Uncertainty induced by hazard attribution methods in building-level flood damage and risk assessment

Estimating monetary flood impacts commonly relies on combining hydrodynamic simulations with building-level damage models to compute scenario-specific damages. These losses are then aggregated across multiple flood scenarios to derive the expected annual damage (EAD), which often informs risk-reduction decision-making. While uncertainties related to hydraulic modelling and damage functions have been widely explored, the impact of methodological choices in the aggregation of flow variables at the building scale (also called hazard attribution) has received limited attention.

In high-resolution flood simulations, multiple computational cells typically overlap a single building footprint. To provide input to damage models, a representative value for each flow variable must be assigned to the building, commonly through a statistical operator such as a selected percentile. This study investigates the influence of this choice for water depth, flow velocity, and duration of inundation on the EAD, comparing it to uncertainties arising from modelled flood wave shape and friction parameterization. The analysis is conducted for a residential flood damage model, INSYDE-BE [1], in the city of Theux located in the Vesdre catchment (Belgium) severely affected by the 2021 European flood. A baseline scenario and two risk-reduction configurations (grey vs. hybrid measures) are evaluated.

Results indicate that water depth attribution dominates the uncertainty, with the choice of percentile resulting in up to twice the relative influence on EAD compared to other major uncertainty sources in the hydrodynamic modeling. In contrast, the selection of a particular statistical operator for the attribution of flow velocity and inundation duration has minimal impact, reflecting that particular attention must be paid to the attribution method for  water depth. For this reason, the water depths-to-building attribution method was calibrated using surveyed data from the study area in order to determine the most appropriate percentile for obtaining representative water depths.

Furthermore, the study explores the effect of incorporating risk aversion factors to address EAD’s tendency to underweight extreme, but low-probability events. Accounting for this factor increases the contribution of highly damaging scenarios. This potentially alters the ranking of mitigation measures, highlighting the importance of considering monetary indicators with caution.

[1] Scorzini, A. R., Dewals, B., Rodriguez Castro, D., Archambeau, P., and Molinari, D.: INSYDE-BE: adaptation of the INSYDE model to the Walloon region (Belgium), Nat. Hazards Earth Syst. Sci., 22, 1743–1761, https://doi.org/10.5194/nhess-22-1743-2022, 2022.