The AI General Circulation Model

The project aims to construct a novel General Circulation Model (GCM) approach using recent developments in ML and data-driven techniques, supported by intra-seasonal atmospheric signals. The aim is twofold. On the one hand, the potential to significantly reduce the run time of short-term forecasts. On the other hand, the possibility to increase the forecast skill for the medium term. The final objective is to produce a prototype model that could be further developed with future funding. The training process will use an ERA5 reanalysis dataset to develop different models based on U-Net or Transformers technologies, focusing on the regional boundaries of the Italian peninsula. This project is funded by IFAB for 18 months. Cineca is involved by providing computational support and a suitable framework to run the model.