Swiss researchers want to use AI to predict weather more quickly and spot signs of floods, landslides and glacier collapses before disaster strikes. To do this, they have collected about 100 petabytes of NASA data on servers directly connected to the Alps supercomputer.
Swiss researchers want to use AI to predict weather more quickly and spot signs of floods, landslides and glacier collapses before disaster strikes. To do this, they have collected about 100 petabytes of NASA data on servers directly connected to the Alps supercomputer.
This is reported by Phys.org.
A team from the Federal Institute of Technology in Zurich copied about 6.5 billion NASA files — data on clouds, precipitation, glaciers, greenhouse gases, and other processes on Earth. The entire array takes up about 100 PB. The researchers compare that to about 20 million full-length movies.
Now they want to use this data to train AI models that will be able to predict the weather faster and look for unusual changes that could indicate future danger.
Conventional weather forecasting requires complex calculations of what's happening in the atmosphere and oceans. Such calculations can take hours for humans. According to Professor Reto Knutti, their model can calculate a forecast for the entire planet for several days in about a minute.
Also, training this model is expensive, but running it is cheap. So the predictions can be recalculated over and over again. This can be done even every few minutes, looking for a dangerous scenario.
“This allows for early warning systems that can save lives,” says Knutti.
One of the project's goals is to teach the system to notice changes in satellite data that are difficult for a person to track manually on a global scale.
A recent example is the disaster on August 26 on the border of Nepal and Tibet. A collapse of ice and rocks caused a massive flood. Later analysis of satellite images showed that the section of the glacier and slope before the disaster began to move faster. This signal in itself did not mean that the collapse would occur on a specific day, but it could mark the area as dangerous and draw the attention of experts to it.
These are the tasks that researchers want to scale with AI. The machine can continuously review huge volumes of satellite observations and look for suspicious changes in thousands of potentially dangerous locations at once.
ETH Zurich also plans to develop its own geospatial AI model within the Swiss GeoLab. It will be trained on data from satellites, drones, sensors and ground stations, in particular to identify areas at increased risk of landslides and other dangerous ground movements.
Previously, dev.ua wrote that Google DeepMind has taught AI to predict weather and extreme events more accurately and much faster. Now researchers at the ETH Zurich want to use a similar approach as an early warning system for disasters.
