Regional Flood Risk Assessment Using Ensemble of GCMs

Abstract

Flood risk assessment is generally carried out at a basin scale by developing hydrologic and hydraulic models with an objective to arrive at hazard/inundation maps for the river segments/reaches. A hydrologic model is used to derive a flood hydrograph corresponding to a specific return period and this is routed through the flood plain of the study area with the aid of a hydraulic model to obtain water surface elevations and inundation extents. This approach is best suited to represent a flood event at a river segment accurately, but the risks associated with floods need to be analyzed on a regional scale for effective flood risk management. As the size of the region increases beyond one basin, this approach fails to realistically represent the flood events across different river segments and basins. The simultaneous occurrences of different return periods on different river segments cannot be captured by this approach. The traditional approach to model these simultaneous occurrences is by using the streamflow records to arrive at spatially correlated stochastic simulations of streamflow. One issue that compounds this problem is the data scarcity in certain regions to accurately estimate the return periods of floods at distinct locations. To address these issues, Impact Forecasting, Aon’s catastrophe model development team, has undertaken a study to simulate stochastic flood events in the Southeast Asian region by considering Malaysia as a case study. The approach involves (i) downscaling of precipitation and temperature data from an ensemble of Global Circulation Models (GCMs), (ii) calibration of a grid-based Rainfall-Runoff (RR) model using available historical data of meteorological variables and streamflow, (iii) providing the downscaled precipitation and temperature data as input to the calibrated RR model to simulate streamflow across the study area and (iv) identifying flood events from the simulated streamflow to extract an exhaustive set of realistic flood events in the region. The approach involves the use of meteorological data of 15,000 years from 7 different GCMs, downscaled to 10 km grids from 100 km resolution. This enables capturing a broader spectrum of potential climate conditions and therefore generating a wide range of possible flood events without relying significantly on in-situ data. The proposed methodology considers the uncertainty inherent in climate models, providing a robust framework for assessing flood risk and results in a more reliable and realistic representation of stochastic flood events over a region. The approach presents a physically based alternative to the commonly used statistical approaches based on extreme value theory and could be a valuable tool for policymakers, researchers, and practitioners in making informed decisions in the face of evolving climate conditions.