The Weather Outlook

Remove ads from site

Quantum
  • Quantum
  • Advanced Member Topic Starter
17 November 2023 09:32:25
Google has produced an AI weather model called graphcast which apparantely does really quite well, exceeding the ECM in verification stats. Its open source so I imagine Brian could get some charts onto the site. https://deepmind.google/discover/blog/graphcast-ai-model-for-faster-and-more-accurate-global-weather-forecasting/ 

ECM has put some graphcast charts on their own website, initialized using their ECM analysis data here:

https://charts.ecmwf.int/products/graphcast_medium-mslp-wind850?base_time=202311170000&projection=opencharts_europe&valid_time=202311170600 


25/26 (850hpa temp) 11 days snow/sleet falling

18/11 (-4) 19/11 (-6) 20/11 (-6) 01/01 (-7) 04/01 (-10) 10/01 (-7) 11/01 (-3) 30/01 (-1) 13/02 (-6) 15/02 (-4) 18/02 (-6)

24/25 10d

18/11 (-6) 19/11 (-6) 23/11 (-2) 22/12 (-5) 04/01 (-5) 05/01 (0)14/02 (0) 15/02 (0)12/03 (-6) 13/03 (-6)

23/24 8d

29/11 (-6) 30/11 (-6) 02/12 (-5) 03/12 (-5) 04/12 (-3) 16/01 (-3) 18/01 (-8)08/02 (-5)

22/23 7d

18/12 (-1)06/03 (-6) 08/03 (-8) 09/03 (-6) 10/03 (-8) 11/03 (-5) 14/03 (-6)

21/22 12d

DEW
  • DEW
  • Advanced Member
17 November 2023 09:49:40
Interesting. For those ignoramuses like myself who didn't realise just how many charts were available from ECMWF the home link is https://charts.ecmwf.int/ . Graphcast is well down the page.
I can calculate the movements of stars, but not the madness of men - Isaac Newton

Chichester 12m asl

Retron
17 November 2023 09:54:57
It's not really AI, of course (none of what people call AI is actually AI), it's more just a very refined database search - or pattern matching.

Computers should be better at that than we are (no unconscious bias for one), but it remains to be seen how useful it actually is!


Leysdown, north Kent
Brian Gaze
17 November 2023 10:11:28
Thanks for the heads up. I'll take a look and see what the options are for integrating.

PS: I'm hoping to have ECM ENS stamps and ICON-D2 online in the next few days.


Brian Gaze

Berkhamsted

TWO Buzz - get the latest news and views 

"I'm not socialist, I know that. I don't believe in sharing my money." - Gary Numan

Rob K
17 November 2023 10:17:37

ECM has put some graphcast charts on their own website, initialized using their ECM analysis data here:

https://charts.ecmwf.int/products/graphcast_medium-mslp-wind850?base_time=202311170000&projection=opencharts_europe&valid_time=202311170600 

Originally Posted by: Quantum 

It's showing a whacking great high to the south and boring westerlies so I imagine it is 100% accurate 🤓


Yateley, NE Hampshire, 73m asl

"But who wants to be foretold the weather? It is bad enough when it comes, without our having the misery of knowing about it beforehand." — Jerome K. Jerome

Retron
18 November 2023 14:29:59
Here's a paper going into considerable detail about the new ML (not AI) model. It uses the ERA5 reanalysis charts as its training source and, on the specifics tested, does indeed provide higher accuracy than the op run. No comparison was made to the ensemble output, but it's interesting to see nonetheless.

The paragraph below really stood out to me - it seems that the ML model will attempt a blended solution if faced with two conflicting scenarios - and that's something we humans have often done! Most winters will see the GFS vs ECM, or similar, and it's often the case that somewhere between the two extremes is what verifies.

The non-linearity of weather dynamics means there is increasing uncertainty at longer lead times, which is not well-captured by a single deterministic forecast. ENS addresses this by generating multiple, stochastic forecasts, which approximate a predictive distribution over future weather, however generating multiple forecasts is expensive. By contrast, GraphCast’s MSE [Mean Squared Error] training objective encourages it to spatially blur its predictions in the presence of uncertainty, which may not be desirable for some applications where knowing tail, or joint, probabilities of events is important. Building probabilistic forecasts that model uncertainty more explicitly, along the lines of ensemble forecasts, is a crucial next step.

https://www.science.org/doi/10.1126/science.adi2336 

The current model day 6 verification stats can be seen below: as is usually the case, ECM remains in the lead, with MetO a close second. It's interesting that improvements to the accuracy have levelled off over the past 10 years, which is no surprise - there are limits as to how far you can go with increasing resolution before it starts backfiring on you!

https://emc.ncep.noaa.gov/users/verification/global/gfs/ops/monthly_long_term/all_models/ 

 


Leysdown, north Kent

Remove ads from site