ORNL’s TRITON: How GPUs and Supercomputers Accelerate Flood Forecasting and Risk Mapping

Oak Ridge National Laboratory (ORNL) researchers have developed TRITON, a GPU-accelerated, open-source hydrodynamic toolkit that dramatically speeds large-scale flood simulations. By harnessing modern graphics processing units (GPUs) and high-performance systems such as Summit and Frontier, TRITON runs complex two-dimensional shallow-water equations much faster than conventional approaches, enabling rapid scenario testing for emergency planning and infrastructure design.

Why this matters

Flooding is among the leading causes of weather-related fatalities and multi-billion-dollar natural disasters worldwide. As a warming atmosphere can hold more moisture and fuel more intense storms, rapid and accurate flood prediction is becoming increasingly critical for communities, emergency managers, and planners. TRITON helps translate weather and terrain data into actionable inundation maps and risk estimates in hours rather than days or weeks.

How TRITON works

TRITON (Two-dimensional Runoff Inundation Toolkit for Operational Needs) is a physics-based model that solves the full 2D shallow-water equations. It combines high-resolution terrain grids (digital elevation models, or DEMs), streamflow hydrographs, and runoff inputs to simulate how water spreads across landscapes and through river networks. TRITON uses DEMs derived from photogrammetry, radar interferometry, or LIDAR; the higher the DEM quality, the more reliable the flood predictions.

“With enhanced modeling capabilities, we can better understand the risks and uncertainties associated with floods, which is particularly important given changing climatic and environmental conditions,” said Shih-Chieh Kao, leader of ORNL’s Water Resource Science and Engineering Group and manager of the ORNL Water Power Program.

Computing at scale

GPUs—originally designed for graphics—provide the parallel processing power that enables TRITON to run very large simulations quickly. The code supports multiple GPU programming models (CUDA for NVIDIA, HIP for AMD, and Kokkos for portability) and can run on a single GPU up to thousands of GPUs on modern supercomputers. For example, using roughly 6,000 GPUs (about one-fifth of Summit’s capacity), the team simulated five months of flooding across the Missouri River Basin in approximately three days.

Real-world applications and results

TRITON has been applied to notable events, including a 10-day simulation of Hurricane Harvey’s flooding in 2017, where the team found precipitation that exceeded previously estimated probable maximum values for that storm. It has also been used to model floods from Hurricanes Laura and Sally (2019) and large-scale riverine flooding in the Missouri River Basin. In 2024, TRITON was used to simulate 10 days of flooding around Asheville, North Carolina, on a separate high-performance system.

TRITON-Lite and operational use

To support faster, resource-efficient operational needs, the team developed TRITON-Lite—an AI-based surrogate model that produces inundation maps (depths and extents) more quickly while maintaining useful accuracy for many response and planning tasks. This surrogate can help deliver timely flood forecasts to emergency managers and the public when compute resources are limited.

Challenges and next steps

Despite advances, challenges remain: precise local rainfall forecasting is still difficult, calibration and validation data for floods are often scarce, and satellite imagery can be obscured by clouds during active events. Large simulations also produce terabytes of output, creating data-management, compression, and archival needs. ORNL’s work has been recognized with computational awards to continue expanding simulations and refine models under future climate scenarios.

TRITON is open source and available at https://code.ornl.gov/hydro/triton for researchers and agencies that wish to test, contribute to, or deploy the toolkit.

Implications: Faster, higher-fidelity flood modeling enables more robust early warning systems, better-informed infrastructure and land-use decisions, improved evacuation planning, and clearer assessments of how flood risk may change under a warming climate.