge of climatology and meteorology because I never worked in that field. However, I have a more than passing interest in those subjects and I know enough to recognize that he knows what he is talking about and can explain it in layman's terms. Your side of the discussion is quite a bit different and is why I joined in. I was curious about what it was that you were trying convince Frank about with your postings. I could understand Frank's points but yours were less clear. Unfortunately you have lumped me into that same "old school" category because, like Frank, I pointed out gaps in data that need to be filled and, apparently, won't in your AI/ML system because you have deemed it irrelevant and unneccessary. Sorry, but if we are supposed to acknowledge your expertise with out, perhaps, fully understanding it then you should, at least, do the same with experts (and I don't mean me, in this case) outside your own field. Some of the comments you have made show that you really don't know much about the dynamics of the atmosphere, nor do you care. However there is legitimate reasons and importance for what you dismiss as irrelevant, unnecessary and archaic.

By the way much of the historical data that your AI/ML systems will be using were generated by "old school" scientists following that annoying and archaic "Scientific Method" that took time and much debate and peer review. I wonder of what value that old data would be if they hadn't followed that process and what quality of data the machine learning would have to learn from? I seem to recall an expression ("GIGO") from the early days of computers and I do wonder how one would determine if the output was "GO" without understanding that the input was "GI" because you don't "...need to understand, we just need to test predictions and be right more often." You have never explained how you would know when you are right when you don't understand, and don't need to understand, the process you are analysing.

 

[franksingleton]

The problem talking to you is that you're unwilling to try to understand what I'm saying and you're wedded to your old methods. I've been quite consistently trying to help you to understand and you've demonstrated in almost every post that you're not even willing to consider modern data science methods as a possibility. You rule things out without even understanding what I was saying, based on old and outdated assumptions about data and processing.

You keep saying you're interested in the topic, so I can only assume you're oblivious to your closed mindedness.

Not at all. ECMWF and the Met Office have far younger and better scientists than I am or was. I note what ECMWF is doing. I can see why it works. Can you produce an example to prove what you are claiming?
We know that AI is a powerful tool. It has been shown to have potential in weather prediction. Do you know of any examples where AI has shown the potential to replace NWP when using data sets similar to those that you describe.

PS. I drafted and posted this before I had read #84 with which I fully agree. Like most meteorologists, I am an experimentalist at heart.

 

[lustyd]

I can see that he has been trying to understand your point of view

We'll agree to disagree there. He's ignored everything I've said and rejected any and all new ideas out of hand. I have the greatest respect for Franks knowledge within his domain, but this is not his domain and that is very obvious to those who are in this domain. The very idea of indirect measurement is alien to him as has been demonstrated throughout the thread.

You have contemptively dismissed legitimate counter-points by saying to the effect that it is "old school"

I don't believe I have, could you give an example of where someone has understood what I've actually said and I've dismissed it? Most of the thread has been people not even reading my posts, let along taking the time to understand and respond appropriately. Much of it has been Frank trying to mansplain how traditional methods don't work that way despite talking about new methods.

By the way much of the historical data that your AI/ML systems will be using were generated by "old school" scientists following that annoying and archaic "Scientific Method"

Some of it is, yes. Most of it won't be data used for weather forecasting, although that will be a base data set. Much of it will be from other areas created by scientists, and much of it will be data sets previously considered unrelated, and some will be sourced from what would traditionally be seen as unreliable sources. Our methods are well proven in many industries including forecasting (as the original post demonstrated). The stuff you'll see over the next couple of years will be orders of magnitude better, and will rely less and less on physical modelling.

I seem to recall an expression ("GIGO") from the early days of computers

Indeed, me too. But there's a difference between garbage and unrefined data. Just because you don't understand the link between the data and the result, it doesn't mean that there isn't one. None of the data sets mentioned by me would be considered garbage, they're all perfectly valid data from valid and well understood sources. Those sources will need processing to remove outliers, but the data sets are good. From what I can tell though, the problem on this thread is that people aren't able to understand how we're using them or the link to forecasting. That's fine, you won't need to understand how the forecasts are made, you'll just gain access to better products.

