Shen et al. (2025) Physically Consistent Runoff Simulation in Mountainous Catchments Using a Time-Varying Gated Hybrid XAJ–LSTM Model

⚠️ Warning: This summary was generated from the abstract only, as the full text was not available.

Identification

• Journal: Water
• Year: 2025
• Date: 2025-12-11
• Authors: Hayley H. Shen, Linyao Dong, Wenjian Tang, Yujie Zeng
• DOI: 10.3390/w17243507

Research Groups

Not specified in the provided text.

Short Summary

This study developed a time-varying gated hybrid model (XAJ–LSTM) integrating the Xinanjiang (XAJ) model with a Long Short-Term Memory (LSTM) network to improve rainfall-runoff prediction accuracy and physical consistency in mountainous catchments, demonstrating superior performance over individual models.

Objective

• To develop and evaluate a time-varying gated hybrid model (XAJ–LSTM) that integrates a physically based model (Xinanjiang) with a deep learning model (LSTM) to enhance runoff prediction accuracy and physical consistency in mountainous catchments.

Study Configuration

• Spatial Scale: 17 small to medium mountainous catchments in Shi Yan and En Shi, Hubei Province, China.
• Temporal Scale: Hourly data from 2015 to 2023 (9 years).

Methodology and Data

• Models used: Xinanjiang (XAJ) model, Long Short-Term Memory (LSTM) network, and the proposed time-varying gated hybrid XAJ–LSTM model.
• Data sources: Hourly rainfall, temperature, potential evapotranspiration, and runoff data from observations.

Main Results

• The XAJ–LSTM hybrid model achieved superior performance with mean Nash-Sutcliffe Efficiency (NSE) of 0.971 ± 0.020 and Kling-Gupta Efficiency (KGE) of 0.962 ± 0.024, outperforming both individual XAJ and LSTM models.
• The time-varying gating parameter λ(t) showed a negative correlation with discharge in approximately 80% of the catchments, indicating adaptive adjustment between the physical and data-driven components.
• The coupled model accurately reproduced both high-flow and low-flow processes, with deviations in flow duration curves generally within ±5%.

Contributions

• Introduction of a novel time-varying gated hybrid modeling framework (XAJ–LSTM) that effectively integrates physically based and data-driven approaches for rainfall-runoff simulation.
• Demonstration of improved runoff prediction accuracy, stability, and physical interpretability in mountainous catchments compared to standalone physically based or deep learning models.
• Provision of insights into the adaptive interaction between physical and data-driven components through the dynamic gating mechanism.

Funding

Not specified in the provided text.

Citation

@article{Shen2025Physically,
  author = {Shen, Hayley H. and Dong, Linyao and Tang, Wenjian and Zeng, Yujie},
  title = {Physically Consistent Runoff Simulation in Mountainous Catchments Using a Time-Varying Gated Hybrid XAJ–LSTM Model},
  journal = {Water},
  year = {2025},
  doi = {10.3390/w17243507},
  url = {https://doi.org/10.3390/w17243507}
}