Enhancing runoff predictions in data-sparse regions through hybrid deep learning and hydrologic modeling

Area of Science:

• Hydrology
• Climate Science
• Data Science

Background:

• Climate change intensifies extreme weather, necessitating accurate flood forecasting.
• Data scarcity in regions like the Chaersen Basin challenges traditional hydrologic models.
• Existing models struggle with insufficient observational data for reliable predictions.

Purpose of the Study:

• To develop a hybrid model combining deep learning (Informer) and physical hydrological simulation (WRF-Hydro) for improved runoff prediction.
• To address flood forecasting challenges in data-sparse environments.
• To leverage transfer learning to bridge data gaps in hydrological modeling.

Main Methods:

• A hybrid model integrating the Informer deep learning model with the WRF-Hydro hydrological model was developed.
• The Informer model was initially trained on the CAMELS dataset and applied to the Chaersen Basin using transfer learning.
• WRF-Hydro was integrated with Global Forecast System (GFS) data for comparative analysis and refinement.

Main Results:

• The hybrid model demonstrated significant improvements in runoff prediction accuracy compared to individual models.
• Nash-Sutcliffe Efficiency (NSE) and Index of Agreement (IOA) metrics showed substantial increases.
• Optimal performance was achieved when the Informer model contributed 60%-80% to the hybrid model.

Conclusions:

• The hybrid Informer-WRF-Hydro model effectively enhances flood forecasting precision in data-sparse regions.
• The integration of deep learning pattern recognition with physical modeling provides a robust solution for complex hydrological dynamics.
• This approach offers a promising strategy for improving flood prediction accuracy in challenging environments.