Researchers Unveil AIFS System at ECMWF, Leveraging Graph Neural Networks
gpu
| Source: Mastodon | Original article
Researchers unveil AIFS, a rapid forecasting system. It generates 10-day forecasts in minutes using neural networks.
Researchers at the European Centre for Medium-Range Weather Forecasts (ECMWF) have published a system combining graph neural networks with a sliding window transformer, known as AIFS. This system, trained on 40 years of ERA5 data, can produce 10-day forecasts in approximately 2.5 minutes on a single GPU.
As we have previously discussed the importance of domain knowledge in building AI systems and the role of memory in AI assistants, the AIFS system represents a significant development in this area. The ability to generate accurate forecasts quickly is crucial for various applications, including weather forecasting and climate modeling.
The AIFS system's performance and efficiency make it an important advancement in the field, and its potential impact on weather forecasting and related areas will be worth watching. With the launch of AIFS and its potential for future development, including an ensemble version, the weather forecasting community will likely be closely following its progress and applications.
Sources
Back to AIPULSEN