Regarding expertise, I don't choose to share details of my work on this forum, much of it is classified anyway and knowing my job and experience won't change your understanding of what I do.

 

[Marsali_1]

We'll agree to disagree there. He's ignored everything I've said and rejected any and all new ideas out of hand. I have the greatest respect for Franks knowledge within his domain, but this is not his domain and that is very obvious to those who are in this domain. The very idea of indirect measurement is alien to him as has been demonstrated throughout the thread.

I don't believe I have, could you give an example of where someone has understood what I've actually said and I've dismissed it? Most of the thread has been people not even reading my posts, let along taking the time to understand and respond appropriately. Much of it has been Frank trying to mansplain how traditional methods don't work that way despite talking about new methods.


Some of it is, yes. Most of it won't be data used for weather forecasting, although that will be a base data set. Much of it will be from other areas created by scientists, and much of it will be data sets previously considered unrelated, and some will be sourced from what would traditionally be seen as unreliable sources. Our methods are well proven in many industries including forecasting (as the original post demonstrated). The stuff you'll see over the next couple of years will be orders of magnitude better, and will rely less and less on physical modelling.

Indeed, me too. But there's a difference between garbage and unrefined data. Just because you don't understand the link between the data and the result, it doesn't mean that there isn't one. None of the data sets mentioned by me would be considered garbage, they're all perfectly valid data from valid and well understood sources. Those sources will need processing to remove outliers, but the data sets are good. From what I can tell though, the problem on this thread is that people aren't able to understand how we're using them or the link to forecasting. That's fine, you won't need to understand how the forecasts are made, you'll just gain access to better products.

Regarding expertise, I don't choose to share details of my work on this forum, much of it is classified anyway and knowing my job and experience won't change your understanding of what I do.

1: I still maintain you are wrong. I don't get the impression that Frank is averse to indirect measurement as a source of data, or other forms of data beyond what is currently used and he certainly hasn't ignored you. He's asked questions that you haven't answered. I think he, like me, would like a good clearly explained example of how, for example, your indirect measurents would work and how that is an improvement over current data sets. The fact that you think you have been clear in your posts doesn't mean that they have been clear to the reader. I don't see anywhere that he has suggested that he thinks he knows your domain but I see much in your posts that exhibits either lack of knowledge of his domain or outright dismissal of it.

2: Look back over the posts you have made. If you can't see the tones of dismissal then that suggests that you might be tone deaf. Try looking for terms such as "mansplain".

3: You have completely missed my point.

4: Again, you have missed my point. When you say "...the problem on this thread is that people aren't able to understand how we're using them or the link to forecasting..." is that the reader's fault or the writer's?

5: I did not ask you to provide a CV or release trade secrets. I said "...Sorry, but if we are supposed to acknowledge your expertise with out, perhaps, fully understanding it then you should, at least, do the same with experts (and I don't mean me, in this case) outside your own field...". Surely, with your expertise, you can engage in a discussion without compromising classified information and in a way that helps us better understand your point.

 

[franksingleton]

1: I still maintain you are wrong. I don't get the impression that Frank is averse to indirect measurement as a source of data, or other forms of data beyond what is currently used and he certainly hasn't ignored you.

Indeed; in fact the vast majority of data used in NWP models come from indirect sources. These include infra red radiance data over the oceans and microwave data globally. Together these provide about 50% of the value of the forecasts globally. There are GNSS Radio Occultation data providing a little under 10% of the value.

 

[GHA]

google joining in >
Google Reveals New A.I. Model That Predicts Weather Better Than the Best Traditional Forecasts

 

[franksingleton]

google joining in >
Google Reveals New A.I. Model That Predicts Weather Better Than the Best Traditional Forecasts

The ML model forecast is still using the ERA5 data set and still requires an IFS analysis as a starting point. The ECMWF ENS, like all NWP ensembles is limited by the number of ensembles members that can be produced and by human, subjective choices regarding the data input and, more importantly, variations in the model formulation. These ensembles use a coarser grid in order to produce results in a timely manner. The ML ensemble can obviously produce many more forecast runs so enabling better assessment of uncertainty. I do not think that the deterministic forecast will be markedly better.

