Swiss researchers copy NASA climate data to supercomputer

Researchers have copied large amounts of publicly available climate and environmental data from NASA onto a Swiss supercomputer to train artificial intelligence models and to keep a safeguard copy amid concern over US funding cuts. Switzerland's Federal Institute of Technology Zurich, known as ETH, announced late last week that its researchers had copied about 100 petabytes of NASA data to the Swiss National Supercomputing Centre in Lugano.

Reto Knutti, a professor of climate physics and head of ETH's Center for Climate Systems Modeling, said it took about a year to copy roughly six billion files. He described the volume as comparable to about 20 million feature-length films. Researchers also plan to copy large amounts of data from the US National Oceanic and Atmospheric Administration.

The datasets were collected over decades and include information on Earth systems, including greenhouse gases, clouds, precipitation and ice sheets. Researchers plan to use them to train AI models for faster and more reliable forecasting in fields such as weather, climate and natural hazards.

The researchers said US funding cuts were not the main reason for the project, but partly motivated it. ETH professor and former NASA chief scientist Thomas Zurbuchen, who began the transfer, said in a statement that NASA and NOAA measurement programmes had played a key role in global knowledge of Earth systems. Knutti said the United States had not restricted access to the data so far, but added that decisions by the US administration can be quick and not always obvious. He stressed that the data copied so far had been publicly available.

The copied data is being stored at the Swiss National Supercomputing Centre, which is also home to Alps, described by ETH as one of the world's most powerful supercomputers. ETH said the aim is to combine the large data collection with Alps' computing power so researchers can analyse it with AI-based methods and train new AI models.

Knutti said some data-driven forecasting models, known as foundation models, are already outperforming traditional models in prediction skill and can be about 1,000 times faster than physics-based models. He said a global weather forecast for several days can now be run in about a minute, instead of taking hours. Faster forecasting can be important for agriculture, hydropower and protection from natural hazards such as floods, storms and landslides. Knutti said improved computing and machine-learning tools could help researchers recognise important patterns more quickly and make more accurate predictions about climate change impacts over time.