A Bayesian decision framework to support flood anticipatory actions in the urban data scarce city of Alexandria, Egypt

Abstract

Ensemble prediction systems (EPS) have been proposed to quantify uncertainty in forecasts, but to what extent they are useful for supporting flood anticipatory actions in an urban data-scarce city has not been fully explored. This research uses a Bayesian decision theory framework to support sequential decisions for reducing flood impacts. The predictive information is derived from probability distributions of flood depth simulated from a coupled ensemble Weather Research and Forecasting (WRF) and hydrodynamic MIKE urban inundation model. A damage function is used to value user actions and expected damages. Posterior probabilities are computed using prior probabilities and expected damages to select an action which minimises the expected losses.The analysis is done for the Egyptian coastal city of Alexandria, which experiences extreme rainfall and pluvial flooding from winter storms resulting in disruptions, damages and loss of lives. The decision framework supports anticipatory actions which can be taken 12-72 hours before an event. These include cleaning drains, dispatching pump trucks to critical flood locations before events, and proactive pumping to increase storage.Results suggest the use of a probabilistic decision framework can help support mitigating actions and reduce the occurrence of false and missed alarms. However, it depends on the combination of event intensity and probability (e.g. high intensity, low probability) the specific action and the loss function used. This approach helps decision-makers understand the value of probabilistic forecasts and models to trigger actions for improved decision support.