Development and assessment of dynamic storage cell codes for flood inundation modelling
Hunter, Neil Martin
PublisherUniversity of Bristol
MetadataShow full item record
Since 1962 storage cell codes have been developed to simulate flow on fluvial and coastal floodplains. These models treat the floodplain as a series of discrete storage cells, with the flow between cells calculated explicitly using some analytical flow formulae such as the Manning equation. Recently these codes have been reconfigured to use regular Cartesian grids to make full use of widely available high resolution data captured from remote sensing platforms and stored in a raster GIS format. Such raster-based storage cell codes have many of the advantages over full two-dimensional depth averaged schemes but without the computational cost, however their typical implementation results in a number of fundamental limitations. These include an inability to develop solutions that are independent of time step or grid size, and an unrealistic lack of sensitivity to floodplain friction. In this thesis, a new solution to these problems is proposed based on an optimal adaptive time step determined using a Courant-type condition for model stability. Comparison of this new adaptive time step scheme to analytical solutions of wave propagation/recession on flat and sloping planar surfaces and against field measurements acquired for four real flood scenarios demonstrates considerable improvement over a standard raster storage cell model. Moreover, the new scheme is shown to yield results that are independent of grid size or choice of initial time step and which show an intuitively correct sensitivity to floodplain friction over spatially-complex topography. It does, however, incur a prohibitive computation cost at model grid resolutions less than 50 m. This primary research is supplemented by an examination of the data and methods used to apply, and in particular calibrate, distributed flood inundation models in practice. Firstly, different objective functions for evaluating the overall similarity between binary predictions of flood extent and remotely sensed images of inundation patterns are examined. On the basis of the results presented, recommendations are provided regarding the use of various measures for hydrological problems. Secondly, the value of different observational data types typically available for calibrating/constraining model predictions is explored within an extended Generalised Likelihood Uncertainty Estimation (GLUE) framework. A quasi-Bayesian methodology for combining these individual evaluations that overcomes the limitations of calibration against any single measurement source/item is also presented.