Low-fidelity Hydrodynamic Model-based Method for Efficient Flood Inundation Modelling

© 2021 Qi YangFlood is one of the most devastating natural hazards, as it often causes fatalities and damages to infrastructure. To develop strategies for flood risk mitigation, flood inundation models are often used to provide useful information for assessments of potential impacts of floods. Two-dimensional (2D) hydrodynamic models are commonly used for flood inundation simulations. However, they can be computationally intensive when used to simulate many flood events, for example for uncertainty analysis, or to simulate very large floodplains. To improve computational efficiency, data-driven models based on machine learning techniques and conceptual models based on simplified water-filling concepts have been developed. Data-driven models appear as black-box models and are yet to be used by many practitioners with confidence. Simplified conceptual models are generally not designed to simulate the temporal propagation of floods and are often only applied to estimate maximum/final flood extent and floodwater levels. In Australia, 2D hydrodynamic models have been established for many important catchments. There is potential to build on these existing models and develop methods to speed up flood inundation simulations. In this MPhil thesis, a new modelling method, LoHy+, is proposed based on existing 2D hydrodynamic models, to produce an efficient simulation of flood extent and depth with time. The method first develops a low-fidelity 2D hydrodynamic model (LFM) with coarse mesh based on an existing high-fidelity 2D hydrodynamic model (HFM). The aim of the LFM is to produce reasonably accurate simulation of water levels within the main river channels while tolerating poorer simulation elsewhere in the floodplain. Next, the method develops a Mapping Module by using training data to establish relationships between water levels in both river channels and across the floodplain generated using the HFM and water levels in the river channels generated using this LFM. In subsequent applications, the LFM is run first, and the Mapping Module is applied to estimate flood inundation within the entire model domain. The implementation of the LoHy+ is demonstrated using a real-world catchment located in the southern Murray Darling Basin, Australia. A fully calibrated HF MIKE21 FM hydrodynamic model is available for the catchment. The performance of the LoHy+ method is evaluated against simulation from the high-fidelity hydrodynamic model. There is a good agreement between results from the LoHy+ method and the original high-fidelity 2D hydrodynamic model. The new method is much more efficient and can simulate the spatiotemporal evolution of flood inundation with reasonable accuracy. It is potentially a useful tool for applications that require many model runs or long simulation durations