Extreme weather events that were once considered rare are now increasingly common, which challenges the traditional definition of “100-year events.” In response, researchers at the University of Washington have developed a new AI-powered climate model, Deep Learning Earth SYstem Model (DLESyM), that simulates Earth’s climate up to 1,000 years using minimal computing power, making advanced forecasting more accessible.
Unlike traditional models, DLESyM was trained on short-term data but still effectively captures seasonal and interannual variability. It performs comparably to leading supercomputer-based models in simulating phenomena like tropical cyclones and atmospheric blocking. This innovation reduces the carbon footprint of climate modeling and democratizes access to powerful forecasting tools for researchers worldwide.