Satellite image of Hurricane Sandy (Credit: NASA/Goddard/MODIS Rapid Response Team)
SUBTOPIC

Extreme Weather Events

Every hour of lead-time on a hurricane or supercell is a life saved — and a race against the solver's clock.

Image: NASA/Goddard/MODIS Rapid Response Team

Where Physics Still Beats Pattern Recognition

Extreme weather forecasting is the sharpest edge of meteorology: hurricanes, supercells, and record-breaking heat or cold events. Recent research shows that AI weather models, while fast and accurate for average conditions, systematically underestimate the intensity and frequency of unprecedented extremes because they are trained on historical data that, by definition, contains no record of events beyond the record. Malgukke architectures are built to run physics-based ensemble models and AI systems side by side, giving forecasters both speed and the physical fidelity that extremes demand.

HPC Solution Architecture: Storm to Warning

A hybrid physics/AI pipeline for cyclone and severe convective storm prediction:

Storm Detection Satellite + Doppler radar Ensemble Core CPU Nodes Non-hydrostatic solver GPU Nodes AI trajectory ensemble Extreme-Value Cross-Check Physics vs. AI divergence Impact Modeling Storm surge, wind field, hail risk Early Warning Local alerts, evacuation lead-time Verified outcomes retrain ensemble weighting for the next event

Running on the TOP500: Dogwood & Cactus

TOP500-RANKED — NOAA WCOSS

Dogwood (Manassas, VA) & Cactus (Phoenix, AZ) — NOAA

NOAA's twin operational supercomputers, each delivering 12.1 petaflops, were ranked No. 49 and No. 50 on the TOP500 list at their 2022 deployment and remain NOAA's primary machines for the National Weather Service today. They were built specifically to run the Hurricane Analysis and Forecast System (HAFS), the Global Forecast System (GFS), and severe-storm models — the exact extreme-weather workloads described on this page.

Yes — these are real, currently operating TOP500-ranked systems (as of their most recent published ranking), purpose-built for hurricane and severe-storm forecasting rather than for climate research generally.

#49/#50
TOP500 Rank at Deployment
12.1×2
Petaflops (combined)

Voices from the Field

Researchers examining where AI and physics-based models diverge on extreme events:

Zhongwei Zhang

Institute of Statistics, Karlsruhe Institute of Technology (KIT)

On why AI weather models still systematically underestimate record-breaking heat, cold, and wind extremes.

Read article
Johannes Brandstetter

Johannes Kepler University Linz / Microsoft AI for Science

On "Aurora", an AI foundation model that outperformed operational centers in forecasting every hurricane of 2023.

Read article