Knowledge-Graph-Augmented Chronos-2 for HEC-RAS Surrogate Forecasting

Edward Holmberg; Elias Ioup; Mahdi Abdelguerfi

cs.LGSep 18, 2026

Knowledge-Graph-Augmented Chronos-2 for HEC-RAS Surrogate Forecasting

Authors: Edward Holmberg, Elias Ioup, Mahdi Abdelguerfi

Organizations: Canizaro-Livingston Gulf States Center for Environmental Informatics New Orleans, LA, USA · Center for Geospatial Sciences Naval Research Laboratory Stennis Space Center, Mississippi, USA

Abstract

We investigate whether coupling a time-series foundation model to hydraulic project knowledge improves surrogate forecasting of HEC-RAS water-surface elevation (WSE). We present KG-Chronos-2, which combines a frozen Chronos-2 predictor with exact-state residual decoding, graph-conditioned historical retrieval, and input-aligned correction. We compare the method with persistence, a residual LSTM, project-conditioned recurrent GeoFNO, a hydraulic DCRNN-style model, and frozen Chronos-2. Task-specific fitting uses the 2008 simulation. Evaluation covers 64 fixed 24-hour windows from the 2011 and 2002 simulations at 4,675 cross sections in 71 reaches on a shared geometry. KG-Chronos-2 achieves event-balanced root-mean-square error 0.246970 in native WSE units. It reduces RMSE by 14.13% relative to frozen Chronos-2, 29.38% relative to the hydraulic DCRNN-style model, and 39.54% relative to recurrent GeoFNO. The 95% hierarchical-bootstrap interval for its event-balanced RMSE difference from frozen Chronos-2 is [-0.075177, -0.016317]. KG-Chronos-2 also achieves the lowest active-window and final-lead RMSE among the six completed systems. These results support coupling a frozen temporal predictor to project knowledge for warm-start HEC-RAS forecasting on the fixed benchmark.

Figures & tables

Panel of Figure 1 — Fig. 1 : KG-Chronos-2 forecast path. Frozen Chronos-2 generates an exact-state baseline. The hydraulic knowledge graph conditions historical-analog retrieval and fusion, then maps prescribed inputs to reaches for gated residual correction. Both the baseline and analog forecast start from $\bm{y}_{t}$ . Each forecast contains 24 hourly WSE fields at all 4,675 cross sections.

Knowledge object	Link and forecasting role
Cross-section catalog	Identifier $\rightarrow$ reach membership; aligns outputs and localizes analog reconstruction.
Reach graph	Same-river order $\rightarrow$ approximate adjacency; supplies directional descriptors and distance weights.
Static attributes	Reach $\rightarrow$ 14 descriptors; reweights adjacency and conditions correction.
Input catalog	Native input identifier $\rightarrow$ receiving reaches; aligns seven channels across events.
Historical bank	Source event/origin $\rightarrow$ response trajectory; supplies training-segment analogs.

TABLE I : Hydraulic knowledge in KG-Chronos-2

System	Benchmark role	Forecast-time information	Fitting and generation
Persistence	Constant-state reference.	Exact current WSE.	Repeat the field at each lead.
Residual LSTM	Task-trained temporal model.	Latent WSE increments and exact current field.	Fit recurrent encoder and decoder; decode 24 predicted increments.
Recurrent GeoFNO	Published-method hydraulic reference.	WSE history, coordinate, static attributes, and prescribed input summaries.	Fit GRU/spectral blocks; recursively advance 24 hourly states.
Hydraulic DCRNN-style	Graph-recurrent hydraulic model.	WSE history, cross-section graph, static attributes, and prescribed input summaries.	Fit diffusion-GRU encoder and decoder; roll out native-field increments.
Frozen Chronos-2	Pretrained temporal reference.	WSE increment history and exact current field.	Fit scaling/PCA and an increment multiplier; retain pretrained weights.
KG-Chronos-2	Proposed knowledge-coupled surrogate.	WSE history, reach graph, attributes, source analogs, and prescribed inputs.	Retain Chronos; fit retrieval fusion, response alignment, and gated correction.

TABLE II : Six-system benchmark. Every system predicts 24 hourly WSE fields at all 4,675 cross sections. The recurrent GeoFNO and DCRNN-style models receive reach-level input summaries; KG-Chronos-2 retains channel-specific input alignment.

Plan	Year	Frames	Role
p07	2008	8,737	Source training, validation, and chronological source test
p06	2011	8,737	32 cross-event forecast windows
p08	2002	8,713	32 cross-event forecast windows
Cadence: 1 hour; domain: 4,675 cross sections, 71 reaches
Nominal history: 512 hours; forecast: 24 hourly WSE fields

TABLE III : Dataset and common forecast protocol

System	EB RMSE	EB MAE	EB active RMSE	EB 24-h RMSE
Persistence	0.492611	0.229447	0.654633	0.789602
Residual LSTM	0.483055 $\pm$ 0.005266	0.234526	0.637714	0.778346
Recurrent GeoFNO	0.408510	0.138732	0.586309	0.656622
Hydraulic DCRNN-style	0.349729 $\pm$ 0.012500	0.119977	0.470438	0.558798
Frozen Chronos-2	0.287613	0.085292	0.431725	0.493318
KG-Chronos-2	0.246970	0.082985	0.351510	0.434783

TABLE IV : Six-system results in native WSE units. EB averages event-level metrics equally. LSTM and DCRNN-style entries average three individual-seed metrics;

\pm

gives the sample RMSE standard deviation. Other rows use one forecast record. Bold marks the lowest mean error.

