Enhancing hazard mapping, emergency response comprehension, and life-loss assessment of flood events: a study on probabilistic scenarios and empirical analysis

Abstract

Floods are the most recurrent natural disaster on a global scale, with a significant increase in the frequency of extreme events driven by climate change. This growth, combined with intensified urbanisation and the resulting expansion of population exposure, substantially raises associated risks, leading to a greater number of people affected and an increase in fatalities. Despite this, a reduction in the average number of individuals impacted per event has been observed, indicating progress in risk mitigation, particularly through investments in Early Warning and Evacuation Systems (EWESs). The effectiveness of these systems can be evaluated using life-loss estimation models, which provide insights into the potential impacts of floods under different operational conditions of EWESs. However, these models still face significant limitations, including the insufficient representation of hydraulic uncertainties, mainly due to the high computational cost of high-resolution probabilistic simulations; the lack of probabilistic approaches that account for temporal and spatial variability in population distribution; and the limited representativeness of agent behavior (authorities and population), which is often based on studies focused on the United States. This thesis aims to explore and address these limitations by advancing the application and understanding of life-loss estimation in flood events across four main chapters. First, life-loss modelling is introduced through the evaluation of different EWES alternatives, considering a hypothetical dam-break scenario upstream of a densely urbanised valley. The analysis identifies, by sector along the valley, the optimal moments and critical minimum lead times for issuing alerts that minimise fatalities. Next, a probabilistic flood mapping method for the failure of an earthen dam is developed using a computationally optimised approach. In a fictional case study, it is observed that the inundation area derived from deterministic analysis corresponds to the area associated with a 40% exceedance probability in probabilistic scenarios. In the third stage, a new methodology is proposed for probabilistic life-loss estimation that integrates, under a probabilistic framework, both hydraulic uncertainties and the temporal and spatial variability of population distribution. Applied to the same case study, the analysis reveals that although deterministic estimates tend to produce higher median values than the probabilistic weighted means, accounting for variability, expressed through standard deviation, can reverse this outcome. Finally, the performance of EWESs during the 2021 flood in Germany is analysed based on surveys conducted with affected residents. The study identifies substantial gaps between the practical implementation of the systems and theoretical recommendations. Despite the timely issuance of alerts, systemic communication failures undermined public response, and it is estimated that alternative EWES strategies could have reduced human losses by up to 80%. The findings, proposed methodologies, and discussions developed in this thesis contribute to improving the simulation tools used to assess the impacts of EWESs. They provide valuable insights for the development of more robust, context-sensitive models that can produce more accurate estimates of human impacts during flood events.