Researchers in Switzerland are working on a new way to forecast natural disasters. They are using artificial intelligence trained on huge amounts of NASA climate data.
The project is based at the Federal Institute of Technology Zurich, known as ETH. It uses one of the most powerful supercomputers in the world, called Alps.
Scientists copied about 100 petabytes of public NASA data onto servers next to the supercomputer. That is close to six billion individual files.
Reto Knutti, a climate physics professor at ETH, said the copying process took about a year. He compared the data size to around 20 million feature length films.
Why Speed Matters for Weather Forecasting
Traditional forecasting relies on complex math equations to simulate the oceans and atmosphere. That process can take hours to complete.
AI based statistical models work differently. They spot patterns in data instead of running long calculations.
Knutti said a statistical model can produce a full global forecast for several days in about a minute. He called it a shift in how forecasting works.
The models cost a lot to train at first. But once built, they are cheap and fast to run again and again.
That means scientists can run forecasts every few minutes instead of waiting hours. It gives them more chances to catch a dangerous weather pattern early.
Spotting Disasters Before They Happen
The researchers say the technology is not just for weather. It could also help detect slow moving geological hazards.
Landslides and glacier collapses often show warning signs in satellite data. The challenge has been having enough computing power to catch those signs in time.
Knutti pointed to the Swiss village of Blatten as an example. A glacier collapse destroyed the village in May last year.
Satellite data had shown signs of the coming collapse more than a year in advance. Authorities evacuated the village about a week before it happened, which prevented casualties.
Researchers also looked at the glacial collapse on the Nepal-China border on August 26. The journal Nature reported that satellite images showed early warning signs before the disaster occurred.
Those signs could have flagged the area as one needing closer monitoring, according to the report. Knutti said it shows how satellite data could support systems built specifically for disaster prevention.
Thomas Schulthess leads the Swiss National Supercomputing Centre in Lugano. He said having NASA's climate data plugged directly into the supercomputer opens new scientific possibilities.
He said the setup lets scientists explore ideas they might not have considered before. Moving data quickly next to the supercomputer is central to that.
"It matters whether you can move the data within a few seconds or whether you have to wait days," Schulthess said.
The researchers say the next phase of the project is focused on making better sense of the data. They hope the tools can eventually be used for early warning systems worldwide.
Nepal's floods last month left thousands dead or missing. Researchers say faster AI forecasting tools may help authorities respond earlier to similar events in the future.
