FourCastNet

[Jason N]

Have any of you had any experience with this or know where any active data can be looked at? I am doing some preliminary research into it and downloaded a paper which I am currently reviewing, but I suspect some of you know more about it, opinions? good/bad Pro's/Cons etc.

 

[rdale]

I’d be interested in hearing what you find… So far the only areas that they have published their work is in ML/AI-type forums, and not in meteorological conferences or publications. It only produces a handful of output variables/

I don’t think any meteorologists are part of the model’s development either.

 

[Jason N]

I’d be interested in hearing what you find… So far the only areas that they have published their work is in ML/AI-type forums, and not in meteorological conferences or publications. It only produces a handful of output variables/

I don’t think any meteorologists are part of the model’s development either.

I will continue to Dig around. It's not exactly a NEW,NEW thing, the paper I am reading is a year old, but its new enough and the concept is interesting to me, but there is a lot I do not know either. I'd like some other folks to chime in that are knee deep into modeling to see what they think or what their opinion is, .25Deg resolution, generating 96hr forecasts in 2 seconds is uhh... well, potentially useful!.. but there has to be some drawbacks I am not even thinking about or considering, thus the reach out.

 

[Jason N]

Based on what I have read so far, it appears to be targeted to the energy Sector, wind prediction, extreme Wx events (tropical and Pineapple Express AR Events), while saving significant computational energy at the sacrifice of less Vertical layers than say ECMWF(IFS) which this paper compares it against. So it suggests a home computer with a single GPU could run a 18km or 30km in seconds, which could be useful to a targeted set of users looking for only certain things. I don't "think" it would be necessarily useful to chasers or Svr Storms forecasters unless the parameters and resolution were to resolve for convection specifically. I think there are some interesting implications of this approach though, but this thought is based on a quick overview of the paper and not a fundamentally deep grasp of this kind of modeling.

 

[Jason N]

If anyone wants to look more into this, I am pasting a link to the paper I have been reading. If the moderators don't approve of the link, I'll understand.

FourCastNet: A Global Data-driven High-resolution Weather Model using Adaptive Fourier Neural Operators - CaltechAUTHORS