Streamflow Prediction Using Complex Networks.

Area of Science:

• Hydrology and Water Resources Engineering
• Complex Systems Analysis
• Time Series Forecasting

Background:

• Reliable streamflow prediction is essential for water resource management, environmental monitoring, and ecosystem health.
• Existing streamflow prediction methods often face challenges in capturing complex temporal dynamics.
• Complex networks offer a novel framework for analyzing and modeling time series data.

Purpose of the Study:

• To develop and evaluate a complex networks-based approach for daily streamflow prediction.
• To assess the prediction accuracy using established statistical metrics.
• To investigate the relationship between prediction accuracy and catchment characteristics.

Main Methods:

• Streamflow time series were transformed into complex networks where timesteps are nodes and streamflow differences define links.
• A critical distance threshold was identified to form the network structure.
• Clustering coefficient (CC) was calculated and used in a nearest neighbor search for prediction.

Main Results:

• The complex networks approach was applied to 142 stations in the contiguous United States, yielding prediction accuracies with correlation coefficients (R) from 0.05 to 0.99.
• Normalized root mean square error (NRMSE) ranged from 0.1 to 12.3, and Nash-Sutcliffe efficiency (NSE) ranged from -0.89 to 0.99.
• Prediction accuracy showed a stronger relationship with the coefficient of variation of flow than with drainage area or mean flow.

Conclusions:

• The proposed complex networks-based method provides a promising tool for accurate streamflow prediction.
• The approach demonstrates effective application across diverse hydrological conditions in the United States.
• Further research can extend this methodology to other hydrologic time series forecasting applications.