I am also using GraphCast. I used the dataset of dm_graphcast and used normalized data. For the final predicted data, after calculation by RMSE, there is a certain gap with the paper's value, which is relatively obvious. Have you reproduced the data in the paper?
