Connected Extremes: Capturing Dependence in Global-scale Flood Risk Assessments

Flooding is among the most frequent natural hazards worldwide and causes disastrous impacts every year. Coastal regions are particularly vulnerable to these impacts due to their low elevation, dense population, and critical role as important socioeconomic centres. Due to socioeconomic development, these areas are projected to face greater flood impacts in the coming century. In addition, a warming climate is projected to increase the frequency and intensity of flood events, thereby amplifying flood hazards and further increasing associated damages and risks. These concerns highlight the urgent need for improved flood risk assessments to support designing effective risk management strategies that can reduce flood risk and facilitate adaptation to a changing climate. Recent years have seen increased efforts from the scientific community in developing models and methods for estimating large-scale flood risk. However, most large-scale coastal flood hazard modelling is still based on a small set of design events with spatially homogeneous return periods. This approach fails to address the spatial patterns of real flood events. Moreover, such assessments are limited by using short observational or modelled sea level records, which are insufficient to characterise the full range of possible extreme events. Flooding in coastal areas can also be influenced by other meteorological and hydrological factors, such as intense rainfall and high river discharge. When these drivers occur simultaneously or in close succession, their interactions can lead to a compound flood event. While our understanding of the multivariate dependence between compound flood drivers has improved over the past decade, spatial dependence is often overlooked or only considered within river tributaries of a single catchment. To address these issues, I first develop a novel approach to characterise spatial dependence of historic events by applying a multivariate conditional statistical model to 40-year reanalysis sea levels at 17,394 modelled coastal stations. These estimated dependence structures are then incorporated into a Monte-Carlo simulation to develop a large set of synthetic extreme events at the global scale over 10,000 years under current climate conditions. The validation results show that the synthetic event set can capture a wide range of extreme sea level events with realistic spatial patterns, which can be used to develop spatially dependent hazard maps for deriving better risk estimates. Next, I develop an event-based probabilistic coastal flood risk framework by combining the synthetic event set, a precomputed suite of inundation maps, and existing vulnerability and exposure datasets. These event-based risk estimates are then compared to those derived from the traditional assessment based on spatially homogeneous assumptions. Results show that the spatially homogeneous approach estimates lower damages for relatively low return periods, while higher damages are estimated for medium-to-large return periods. Spatial dependence has minor effects on EADs but a higher 200-return-year damage is estimated for 76% of global countries by the spatially homogenous approach. I also demonstrate the added value of the event-based approach by showing the damage distribution of combined maximum annual damages at a subnational scale for each continent. Lastly, I extend the multivariate conditional statistical framework to capture both spatial and multivariate dependence between coastal and river floods. This framework is applied to sea level and river discharge observations along the U.S. coastlines. Results show that the West coast is identified as the coast with the strongest spatial correlation of compound flooding, while the Gulf of Mexico shows the weakest spatial correlation and the East coast features moderate spatial correlation. The models and results presented in this thesis enable more robust risk estimates, thereby supporting risk management through prioritising adaptation efforts, guiding large-scale spatial planning, developing transnational risk financing plans, and enabling more accurate insurance premium pricing.