google-deepmind / google-deepmind/weathernext
Issues about using ECMWF Open Data for Graphcast/GenCast Operational Input
- Dominant language
- Python
- Stars
- 7.7k
- Forks
- 986
- PR merge metrics
- No merged PRs in 30d
Description
Hi,
I am looking into use ECMWF Open data [1] as operational Graphcast / GenCast Input instead of ECMWF Operational Archive (HRES-f0). ECMWF confirmed it's feasible [2].
[1] https://www.ecmwf.int/en/forecasts/datasets/open-data
[2] https://www.ecmwf.int/en/about/media-centre/aifs-blog/2024/run-ai-models-yourself-ecmwf-open-data
However I observed notable discrepancies between the ECMWF Open Data and the ECMWF Operational Archive (HRES-f0). Specifically, differences of up to 8°C in some locations at T0 initial conditions.
I wonder if you have benchmarked Graphcast / GenCast against ECMWF Open data?
Also, Would you recommend using ECMWF Open Data as a substitute for running operational Graphcast / GenCast? Thank you!
Regards,
Contributor guide
Assessment
This issue has not been assessed yet.