National Weather Service Moves Supercomputing to Cloud for AI

The National Weather Service is migrating its core supercomputing infrastructure to the cloud, a move designed to accommodate the rapidly evolving landscape of AI-powered weather prediction and provide greater operational flexibility.

The agency currently operates two on-premises supercomputers that power all its weather models through a system called the Weather and Climate Operational Supercomputing System (WCOSS). With that equipment nearing the end of its lifecycle, NWS is seizing the opportunity to modernize rather than simply replace aging hardware.

Why it matters

Traditional weather models and AI prediction systems have fundamentally different computational requirements. Fixed on-premises infrastructure struggles to reallocate resources quickly when new AI capabilities emerge or severe weather demands surge capacity. Cloud environments can shift that balance dynamically, allowing meteorologists to deploy improved forecasting tools faster and respond to urgent weather events more effectively.

A phased transition to Google Cloud

NWS recently selected Google Cloud as its primary high-performance computing provider for WCOSS. David Michaud, director of NWS central operations, told FedScoop the agency prioritized flexibility over raw computing power alone.

"We wanted to look beyond just trying to build the fastest system that we could or the largest system that we could," Michaud said. "We were really looking at a balance between the computing that we have and the flexibility that we would get moving to the cloud."

The agency is taking a risk-reduction approach by establishing a smaller early-access environment first, running performance tests before full-scale deployment. NWS expects the complete at-scale cloud environment to be operational by early 2027.

Returning AI models to their native environment

The cloud migration will actually reunite some of NWS's AI models with the infrastructure they were designed for. Richard Bandy, director of NWS's Meteorological Development Laboratory, explained that several existing AI-based models were developed in the cloud using Google DeepMind solutions, then moved to on-premises systems for operational use.

Those models include AIGFS, a global model that produces forecasts with less computational overhead, and AIGEFS, an ensemble model providing probabilistic forecasts. The agency has also deployed Hybrid-GEFS, which combines traditional numerical weather prediction with AI for improved accuracy.

The Warn-on-Forecast System, designed to accelerate warnings for tornadoes, severe thunderstorms, and flash floods, has been running in demonstration mode but lacked an operational cloud environment for full deployment. The WCOSS migration solves that constraint.

The future of hybrid forecasting

NWS envisions a future where traditional physics-based models work alongside AI systems, leveraging the strengths of both approaches. Bandy noted that while AI models can run significantly faster, the agency doesn't plan to abandon numerical weather prediction entirely.

The cloud transition is part of broader modernization efforts at NWS, including migration of weather data to cloud platforms called NWS HIVE and NWS CIRRUS, which will give forecasters mobile access to critical information outside traditional office environments.

These details were first reported by Madison Alder at FedScoop.