Comparing flood susceptibility estimation methodologies

Abstract

Recent studies have shown that flood risk contributes to a major part of the total risk caused by natural hazards in Western Europe. Especially in the United Kingdom, flood risk has been identified as a substantial threat, which is likely to continue to grow due to (coastal) urbanisation, sea level rise and climate change. A separate study has shown that common analysis practice may underestimate flood risk due to the exclusion of compound flood events. These are flood events with multiple flood causes, such as a combined coastal and pluvial flood event. Floods are often caused by a complex system of multiple dependent flood driving mechanisms, which vary and interact as time progresses. Various fundamentally different methodologies have been developed to estimate the susceptibility of an area to floods. There is no singular best methodology which can be applied to every problem. Due to the complex nature of flood events, it is vital to understand these methodologies and their characteristics to represent the susceptibility to flooding correctly. Misrepresentation of flood susceptibility can lead to inefficient decision making, or worse, insufficient flood safety. The goal of this research is to compare several flood susceptibility estimation methodologies (including coastal, pluvial, and compound flooding), to observe the results and clarify notable differences in outcomes. A case study is conducted on the coastal town of Eastbourne, which has various common, but non-trivial characteristics. The flood susceptibility of Eastbourne is estimated using various estimation techniques, such as statistical analysis and numerical modelling. Not only did this give insight into the flood susceptibility of Eastbourne, but it also illustrated the capabilities of the selected methodologies. Eastbourne has a history of coastal and pluvial/fluvial flooding. In order to include the physical mechanisms of these flood types, it was decided to use relatively computationally expensive flood susceptibility estimation methodologies. Hence, use was made of sensibly chosen storm scenarios, based on extreme value analysis. The statistical analysis started with a general exploratory data analysis in order to find notable dependence structures between variables. Afterwards, two variants of extreme value analysis on reanalysed historical data were used to create "smart" scenarios. First of all, the peak over threshold approach (POT) was used to create scenarios in which a singular flood type was dominating. Secondly, the conditional approach was applied to create compound storm scenarios. In total, 8 different scenarios were created. These scenarios gave the input conditions for the models. At the coastal boundary, a spectral action balance resolving wave model SWAN was used to estimate close shore conditions based on offshore scenario input. These nearshore conditions where then used to estimate overtopping rates into the inundation domain via three different overtopping methodologies. First of all, a empirical equation (the "new" overtopping formula of the EuroTop manual) has been applied. Secondly, a Gaussian process emulator (Bayonet GPE), an approach similar to a neural network, has been applied. And lastly, use was made of a numerical hydrodynamic model SWASH to simulate wave transformation and overtopping. This last approach is especially interesting since numerical models have the ability to improve upon spectral action balance resolving wave models. Numerical models can implicitly account for infra gravity waves, a physical mechanism that is known to impact overtopping rates, something that spectral action balance resolving models struggle with. The numerical hydrodynamic model HEC-RAS, is used for the simulation of inundation caused by the storm scenarios. All 8 scenarios are calculated twice, in order to compare simulations made with two sets of governing equations: The diffusive wave approximation of the shallow water equations (DSW) and the full shallow water equations (SWE), also known as the Saint Venant equations. The DSW approach has the advantage of being relatively computationally efficient, whilst SWE approach has the benefit of including a more accurate representation of physics. Since the DSW approach is more computationally efficient, it allows for solving on a higher spatial resolution. The model was applied to the part of Eastbourne, which is deemed to have the highest flood risk. It was then fed with overtopping discharges at the coastal flood defences and used precipitation events as defined by the respective scenarios. After the simulation, flood maps have been generated to compare the inundation patterns. Inundation curves have been made as well, for a more objective and apprehensible comparison of the severity of the flood events. The overtopping results showed that all three approaches agreed reasonably well, and were able to estimate plausible overtopping rates. There were no overtopping measurements for validation purposes, but the overtopping magnitude does agree with the coastal flood history of Eastbourne. The numerical method is more advanced than the empirical formula and the GPE. However, the results also showed a large sensitivity with regards to the (vertical) discretisation. Furthermore, the numerical approach, as applied in this research, did require explicit inclusion of refraction and fitting of breaking behaviour based on computationally expensive simulations, in order to give reasonable estimations. The numerical approach is promising and has the potential to become more accurate than the other methods considered. However, in its current state the method does seem as uncertain as the other considered approaches, whilst being more complex to implement. Lastly, regarding the comparison of governing equations for the estimation of inundation, it was found that the low spatial resolution SWE approach systematically underestimated inundation severity with regards to the high spatial resolution DSW approach. In addition to this, it was found that the times between the storm peaks and the inundation peaks was systematically lower for the SWE simulations than for the DSW simulations. Both these phenomena can be explained as a consequence of using a different spatial discretisation. When using the same spatial discretisation these effects disappear and the inundation curves agree well. There are, however, still some smaller local differences between simulations visible on the inundation maps. In the case of Eastbourne, modelled as described in this report, it is thus more important to accurately include topological features than to accurately represent the flow physics. This could, however, be different when considering higher spatial resolutions.