ESRL to Host Hurricane Computer
A powerful new supercomputer is heading to ESRL this spring, to help scientists improve hurricane forecasts. The Linux cluster will be based on Intel’s latest member of its Xeon line, code-named Nehalem.
“This will be a development machine,” said Leslie Hart, a High-Performance Computing leader in ESRL’s Global Systems Division. “The idea is to develop and improve the next-generation models for hurricane prediction.”
Hart said the new computer cluster will have at least 5,000 cores and will occupy about 15 racks, each 84 inches tall. He and his colleagues hope the “Nehalem Jet” is ready to churn out experimental hurricane track and intensity forecasts by August, when the US hurricane season approaches its peak.
Steve Koch, Director of ESRL’s Global Systems Division, said ESRL will host the new computer, but researchers from across government and academia will be able to access it for hurricane research. The Nehalem Jet will likely be used to run a variety of research models, from ESRL’s experimental FIM (the Flow-following, finite-volume Icosahedral Model) to the National Center for Environmental Prediction’s Hurricane WRF (Weather Research and Forecasting) model.
Last season, ESRL researchers and colleagues experimented with FIM (a global model) using model grid resolutions of 15 and 30 km, to generate hurricane track forecasts, including running ensembles of models. FIM forecast the track of Hurricane Ike and several other tropical storms last year with impressive skill several days before landfall. However, running the experiment required intense computing power, and ESRL scientists relied on powerful computers at the Texas Advanced Computing Center. In the future, the new Nehalem Jet computer will enable more accurate forecasts from an improved higher-resolution version of FIM, and some of that work can be conducted in-house more easily.
Running at even higher resolution than FIM is Hurricane WRF, a limited-area regional model that is planned to couple separate models of the ocean, atmosphere, and waves. Computer-intensive experiments may help researchers improve the ocean models, for example, or develop higher-resolution, larger ensembles of models to provide forecasters at NOAA’s National Hurricane Center with improved measures of forecast uncertainty.
The new computer will also let researchers experiment with incorporating Doppler radar and other data collected by the NOAA P-3 hurricane hunter airplanes. In theory, those data could improve forecasts, but it is difficult to quickly quality-check and assimilate them into models.
Funding for the new computer and supportive infrastructure—more than $6 million—came from a late decision by the Bush Administration to boost the $4 million budget of NOAA’s Hurricane Forecast Improvement Project to about $17 million. The new computer will triple the computing power of NOAA’s Research and Development High-Performance Supercomputer, hosted by ESRL, Koch said.