According to the NOAA National Severe Storms Laboratory, flooding causes more damage than any other severe weather-related event, an average of $5 billion per year in the U.S. It kills more people each year than tornadoes, hurricanes or lightning, and it occurs in every U.S. state and territory. It’s a threat experienced worldwide, anywhere that receives rain.
A flood event in Iowa back in 2008 cost $10 billion in damages to the state, prompting the establishment of the Iowa Flood Center (IFC) at the University of Iowa, the first center in the country for advanced flood-related research and education.
Simplified 2-D models are the current state-of-the-art for predicting flood wave propagation, but an IFC team led by UI professor George Constantinescu is creating 3-D non-hydrostatic flood models to more accurately simulate how floods spread across land. These 3-D models can also be used to assess and improve the predictive capabilities of the 2-D models already in use.
The researchers used Titan, one of the world’s most powerful supercomputers, to perform a high-resolution 3-D simulation of a dam break in a natural environment—allowing them to map precise water levels for actual flood events over time. "Flood events, like those generated by dam breaks, can be computationally very expensive to simulate," Constantinescu said. "Previously, there wasn't enough computer power to do these kinds of time-accurate simulations in large computational domains, but with the power of high-performance computing and Titan, we are achieving more than was previously thought possible."
Titan is the 27-petaflop Cray XK7 at the Oak Ridge Leadership Computing Facility (OLCF), a U.S. Department of Energy facility located at the Oak Ridge National Laboratory in Tennessee.
The team's simulations also highlighted prediction inaccuracies that may occur in using 2-D models, which make simplified assumptions about some aspects of flow. These include the inaccuracies about the amount of surface area flooded and the length of time that dangerous flood levels will last, as well as underestimation of the speed at which floods spread and overestimation of the time at which waves reach their highest point.
Titan simulated two hypothetical dam failures using computational fluid dynamics software STAR-CCM+, which features a volume-of-fluid method to track the areas where water meets air. The researchers employed a computational grid of about 35-50 million cells for each simulation, and used 2,500 of Titan's CPU processors to achieve peak simulation performance. The Reynolds-averaged Navier-Stokes (RANS) method, a widely-used set of equations for modeling turbulent flows, was also employed.
Constantinescu said that simulations of full flooding events over larger physical regions will be possible as computers become faster and more powerful. He noted that Summit, a next-generation OLCF supercomputer scheduled to come online in 2018, will open new possibilities for research. "Eventually, things we previously had to do by hand, such as generating a high-quality computational grid, will just be part of the typical software package," he added.