Institute of Hydrology
Topic
Data-Driven Urban Flood Prediction Using Hydrologic Modeling
PhD. program
Water Resources Engineering
Year of admission
2025
Name of the supervisor
MSc. Saeid Okhravi, PhD.
Contact:
Receiving school
Faculty of Civil Engineering STU
Annotation
Climate change is intensifying rainfall patterns, increasing urban flood risks exacerbated by urbanization and aging drainage systems. In 2021, floods caused global economic losses of $82 billion, highlighting the need for improved mitigation strategies. As traditional structural measures prove insufficient, non-structural methods such as flood forecasting and early warning systems are gaining prominence.
Flood forecasting typically relies on physically-based numerical models to produce high-resolution inundation maps. While accurate, these models are computationally expensive, data-intensive, and unsuitable for real-time applications. In contrast, data-driven approaches like Artificial Neural Networks (ANNs) offer computational efficiency and require only input/output data for training. When trained on pre-simulated hydrodynamic scenarios, ANNs can predict inundation maps with resolutions comparable to numerical models, offering a scalable alternative for flood prediction.
The proposed research, aims to establish a physics-informed machine learning (PIML) framework to address these challenges. By integrating hydrologic and hydraulic models with AI and ML algorithms, the framework will utilize a synthetic database generated from physically-based simulations and real-world field data. The PIML model will incorporate governing equations to enhance prediction accuracy while significantly reducing computational costs.
The ultimate goal is to replace traditional numerical flood models with a fast, robust, and scalable system capable of real-time inundation mapping for settlement planning and early warning systems, thereby mitigating socio-economic impacts of urban floods.
Flood forecasting typically relies on physically-based numerical models to produce high-resolution inundation maps. While accurate, these models are computationally expensive, data-intensive, and unsuitable for real-time applications. In contrast, data-driven approaches like Artificial Neural Networks (ANNs) offer computational efficiency and require only input/output data for training. When trained on pre-simulated hydrodynamic scenarios, ANNs can predict inundation maps with resolutions comparable to numerical models, offering a scalable alternative for flood prediction.
The proposed research, aims to establish a physics-informed machine learning (PIML) framework to address these challenges. By integrating hydrologic and hydraulic models with AI and ML algorithms, the framework will utilize a synthetic database generated from physically-based simulations and real-world field data. The PIML model will incorporate governing equations to enhance prediction accuracy while significantly reducing computational costs.
The ultimate goal is to replace traditional numerical flood models with a fast, robust, and scalable system capable of real-time inundation mapping for settlement planning and early warning systems, thereby mitigating socio-economic impacts of urban floods.