GraphCast model?
-
So, now that google open-sourced the GraphCast AI model. Can we expect windy to experiment with, or implement this model too? It looks very interesting to me. They say it is faster and more accurate than ECMWF.
-
I agree, it would be interesting to have this model in Windy. If Google is to be believed it's even more accurate than ECMWF.
-
Buenas tardes,
+1.
¿Podéis responder, por favor?
Gracias,
Saludos
-
@Oxidiser I’d also like to see Graphcast added into Windy, along side the other global model forecasts.
Google has added it to GitHub for the public.
It’ll only get better as Google works on it more.
https://github.com/google-deepmind/graphcast
Keep growing the Windy platform team. The more accurate global models Windy has, the better for all of us.
-
@anthony10 I did not notice the github repository yet. Very cool of them!
Hope we get some response from windy soon!
-
I would also like to see this incorporated if possible as it has piqued my interest in weather forecasting! And yes, Google does claim it to be the most accurate in the world for 5-10 days. That combined with their 24-hour models represent the best forecasting in the world.
Granted, every media organization cites 5 studies proclaiming their own forecasting as the most accurate. So maybe it should all be taken with a grain of salt.
-
A anthony10 referenced this topic on
-
Google says it is the most accurate…
The European Center for Medium range Weather Forecast ECMWF carry tests on several Machine Learning models including their own ML model called AIFS and GraphCast too :
https://charts.ecmwf.int/?facets={"Product type"%3A["Experimental%3A Machine learning models"]}At bottom of this page you find the error comparison of these AI models compared with the one of IFS which is the HRES model shown in Windy as ECMWF model. The RMS error is computed for 500hPa geopotential heights, an important parameter for synoptic forecast on 10 days.
https://charts.ecmwf.int/products/plwww_3m_fc_aimodels_wp_mean?area=Northern Extra-tropics¶meter=Geopotential 500hPa&score=Root mean square error -
@idefix37 This would be an incredible addition! Hopefully windy is able to implement it soon!
-
Graphcast looks exciting and verification stats appear to support an increase in forecast accuracy out to 10 days in comparison to EC HRES. However, you might be a little disappointed with the current Graphcast resolution should Windy support implementing this model into the current model suite. Res EC HRES ~9km GRAPHCAST ~30km
HRES
Graphcast
