After the storm: Simulating and visualizing extreme weather with XSEDE

by Katherine Kendig

Brian Jewett, research scientist in the Department of Atmospheric Sciences at Illinois, used to be a storm chaser. Or perhaps storm “racer” is a better term: On cold winter days when others were bundling up or staying home, he would head out to the Peoria airport in anticipation of a snowstorm. If he stayed ahead of the snow, he and the “PLOWS” team (short for ProfiLing Of Winter Storms) were piloted skyward in an NSF-funded ex-Navy plane to take radar and cloud-probe measurements that could give forecasters a better foundation for snow predictions. If, however, snow hit the runway before he did, he was out of luck: de-icing the plane upset the aircraft’s sensitive scientific instruments, so takeoff post-snowfall was out of the question.

Of course, flying through the snow isn’t all Jewett does (and he’d be the first to tell you that those airbound adventures are a lot safer than they sound: “There’s no white-knuckle element,” he insists). Jewett spends most of his time on solid ground, studying both tornadoes and winter storms. And when he’s not gathering data in the field, he’s working with resources and experts from the Extreme Science and Engineering Discovery Environment (XSEDE), a nationwide collection of advance digital resources and support lead by the National Center for Supercomputing Applications (NCSA) at Illinois, to analyze, simulate, and visualize severe storms.

One of Jewett’s current projects involves simulating a catastrophic tornado from 2011—the Joplin, MO, tornado, which struck on May 22, 2011, and left over 150 dead. With so much of the science of weather focused on using the present to predict the future, it may seem strange to look back at the unchangeable past. Jewett suggests, however, that modeling storms that have already struck allows scientists to enhance their knowledge in meaningful ways, improving our understanding of how and why tornadoes strike. Take the Joplin tornado, which occurred when two small storms merged with a larger storm. Would there have been a tornado without that merger? Using simulations, Jewett and Kevin Van Leer, whose Master’s thesis at Illinois developed from his work studying Joplin, can answer that question. By altering the conditions in their model, they discovered that if the small storms had been weaker, there would have been no tornado—but, unexpectedly, if they hadn’t existed at all, the twister would have found another way to form. And it’s not just weather that can be altered in simulations: When modeling a Chicago snowstorm, the PLOWS team decided to figure out the effect Lake Michigan had on the snow—by removing it from the model entirely.

In the old days, Jewett says, scientists had to search through hundreds of real storm records trying to determine the consequences of specific conditions on the progression of a particular storm. Now, they can use models. For Jewett, XSEDE makes all these simulations possible. Jewett remembers questioning whether the work he needed done could be accomplished on a local computer cluster—but the answer was no. “These are big problems, scaling to bigger and bigger problems,” he says. “And they’re computationally intensive. We really do need XSEDE.”

Jewett and his team utilize the Stampede and Stampede2 supercomputers and the Ranch mass storage system housed at the Texas Advanced Computing Center (TACC) in Austin. Jewett believes the ability to work with computing resources at XSEDE’s scale is highly valuable for future scientists; that’s why he teaches two courses—Introduction to Computational Geosciences and, at the graduate level, Numerical Fluid Dynamics—that introduce students to XSEDE resources like Jetstream (a cloud environment hosted by TACC and Indiana University) and the very same supercomputers Jewett conducts his research on. “The idea is to expose them to supercomputer center resources, so they’ll know what questions to ask” when conducting their own research with powerful computing resources.

In addition to compute resources, Jewett works with two kinds of experts through XSEDE’s Extended Collaborative Support Services (ECSS). David O’Neal of the Pittsburgh Supercomputing Center has helped Jewett and his students run their systems efficiently, particularly when it comes to parallelization. “XSEDE has done their own benchmarking,” on optimal parallelization practices, Jewett notes, and O’Neal brings that understanding to Jewett’s project. O’Neal also helps Jewett store outputs in a cost-effective manner—something Jewett says scientists normally consider a hassle—which then allows Jewett to interact with his data at a more detailed level. Now, Jewett is also working with Dave Bock, a visualization expert at NCSA, to better understand the lead-up to the Joplin tornado. Bock is a “visualization wizard,” says Jewett, who previously worked with NCSA’s Advanced Visualization Lab to bring his and Van Leer’s Joplin simulations to life in 2016. This time around, he has additional data (in part thanks to O’Neal) that will help fine-tune the process. “We’re going to find out new things this way,” Jewett predicts.

Flying through snowstorms is undoubtedly exhilarating, but modern computing capabilities make the power and possibility of computer simulations just as exciting—and a fair bit warmer. “XSEDE is a fabulous resource,” Jewett says, adding that today’s technology makes scientists wonder what leaps of understanding might have been achieved earlier if brilliant people like Ted Fujita (namesake of the Fujita scale for measuring tornado intensity) had had access to resources like those offered by XSEDE. “He had such partial data,” Jewett says of Fujita. “What if he had had supercomputers?”

Profiling of Winter Storms is supported by the National Science Foundation through award AGS-1247404.

Disclaimer: Due to changes in website systems, we've adjusted archived content to fit the present-day site and the articles will not appear in their original published format. Formatting, header information, photographs and other illustrations are not available in archived articles.