Top NSF petascale supercomputer and expert staff accelerate discoveries for nation's scientists

Published on May 27, 2014 by Aaron Dubrow

A snapshot from the Penn State University real-time hurricane analysis and forecast system. Credits: Fuqing Zhang and Yonghui Weng, Meterology, Penn State University

Better Hurricane Forecasting

Using Stampede, Fuqing Zhang and his team of weather modelers at Penn State re-forecasted all of the tropical storms that occurred between 2008-2012, incorporating high-resolution airborne radar observations from the inner core of the storms. Averaged over 100 storms, the system reduced Day-1-to-Day-5 intensity forecast errors by 15-40% compared to the National Hurricane Center's official forecasts.

This approach has shown great promise for future hurricane systems, but requires additional computation. The problem isn't simply creating accurate models — its assimilating large amounts of Doppler radar data in real-time and merging it with information on historical precedents, as well as physical models of hurricane formation. To accomplish this data-intensive workflow, Zhang made extensive use of Stampede's Intel Xeon Phi co-processors.

"The increased computing power of Stampede has allowed us to run numerous sensitivity experiments for hurricane models at a higher resolution, allowing us to see details more clearly," Zhang said. "Especially for the hybrid data assimilation system, the improved computational performance of Stampede over previous supercomputer platforms gives us more flexibility in configuring the domain size and grid spacing that will be used."

Implementation of Zhang's new hurricane prediction system will allow emergency management officials, the private sector, and the general public to make more informed decisions during major storms, minimizing the losses of life and property.

The methodology of incorporating airborne Doppler measurements was fully adopted by the National Oceanic and Atmospheric Administration's operational hurricane prediction model, HWRF, in 2013. This breakthrough in hurricane prediction recently received the 2014 Banner Miller Award bestowed by the American Meteorological Society.

Back to overview >>