APPLICATIONS

The National Center for Supercomputing Applications (NCSA) has developed tools and is providing compute resources to enable atmospheric scientists to automatically produce high-resolution storm forecasts in response to real-world conditions. Hundreds of computer forecasts have been triggered by storm watches and weather discussions over the past several months. "The idea is that when a critical need exists, we need to provide resources quickly," said Jay Alameda, leader of the middleware development team at NCSA. The triggered forecasting capability is being rolled out as part of the 2007 Spring Experiment, hosted by the NOAA Hazardous Weather Testbed. The experiment, which runs during the storm-heavy season from mid-April through June, provides forecasters with a first look at the latest atmospheric science research and technology and familiarizes researchers with the needs and challenges of forecasters in the field. As part of the spring experiment's effort to improve forecasting of severe weather, NCSA has demonstrated the capability to trigger computational simulations in response to real-world conditions. When the NOAA Storm Prediction Center in Norman, Oklahoma, issues a severe thunderstorm or tornado watch, or a discussion of current conditions points to the potential for storms to develop, this information automatically initiates computational forecasts using the WRF forecasting code on one of the compute systems (currently Tungsten or Mercury) that NCSA contributes to the TeraGrid. "One of the things that's really unique about this is the unpredictability," said Brian Jewett, an atmospheric scientist at the University of Illinois at Urbana-Champaign who is working closely with NCSA on the development of advanced tools and technologies for weather forecasting. Jewett explained that watches can come at any time, and they often come in batches. So multiple simulations can be spawned around the clock. NCSA's Ensemble Broker software is designed to manage multiple computational jobs. To support the triggered simulation capability, NCSA enhanced the Ensemble Broker so it can submit jobs to the Moab cluster workload management package. The Ensemble Broker also stores numerous aspects of workflow submission and execution, which Jewett said makes it easier for researchers to reproduce the steps they took to get their results and facilitates collaborating with others. Currently, a single forecast is trigged for each watch or discussion; the forecast is carried out at two resolutions -- one with a grid spacing of 18 kilometers and the other at a high resolution of 2 kilometers. Details on the triggered forecasts are available online at its Web site. Because so many variables affect weather, forecasts are very sensitive to subtle changes in conditions. Atmospheric scientists often use ensemble forecasting to account for these uncertainties, carrying out multiple forecasts with slightly different conditions and configurations. Looking at the results of all of the simulations often provides forecasters with a best estimate for the ensuing weather conditions and gives a measure of the confidence in that forecast. The Ensemble Broker was designed to manage the thousands of jobs required for ensemble forecasting and other parameter studies. "It's exciting for me to see the capabilities," Jewett said. "It's really general, it's not weather specific, and it's deceptively powerful." For the spring experiment, all of the triggered simulations are being carried out on NCSA systems, but work has begun toward dynamic job deployment, meaning that once a forecast is submitted the best resource would be identified and the calculations would be carried out on whichever TeraGrid resource could complete the work most quickly. This type of dynamic deployment could be key during a national emergency -- whether it's a hurricane heading for the coast or an infectious disease spreading from person to person. NCSA's work on tools to support advanced weather modeling and forecasting is supported by the National Science Foundation. NCSA is a partner in LEAD (Linked Environments for Atmospheric Discovery), an NSF-supported effort to create an integrated, scalable framework for identifying, accessing, preparing, predicting, managing, analyzing, and visualizing a broad array of meteorological data and model output independent of format and physical location and in a dynamically adaptive manner.