Analysis of the Uncertainty Associated With Numerical Schemes in Flood Inundation Models (PAP014819)
Thomas D. M. Willis, Nigel G. Wright, P. Andew Sleigh
Recent evaluations of 2D models have focused on benchmarking computer codes and analysis of model outputs with respect to model inputs, focusing on the model as a black box, in order to understand sources of input uncertainty and how they propagate through to results. In this work, the influence of the numerical representation of the model on the results and the associated uncertainty is evaluated in a systematic way. For a test case of a canal embankment outburst event in Coventry the LISFLOOD code, which contains modules within the framework with different levels of physical representation, is tested against a range of uncertain inputs that are typically considered in a modeling exercise. The analysis is extended to show how traditional evaluation techniques such a comparison of extent do not capture the full uncertainty range of the modules. Instead, the use of risk based methods, in this case a cost of damage model shows how local variations in model results are rarely captured in measures of extents, but can be enhanced with exposure based methods. Overall, the level of physical representation is shown to be critical to model results.