[Blog] WeatherMesh-5c

[jacobbieker]

Arxiv/Blog/Paper Link

https://windborne.notion.site/wm-5c-release-notes

Detailed Description

Context

Continous forecasting is one of the ways ML models should be able to work well andd beat NWP systems at

[prashantpandeygit]

hello @jacobbieker , I just read the weathermesh-5c paper and i would like to work on the enhancement, for the implementation, to add the continuous updation, the 'graphweatherforecaster' is the one to implement here according to me, if you would like i thought of implementing it as adding a method to the forecaster for forward pass, then use its predictions as inputs, then add training support later. What do you think of this approach for the implementation of this paper? I might be wrong here and must be missing things, i just wanted to know how you would suggest approaching this implementation, thanks!

[jacobbieker]

Hi,

The current model can do continuous forecasting, its easy enough to run it multiple times. I don't think there is anything to specifically add for this model, as the parts already exist in GraphWeather.

Thanks,
Jacob

[prashantpandeygit]

thanks for the quick reply @jacobbieker , i got the continuous forecasting is done using looping, just a question that if it is easy to run it multiple times, then what would the model benefit if it is implement to WM-5c ?

[jacobbieker]

The comment on continuous forecasting was more why this model is interesting and the WM-5c is an interesting operational model. Its not something we particularly need to add here. #153 and #150 are both more interesting details of models that we can implement, and would be interesting to implement. WM-5c is more seemingly just productionizing the previous ones.