Abstract

Historically, the prediction of flooding to aid flood risk management and mitigation has relied largely on statistics of past flood events. Physics-based computer simulation of flood hydrodynamics—the ever-changing nature of a potentially deadly flood and the factors influencing its progression and effects—could greatly improve the reliability of predictions, but high computational demands have limited its application, particularly for large-scale, high-resolution, and long-duration flood simulations. The good news is that the world-class computing power at Oak Ridge National Laboratory has enabled ORNL researchers to overcome those limitations.

In speaking about the open-source hydrodynamic model TRITON (Two-dimensional Runoff Inundation Toolkit for Operational Needs) (https://triton.ornl.gov), I will describe a multi-year effort to scale up hydrodynamic inundation modeling using ORNL’s state-of-the-art high-performance computing systems, including Frontier, the world’s second fastest supercomputer. The TRITON model is based on two-dimensional shallow water equations, and its ability to process data has been accelerated by use of ORNL computer systems equipped with graphics processing units (GPUs), which were first used for video games.

ORNL researchers have demonstrated that TRITON can be used to simulate on ORNL supercomputers the spring 2019 Missouri River Basin flood, a major historical event that devastated midwestern U.S. communities and caused billions of dollars in infrastructure and agricultural losses. This event was reconstructed using a large computational domain covering the entire Missouri River Basin (529,000 square miles in all or parts of 10 states, or ~800,000 km²), resulting in 3.3 billion and 33 billion grid cells at 30-meter and 10-meter spatial resolutions, respectively. The simulated inundation using data on the massive flood was benchmarked against high-water mark datasets, satellite-derived flood maps, river stage observations, and other high-fidelity inundation metrics. 

Results demonstrate that, by harnessing GPU-equipped supercomputers at ORNL (including Summit), large-scale, multi-day flood events can be simulated by hydrodynamic models within a few hours. This work lays the foundation for future advancements in flood modeling, enabling more effective flood risk management and mitigation strategies, such as accurate flash flood warnings and the relocation of people living on the floodplain.

Biographical Sketch

Dr. Shih-Chieh Kao is a distinguished researcher and group leader of the Water Resource Science and Engineering Group in the Environmental Sciences Division at ORNL. He also serves as program manager for the ORNL Water Power Program, for which he oversees various research projects supported by the Water Power Technologies Office of the Department of Energy. His research areas include high-performance computing, hydrologic modeling, flood simulation, hydro-climate impact assessment, and hydropower resource evaluation. 

Since 2011, he has been leading the SECURE Water Act Section 9505 Assessment, which examines the effects of climate change on national hydropower production. He also contributed to a National Academies of Sciences, Engineering, and Medicine study on modernizing probable maximum precipitation estimates in a changing climate. He holds three degrees in civil engineering—B.S. and M.S. degrees from National Taiwan University and a Ph.D. from Purdue University.

Dr. Kao is a co-creator of the TRITON inundation model and the Dayflow dataset. He has served as a frequent reviewer for more than 20 scientific and engineering journals and has received several distinguished awards. They include the 2008 Purdue Civil Engineering Best Dissertation Award, the 2009 Journal of Hydrologic Engineering Outstanding Reviewer Award, the 2013 ICSH Statistical Hydrology Best Paper Award, and the 2020 Platform for Advanced Scientific Computing Best Paper Award. In 2023, he was named a fellow of the Environmental & Water Resources Institute of the American Society of Civil Engineers.