PhD Candidate: Bekir Demiray
Abstract
Accurate streamflow prediction is crucial for flood forecasting, water management, and climate resilience. While deep learning has advanced hydrological modeling, most research remains focused on daily predictions and region-specific models, limiting generalizability. Additionally, reliance on low-resolution datasets affects forecasting accuracy. This dissertation addresses these challenges through data enhancement, streamflow prediction, and multi-task learning. It explores deep learning-based super-resolution techniques to improve spatial and temporal resolutions, enhancing environmental datasets for hydrological applications. It also develops high-resolution streamflow forecasting models, tackling sub-daily prediction, generalization, and interpretability. Finally, it investigates multi-task learning (MTL) in hydrology, integrating multiple hydrological predictions within a unified framework. By combining data enhancement, high-resolution forecasting, and MTL, this research advances AI-driven hydrology, providing improved models for flood mitigation, water management, and climate adaptation.
Advisor: Ibrahim Demir