Flood threats towards urban areas are escalating due to climate change, which increases the
frequency of extreme rainfall events, and due to the growth of sealed surfaces caused by
urbanisation. One-dimensional (1D) hydrodynamic and two-dimensional (2D) hydrodynamic
urban flood models are used worldwide to better understand how the flood risk can be reduced at
an acceptable level. 2D models are preferred over 1D models due to higher simulation accuracy,
but this benefit comes at a computational expense that hinders the use of 2D models for real-time,
high-resolution and large-scale modelling. To optimise the computational efficiency of 2D
models, this PhD study developed a sub-model approach to rapidly identifying the relevant areas
that may cause flooding of the targeted object (e.g. infrastructures in a city or in a specific
district, or a single building), excluding the irrelevant areas (i.e. 2D model grids), and thus
resulting in a significantly reduced number of computational cells and hence a much faster
computation.
The tailor-made sub-models for fast simulations are obtained through the following
programming steps: i) computationally important sinks are identified from basin-wide detected
sinks by referring a Volume Ratio Sink Screening, while accumulated effects of volume losses
from eliminated sinks are still controlled; ii) the drainage basin area is discretized into a
collection of sub-impact zones based on the spatial distribution of important sinks, and a 1D
surface network is delineated accordingly; iii) the link-based fast-inundation algorithm is
programmed for fast computation of the basin-wide flow conditions using 1D static routing; iv)
according to the target objects, relevant sub-impact zones are identified by tracing 1D flow
conditions; v) the critical computational cells for a 2D hydrodynamic model can be extracted
based on these identified sub-impact zones, suggesting the reduced domain as well as the
optimized boundary condition for model configuration. The suggested method was validated by
five model experiments, using MIKE FLOOD as a reference for full 2D hydrodynamic models.
The results revealed that the sub-model approach yields comparably accurate results (flood
extents, depths and flow velocities) while offering a significantly improved computational
efficiency with robust performance, compared to the modelling for the full basin.
The sub-model approach identifies a promising solution to the realization of real-time
applications of high-resolution 2D urban flood models at a large scale. Future research for
implementation in real-time forecasting system including real-time weather radar monitoring,
parallel computing, adaptive-grids, enhanced 1D network representations is recommended for
future research.