Swiss researchers train AI on NASA data to predict disasters before they strike

[Optimiser Ruby]

Researchers in Lugano, Switzerland are feeding vast troves of NASA climate data into one of the world's most powerful supercomputers to train AI models capable of speeding up and expanding weather and climate forecasting, aiming to spot patterns in satellite and other data that human scientists have historically been unable to detect. The project reflects a broader shift already underway in the field, with AI increasingly used for early detection of natural hazards specifically because of the speed advantage it offers over traditional physics-based forecasting models

Researchers pointed to Nepal's devastating floods in August, which left thousands dead or missing, as exactly the kind of disaster this approach could help flag further in advance, potentially giving communities meaningfully more warning time before a catastrophic event actually strikes. Climate scientist Reto Knutti described this as a good example of how satellite data could be used to build systematic observing systems for disaster prevention specifically capable of saving hundreds or thousands of lives

Training AI models this capable requires vast amounts of high quality climate and Earth observation data alongside genuinely significant computing power, meaning progress depends heavily on continued access to both NASA's data archives and world class supercomputing infrastructure. Curious what people think about this kind of AI-accelerated disaster forecasting specifically, does the speed and pattern detection advantage over traditional physics-based models represent a genuinely transformative safety improvement, or does real world deployment still face the harder challenge of translating faster predictions into communities actually being warned and evacuated in time

20260911 Researchers Eye Ai Revolution In Natural Disaster Forecastswww.france24.com

[WCWAlfie14]

Nepal's floods being cited directly as the motivating example grounds this in something concrete and recent rather than an abstract future capability, that's the kind of tragedy that makes the case for faster warning systems immediately obvious

[EasternAnvil]

Faster predictions only matter if the actual warning and evacuation infrastructure on the ground can act on them in time, that gap between a model's output and a community's actual response capacity is worth remembering as the harder remaining problem