Ma et al. (2026) A spatiotemporally differentiated hybrid hydrological modeling strategy with dynamically adaptive runoff generation modes

Identification

• Journal: Journal of Hydrology
• Year: 2026
• Date: 2026-04-10
• Authors: Xiaozan Ma, Yufei Ma, Qin Ju, Cuishan Liu, Junliang Jin, Yao Xiao, Haowen Liu, Guoqing Wang
• DOI: 10.1016/j.jhydrol.2026.135480

Research Groups

• National Key Laboratory of Water Disaster Prevention, Nanjing Hydraulic Research Institute, Nanjing, China
• College of Hydrology and Water Resources, Hohai University, Nanjing, China
• Yangtze River Conservation and Green Development Research Institute, Nanjing, China
• Research Center for Climate Change, Ministry of Water Resources, Nanjing, China
• Academy of Environmental Planning and Design, Co., Ltd. Nanjing University, Nanjing, China
• Key Laboratory of Fluid and Power Machinery (Xihua University), Ministry of Education, Chengdu, China

Short Summary

This study develops a spatiotemporally differentiated hybrid hydrological model (STHM) that dynamically adapts runoff generation modes using machine learning to improve flood forecasting in small-to-medium basins. The STHM demonstrates superior and more stable simulation performance compared to conventional models, effectively capturing event-dependent runoff processes.

Objective

• To develop a spatiotemporally differentiated hybrid hydrological model (STHM) that dynamically switches among infiltration-excess, saturation-excess, and vertical mixed runoff modes to accurately represent event-dependent runoff generation for improved flood forecasting in small-to-medium hilly basins.

Study Configuration

• Spatial Scale: Small-to-medium hilly basins; case study in the Tunxi Basin.
• Temporal Scale: Event-dependent runoff generation for flood forecasting; calibration and validation periods for model performance assessment.

Methodology and Data

• Models used:
  • Spatiotemporally differentiated hybrid hydrological model (STHM) integrating machine learning (Random Forest classification) with classical runoff generation formulations (infiltration-excess, saturation-excess, vertical mixed runoff).
  • Four conventional hydrological models (benchmarks).
  • SHapley Additive exPlanations (SHAP) for interpretability analysis.
• Data sources: Not explicitly listed, but implied inputs include rainfall characteristics, antecedent wetness, and soil moisture conditions.

Main Results

• The Random Forest classification model achieved a balanced accuracy of 86.65% on the test set, demonstrating high accuracy and robustness with consistent recall across the three runoff generation modes.
• The STHM yielded the best and most stable simulation performance among all models, with Nash-Sutcliffe Efficiency (NSE) values of 0.855 during calibration and 0.850 during validation.
• The STHM exhibited strong adaptability across various flood event types.
• SHAP analysis confirmed distinct responses of the three runoff generation modes to rainfall characteristics and soil moisture conditions, supporting the need for dynamic strategies.

Contributions

• Proposes a novel spatiotemporally differentiated hybrid hydrological model (STHM) that dynamically adapts runoff generation modes based on event conditions, addressing limitations of fixed schemes in existing models.
• Integrates machine learning with classical hydrological theory to provide a physically meaningful and effective framework for flood forecasting.
• Achieves superior and more stable flood simulation performance in small-to-medium basins, enhancing accuracy and robustness.
• Provides insights into the necessity of differentiated and dynamic runoff generation strategies through interpretable machine learning analysis.

Funding

• No funding information is provided in the given paper text.

Citation

@article{Ma2026spatiotemporally,
  author = {Ma, Xiaozan and Ma, Yufei and Ju, Qin and Liu, Cuishan and Jin, Junliang and Xiao, Yao and Liu, Haowen and Wang, Guoqing},
  title = {A spatiotemporally differentiated hybrid hydrological modeling strategy with dynamically adaptive runoff generation modes},
  journal = {Journal of Hydrology},
  year = {2026},
  doi = {10.1016/j.jhydrol.2026.135480},
  url = {https://doi.org/10.1016/j.jhydrol.2026.135480}
}