Abstract :
[en] Floods are among the most frequent and damaging natural hazards worldwide, causing substantial losses to human lives, infrastructure, economic activities, and social well-being. Climate change, urbanization, and increasing concentrations of exposed assets have contributed to a growing flood risk in many regions, highlighting the need for effective flood risk management strategies. Reliable flood damage models play a central role in these efforts by supporting risk assessments, cost-benefit analyses of mitigation measures, emergency planning, recovery strategies, and climate adaptation policies.
Despite considerable advances in flood damage modelling, important challenges remain, particularly for flash floods. Compared with riverine floods, flash floods are characterized by rapid onset, high flow velocities, and significant debris and sediment transport, resulting in damage mechanisms that are often more complex and less understood. Existing flood damage models primarily focus on hazard and physical vulnerability variables, while the influence of preparedness, coping capacity, and emergency response remains insufficiently explored. Furthermore, many models are developed using data from a single event or geographical context, raising questions regarding their transferability, generalizability, and applicability across regions. At the same time, the scarcity of high-quality post-event damage data continues to constrain both the development and validation of flood damage models.
This thesis addresses these challenges through the analysis of the extreme July 2021 flood event that affected Belgium, Germany, and the Netherlands. A unique harmonized post-event database was compiled from detailed field surveys conducted in the affected regions, providing information on observed residential flood damage, hazard characteristics, building attributes, mitigation measures, and indicators of preparedness and coping capacity.
The overall aim of the thesis is to improve the understanding and modelling of residential flash-flood damage by identifying the factors that control damage, evaluating the transferability of damage relationships across regions, assessing the benefits of integrating heterogeneous datasets, and investigating the applicability of synthetic modelling approaches under flash-flood conditions.
To achieve this aim, four research questions are addressed. First, the thesis investigates whether indicators related to community preparedness, coping capacity, and emergency response can improve the predictive performance of residential flood damage models and contribute to explaining severe damage patterns observed during the July 2021 flash floods. Second, it examines the principal drivers of residential flash-flood damage and evaluates the extent to which these drivers remain consistent across different regions affected by the same event. Third, it assesses how the integration of cross-regional datasets influences model predictive performance and uncertainty, while exploring the feasibility of developing harmonized transboundary flood damage models. Finally, the thesis investigates whether a synthetic flood damage model originally developed for riverine flooding can be adapted to represent residential flash-flood damage and reproduce observed damage.
Methodologically, the thesis combines empirical and synthetic modelling approaches. The empirical analyses employ a range of statistical and machine-learning techniques, including binomial logistic regression, multilinear regression, elastic net regression, random forests, conditional inference trees, and Bayesian networks.
By integrating empirical and synthetic perspectives, the thesis contributes to advancing flood damage modelling in four key areas. First, it shows that community-level coping-capacity factors are insufficient to explain severe residential flood damage at the building level, and that explanatory variables at the household level are necessary. Second, it reveals that several drivers of flood damage remain consistently important across regions, including the building size and structure, as well as socio-economic factors. Third, the thesis demonstrates that probabilistic empirical flood damage models can be generalised to multiple regions if transnational, heterogeneous datasets are available for model training. The cross-regional model performs almost as well as models trained in individual regions. Finally, this research presents a replicable methodology for adapting a synthetic riverine damage model to represent the distinctive damage processes typical of flash floods. Collectively, these contributions support the development of more robust, transferable, and physically meaningful flood damage models, particularly in transboundary contexts, and contribute to improving flood risk assessment and management under increasingly uncertain future conditions.
