Identification of key factors contributing to damage caused by flash floods to homes during the 2021 floods
A study conducted by researchers at the University of Liège, in collaboration with the GFZ Helmholtz Centre in Potsdam and the University of Potsdam, identifies the main determinants of damage to residential buildings during flash floods, based on data collected following the floods of July 2021. These events provided researchers with a rare opportunity to collect large-scale empirical data on the factors influencing damage to homes during flash floods, a type of flooding that remains insufficiently documented compared to conventional river floods, and for which existing damage models remain largely inadequate.
T
he catastrophic floods of July 2021 in the Vesdre valley in Belgium and the Ahr valley in Germany caused more than 200 deaths and approximately €40 billion in damage across Europe. These events provided researchers with an unprecedented opportunity to collect empirical data on the factors influencing damage to homes during flash floods, a type of flooding that remains under-documented compared to conventional river floods.
"Flash floods differ from ordinary river floods in their sudden onset, high flow velocities and the transport of debris and sediment," explains Daniela Rodriguez Castro, a PhD student at the HECE laboratory (School of Engineering) at the University of Liège. "These characteristics make the damage processes fundamentally different, and existing models, developed mainly for moderate river floods, struggle to capture them accurately." To address this gap, teams from the University of Liège and the University of Potsdam conducted independent field surveys following the 2021 floods. In Belgium, 420 residential buildings were surveyed in the Vesdre valley; in Germany, 277 in the Ahr valley in Rhineland-Palatinate and 332 in North Rhine-Westphalia. In total, thirty potentially explanatory variables were harmonised between the two countries, enabling a cross-border analysis at the building level for the first time.
"The analysis, based on machine learning approaches, confirms that water height is the most decisive factor for structural damage, regardless of the model used or the region considered," continues the PhD student. "This result is consistent with the existing scientific literature. However, the study highlights that other variables play a significant role when data from several regions are combined." For building damage, floor area, building type – particularly the number of facades exposed to water – and wall materials emerge as important factors. The presence of sediment in floodwaters, often underestimated in conventional models, also contributes to damage by generating additional costs for cleaning and removal. The presence of elderly people in the household is also a vulnerability factor, likely due to their reduced ability to implement emergency measures.
For damage to household contents, tenancy status – whether the person is a tenant or an owner – and the emergency measures taken during the event prove to be significant predictors, in addition to water depth and floor area. These results suggest that tenants, who typically have fewer household goods, suffer on average less damage to their contents than owners.
Furthermore, a supplementary study conducted exclusively in the Vesdre Valley examined whether resilience variables assessed at the level of municipalities or statistical sectors (such as the surprise effect, the saturation effect on emergency services, or the hydrological rarity of the event) could improve damage classification. While these variables reveal marked disparities between municipalities, their statistical contribution to the prediction of individual damage proves to be insignificant. This result can be partly explained by the extreme nature of the event, against which building-level protective measures proved largely ineffective, and by the fact that risk awareness does not necessarily translate into concrete protective actions. "This work highlights the importance of developing damage models specific to flash floods, incorporating variables related to the intensity of the phenomenon—including sediment transport and flow dynamics—as well as the physical characteristics of the exposed buildings," concludes Benjamin Dewals, a physics engineer at ULiège.
Scientific references
- Rodríguez Castro, D., Rafiezadeh Shahi, K., Sairam, N., Fischer, M., Samprogna Mohor, G., Thieken, A., Dewals, B., & Kreibich, H. (2025). Key Drivers of Flash Flood Damage to Private Households. Journal of Flood Risk Management, 18, e70088. https://doi.org/10.1111/jfr3.70088
- Rodríguez Castro, D., Cools, M., Roucour, S., Archambeau, P., Molinari, D., Scorzini, A. R., Dessers, C., Erpicum, S., Pirotton, M., Teller, J., & Dewals, B. (2025). Can macro- or meso-scale coping capacity variables improve the classification of building flood losses? Natural Hazards, 121, 1–31. https://doi.org/10.1007/s11069-025-07123-4
Contacts
Partners
- University of Liège – Hydraulics in Environmental and Civil Engineering (HECE) and Local Environment Management and Analysis (LEMA)
- GFZ German Research Centre for Geosciences
- University of Potsdam
This research was conducted in part in the framework of the Interreg Flash Flood Breaker project, co-funded by the European Regional Development Fund (ERDF)
