Enhanced streamflow forecasting using hybrid modelling integrating glacio-hydrological outputs, deep learning and...

Area of Science:

• Hydrology
• Glaciology
• Artificial Intelligence
• Water Resource Management

Background:

• Accurate snow and ice melt dynamics are crucial for managing water resources and assessing flood risks in high-mountain river basins.
• Inaccessible terrain limits direct measurements, necessitating advanced modeling techniques.

Purpose of the Study:

• To develop and evaluate an AI-enabled hybrid approach integrating glacio-hydrological model outputs with machine learning and deep learning for enhanced runoff forecasting.
• To compare the predictive performance of different AI models and hybrid configurations against traditional methods.

Main Methods:

• Utilized a hybrid approach combining a glacio-hydrological model (GSM-SOCONT) with various machine learning and deep learning techniques (CNN-LSTM).
• Trained standalone deep learning models and hybrid models using meteorological data and glacio-hydrological outputs representing snow and ice melt.
• Applied multi-scale analysis with wavelet transformation and feature permutations to optimize the hybrid model.

Main Results:

• The standalone deep learning model (CNN-LSTM) outperformed traditional models using only meteorological data.
• A hybrid model (CNN-LSTM14) using glacier-derived features achieved high performance metrics (NSE=0.83, KGE=0.88, R=0.91) during validation.
• The final optimized hybrid model (CNN-LSTM19) significantly improved predictive accuracy, especially for high-flow events (NSE=0.97, RMSE=442).

Conclusions:

• AI-enhanced hydrological models offer superior accuracy in runoff forecasting compared to traditional methods, even with limited direct measurements.
• The developed hybrid approach provides reliable and actionable insights for effective water resource management and flood risk mitigation in data-scarce mountain regions.