Gao et al. (2026) A comparative study of deep learning and process-based models for hourly flood simulation under a unified input and optimization framework

Identification

• Journal: Environmental Modelling & Software
• Year: 2026
• Date: 2026-07-16
• Authors: Shuai Gao, Zhenxu Chen, Xueying Xie, W W Huang, W CHEN, Wuxi Chen, Y Huang
• DOI: 10.1016/j.envsoft.2026.107101

Research Groups

College of Civil Engineering, Fuzhou University, China

Short Summary

This study presents a unified input and optimization framework to compare the performance of Process-Based Models (PBMs) and Deep Learning Models (DLMs) in hourly flood simulation.

Objective

• Investigate the relative accuracy of PBMs versus DLMs in hourly flood simulation under a standardized input and optimization framework.

Study Configuration

• Spatial Scale: Seven Chinese catchments.
• Temporal Scale: Hourly time step.

Methodology and Data

• Models used:
  • Gated Recurrent Unit (GRU)
  • Temporal Convolutional Network (TCN) for DLMs
  • HBV model
  • Tank model for PBMs
• Data sources: Historical hydro-meteorological data from the seven Chinese catchments.

Main Results

• The GRU model achieved superior simulation accuracy when fed with complete flood process as input compared to using sliding-window samples.
• Adam is an effective method for parameter calibration within PBMs.
• Under unified input and optimization conditions, DLMs generally achieved superior simulation performance to PBMs across most catchments.

Contributions

This study provides a systematic evaluation framework that standardizes the input architecture and optimization algorithms for PBMs and DLMs in hourly flood simulation. The results highlight the potential of DLMs in achieving higher accuracy compared to traditional PBMs.

Funding

• This research was funded by [project/program name] with reference code [reference code].

Citation

@article{Gao2026comparative,
  author = {Gao, Shuai and Chen, Zhenxu and Xie, Xueying and Huang, W W and CHEN, W and Chen, Wuxi and Huang, Y},
  title = {A comparative study of deep learning and process-based models for hourly flood simulation under a unified input and optimization framework},
  journal = {Environmental Modelling & Software},
  year = {2026},
  doi = {10.1016/j.envsoft.2026.107101},
  url = {https://doi.org/10.1016/j.envsoft.2026.107101}
}