NCAR Researchers Chasing Subseasonal Forecasts with AI
A 2-to-8-week weather forecast could change how entire industries make operational, financial, and risk decisions and researchers at the National Center for Atmospheric Research (NCAR) are using AI to make that elusive window more attainable.
This brilliant team is advancing CAMulator, a machine-learning emulator of the Community Atmosphere Model. The performance is striking: CAMulator can generate 480 simulated years per day - about 350× faster than CAM6. For real-time research, it can produce a subseasonal forecast in under three minutes and generate a 20-year hindcast suite in about 12 hours.
Why does that matter outside atmospheric science? The 2-to-8-week forecasting window sits directly inside critical business planning cycles.
Better information about whether coming weeks are likely to be hotter, colder, wetter or drier could ultimately help:
1) Energy companies and utilities anticipate demand, generation conditions and operational risks.
2) Agriculture make better-informed irrigation, planting, harvesting and input decisions.
3) Water managers prepare earlier for changing supply and demand conditions.
4) Infrastructure-dependent businesses improve contingency planning around weather-driven disruptions.
CAMulator is already demonstrating skill in capturing large-scale patterns such as El Niño, the Pacific–North American pattern and the North Atlantic Oscillation. #NCAR is now developing subCESMulator, which adds interactive land and ocean variables in pursuit of greater subseasonal forecasting skill.
There is another important impact: economics of access. These AI emulators require dramatically less computing power than traditional simulations and are designed to be usable on a modern laptop. That means more scientists can experiment, test hypotheses and accelerate discoveries without requiring continuous access to a supercomputer.
CO-LABS champions the National Center for Atmospheric Research because this is exactly what sustained investment in federally funded labs makes possible: foundational science, world-class modeling and AI expertise being combined to create capabilities with potentially enormous economic value. The research is supported by the U.S. Department of Energy (DOE) and NCAR; continued investment is essential if the U.S. intends to lead the rapidly accelerating global competition to apply AI to weather and Earth-system prediction.
Comments