Contact: Barbara McGehan FOR IMMEDIATE RELEASE
(303)497-6286 5/1/95
Computer scientists have come up with a new and less expensive way of generating national-scale weather forecasts, the National Oceanic and Atmospheric Administration announced today. For the first time in computing history, a weather forecast model normally run on a large multiprocessor supercomputer has been run on a regular basis on a Massively Parallel Processor.
Using a simulated operational environment, computer scientists at NOAA's Forecast Systems Laboratory in Boulder, Colo., have successfully demonstrated the feasibility of running complex weather forecast models on Massively Parallel Processor (MPP) computers. The scientists use real-time weather data and make real weather forecasts every three hours on a regular basis, but in a research setting. According to Tom Henderson of the Forecast Systems Laboratory, "We believe we are the first to demonstrate quasi-operational use of an MPP system for national- scale weather forecasting."
MPP technology offers new opportunities for achieving affordable performance to meet the increasingly complex computing requirements of weather forecasting models. MPP systems use relatively inexpensive commodity microprocessors and memory connected with some form of high-performance communication network. High performance is achieved by using a large number of processors. In contrast, traditional vector supercomputers are built from expensive custom circuits and memory that rely on "state-of-the-art" semiconductor technology.
Work is currently under way at FSL to provide the aviation community with improved short-range weather forecasts. These forecasts are made using numerical weather prediction models that run on a variety of computing platforms. However, in a time of decreasing funds, FSL is developing expertise in using MPP systems in order to demonstrate the cost-effectiveness and operational capability of MPP technology. The long-term goal is to use MPP systems in operational environments before the end of the decade.
The Rapid Update Cycle (RUC) model which runs operationally at the National Meteorological Center in Suitland, Md., and produces national-scale weather forecasts, was used for this demonstration that was run on an Intel Paragon MPP.
The RUC is a mesoscale analysis and forecast system that operates on a three-hour cycle to provide frequent updates of atmospheric conditions and short-range national forecasts out to 12 hours. The model covers the continental United States and is run on a 60 km grid.
Funding for this project is being provided by the Federal Aviation Administration's Aviation Weather Development Program.