Has somebody broken the Met Office weather model?

[Ammonite]

If only PredictWind had more consumer friendly business practices! I’ll never go back and I warn people away.
They bill your credit card unless you cancel, and if you cancel they cut the service immediately without refund. Bunch of crooks.

I can't comment on your experience but given they provide a free platform which offers pretty much everything I need, my experience couldn't be further from yours

 

[lustyd]

I can't comment on your experience but given they provide a free platform which offers pretty much everything I need, my experience couldn't be further from yours

You’re not a customer then the free platform has less in it than Windy

 

[Ammonite]

You’re not a customer then the free platform has less in it than Windy

You said consumer not customer
...also I don't think Windy has a tabular format that allows the forecast models to be compared side by side which is my favourite feature, although I stand to be corrected on this

 

[ylop]

...also I don't think Windy has a tabular format that allows the forecast models to be compared side by side which is my favourite feature, although I stand to be corrected on this

it does. Get the forecast for the location - tap the bottom of screen like you are going to change the model and bottom left is a compare… may be a paid feature.

 

[Ammonite]

it does. Get the forecast for the location - tap the bottom of screen like you are going to change the model and bottom left is a compare… may be a paid feature.

I've just checked and it's available as part of the premium (paid) option, not the free one. Thanks

 

[Boathook]

A report in today's i about weather app accuracy. Met Office came out OK/ best for planning ahead and the rest no better, even struggling for the next day.

I do think that these weather apps need a human forecaster to keep any eye on what is being produced by persumably AI.

 

[Beneteau381]

All models have been predicting many days with the difference between wind and gust over 10kts, sometimes 20kts. I have found this to be accurate on the W Coast of Scotland this year. It is therefore essential to look at the gust forecast and not the steady wind forecast when passage planning. These are not occasional gusts, the wind has been blowing at or near the indicated gust speed for maybe 50% of the time.

So reef early and reef deep !


- W

Meteo, xcweather and windy have been doing way out forecasts for the last few weeks here in Portugal

 

[franksingleton]

We were in Brittany, as far south as La Turballe for much of July, August, early September. For long periods all the forecast models were erratic. They differed markedly from each other and were inconsistent from one day to the next. There were spells, around the second week of August and the second week in September when forecasts seemed to be more consistent with each other and with themselves from day to day. I have no explanation for this.

There were some slightly confusing comments about grid lengths and predictability in earlier posts. Global models use grid lengths of between about 9 and 13 km. (PredictWind “proprietary”, aka CSIRO CCAM, uses 50km.)
All models contain smoothing of about 5 grid lengths so that effective resolution is between 50 and 80km.
Detailed models, eg ICON-EU, UKV, AROME, HRRR, are nested in global models and can produce nothing greatly different.
The area covered by, say, UKV is small. Weather systems move. Therefore such models are really only useful to us for short periods. I might look at them for the next 12 hours. Beyond that, and maybe for even shorter periods, they are indicative only.-

 

[Puffin10032]

A report in today's i about weather app accuracy. Met Office came out OK/ best for planning ahead and the rest no better, even struggling for the next day.

I do think that these weather apps need a human forecaster to keep any eye on what is being produced by persumably AI.

My understanding is that the weather apps simply take a feed from each model's main run. When the Met Office produce a forecast they do a whole series of runs (termed an ensemble) not only from their own but also the European model. The degree to which all these runs agree allows them to put a confidence value on the forecast which is missing from the apps although you can get a hint of it by comparing the Met Office and ECMWF models on Windy.

 

[franksingleton]

A report in today's i about weather app accuracy. Met Office came out OK/ best for planning ahead and the rest no better, even struggling for the next day.

I do think that these weather apps need a human forecaster to keep any eye on what is being produced by persumably AI.

National weather services and ECMWF run their models. The various apps that we see are beyond their control except, of course, for those appearing on their own sites. AI might be being used to generate apps but I am not aware of that being the case. I suspect that they are using some form of model output statistics.

 

[lustyd]

Some apps are starting to use AI to refine local weather. Where local measurements are available it's easy enough to compare expected and observed and see patterns. It's also possible to take a Navionics approach and collect weather data through the app and use this for machine learning.
Whether it's an effective approach time will tell.

 

[franksingleton]

My understanding is that the weather apps simply take a feed from each model's main run.

I believe that is so.

When the Met Office produce a forecast they do a whole series of runs (termed an ensemble) not only from their own but also the European model.

As far as I know, ensembles are produced by each service using only their own model output.

The degree to which all these runs agree allows them to put a confidence value on the forecast which is missing from the apps although you can get a hint of it by comparing the Met Office and ECMWF models on Windy.

The danger in using Windy or Ventusky to compare ECMWF, UK UM. ICON, CMC GEM, US Navy NAVGEM, Meteo France Arpege is that this is too small an ensemble.

 

[AntarcticPilot]

National weather services and ECMWF tun their models. The various apps that we see are beyond their control except, of course, for those appearing on their own sites. AI might be being used to generate apps but I am not aware of that being the case. I suspect that they are using some form of model output statistics.

I have read recently that the Met Office has incorporated AI in their models. However, this is NOT AI such as LLMs like ChatGPT - it's physics based AI modelling used to speed up processing, a technique that is increasingly widely used where speed of processing of problems with high dimensionality is required. Such techniques incorporate mechanisms to ensure that the results are explainable in physical terms.

 

[lustyd]

it's physics based AI modelling

Just to be clear though, AI is statistics, not physics. If it's physics, it's just modelling.

 

[franksingleton]

Change is occurring in weather prediction at an alarming pace. It is all too easy for expectation to exceed practicality.
As lustyd says, AI is just statistics. The atmosphere is a turbulent system. Whatever prediction system is used, operational weather prediction will be limited by the lifetime of weather features of all sizes from showers to monsoons. Predictability will still be limited by the physical reality of the atmosphere.

 

[Stemar]

The YouTube Met office channel is worth a look. You get what's happening and why, including the degree of certainty. It won't tell you that it's going to rain in Gosport, but not in Southampton but, if you understand what's happening, it helps to know hoe much you can trust your preferred weather app when it does tell you.

 

[lustyd]

Simonjk used to do weekly forecasts on this forum and with videos to explain which were extremely valuable. Not seen them for a while and his channel seems to be aviation now so I guess he moved on

 

[franksingleton]

The YouTube Met office channel is worth a look. You get what's happening and why, including the degree of certainty. It won't tell you that it's going to rain in Gosport, but not in Southampton but, if you understand what's happening, it helps to know hoe much you can trust your preferred weather app when it does tell you.

The Met Office app can resolve detail to about7km - at least for the next few hours. It will resolve small detail further ahead, but less accurately. The website version, UK rainfall radar map - Met Office .
gives the last 1-2 days actual radar imagery. That should be at less than 5 km res.

 

[capnsensible]

Whenever you hear this term in a forecast, this will haunt you.

 

[FairweatherDave]

Simonjk used to do weekly forecasts on this forum and with videos to explain which were extremely valuable. Not seen them for a while and his channel seems to be aviation now so I guess he moved on

I get a weekly weekend sailing forecast e mail from Simon, from something I must have clicked on on his site. It's an e mail not a video but useful all the same.