Including stochastic rainfall distributions in a probabilistic modelling approach for compound flooding due to tropical cyclones

Abstract

Hurricanes impose great threats on coastal communities in terms of hazard and impact. Recent hurricanes like Harvey (Texas, 2017) and Idai (Mozambique, 2019) emphasize the global character of this threat. In the U.S., Hurricane Harvey tied with 2005's Hurricane Katrina as the costliest tropical cyclone on record, inflicting $125 billion in damage. Harvey hovered above the state of Texas for six days in 2017, making it the longest landfalling storm in Texas history. During this time, over 600 millimeters of rain had fallen in most of the Houston area, with extreme observations showing 1500 millimeters of rain. This, in combination with surge and river discharges, caused inundation for over one-third of Houston. Current generation coastal flood early warning systems are often not designed to take into account the compound effects of pluvial, fluvial and marine flooding (e.g. the ADCIRC + SWAN model deployed by the Coastal Emergency Risks Assessment group ignores pluvial flooding). Moreover, the advanced models applied in these systems are computationally demanding and can therefore not be used in probabilistic real-time forecasting applications in order to include uncertainty in meteorological conditions. Furthermore, despite the importance of tropical cyclone rainfall, this field is not completely understood. Tropical cyclone rainfall distributions are dependent on best track data parameters (e.g. storm motion velocity), but also on environmental elements (e.g. topography). Current modelling approaches are either computational inefficient (e.g. large-scale climate models), are location-specific or are dependent on parameters not always available in archives. Therefore, there is need for a simple generic parametric rainfall model. Main focus of this research was the derivation of a methodology for assessing the joint probability of fluvial, pluvial and marine flooding. For this study, a case study is carried out for Houston, Texas. The TCWiSE tool is used to generate a set of synthetic hurricane tracks based on historical data and a Monte Carlo sampling method. A derived parametrization of tropical cyclone rainfall is used to generate a spatial tropical cyclone rainfall field. Furthermore, a Delft3D-FM model is used to generate offshore water level time-series for the synthetic hurricanes. A combination of SFINCS and Delft-FIAT is used to make an assessment based on both hydrodynamics and exposure for every single generated hurricane. The model train is capable of carrying out a flood risk assessment, derive flood maps for given return periods (e.g. 1 in 100-year flood) and make an exposure assessment for the joint occurrence of pluvial, fluvial and marine flooding. This framework can potentially be a useful tool for future policy- and decision making.