🐛[BUG]: IFS DataSource using t2m, msl, tcwv from 12h behind instead of current Initial Conditions

[david5010]

Version

source-main

On which installation method(s) does this occur?

No response

Describe the issue

I've rewritten part of the IFS initial conditions, but I realized that some of the parameters are using past 12h, but I don't remember the paper mentioning that FourCastnet uses it. Are those parameters there for GraphCast and/or other models? Looking at Earth2Studio, I don't see them doing this either

    channel_data = [
        _get_channel(
            c,
            u10m=dataset_0h.u10,
            v10m=dataset_0h.v10,
            u100m=dataset_0h.u100,
            v100m=dataset_0h.v100,
            sp=dataset_0h.sp,
            # ? Why is it 12h past? Should we switch it?
            t2m=forecast_12h.t2m,
            msl=forecast_12h.msl,
            tcwv=forecast_12h.tcwv,
            t=dataset_0h.t,
            u=dataset_0h.u,
            v=dataset_0h.v,
            r=dataset_0h.r,
            z=dataset_0h.gh * 9.81,
        )
        for c in channels
    ]

Environment details

[nbren12]

IIRC I was unable to find t2m, msl, or tcwv in the actual analysis files (t=0) from ECMWF, but I did find them in the 12 hour forecasts initialized 12 hours prior, so decided to use that for expedience.