I am pretty sure that AI forecasts will be introduced just as soon as all the necessary testing, product generation and interfaces with users have all been put in place. As yet, I have no information about the use of raw data rather than the IFS analysis.

 

[GHA]

There's a quote somewhere about not really being able to see the enormity of revolutions at the time, only in retrospect. Many revolutions are happening right now, AI's & crypto are massively changing society as much as the printing press or internet. Interesting times.

 

[franksingleton]

Anyone wishing to see the revolution in weather prediction might like to start with ECMWF Newsletter and AIFS Blog . The downside is that you will be bemused by the inevitable use of acronyms. As well as continuing developments in physical modelling, it provides an insight into AI/ML.
It is clear that ECMWF is working closely with its member countries.

Some points that I have picked up are -
The use of raw data in ML is being studied but I did not get any idea about progress.
They see a future role for physical models, particularly in fine scale, relatively short term prediction.
There is a readable account of AI/ML ensembles.

 

[Trident]

There is another saying - which I shall paraphrase - that you cannot be expert in a subject if you cannot explain it to a child in terms they understand. As the majority of readers are not AI experts perhaps a more basic explanation of how the OP expects the systems he works on to provide better results and why .

I will of course be very happy if in the next few years AI or anything else gives me forecasting I can reply on . I don't actually expect forecasts I could be certain to sail by but knowing if it will actually be dry the day I plan to paint the boat and not show bright sun when I look at 7am and then by the next update at 12 tell me it will be a downpour all afternoon. The BBC and to a lesser degree the Met office apps are very good at that rather annoying form of prediction even for just a few hours ahead

 

[franksingleton]

There is another saying - which I shall paraphrase - that you cannot be expert in a subject if you cannot explain it to a child in terms they understand. As the majority of readers are not AI experts perhaps a more basic explanationi of how the OP expects the systems he works on to provide better results and why .

Yes, but this thread is about AI/ML and started with a fairly technical url. There is enough in plain language for anyone using computer forecasts. Anyone getting past the OP will find something of interest here. Others may or may not get some bits that will provide some background to what could well be a major advance in weather prediction.

I will of course be very happy if in the next few years AI or anything else gives me forecasting I can reply on . I don't actually expect forecasts I could be certain to sail by but knowing if it will actually be dry the day I plan to paint the boat and not show bright sun when I look at 7am and then by the next update at 12 tell me it will be a downpour all afternoon. The BBC and to a lesser degree the Met office apps are very good at that rather annoying form of prediction even for just a few hours ahead

My personal assessment is that AI forecasts will not provide significantly better deterministic forecasts but will, usefully, provide them quicker, therefore more usefully. An improvement in ensembles, probability forecasts is likely. For today and the next few days, I use the Met Office app. I cannot, in all honesty, say that I have encountered the situation that you describe. We require a window of about 4 hours to walk to the village, and back, do a little shopping and have a coffee. My decisions are along the lines of, do we go today, tomorrow or the day after? Do we need an umbrella? Over the past couple of years, we have never got it wrong. Sometimes we have cut it a little fine. On these time and distance scales AI will add nothing. It might produce the data input to the app more efficiently in terms of computer resources.

 

[franksingleton]

Despite the plethora of annoying adverts, Weather Tech Revolution! How AI is Changing Forecasts Forever - Qhubo is a sensible article. Like others it does suggest greater improvements in prediction accuracy than what , I believe, is achievable. There will be an improvement in timeliness and, probably, ensemble predictions. These will benefit government operations, commerce etc but have little direct effect on most of us who use deterministic forecasts.

As someone interested in words, I have been a little bemused by the use of “Advancement” in the thread title. The writer uses the same word. Why not use “Advance”? Well, apparently, “advance describes individual and tangible improvements,, while advancement refers to the overall process of improvement. Also, advancement is more used with the preposition of whereas it advance is used with the preposition in.”
All a bit too subtle for me

 

[zoidberg]

It's almost that time of year when one is encouraged to look back so, on a slightly lighter note, here's a little retrospective....

