Can WRF Be Driven by Outputs from Large AI Weather Models?

[view_2006]

Hello WRF Community,

I have been exploring the possibility of using outputs from large AI-based weather models to drive the WRF model. My idea is to use AI model forecasts (e.g., pangu, graphcast, fourcastnet, etc.) as input to initialize WRF simulations.

I know that the WRF model can be run using FNL or ERA5 data, but I am not sure whether it is applicable to AI models. The output variables of AI models are generally u-wind, v-wind, geopotential, specific humidity, temperature; and mean slp, u10, v10, t2m, etc.

I’d greatly appreciate your thoughts and advice.

Best regards,
Feng

 

[Ming Chen]

Feng,
If the AI model outputs include all the variables required for running WRF, then yes you can use AI output to drive WRF.

Please find all the variables required for running WRF in this document.

 

[view_2006]

Feng,
If the AI model outputs include all the variables required for running WRF, then yes you can use AI output to drive WRF.

Please find all the variables required for running WRF in this document.

Hello Ming,

Thanks for your reply. I checked the document and found that soil informations were not provided by the AI models. Is it possible to obtain soil information from other forecastings to complement the AI predictions?

Many thanks for your help.

 

[Ming Chen]

Yes I believe you can use soil data from other sources.

Please try and let me know how it works.