Akkala et al. (2025) Improved Streamflow Forecasting Through SWE-Augmented Spatio-Temporal Graph Neural Networks

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

Identification

• Journal: Hydrology
• Year: 2025
• Date: 2025-10-11
• Authors: Akhila Akkala, Soukaïna Filali Boubrahimi, Shah Muhammad Hamdi, Pouya Hosseinzadeh, Ayman Nassar
• DOI: 10.3390/hydrology12100268

Research Groups

Not specified in the provided text.

Short Summary

This study comparatively evaluates statistical, machine learning, and deep learning models for streamflow forecasting in snowmelt-dominated basins, demonstrating that Spatio-Temporal Graph Neural Networks (STGNNs) with integrated Snow Water Equivalent (SWE) data achieve superior accuracy and provide a scalable forecasting approach.

Objective

• To comparatively evaluate statistical, machine learning (Random Forest), and deep learning models (LSTM, GRU, STGNN) for streamflow forecasting in the Upper Colorado River Basin (UCRB).
• To assess the impact of integrating meteorological variables, particularly Snow Water Equivalent (SWE), and spatial dependencies on predictive performance.

Study Configuration

• Spatial Scale: Upper Colorado River Basin (UCRB), utilizing data from 20 monitoring stations, with a critical downstream node at Lees Ferry.
• Temporal Scale: 30 years of historical data.

Methodology and Data

• Models used: Statistical models, Random Forest (RF), Long Short-Term Memory (LSTM), Gated Recurrent Unit (GRU), Spatio-Temporal Graph Neural Network (STGNN).
• Data sources: 30 years of observational data from 20 monitoring stations, including meteorological variables and Snow Water Equivalent (SWE).

Main Results

• The Spatio-Temporal Graph Neural Network (STGNN) achieved the highest accuracy among all models, with a Nash–Sutcliffe Efficiency (NSE) of 0.84 and Kling–Gupta Efficiency (KGE) of 0.84 in the multivariate setting at the Lees Ferry node.
• Integrating Snow Water Equivalent (SWE) into predictions reduced the Root Mean Square Error (RMSE) by 12.8% compared to univariate setups.
• Seasonal and spatial analyses indicated that the greatest improvements in forecasting performance occurred at high-elevation and mid-network stations, where snowmelt dynamics significantly influence runoff.

Contributions

• This study provides a comprehensive comparative evaluation of various advanced models for streamflow forecasting in snowmelt-dominated basins.
• It demonstrates the superior performance of spatio-temporal learning frameworks, particularly STGNNs, for streamflow forecasting.
• It highlights the critical importance of integrating Snow Water Equivalent (SWE) and leveraging spatial dependencies for enhanced predictive accuracy.
• The findings establish STGNNs as a scalable and physically consistent approach for streamflow forecasting under variable climatic conditions.

Funding

Not specified in the provided text.

Citation

@article{Akkala2025Improved,
  author = {Akkala, Akhila and Boubrahimi, Soukaïna Filali and Hamdi, Shah Muhammad and Hosseinzadeh, Pouya and Nassar, Ayman},
  title = {Improved Streamflow Forecasting Through SWE-Augmented Spatio-Temporal Graph Neural Networks},
  journal = {Hydrology},
  year = {2025},
  doi = {10.3390/hydrology12100268},
  url = {https://doi.org/10.3390/hydrology12100268}
}