Mu et al. (2026) Prior-Informed Directed-Lag Graph Neural Residual Learning for Multi-Step Streamflow Forecasting

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

Identification

• Journal: Hydrology
• Year: 2026
• Date: 2026-09-29
• Authors: Liang Mu, Zhiguo Yu, Hongmin Zhang, Junxi Chen
• DOI: 10.3390/hydrology13100267

Research Groups

• Department of Hydrology and Water Resources Engineering, University of Natural Resources and Life Sciences (BOKU)
• GloFAS Consortium

Short Summary

This study proposes a novel graph neural network model, PI-DLGNR, to improve multi-step streamflow forecasting by incorporating prior knowledge and learning delayed dependence between gauges. The model achieves state-of-the-art performance on the GloFAS reanalysis series in the Yangtze River Basin.

Objective

• Develop an accurate multi-step streamflow forecasting model that preserves antecedent streamflow memory and accounts for delayed dependence between gauges.

Study Configuration

• Spatial Scale: Catchment scale, focusing on the Yangtze River Basin.
• Temporal Scale: Daily to weekly time steps, with a focus on short-term forecasting (up to 7 days ahead).

Methodology and Data

• Models used: Graph neural network (GNN) framework, specifically PI-DLGNR, which combines VAR-Ridge backbone with directed-lag graph residual branch.
• Data sources: GloFAS reanalysis series for the Yangtze River Basin.

Main Results

• PI-DLGNR achieves NSE values of 0.998, 0.975, and 0.900 at Steps 1, 3, and 7, respectively, outperforming VAR-Ridge backbone.
• Relative to VAR-Ridge, MAE decreases by 1.47%, 0.84%, and 0.72% at Steps 1, 3, and 7, while RMSE decreases by 0.34%, 0.03%, and 0.01%.

Contributions

• The study contributes a novel graph neural network model that effectively incorporates prior knowledge and learns delayed dependence between gauges for multi-step streamflow forecasting.
• PI-DLGNR outperforms existing models on the GloFAS reanalysis series in the Yangtze River Basin.

Funding

• This research was supported by the European Union's Horizon 2020 research and innovation program under grant agreement No. 776978 (GloFAS).

Citation

@article{Mu2026PriorInformed,
  author = {Mu, Liang and Yu, Zhiguo and Zhang, Hongmin and Chen, Junxi},
  title = {Prior-Informed Directed-Lag Graph Neural Residual Learning for Multi-Step Streamflow Forecasting},
  journal = {Hydrology},
  year = {2026},
  doi = {10.3390/hydrology13100267},
  url = {https://doi.org/10.3390/hydrology13100267}
}