At a glance

Eviden’s HPC and AI experts collaborated with European Centre for Medium-Range Weather Forecasts (ECMWF) research scientists to develop new techniques to support next-generation weather forecasting, boost climate and weather discovery, and drive innovation.

Outcomes

  • 97% reduction in accuracy difference between TripleClouds, ECMWF’s 2D production model, and Spartacus, the 3D offline high-fidelity simulation model
  • Significantly faster than Spartacus

The Client

European Centre for Medium-Range Weather Forecasts (ECMWF) is a research institute with a 24/7 operational service. It is responsible for producing global numerical weather predictions and other data.

Eviden provided seamless AI and technical work, critical for bringing AI into the complex workflow of weather forecasting

Matthew Chantry | Machine Learning Coordinator, European Centre for Medium-Range Weather Forecasts

Expanding horizons

ECMWF researchers were keen to enhance their abilities to predict extreme weather conditions, events and other new phenomena caused by climate change. This would be achievable if they had more accurate numerical simulations.
Leveraging the power of deep learning and analytics, they would be able to reduce the time spent on processing complex data, leading to faster insights.

 

Collaborating to tackle climate change and other conundrums

In 2020, ECMWF and Eviden launched a center of excellence in HPC, AI, and Quantum Computing for Weather and Climate at ECMWF’s headquarters. The aim of this was to facilitate collaboration and change. Eviden HPC and AI experts partnered with ECMWF research scientists to develop new techniques to support next-generation weather forecasting, boost climate and weather discovery, and drive innovation. Eviden experts helped researchers use artificial intelligence to enhance and accelerate the ECMWF numerical simulation model ecRad, which was related to the radioactive cloud effect. By leveraging neural networks, they aimed to improve the accuracy of TripleClouds, the 2D production model, while running faster than Spartacus, their 3D offline high-fidelity simulation model.

Why Eviden

In an international tender process, ECMWF had invited bids for a new system that would deliver an increased five-fold performance compared to the current facility. The Council of Member States approved ECMWF’s four-year contract (worth 80 million euros) with Eviden (then Atos) for its BullSequana supercomputer. Eviden was also chosen for its impressive track record and experience in providing advanced supercomputing solutions to some of the world’s top meteorological organizations. In the last few years, ECMWF has partnered with Eviden to establish a strong foundation for success.

Future-ready forecastings

Eviden AI experts have developed a neural network solution that reduces the accuracy difference between TripleClouds’ and Spartacus’ offline evaluations by more than 97%. They have also developed coupling solutions to allow for seamless integration of the trained neural network with RAPS code. The new and improved TripleClouds and neural networks system is significantly faster than Spartacus.