A US-France research team published in Physics World an AI-augmented rare-event
sampling (RES) climate simulation framework that uses AI to pre-screen
meteorological fields likely to spawn heatwaves or extreme cyclones and runs
full physical simulations only for those high-risk windows, cutting compute to
about 1/1,000 of full-domain simulation. Benchmarked on the PlaSim simplified
climate model, the method produced mid-latitude heatwave frequency estimates
closely matching full simulations and is applicable to analysis of current
European heatwaves. ETH Zurich researchers say the approach remedies narrow
screening in traditional RES, can be expanded to heavy rainfall and typhoon
prediction, and is adaptable to higher-resolution climate models.
