Why WB2ERA5 data source is limited for certain variables

[uturuncoglu]

Consider about following code,

forecast = "2011-08-29"
timestamp = datetime.strptime(forecast, "%Y-%m-%d")
ds_input = WB2ERA5(cache=True)
da_input = ds_input([timestamp],['msnswrf'])

If I try to get a variable like msnswrf it is not working and giving error like variable id msnswrf not found in WB2 lexicon. I check the WB2Lexicon and it does not have the variable that I am requesting even if it is exit in the WB2ERA5. I wonder why this is limited. At this point, I am trying to use CBottle3D to generate some variable but it seems that I could not generate msnswrf (mean surface net shortwave radiation) and it is limited to for certain variables. Is there any way to generate downwelling long wave radiation and net shortwave radiation with this way. Maybe there is another way to do it but not sure. Any suggestion would be helpful at this point.

[NickGeneva]

Hi @uturuncoglu

With data sources, we have a conservative strategy when adding supported fields. Namely we add variables on a as needed basis, the reason being our attempt to maintain consistency with how things are labeled / parameterized in the package.

So in this case if there is a need for surface net shortwave radiation from wb2 please open a feature request (or a PR if you decide to add it yourself). Its pretty easy to add variables and we are happy to do it when its needed! We just need to double check how consistent it is with the output a cbottle and adjust accordingly.

There also a user guide on our intentions with the lexicon, if curious:
https://nvidia.github.io/earth2studio/userguide/advanced/lexicon.html

Thanks!

[uturuncoglu]

@NickGeneva Sounds good. Thanks for your help. I could try to add new variable and try to create a PR.

I am not sure it is possible or not but I am plaining to run GraphCast and then fill the missing variables like net shortwave and downwelling longwave radiation components with CBottle.

I also wonder if there is a plan to bring interface for GenCast ensemble prediction. That would be also really useful since it accepts sea surface temperature as input. It is temporal resolution is 12 hours and relatively coarse but maybe InterpModAFNO could be also used to increase the temporal resolution.

My personal experience it is very easy to access models through the earth2studio interface and it makes easy to mix and match different models to create more sophisticated ones and experiment different ideas. So, thanks for such a great tool.