Home // COCE 2025, The Second International Conference on Technologies for Marine and Coastal Ecosystems // View article

Accelerating HEC-RAS: A Recurrent Neural Operator for Rapid River Forecasting

Authors:
Edward Holmberg
Pujan Pokhrel
Elias Ioup
Ken Pathak
Steven Sloan
Kendall Niles
Jay Ratcliff
Maik Flanagin
Christian Guetl
Julian Simeonov
Mahdi Abdelguerfi

Keywords: Fourier Neural Operator, Surrogate Modeling, HEC-RAS, Gated Recurrent Units.

Abstract:
Physics‑based solvers such as HEC‑RAS provide high‑fidelity river forecasts but are too slow for on‑the‑fly decision‑making during floods. We present a deep‑learning surrogate that treats HEC‑RAS as a data generator and couples a Gated Recurrent Unit (GRU) for short‑term memory with a geometry‑aware Fourier Neural Operator (Geo‑FNO) for long‑range spatial coupling. Trained on 71 reaches of the Mississippi River Basin and evaluated on a year‑long hold‑out, the surrogate achieves a median absolute stage error of 0.28 ft. For a full 71‑reach ensemble forecast, it reduces wall‑clock time from 139 to 40 minutes (~3.5×). By reading native HEC‑RAS files and operating on a compact eight‑channel feature interface, the model delivers operational speed while preserving fidelity, enabling rapid “what‑if” ensemble guidance.

Pages: 38 to 45

Copyright: Copyright (c) IARIA, 2025

Publication date: October 26, 2025

Published in: conference

ISBN: 978-1-68558-329-3

Location: Barcelona, Spain

Dates: from October 26, 2025 to October 30, 2025