'A long time ago, on an airfield far, far away...'

A bunch of young RAF aircrew would gather, in the early 70s, for 'Morning Prayers' at 0730hrs in their squadron crewroom in faraway Cornwall. Boss's Messages, Flight Safety Brief, and Flying Programme changes would always be preceded by the Station Met Officer's briefing. The guys would typically be dozy - due to e.g. baby's teething/just got back from a Truro Nurses' Party/well hungover - and would listen with droopy eyelids and half an ear open for the bits that were 'dreckly' and personally relevant.

Now, the StnMetO was a good guy. We were a long way west from the Met Office admin boffins in their comfortable offices in Bracknell ( e.g. Frank S. ) and sometimes Frank's Official Forecast was, like EVRI parcels, late/delayed/misdelivered.....so he often had to 'wing it'. And he was good at that.

When he was a bit short of reliable data, he'd often ask our Boss if one of our aircraft, during its climb-out to Height ( >30,000' ), could climb out to the west over the Western Approaches..... and radio back at 5000' intervals the wind speed/velocity, the temperature, cloud type and amount, and any airframe icing. Which we were happy to do. Sometimes we could also do this from well west of Ireland and the Hebrides. Essentially, he had his own private 'ascent' using Canberras and Nimrods instead of met balloons. This helped him offer more reliable 'landing forecasts', or Terminal Airfield Forecasts/TAFs, which was 'dreckly' relevant to our safety.

One morning he let us into a secret. "Statistically," he said, "we meteorologists know that if we forecast for today exactly what we had here yesterday, our 'success rate' is higher than the Bracknell computers' score. But don't tell anyone I told you...."

We used, bleery-eyed, to observe him as he came in through the outside door:
If he was in shirt-sleeves - it would be fine and warm. If he wore an overcoat - it would be cold. If his coat was wet - it indicated rain. And if he didn't turn up at all - it was thick fog.

I used that relationship to get invites onto the RAF's handful of racing yachts, because I was always ( pre-electronics ) able to get a fix of position even in bad weather AND I was able to wheedle out of the StnMetO a detailed forecast - several scenarios - for the route of whichever RORC race was on that weekend. I scarely knew, back then, one end of a boat from another but I - or rather my CONSOL chart and my bespoke met folder - were most welcome.

 

[GHA]

As someone interested in words, I have been a little bemused by the use of “Advancement” in the thread title. The writer uses the same word. Why not use “Advance”? Well, apparently, “advance describes individual and tangible improvements,, while advancement refers to the overall process of improvement. Also, advancement is more used with the preposition of whereas it advance is used with the preposition in.”
All a bit too subtle for me

Web chatrooms are in many ways global repositories of the multitude of human biases which affect each and every one of us . No point coming up with data and evidence, well studied that it just makes thinks worse & the heels dig in deeper. AI to the rescue. From an analysis of post 86..

Answer
The text you've provided contains several examples of cognitive biases and logical fallacies. Let's examine them:

Confirmation Bias​

Confirmation bias is evident in the way the speaker interprets and remembers information that confirms their preexisting beliefs while dismissing contradictory evidence.
Example: "He's ignored everything I've said and rejected any and all new ideas out of hand." This statement suggests that the speaker is focusing on instances where their ideas were rejected, potentially overlooking any instances where their points were considered.

In-Group Favoritism​

The speaker demonstrates in-group favoritism by implying that those within their domain have a superior understanding compared to outsiders.
Example: "This is not his domain and that is very obvious to those who are in this domain." This statement suggests a bias towards those within the speaker's field of expertise.

Dunning-Kruger Effect​

The Dunning-Kruger effect is apparent in the speaker's confidence in their understanding and dismissal of others' expertise.
Example: "The very idea of indirect measurement is alien to him as has been demonstrated throughout the thread." This statement implies that the speaker believes they have a superior understanding of the concept compared to others.

Fundamental Attribution Error​

The speaker attributes others' disagreement to their lack of understanding rather than considering alternative explanations.
Example: "Most of the thread has been people not even reading my posts, let along taking the time to understand and respond appropriately." This statement assumes that disagreement stems from a lack of understanding rather than valid counterarguments.

