Evaluating a new approach to global stochastic flood risk mapping

Abstract

Flooding is one of the most impactful natural disasters worldwide, causing billions of dollars in damages and affecting millions of people each year. Flood risk assessment generally relies on hazard maps, representing the water depth associated with a flow (or precipitation amount) with a set probability of occurrence. However, inundation hazard maps assume the magnitude of the event to be constant throughout the model domain, not accounting for the spatial characteristics of events. This limitation makes hazard maps more appropriate for local scale risk assessment, while accurately evaluating risk at larger scales should include the simulation of flood events with realistic spatial patterns. This is mainly because the probability of a 1-in-X years flow simultaneously affecting a whole continent is likely to be much lower than 1-in-X years. As a consequence, using a methodology that does not take spatial correlation between sites into account might lead to misestimating economic losses or population exposure. Stochastic generation of flood events using statistical methods can help us to have a better view of risk, while keeping computational costs lower than full hydrological simulations. These models currently rely on historical data to derive spatial correlation between locations, and are therefore limited by gauge availability, making them not applicable in data-scarce regions.
In this thesis, we propose a methodology that uses freely available discharge estimates from global hydrological models and rainfall data from reanalysis as input to a conditional exceedance model to generate flood event sets. First, we test this approach by comparing it to a more standard gauge-based model in different areas of the world. Secondly, we develop a case study model in Southeast Asia, simulating 10,000 years of events. We calculate the resulting population exposure and validate it against observations from EM-DAT (the Emergency Events Database from the Centre for Research on the Epidemiology of Disasters). Finally, we calculate the economic losses for each country in our study region and validate our results against observations.
The results from our analysis show that it is possible to estimate population exposure and economic losses that agree with observations using this model-based methodology. Although further research is needed to test and validate this approach in other regions, this paves the way to develop a fully global stochastic flood model, that includes risk estimation for data scarce regions.

Date of Award	1 Oct 2024
Original language	English
Awarding Institution	University of Bristol
Supervisor	Paul D Bates (Supervisor) & Jeff Neal (Supervisor)