The predictions may have some errors

[LeBronQ]

Hello! We have obtained some strange prediction results. Here is our experiment configuration:
Model: "GraphCast - ERA5 1979-2017 - resolution 0.25 - pressure levels 37 - mesh 2to6 - precipitation input and output.npz"
Input: 0.25*0.25 resolution, 37 levels, 2023-05-02 18:00 and 24:00 (2023-05-03 00:00) ERA5 data
step:40

We trimmed the forecast results for the Brazil region and found that there is a significant gap between the 2m_temperature and Precipitation indicators in forecast and the actual situation. Here is the result of Acre state as an example. Temperature is measured in Celsius degrees and the precipitation is measured in mm.

Looking forward to your reply!