Author(s): Behzad Jamali; Reinier Koster; Monique Retallick; Mark Babister
Linked Author(s): Mark Babister, Monique Retallick, Behzad Jamali
Keywords: Fast Flood Modelling 1D and 2D Hydraulic Models Parametrising models
Abstract: Two-dimensional (2D) hydrodynamic models have a high computational cost and usually take hours to simulate a flood event. Despite the advancement in the computational speed of modern computers, especially in parallel computing, running multiple flood simulations can be costly (Jamali et al., 2019). For example, the Australian Bureau of Meteorology’s Hydrological Forecasting System is using ensemble rainfall forecasting for real-time flood warning which has a short prediction lead time (Velasco-Forero et al., 2023). Additionally, 2D hydraulic models cannot be used in Monte-Carlo simulations which usually require 5 to 20 thousand runs. There has been a growing number of studies that use machine learning techniques or conceptual methods to emulate 2D hydrodynamic models. These models usually lack generalisation, suffer from an overfitting problem. Conceptual flood models that rely only on water balance or simplified forms of the Shallow Water Equations are generally not accurate where flow behaviour is dominated by momentum terms. 1D hydraulic models are fast but are not easy to setup and calibrate. They are also limited in predicting accurate map outputs such as flood extent and depth. This study introduces a new fast flood modelling tool that is one to two orders of magnitude faster than the full 2D hydrodynamic models. The fast model is based on the 1D hydraulic model solving St Venant equations but capable of emulating 2D Shallow Water equations. The advantage of the proposed model is that the 1D model is directly parametrised from a few the 2D model simulation results and therefore, it does not require any additional data. A semi-automatic procedure is developed to easily setup the model from the calibrated 2D model setup. Additionally, a semi-automatic procedure is developed to map the fast flood model outputs taking advantage of existing 2D model grids. This innovation makes it possible to setup 1D models without having any cross-sectional data while accurately simulating 2D model behaviour. The application of the model is demonstrated in a riverine flooding in a large catchment in Australia for which a calibrated 1D-2D TUFLOW hydraulic model is available. Results showed that the fast model is capable of reproducing 2D model results with high accuracy and for a range of flood events. The model was also successful in predicting design levels that were not used in the parametrisation process. By using the full physical equations, the fast model proved to have excellent generalisation power. Therefore, it is recommended for Monte-Carlo or real-time flood forecasting applications.
Year: 2025