Author(s): Saba Mirza Alipour; Joao Leal
Linked Author(s): João Bento Leal
Keywords: No Keywords
Abstract: Flood modelling process is always incorporated with uncertainty and capturing the uncertainty can be challenging due to the heavy computational requirements. Surrogate models developed based on physical models using machine learning techniques (like Support Vector Regression, SVR) can be used as fast and reliable tools to overcome the computational constrains. Therefore, this study aims to present the uncertainty of 100-year flood water level using intensive Monte Carlo (MC) simulations and to show the applicability and potential of surrogate models for intensive task like uncertainty analysis. The results show that the surrogate model easily reproduces a large number of simulations (more than 25,000) that are required for a converged MC, which would not be feasible using the physical model.
Year: 2022