He et al. (2026) Event-Aware Validation of Short-Term Water-Level Forecasting in a Single Data-Scarce Gauged Catchment: Implications for Local Flash-Flood Risk Management

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

Identification

• Journal: Water
• Year: 2026
• Date: 2026-09-29
• Authors: Xubin He, Xinyan Jing, Zhonglin Yang, Zhiming Yan, Mengyuan You, Yizhou Yang, Rudong Huang, Weiwei Yu, Qinke Sun, Xinsong Chen, Jiayi Fang
• DOI: 10.3390/w18192424

Research Groups

• Department of Hydrology, University of [Unknown]
• [Other research groups involved in the study are not specified]

Short Summary

This paper evaluates the performance of seven models for short-term water-level forecasting at a single hydrological station, highlighting the importance of event-aware assessment and robustness evaluation.

Objective

• Investigate the effectiveness of different machine learning models for predicting water levels at Hongjiata Hydrological Station with overlapping forecasting windows.

Study Configuration

• Spatial Scale: Local (Hongjiata Hydrological Station)
• Temporal Scale: 2016–2025, with a focus on 2024 and 2025 events

Methodology and Data

• Models used:
  • Ridge
  • Convolutional Neural Network-Long Short-Term Memory (CNN-LSTM)
  • Other five models (not specified)
• Data sources: Telemetry elevations, event archives from 41 events at Hongjiata Hydrological Station

Main Results

• The best-performing model for 1 h forecasting was Ridge with an RMSE of 0.107 m and NSE of 0.908.
• CNN-LSTM achieved a similar performance (RMSE = 0.119 ± 0.012 m) but showed variability across configurations.
• Event-aware assessment revealed differences in performance between models, particularly for matched 1 h targets from six events.

Contributions

• This study provides an event-aware evaluation framework for local water-level forecasting and highlights the importance of robustness assessment in model selection.
• The results demonstrate that CNN-LSTM can achieve competitive performance with Ridge, but further investigation is needed to establish cross-catchment generalization and warning-detection skill.

Funding

• [Funding information not provided]

Citation

@article{He2026EventAware,
  author = {He, Xubin and Jing, Xinyan and Yang, Zhonglin and Yan, Zhiming and You, Mengyuan and Yang, Yizhou and Huang, Rudong and Yu, Weiwei and Sun, Qinke and Chen, Xinsong and Fang, Jiayi},
  title = {Event-Aware Validation of Short-Term Water-Level Forecasting in a Single Data-Scarce Gauged Catchment: Implications for Local Flash-Flood Risk Management},
  journal = {Water},
  year = {2026},
  doi = {10.3390/w18192424},
  url = {https://doi.org/10.3390/w18192424}
}