Advancements in flood risk understanding by ORNL supercomputer tool

Flooding, often triggered by heavy and sustained rainfall, is responsible for a growing number of billion-dollar weather-related disasters worldwide, according to NOAA records. Understanding this risk has become increasingly critical as extreme weather events intensify.

Shih-Chieh Kao, leader of the Water Resource Science and Engineering Group at Oak Ridge National Laboratory (ORNL) and program manager for the ORNL Water Power Program, explained ORNLs recent progress in flood modeling. ORNL has developed a high-performance computational tool that leverages the Summit and Frontier supercomputers, utilizing thousands of graphics processing units (GPUs) to accelerate flood simulations.

Initially designed for gaming, GPUs provide the processing power necessary for Kaos team to perform complex calculations rapidly. Frontier is currently the worlds second-fastest supercomputer, while Summit, once ranked fastest, was decommissioned in 2024. Enhanced modeling allows us to better assess flood risks and uncertainties, particularly as climate and environmental conditions evolve, Kao stated.

The significance of this work became clear during the catastrophic flash flood along the Guadalupe River in Texas on July 4, 2025, which claimed at least 135 lives, including 27 campers and counselors at Camp Mystic. The floodwaters rose 26 feet in 45 minutes, far exceeding the 5 inches predicted by meteorologists and resulting in widespread fatalities.

Kao emphasized the connection between climate change and increased flooding. Warmer air holds more moisture, which can fall as intense rainfall over short periods. He compared the Texas flood to frequent flash floods in Taiwan, highlighting the dangers of sudden water surges even under seemingly calm conditions.

To address these challenges, ORNL developed TRITON (Two-dimensional Runoff Inundation Toolkit for Operational Needs), a physics-based, open-source hydrodynamic model. TRITON simulates flood dynamics using detailed topographical data, such as digital elevation models (DEMs), and streamflow hydrographs to predict water spread and timing. High-quality DEMs, created through LIDAR or other remote sensing techniques, improve the models accuracy.

TRITON can run on multiple computing platforms, from a single GPU to large supercomputers, and uses the full 2D shallow-water equations to model flood propagation. The toolkit has been applied to major events, including Hurricane Harvey (2017), Hurricane Laura and Hurricane Sally (2019), and extensive Missouri River Basin flooding in 2019. The Missouri River simulation, using 6,000 GPUs, modeled five months of flooding in just three days, representing the largest hydrodynamic inundation simulation to date.

TRITON supports emergency preparedness by evaluating numerous flood scenarios, identifying high-risk areas, and guiding protective measures. The tool has also been enhanced through TRITON-Lite, an AI-based surrogate model that efficiently generates flood inundation maps for timely operational use.

The toolkit helps anticipate the impacts of climate-driven extreme floods, informing flood alarms, zoning, building requirements, and insurance needs. Calibration and validation remain challenging, particularly when satellite images are obscured by clouds during floods, and simulations can produce massive amounts of data requiring careful management.

TRITON and TRITON-Lite have earned ORNL recognition, including a DOE INCITE Leadership Computing Award, providing millions of computing hours on Frontier and Aurora for expanded simulations. These models offer insights into historical and future floods, improving strategies for risk management and mitigation as the planet warms.

Images from affected areas, such as the Guadalupe River near Camp Mystic, illustrate the devastating impact of flash floods, underscoring the importance of predictive modeling and preparedness.

Author: Caleb Jennings