Title of the project
Development of Machine Learning Applications for Real-Time River Flood Forecasting
Type
PhD research
Duration
2022-2026
Persons involved
ir. Leon S. Besseling (PhD candidate)
dr. ir. A. Bomers (Daily supervisor)
dr. J.J. Warmink (Co-supervisor)
prof. dr. S.J.M.H. Hulscher (Promotor)
Funding of the project
NWO – Simon Stevin Meester grant of S.J.M.H. Hulscher
Summary of the research
In river flood modelling, the most common models use a 1D-2D coupling of the river and hinterland respectively. These are accurate, but time-consuming. In an emergency situation, running such a model several times to determine evacuation strategies is impossible. To assess flood risk and aid decision making, a real-time flood forecasting system is desired, in which ensemble model predictions and uncertainty analysis allow for reviewing multiple scenarios and making optimal decisions.
In this research, the field of machine learning will be applied to real-time flood risk management in rivers. Machine learning techniques are data-driven and do not contain representations of physical descriptions from the original complex models, so they promise fast computation times. This research focuses specifically on dike breach events, since breaches are often unexpected and have large consequences. The goal is to create a fast machine learning model that can predict flood inundation in a hinterland regardless of the breach location and of the outflow hydrograph. This would allow decision makers to evaluate multiple scenarios during an emergency in real-time, and make appropriate decisions.
Keywords
Machine learning, hydrodynamic modelling, dike failure, flooding and inundation
More information
Leon Besseling
Room: Horst-Ring W209
E-mail: l.s.besseling@utwente.nl