Prepping for the STORM: Overhauling storm-surge code

For the past 15 years, Clint Dawson has been predicting hurricane storm surge. Dawson is the director of the Computational Hydraulics Group at the Institute for Computational Engineering and Sciences (ICES) at The University of Texas at Austin, US. Using computational methods that detail the surge location and depth, Dawson and his collaborators have studied storm surge for every hurricane strike in the US since the late 1990s, and helped Texas emergency managers develop hurricane evacuation plans.

Predicting storm surge currently requires feeding hurricane data into a computational program called ADCIRC, which employs high-performance computers to efficiently generate data about potential outcomes. However, in the drive for increased performance, computational hardware is becoming more and more diverse. For the predictive program to get the best performance across systems, the code must keep up with the changes. “The idea is how do we keep [the program] up to date and modernize it for the next generation,” Dawson says.

Funded by a $3 million National Science Foundation grant, Dawson and US collaborators at Louisiana State University (LSU), the University of Notre Dame in Indiana, and the University of North Carolina at Chapel Hill (UNC) are overhauling ADCIRC into version 2.0, dubbed STORM. The new version is designed to perform more efficiently across a variety of computer hardware architectures. The ultimate goal is to double the speed of ADCIRC, enabling version 2.0 to predict storm surge within an hour of receiving data inputs.

The four-year grant is split between Dawson and collaborators: Hartmut Kaiser, The Stellar Group, LSU; Joannes Westerink, Computational Hydraulics Laboratory, Notre Dame; and Richard A. Luettich, Coastal Hazards Center, UNC. Craig Michoski, a researcher in ICES Computational Hydraulics Group, is a co-principal investigator.

Since its development in the mid-1990s, ADCIRC has been widely used to simulate and predict water flow in US coastal areas. The National Oceanic and Atmospheric Administration, US Army Corp of Engineers, and academic researchers all use the program to help inform their work. Although ADCIRC is primarily for storm surge prediction, the governing equations that describe fluid flow can be applied to other research questions as well. During the Deep Water Horizon oil spill in 2010, Dawson used ADCIRC to predict oil dispersal paths up to three days in advance.

For each fluid flow scenario, the ADCIRC system analyzes the interaction between static elements, such as coastal and undersea topography, and dynamic ones, such as how a hurricane influences water height and water velocity. ADCIRC calculates the variable interactions and their extrapolations through partial differential equation computations, which run multiple times using plausible ranges of variables to produce a selection of potential scenarios. The most likely prediction, a collection of data describing the final fluid behavior at a point in time, is the final result. In the case of storm surges, this information is used to inform emergency response and evacuation plans, and to create maps.

The STORM program will maintain the same ADCIRC functionality, but the code driving it – High Performance ParallelX (HPX) – will provide a completely new foundation for the algorithms, Dawson says. The HPX code is designed to be flexible and easily comingle with other code types, adapting to diverse computer architectures. By rewriting the code using HPX, Dawson and collaborators are ensuring STORM will not only run more efficiently on today’s supercomputing systems, but also be well equipped to handle inevitable changes.

“Where we hope to be in four years is to have a whole new code and a whole new piece of software. And it’s going to be a lot of work but it’s also necessary work if you want to keep your software useful for the next generation,” says Dawson. At the same time, Dawson says turning ADCIRC into STORM will be an exercise in understanding the history and composition of the original code, which could help in constructing STORM and other programs.

“If you don’t do these kinds of projects you lose all this memory of how you got to this point," Dawson says. "We’re really fortunate to have this opportunity to take all of the lessons that we’ve learned and to put them into a new piece of software."