A team of researchers from the Department of Energy’s Oak Ridge National Laboratory and Tennessee Technological University have created a 2D, open-source flood inundation model designed for a multiarchitecture computing system. The Two-dimensional Runoff Inundation Toolkit for Operational Needs, or TRITON, can use multiple graphics processing units, or GPUs, to model flooding more quickly and accurately than existing tools.
Flood modeling is an essential part of emergency preparedness and response. However, models must be both fast and accurate — returning simulation results in a matter of minutes — to be useful tools for decision-making and planning. The higher the model’s resolution, the more computational power it takes to run, so organizations may resort to simpler models that sacrifice accuracy for speed. The computational power of GPUs enables calculations by high-resolution models to run more quickly than simpler models that only use CPUs.
As high-performance computing has grown into an indispensable tool for science, it has also become a requirement for modern flood models to leverage the strength of hybrid CPU + GPU architectures. TRITON, the development of which was funded by the Air Force Numerical Weather Modeling Program, is specifically optimized for the multiarchitecture design of supercomputers like the IBM AC922 Summit at the Oak Ridge Leadership Computing Facility.
“The unique thing about TRITON is not just that it uses GPUs — it’s not the only GPU-accessible flood model. But it is customized to use multiple GPUs simultaneously, which makes it suitable for solving flood problems on Summit,” said Shih-Chieh Kao, an ORNL group leader who led the project.
Testing the waters
The team put the model through its paces on Summit to demonstrate its consistency, stability and some of its unique capabilities, such as the runoff hydrograph. This optional data allows TRITON to simulate pluvial floods — that is, local flash floods — in addition to riverine floods. During a riverine flood, a stream or river swells and inundates a floodplain. Using a dataset from the Federal Emergency Management Agency of 100-year flood zones as a benchmark, simulations that used the runoff hydrograph were more accurate than the basic hydraulic model alone.
“In order to really understand flood impact, we need to understand inundation, which includes how deep a river is and accounts for different flood events: riverine and flash floods. Conventional flood models usually only address riverine floods. TRITON can address both and provide more information about the flood impact,” said Kao. “If you have this inundation information, you can overlay it on assets and evaluate which are at risk and which are not.”
In another test case, the team simulated the 2017 flood in the Houston metropolitan area caused by Hurricane Harvey. The simulation covered 10 days and was modeled on two different hardware configurations: one using multiple CPUs and the other using multiple GPUs. The results soundly demonstrated the advantage of a flood model designed to run on a multi-GPU configuration. Even the smallest hardware configuration — one compute node with six GPUs — completed the simulation faster than the most powerful multi-CPU configuration of 64 nodes.
Researchers at ORNL used TRITON to simulate flood inundation in Houston, Texas, and surrounding areas that resulted from Hurricane Harvey in 2017. Light purple indicates shallower water, and dark purple indicates deeper water. Image: Sudershan Gangrade/ORNL
As an open-source toolkit, TRITON is available for free and can be used on a range of computing platforms — from laptops and desktops to supercomputers. Members of the research team are continuously developing new features and are working on algorithms to scale the current capabilities up to an operational level.
“TRITON will be a foundation for us to keep building on, and we call it a toolkit for a reason. We keep building to make it more useful — that’s our vision. As computing power increases, and the prices go down, eventually everyone should have more access to use these capabilities to better simulate floods,” said Kao.
The OLCF is a U.S. DOE Office of Science user facility.
M. Morales-Hernández, Md B. Sharif, A. Kalyanapu, S.K. Ghafoor, T.T. Dullo, S. Gangrade, S.-C. Kao, M.R. Norman, and K.J. Evans. “TRITON: A Multi-GPU open source 2D hydrodynamic flood model.” Environmental Modelling & Software 141 (July 2021): 105034. https://doi.org/https://doi.org/10.1016/j.envsoft.2021.105034.
UT-Battelle manages ORNL for DOE’s Office of Science, the single largest supporter of basic research in the physical sciences in the United States. DOE’s Office of Science is working to address some of the most pressing challenges of our time. For more information, visit https://energy.gov/science.