Fig. 2 : Event-balanced RMSE for six completed systems. Whiskers show the sample standard deviation across three training seeds for LSTM and DCRNN-style. The other systems each supply one reference forecast.

Fig. 3 : Direct KG-Chronos-2 minus frozen Chronos-2 event-balanced RMSE differences. Points and 95% intervals come from the saved 10,000-replicate paired event/window bootstrap. Negative differences favor KG-Chronos-2.

Fig. 4 : Active-window and final-lead RMSE for the same six systems. Active RMSE uses 16 windows selected from pre-origin increments; final-lead RMSE uses lead 24 over all 64 windows. LSTM and DCRNN-style markers average individual-seed metrics.

Explore similar work

Learned Response-Field Inertia Operator for HEC-RAS 2D Water-Surface Elevation Prediction
Jun 4, 2026Edward Holmberg, Elias Ioup, Md Meftahul Ferdaus +2Neural Surrogate Modeling PDE Surrogate Modeling
Canizaro Livingston Gulf States Center for Environmental Informatics, Department of Computer Science, The University of New Orleans, New Orleans, LA 70148, USA · Center for Geospatial Sciences, Naval Research Laboratory, Mississippi, USA · Ocean Sciences Division, Naval Research Laboratory, Mississippi, USA
RACE: Residual-Aware Test-Time Adaptation for Neighbor-Rich Time-Series Foundation Model Forecasting
Sep 30, 2026Hao-Nan Shi, Tong Wu, Chen-Cong Sun +3Residual Learning Time Series Forecasting
School of Artificial Intelligence, Nanjing University · National Key Laboratory for Novel Software Technology, Nanjing University · Nanjing, China
RATL: Learning from Retrieved Residuals for Robust Multivariate Time-Series Forecasting
Sep 3, 2026Yuchen He, Yueyang Cang, Zhiyuan Ning +2Multivariate Time Series Forecasting Residual Learning
Department of Automation, Tsinghua University · State Key Laboratory of Hydroscience and Engineering, Tsinghua University

Jun 4, 2026, cs.LGJ/K move · Enter open · S save

Learned Response-Field Inertia Operator for HEC-RAS 2D Water-Surface Elevation Prediction

Edward Holmberg, Elias Ioup, Md Meftahul Ferdaus, Mahdi Abdelguerfi+1

Canizaro Livingston Gulf States Center for Environmental Informatics, Department of Computer Science, The University of New Orleans, New Orleans, LA 70148, USA · Center for Geospatial Sciences, Naval Research Laboratory, Mississippi, USA · Ocean Sciences Division, Naval Research Laboratory, Mississippi, USA

Neural Surrogate Modeling PDE Surrogate Modeling Spatiotemporal Forecasting Computational Fluid Dynamics

Open paper AskPDF arXiv

This article presents a cross-dataset evaluation of learned native-cell surrogate models for solver-consistent water-surface elevation (WSE) prediction in HEC-RAS 2D. To avoid raster remapping error and information-access confounding, surrogates are evaluated directly on the original nonuniform computational cells under an explicit policy that separates static project inputs, current hydraulic state, project-input forcing, calibration-derived quantities, and future solver-output targets. We introduce the Learned Response-Field Inertia Operator (LRFIO), a no-forcing, increment-based learned surrogate that calibrates an inertial response operator from solved HEC-RAS trajectories and deploys the retained operator through closed-form native-cell rollout. LRFIO evaluates a base-case-first response hierarchy consisting of persistence, global calibrated inertia, and segmented response-field inertia. Segmentation, residual correction, and neuralized inertia are treated as learnable modeling choices, with added complexity retained only when validation evidence justifies its cost. Evaluated across four diverse HEC-RAS 2D benchmarks, LRFIO retains different response structures for different domains, demonstrating adaptive learned complexity. The selector audit shows controlled complexity with a maximum validation regret of 4.30%. During deployment, retained rollout times range from 0.003 s to 0.242 s, and the Beaver Bayou measured-solve comparison gives an estimated 2.75 x 10^4 horizon-normalized speedup over HEC-RAS. These results indicate that the current native-cell increment is a strong solver-conditioned predictive scaffold and that added response-field, neural, or spatial complexity should be retained only when empirically justified.

Abstract

Explore similar work

Learned Response-Field Inertia Operator for HEC-RAS 2D Water-Surface Elevation Prediction

RACE: Residual-Aware Test-Time Adaptation for Neighbor-Rich Time-Series Foundation Model Forecasting

RATL: Learning from Retrieved Residuals for Robust Multivariate Time-Series Forecasting