Korzani et al. (2025) Dynamic CA-ffé: A hybrid 1D/2D fast flood evaluation model for urban flash floods

Identification

• Journal: Journal of Hydrology
• Year: 2025
• Date: 2025-12-09
• Authors: Maziar Gholami Korzani, A. K. Chaudhary, Behzad Jamali, Ana Deletić
• DOI: 10.1016/j.jhydrol.2025.134762

Research Groups

• School of Civil and Environmental Engineering, Queensland University of Technology, Brisbane, Australia
• WMAwater, Sydney, Australia

Short Summary

This paper introduces Dynamic CA-ff´e, a novel hybrid 1D/2D urban flash flood model coupling a static 2D cellular automata overland flow model with a dynamic 1D drainage network model. It demonstrates high accuracy in flood prediction comparable to full hydrodynamic models, while achieving significantly faster simulation times (at least 20 times faster).

Objective

• To evaluate how well the Dynamic CA-ff´e model reproduces flood extent and depth compared to established hydrodynamic models, despite omitting the momentum equation.
• To determine the extent to which the coupling of a static conceptual overland flow and dynamic hydrodynamic drainage model improves the accuracy of a conceptual pluvial flood model.
• To assess the computational advantages and trade-offs of the Dynamic CA-ff´e model, and how key parameters affect its performance.

Study Configuration

• Spatial Scale: Two urban catchments in Australia: Scotchmans Creek (approximately 3.0 km²) and Coogee Bay (2.9 km²).
• Temporal Scale: Rainfall events ranging from 90 minutes to 4 hours 30 minutes duration. Simulation periods up to 7.5 hours. Coupling intervals for model interaction varied from 5 minutes to 20 minutes.

Methodology and Data

• Models used:
  • Dynamic CA-ff´e: A novel hybrid 1D/2D model.
    • 1D component: EPA SWMM (StormWater Management Model) for drainage network, solving full Saint-Venant equations (dynamic wave routing).
    • 2D component: CA-ff´e (Cellular Automata Fast Flood Evaluation) for overland flow, a static DEM-based model based on mass conservation (continuity equation), neglecting the momentum equation.
  • Comparison models: MIKE Flood and TUFLOW (established hydrodynamic models solving full Shallow Water Equations).
• Data sources:
  • Digital Elevation Models (DEMs) for both catchments.
  • Design rainfall events and historical rainfall data.
  • Lot data for sub-catchment delineation (Scotchmans Creek).
  • Reference inundation maps from MIKE Flood and TUFLOW models (treated as ground truth for validation).

Main Results

• Dynamic CA-ff´e achieved high accuracy in predicting flood extent and depth, comparable to established hydrodynamic models (MIKE Flood and TUFLOW).
• Simulation times were significantly reduced, with Dynamic CA-ff´e running at least 20 times faster than hydrodynamic models (e.g., over 60 times faster than MIKE Flood for Scotchmans Creek, and approximately 20 times faster than TUFLOW for Coogee Bay).
• For Scotchmans Creek (SWMM rainfall-runoff hydrology), the model showed an RMSE of approximately 10%, a hit rate (HR) over 80%, a false alarm rate (FAR) below 15%, and a Nash-Sutcliffe efficiency (NSE) between 0.65 and 0.75.
• For Coogee Bay (direct rainfall hydrology), the model exhibited an RMSE of approximately 10%, HR and Critical Success Index (CSI) near 1, and an NSE around 0.8, indicating higher accuracy than the Scotchmans Creek case.
• The Increment Constant (IC) parameter showed independence for values below 0.01 meter, suggesting it's not a calibration parameter but rather related to model resolution.
• The hf const parameter had minimal impact on accuracy within a range of 0.05 to 0.15.
• The coupling interval showed limited sensitivity on accuracy, with optimal ranges identified (5-9 minutes for Scotchmans Creek, up to 15 minutes for Coogee Bay).
• The 1D dynamic drainage model effectively imposed dynamic behavior (momentum equation) on the 2D static overland flow model, resulting in a quasi-dynamic hybrid simulation.
• A static one-directional coupled model (without drainage feedback) significantly overestimated surface water accumulation and showed lower accuracy compared to the dynamic model.

Contributions

• Development of Dynamic CA-ff´e, a novel hybrid 1D/2D urban flash flood model that uniquely combines a static 2D conceptual overland flow model with a dynamic 1D hydrodynamic drainage network model.
• Achieves a balance between computational efficiency and predictive accuracy, providing results comparable to full hydrodynamic models at significantly faster speeds.
• Introduction of a robust bi-directional coupling module that facilitates dynamic flow exchange between fundamentally different static 2D and dynamic 1D models.
• Implementation of an innovative parallelisation technique for the 2D CA-ff´e component, enhancing scalability for large catchments.
• Demonstration that the momentum equation from the 1D drainage model effectively imposes quasi-dynamic behavior on the 2D static overland flow, leading to a more realistic flood simulation.

Funding

• Australian Government through the Australian Research Council (Discovery Project number DP210103704).

Citation

@article{Korzani2025Dynamic,
  author = {Korzani, Maziar Gholami and Chaudhary, A. K. and Jamali, Behzad and Deletić, Ana},
  title = {Dynamic CA-ffé: A hybrid 1D/2D fast flood evaluation model for urban flash floods},
  journal = {Journal of Hydrology},
  year = {2025},
  doi = {10.1016/j.jhydrol.2025.134762},
  url = {https://doi.org/10.1016/j.jhydrol.2025.134762}
}