7-day forecasts are run with the deep learning based weather forecast system Pangu-Weather released by Huawei Cloud (Bi, K.; Xie, L.; Zhang, H.; Chen, X.; Gu, X.; and Tian, Q. (2022): Pangu-Weather: A 3D High-Resolution Model for Fast and Accurate Global Weather Forecast, https://arxiv.org/abs/2211.02556). Though originally trained with ERA5 reanalysis data, we initialize the forecast from operational analysis of DWD.
With the current setup at KIT, a single 7-day forecast with Pangu-Weather consumes 14 Wh of energy. For a 7-day forecast with the ICON model, the energy consumption amounts to approximately 30000 Wh. This simple calculation of course does not include the energy consumption required to generate the training data and to train the model.
The implementation was done by Julian Quinting. The page design and navigation logic was done by Marco Wurth. We acknowledge DWD for providing the operational analysis data and support by the state of Baden-Württemberg through bwHPC.
Pangu-ICON Comparison Maps
Pangu-Weather (initialized from ICON-Analysis)
ICON-Global-Deterministic
{{var_shaded_description}}, {{var_lines_description}}
Run: {{ run_date }}, Valid Date: {{ valid_date }}
Keyboard Navigation (click on map if not working): ← -6h → +6h ↓-Run ↑+Run ␣ Swap Variables ↵ Preload
Page updates models around 09:10/21:10UTC