Recent evaluations of 2D models have analysed uncertainty in data inputs into flood models, but have treated the model code as a black box. In this work, the influence of the numerical representation of the model on the results is evaluated. The purpose is not only to understand the significance of the physical scheme in the model on results, but also the importance of this in respect to other known sources of uncertainty, in particular boundary conditions, calibrated parameters such as Manning’s friction values, DEM accuracy and other more subjective forms of uncertainty associated with the choices used by modellers in constructing models, such as building representation.
To further explore the impact that the level of physical representation has on model output, models were also analysed using risk and exposure based measures. The methods included vulnerability weighted measures and the use of damage curves from the Multi Coloured Manual.
A series of Monte Carlo tests were undertaken for a range of parameters over 3 test cases using the LISFLOOD-FP code. The LISFLOOD-FP code was chosen as it has several formulations for solving 2D floodplain flow within its framework, each with different level of physical representation.
The test cases included two urban events, a culvert overtopping event in Glasgow and canal embankment failure Coventry, and a river overtopping in Mexborough, Yorkshire a rural urban domain. The test cases provided a wider range of hydraulic conditions and are reflected events typically assessed with inundation models to ensure the effect of model bias was removed from the results.
The results for the test cases indicated that the choice of physical representation was the most critical in affecting model results, particularly for the urban test case. However, the interaction between factors and parameters also indicated that for certain scenarios, this becomes less critical to model results. The use of risk based methods also identified areas of variations between parameters sets and numerical schemes that are not identified with traditional model evaluation techniques.
Recent evaluations of 2D models have analysed uncertainty in data inputs into flood models, but have treated the model code as a black box. In this work, the influence of the numerical representation of the model on the results is evaluated. The purpose is not only to understand the significance of the physical scheme in the model on results, but also the importance of this in respect to other known sources of uncertainty, in particular boundary conditions, calibrated parameters such as Manning’s friction values, DEM accuracy and other more subjective forms of uncertainty associated with the choices used by modellers in constructing models, such as building representation.
To further explore the impact that the level of physical representation has on model output, models were also analysed using risk and exposure based measures. The methods included vulnerability weighted measures and the use of damage curves from the Multi Coloured Manual.
A series of Monte Carlo tests were undertaken for a range of parameters over 3 test cases using the LISFLOOD-FP code. The LISFLOOD-FP code was chosen as it has several formulations for solving 2D floodplain flow within its framework, each with different level of physical representation.
The test cases included two urban events, a culvert overtopping event in Glasgow and canal embankment failure Coventry, and a river overtopping in Mexborough, Yorkshire a rural urban domain. The test cases provided a wider range of hydraulic conditions and are reflected events typically assessed with inundation models to ensure the effect of model bias was removed from the results.
The results for the test cases indicated that the choice of physical representation was the most critical in affecting model results, particularly for the urban test case. However, the interaction between factors and parameters also indicated that for certain scenarios, this becomes less critical to model results. The use of risk based methods also identified areas of variations between parameters sets and numerical schemes that are not identified with traditional model evaluation techniques.
