Graduation of Lucas Terlinden-Ruhl

Flood Risk Modeling Aided By Machine Learning Techniques

Supervisors TU Delft: 

• J.A. Álvarez Antolínez PhD,

• Dr. P. Mares Nasarre,

• ir. G. Hendrickx, Delft University of Technology

Supervisors Deltares:

• ir. D. Eilander PhD, 

• ir. A. Couasnon PhD,

Compound floods, which can be attributed to different drivers (pluvial, fluvial, surge, tide and waves), generate a larger flood hazard when drivers co-occur than when they occur in isolation of each other. Previous research has shown how to conduct a compound flood risk assessment, but these are affected by a curse of dimensionality, where a larger number of events need to be simulated to get a good understanding of the response of risk to drivers. This research aims to create a methodology that improves the quantification of compound flood risk by using machine learning techniques. A treed Gaussian process (TGP) can actively learn from the response of damages to drivers to reduce the number of events that need to be simulated. By comparing this approach with a current strategy, which uses a maximum dissimilarity algorithm and scatter interpolation, the research shows a reduction of the computational cost by a factor of 4, an improvement in the root mean square error by a factor of 8, and an improvement in the estimate of expected annual damages (EAD) by a factor of 20. This reduction in computational cost allows for the inclusion of probabilistic variables that are normally assumed constant such as the duration and lag of drivers. A sensitivity analysis demonstrates these variables produce a statistically significant difference in the estimate of EAD and the risk curve. The research also shows the combination of driver magnitudes leading to extreme damage changes when including these additional probabilistic variables. By applying the TGP on multiple outputs, the research demonstrates the TGP is not only applicable to the case study, and can be implemented in current flood risk assessments.