Swiss researchers harness AI and NASA data on supercomputer to speed disaster forecasts

A team from the Federal Institute of Technology Zurich has transferred roughly 100 petabytes of publicly available NASA climate and Earth-observation data onto servers adjacent to the Alps supercomputer in Lugano. Using this unprecedented dataset, they are training artificial-intelligence models that can produce a global weather forecast for multiple days in about a minute, dramatically outpacing conventional equation-based simulations.

Professors Thomas Schulthess and Reto Knutti say the speed and ability to detect subtle patterns could identify precursors to disasters such as floods, landslides and glacier collapses, citing the year-ahead satellite signals that warned of the May 2025 glacier collapse in Blatten. They also note the potential to have flagged the August 2025 Nepal-China glacial disaster earlier. Although training the AI is expensive, running the models is cheap, allowing frequent updates for early-warning systems that could protect thousands of people.