Overconfidence Bias​

The speaker displays overconfidence in their methods and future predictions.
Example: "The stuff you'll see over the next couple of years will be orders of magnitude better, and will rely less and less on physical modelling." This statement shows a high degree of certainty about future developments.

Appeal to Novelty​

The speaker seems to favor new methods simply because they are new, potentially dismissing established methods.
Example: "Much of it will be from other areas created by scientists, and much of it will be data sets previously considered unrelated, and some will be sourced from what would traditionally be seen as unreliable sources."

Defensive Attribution​

The speaker deflects criticism by attributing it to others' lack of understanding rather than considering potential flaws in their own argument.Example: "From what I can tell though, the problem on this thread is that people aren't able to understand how we're using them or the link to forecasting."
By recognizing these biases, we can better understand the limitations of our own thinking and strive for more objective and balanced discussions.

For a reasoned debate mere evidence & data are blunt instruments against that

Cat, meet pigeons...

 

[franksingleton]

It's almost that time of year when one is encouraged to look back so, on a slightly lighter note, here's a little retrospective....

'A long time ago, on an airfield far, far away...'

A bunch of young RAF aircrew would gather, in the early 70s, for 'Morning Prayers' at 0730hrs in their squadron crewroom in faraway Cornwall. Boss's Messages, Flight Safety Brief, and Flying Programme changes would always be preceded by the Station Met Officer's briefing. The guys would typically be dozy - due to e.g. baby's teething/just got back from a Truro Nurses' Party/well hungover - and would listen with droopy eyelids and half an ear open for the bits that were 'dreckly' and personally relevant.

Now, the StnMetO was a good guy. We were a long way west from the Met Office admin boffins in their comfortable offices in Bracknell ( e.g. Frank S. ) and sometimes Frank's Official Forecast was, like EVRI parcels, late/delayed/misdelivered.....so he often had to 'wing it'. And he was good at that.

When he was a bit short of reliable data, he'd often ask our Boss if one of our aircraft, during its climb-out to Height ( >30,000' ), could climb out to the west over the Western Approaches..... and radio back at 5000' intervals the wind speed/velocity, the temperature, cloud type and amount, and any airframe icing. Which we were happy to do. Sometimes we could also do this from well west of Ireland and the Hebrides. Essentially, he had his own private 'ascent' using Canberras and Nimrods instead of met balloons. This helped him offer more reliable 'landing forecasts', or Terminal Airfield Forecasts/TAFs, which was 'dreckly' relevant to our safety.

One morning he let us into a secret. "Statistically," he said, "we meteorologists know that if we forecast for today exactly what we had here yesterday, our 'success rate' is higher than the Bracknell computers' score. But don't tell anyone I told you...."

We used, bleery-eyed, to observe him as he came in through the outside door:
If he was in shirt-sleeves - it would be fine and warm. If he wore an overcoat - it would be cold. If his coat was wet - it indicated rain. And if he didn't turn up at all - it was thick fog.

I used that relationship to get invites onto the RAF's handful of racing yachts, because I was always ( pre-electronics ) able to get a fix of position even in bad weather AND I was able to wheedle out of the StnMetO a detailed forecast - several scenarios - for the route of whichever RORC race was on that weekend. I scarely knew, back then, one end of a boat from another but I - or rather my CONSOL chart and my bespoke met folder - were most welcome.

In the late 1950s, I was flying as an observer in Canberras from RAE Farnborough. I would not, even then, have recognised your statistical rule.

 

[zoidberg]

Gosh, Frank! Are you really THAT old....?




Edit: Laffans Plain/Farnborough was the site of many early ascents by 'met observers'.
This could have been one of Frank's.....

 

[franksingleton]

Gosh, Frank! Are you really THAT old....?




Edit: Laffans Plain/Farnborough was the site of many early ascents by 'met observers'.
This could have been one of Frank's.....

Not as old as Cody;s tree.
I joined the Met Office in 1955 after a PG year at Imperial College.