Lei et al. (2025) Synergizing machine learning and modified physical models for hydrology modeling: A case study of modified SIMHYD and TANK models

Identification

• Journal: Journal of Hydrology
• Year: 2025
• Date: 2025-10-24
• Authors: Xuxin Lei, Lei Cheng, Lu Zhang, Pan Liu
• DOI: 10.1016/j.jhydrol.2025.134452

Research Groups

• State Key Laboratory of Water Resources Engineering and Management, Wuhan University, Wuhan, China
• Department of Hydrology and Water Resources, School of Water Resources and Hydropower Engineering, Wuhan University, Wuhan, China

Short Summary

This study investigates the effectiveness of hybrid hydrological models (HMs) that combine machine learning with original and modified physical models (SIMHYD, TANK) across 569 catchments in the United States. It finds that HMs with modified physical layers offer superior runoff predictability and improved reasoning ability for evaporation and baseflow compared to those with original physical models.

Objective

• To investigate whether hybrid hydrological models (HMs) incorporating modified process-based physical models (PMs) can improve both runoff predictability and reasoning ability (for evaporation and baseflow) compared to HMs using original PMs.

Study Configuration

• Spatial Scale: 569 catchments in the United States.
• Temporal Scale: Not explicitly stated for the specific data used in the study.

Methodology and Data

• Models used: Hybrid Models (HMs) combining machine learning with process-based physical models (PMs); specifically, original and modified versions of SIMHYD and TANK models.
• Data sources: Observational hydrological data from 569 catchments.

Main Results

• Hybrid models (HMs) generally exhibit superior runoff (Q) predictability but inferior reasoning ability (represented by evaporation E and baseflow Qb) compared to standalone physical models.
• HMs incorporating modified physical models demonstrate better predictability and reasoning ability than those wrapping the original physical models.
• For predicted runoff (Q), HMs with modified physical guidance achieved a Nash-Sutcliffe Efficiency (NSE) of 0.58 ± 0.04, outperforming original-based HMs (NSE of 0.56 ± 0.04).
• For simulated evaporation (E) and baseflow (Qb), HMs with modified physical guidance showed increased performance, with mean ΔR2 values of 0.10 and 0.21, respectively.

Contributions

• This study directly demonstrates that improving the physical layer within hybrid hydrological models (HMs) significantly enhances both predictive accuracy and the physical consistency (reasoning ability) of the models.
• It emphasizes that the future trajectory of hydrological synergy depends not only on advancements in data mining techniques but also on continued developments in physical models.

Funding

• Not explicitly stated in the provided text.

Citation

@article{Lei2025Synergizing,
  author = {Lei, Xuxin and Cheng, Lei and Zhang, Lu and Liu, Pan},
  title = {Synergizing machine learning and modified physical models for hydrology modeling: A case study of modified SIMHYD and TANK models},
  journal = {Journal of Hydrology},
  year = {2025},
  doi = {10.1016/j.jhydrol.2025.134452},
  url = {https://doi.org/10.1016/j.jhydrol.2025.134452}
}

Generated by BiblioAssistant using gemini-2.5-flash (Google API)