1. Introduction
Climate change is fundamentally altering the global hydrological cycle, leading to an increase in the frequency, severity, and unpredictability of extreme weather events such as prolonged droughts and catastrophic floods.
Managing water resources under these shifting conditions is an increasingly complex challenge. Water systems are inherently multi-sectoral, requiring a delicate balance between municipal supply, agricultural irrigation, energy generation (hydropower and cooling), and the preservation of natural ecosystems.
Traditionally, water resource management has relied on deterministic hydrological models and static, rule-based operation policies. However, conventional approaches struggle to cope with the deep uncertainties of climate change, non-linear system dynamics, and the rapid decision-making required during extreme events. Furthermore, managing trade-offs across different spatial scales (e.g., local catchments versus regional management areas) and temporal scales (e.g., immediate flood mitigation versus long-term drought resilience) remains a critical bottleneck.
This project aims to transform water resource management by developing an integrated, AI-driven framework that combines advanced predictive modeling with automated, multi-objective decision-making.
The project will explore how recent AI techniques, such as Spatio-Temporal Deep Learning, Multi-Agent Reinforcement Learning (MARL), and Graph Neural Networks (GNNs), can forecast extreme events and optimize water allocation across competing sectors in real time.
Research Challenges and Questions
The student will address several fundamental research challenges at the intersection of environmental science and computer science:
How can deep learning architectures effectively ingest multi-scale climate data to provide high-resolution, reliable forecasts of extreme hydrological events?
How can AI navigate conflicting socio-economic and ecological demands during extreme event scenarios, ensuring equitable water distribution while preventing ecological impacts?
How can graph or hierarchical AI architectures bridge the gap between macro-level climate variability and localized, catchment-specific infrastructure constraints?
How can we design AI decision-making tools that are transparent and explainable so that water managers and stakeholders can confidently adopt their recommendations?
References:
References
Nearing, G., Cohen, D., Dube, V. et al. Global prediction of extreme floods in ungauged watersheds. Nature 627, 559–563 (2024). https://doi.org/10.1038/s41586-024-07145-1
Sit M, Demiray BZ, Xiang Z, Ewing GJ, Sermet Y, Demir I. A comprehensive review of deep learning applications in hydrology and water resources. Water Sci Technol. 2020 Dec; 82(12):2635-2670. doi: 10.2166/wst.2020.369.
Slater et al., 2025: Challenges and opportunities of ML and explainable AI in large-sample hydrology. Philos Trans A Math Phys Eng Sci 31; 383 (2302): 20240287. https://doi.org/10.1098/rsta.2024.0287