Kang et al. (2026) CaMa-Flood-GPU: a GPU-based hydrodynamic model implementation for scalable global simulations

Identification

• Journal: Geoscientific model development
• Year: 2026
• Date: 2026-06-29
• Authors: Shengyu Kang, Jiabo Yin, Dai Yamazaki
• DOI: 10.5194/gmd-19-5623-2026

Research Groups

• State Key Laboratory of Water Resources Engineering and Management, Wuhan University (China)
• Institute of Industrial Science, The University of Tokyo (Japan)

Short Summary

This study introduces CaMa-Flood-GPU, a GPU-based implementation of the Catchment-based Macro-scale Floodplain model (CaMa-Flood) optimized for Graphics Processing Unit (GPU) architectures. The new version achieves an order-of-magnitude speedup over a 192-core CPU baseline and near-linear scaling on multiple GPUs.

Objective

• To develop a scalable, physically consistent global flood simulation tool that can handle large-scale river routing models like CaMa-Flood at resolutions and runtimes previously impractical.
• To address the fundamental challenges in adapting GPU acceleration to global-scale river models, including irregular network topology, runoff interpolation, diagnosing water depth from storage, and communication overhead.

Study Configuration

• Spatial Scale: Global scale with kilometer-resolution (∼1 arcmin) resolution.
• Temporal Scale: Real-time or near-real-time simulations for flood hazard assessment and scenario analysis.

Methodology and Data

• Models used: CaMa-Flood-GPU, a GPU-based implementation of the Catchment-based Macro-scale Floodplain model (CaMa-Flood).
• Data sources: High-resolution DEM elevations, pre-computed sub-grid topographic profiles, and runoff inputs from various data sources.

Main Results

• The new version achieves an order-of-magnitude speedup over a 192-core CPU baseline.
• Near-linear scaling on multiple GPUs with negligible numerical differences from the original model.
• Efficient diagnosis of water depth and inundation extent from storage volume without resorting to computationally expensive 2D simulations.

Contributions

• This study provides a foundation for scalable, physically consistent global flood simulations at resolutions and runtimes previously impractical.
• The new version addresses fundamental challenges in adapting GPU acceleration to global-scale river models.
• The implementation offers flexible customization and extensibility for various scientific applications, data sources, and regional focuses.

Funding

• This research was funded by the State Key Laboratory of Water Resources Engineering and Management (China) and the Institute of Industrial Science (Japan).

Citation

@article{Kang2026CaMaFloodGPU,
  author = {Kang, Shengyu and Yin, Jiabo and Yamazaki, Dai},
  title = {CaMa-Flood-GPU: a GPU-based hydrodynamic model implementation for scalable global simulations},
  journal = {Geoscientific model development},
  year = {2026},
  doi = {10.5194/gmd-19-5623-2026},
  url = {https://doi.org/10.5194/gmd-19-5623-2026}
}