Guo et al. (2026) Data-Driven Downstream Discharge Forecasting for Flood Disaster Mitigation in a Small Mountainous Basin of Southwest China

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

Identification

• Journal: Water
• Year: 2026
• Date: 2026-01-13
• Authors: Leilei Guo, Ming Luo, Rongwen Yao, Qiang Li, Yangshuang Wang, Renjuan Wei, Xianchun Ma
• DOI: 10.3390/w18020204

Research Groups

Not explicitly stated in the provided text.

Short Summary

This study benchmarks data-driven models for short-lead river discharge forecasting in the Fuhu Stream, China, finding that the LSTM model significantly outperforms SARIMAX and XGBoost in accurately predicting both baseflow and flood peaks.

Objective

• To predict downstream flow in the Fuhu Stream, a small mountainous basin, using high-resolution upstream data.
• To benchmark the performance of SARIMAX, XGBoost, and LSTM data-driven models for short-lead river discharge forecasting.

Study Configuration

• Spatial Scale: Fuhu Stream in Emeishan City, China; small mountainous basin.
• Temporal Scale: High-resolution 5-minute time series; dataset spanning from 7 June 2024 to 25 October 2024; short-lead forecasting.

Methodology and Data

• Models used: SARIMAX (Seasonal Autoregressive Integrated Moving Average with Exogenous Regressors), XGBoost (eXtreme Gradient Boosting), LSTM (Long Short-Term Memory).
• Data sources: High-resolution 5-minute time series of upstream precipitation, stage, and discharge observations.

Main Results

• The LSTM model significantly outperformed XGBoost and SARIMAX during the test period (October 2024).
• LSTM achieved the highest coefficient of determination (R² = 0.994) and the lowest prediction errors (RMSE = 0.016 m³/s, MAE = 0.011 m³/s).
• LSTM accurately reproduced both baseflow conditions and multiple flood peaks, while XGBoost and SARIMAX failed to capture low-flow variability and flood peaks.
• XGBoost showed moderate performance, and SARIMAX exhibited a tendency toward mean reversion.
• Feature importance analysis indicated that WSQ/LY sites were critical monitoring nodes for reliable predictions.

Contributions

• Highlights the superior advantage of sequence-learning architectures (LSTM) in modeling nonlinear hydrological propagation and memory effects for short-term discharge dynamics in small mountainous basins.
• Provides a robust data-driven solution for operationalizing early warning systems and supporting decision-making for downstream flood disaster prevention.

Funding

Not explicitly stated in the provided text.

Citation

@article{Guo2026DataDriven,
  author = {Guo, Leilei and Luo, Ming and Yao, Rongwen and Li, Qiang and Wang, Yangshuang and Wei, Renjuan and Ma, Xianchun},
  title = {Data-Driven Downstream Discharge Forecasting for Flood Disaster Mitigation in a Small Mountainous Basin of Southwest China},
  journal = {Water},
  year = {2026},
  doi = {10.3390/w18020204},
  url = {https://doi.org/10.3390/w18020204}
}