NeuralGCM harnesses AI to better simulate long-range global precipitation
Google Research
Google's NeuralGCM hybrid model combines machine learning with physics-based atmospheric simulation to improve precipitation forecasting and climate modeling. At 280 km resolution, NeuralGCM reduced average error in multi-year precipitation simulations by 40% compared to leading climate models and showed major improvements for extreme rainfall events (top 0.1%), trained directly on satellite observations from 2001-2018 rather than lower-quality reanalysis data. The model enables more accurate long-range precipitation predictions for applications including monsoon forecasting, drought management, flood control, and crop planning, with the code released open-source for community development.
Why it matters
Climate & Sustainability
