Closed iluise closed 7 months ago
Hello. It is not possible to get 1hour predictions from GraphCast. The model steps forward in 6h increments. I do not see any way, without retraining, to get a 1-hour prediction.
Yeah, small misunderstanding within the team :) Sorry for the noise. Best,
Ilaria
I would like to modify the module to get hourly predictions from graphcast:
So, what I have done was to change just the
steps
and thelagger
to 1 here: https://github.com/ecmwf-lab/ai-models-graphcast/blob/main/ai_models_graphcast/model.py#L87-L88But I am not sure about how to change the lines here: https://github.com/ecmwf-lab/ai-models-graphcast/blob/main/ai_models_graphcast/model.py#L277-L284
Is it correct to use the
scda
stream for all the values with % 12 != 0 ? Because at the moment I get the plot I post below. Thanks in advance.