AI disaster forecasts use NASA data on Swiss supercomputer

Researchers at the Federal Institute of Technology Zurich say they are using around 100 petabytes of publicly available NASA climate and Earth observation data on servers beside the Alps supercomputer in Lugano to train AI models for faster weather, climate and natural-hazard forecasts.

The data set, copied over about a year and described by climate physics professor Reto Knutti as roughly six billion files, is connected to the Swiss National Supercomputing Centre's Alps machine. Thomas Schulthess, an ETH computational physics professor and head of the centre, said the direct connection could let scientists do work they had not previously imagined.

Knutti said statistical AI models can run a global weather forecast for several days in about a minute, compared with hours for traditional simulations. He said cheaper, faster runs could support more frequent checks for dangerous weather patterns and early warning systems.

The researchers say AI may also help find subtle signs of natural hazards in satellite and other data, including landslides and glacier collapses. Knutti cited the Swiss village of Blatten, evacuated a week before a glacier collapse wiped it out in May last year, as an example of monitoring preventing mass casualties.

Nature reported last week that satellite image analysis showed warning signs before the August 26 glacial collapse on the Nepal-China border. Knutti said such data could potentially be used in systematic observing systems for disaster prevention.