Author(s): Agnieszka I. Olbert; Sogol Moradian

Linked Author(s): Agnieszka Olbert

Keywords: Compound events; Flood forecasting; Flood risk management; Hydrodynamic modeling; Machine learning; Urban floods

Abstract: Floods are one of the most severe natural disasters. An accurate flood forecasting can aid flood risk management efforts and by that reduce or even prevent flood damages. However, flood forecasting is challenging because it relies on forecasted inputs of meteorological and hydrological variables which can be subject to large uncertainties. Hydrodynamic models, if used for flood modelling, are also another source of uncertainties. Compound floods generated by coastal and fluvial drivers acting simultaneously add another layer of complexity and uncertainty. This research demonstrates a novel approach to an integrated flood forecasting system. The system combines statistical, hydrodynamic and machine learning models. In this three-model cascade the joint probability results derived from a best-fit copula function are used to force the hydrodynamic model of urban riverine-estuarine hydrological domain while the hydrodynamic model outputs are used to train and validate the machine learning model. Ultimately the machine learning model is used to forecast the states of water levels, from which the risk of flood can be determined. This research finds that the coupled statistical-hydrodynamic-ML system can be successfully used for flood forecasting.

DOI: https://doi.org/10.3850/iahr-hic2483430201-309

Year: 2024