Search

Flood loss data collection and modeling are not standardized, and previous work has indicated that losses from different flood types (e.g., riverine and groundwater) may follow different driving forces. However, different flood types may occur within a single flood event, which is known as a compound flood event. Therefore, we aimed to identify statistical similarities between loss‐driving factors across flood types and test whether the corresponding losses should be modeled separately. In this study, we used empirical data from 4,418 respondents from four survey campaigns studying households in Germany that experienced flooding. These surveys sought to investigate several features of the impact process (hazard, socioeconomic, preparedness, and building characteristics, as well as flood type). While the level of most of these features differed across flood type subsamples (e.g., degree of preparedness), they did so in a nonregular pattern. A variable selection process indicates that besides hazard and building characteristics, information on property‐level preparedness was also selected as a relevant predictor of the loss ratio. These variables represent information, which is rarely adopted in loss modeling. Models shall be refined with further data collection and other statistical methods. To save costs, data collection efforts should be steered toward the most relevant predictors to enhance data availability and increase the statistical power of results. Understanding that losses from different flood types are driven by different factors is a crucial step toward targeted data collection and model development and will finally clarify conditions that allow us to transfer loss models in space and time. Key Points: Survey data of flood‐affected households show different concurrent flood types, undermining the use of a single‐flood‐type loss modelThirteen variables addressing flood hazard, the building, and property level preparedness are significant predictors of the building loss ratioFlood type‐specific models show varying significance across the predictor variables, indicating a hindrance to model transferability [ABSTRACT FROM AUTHOR]

Flood damage processes are complex and vary between events and regions. State‐of‐the‐art flood loss models are often developed on the basis of empirical damage data from specific case studies and do not perform well when spatially and temporally transferred. This is due to the fact that such localized models often cover only a small set of possible damage processes from one event and a region. On the other hand, a single generalized model covering multiple events and different regions ignores the variability in damage processes across regions and events due to variables that are not explicitly accounted for individual households. We implement a hierarchical Bayesian approach to parameterize widely used depth‐damage functions resulting in a hierarchical (multilevel) Bayesian model (HBM) for flood loss estimation that accounts for spatiotemporal heterogeneity in damage processes. We test and prove the hypothesis that, in transfer scenarios, HBMs are superior compared to generalized and localized regression models. In order to improve loss predictions for regions and events for which no empirical damage data are available, we use variables pertaining to specific region‐ and event‐characteristics representing commonly available expert knowledge as group‐level predictors within the HBM. Key Points: Hierarchical Bayesian model (HBM) captures spatiotemporal variability in flood damage processes better than established depth‐damage functionsRegion‐ and event‐specific characteristics that explain the variability in damage processes improve spatiotemporal transfer of the HBMThe HBM is tested for transferability using empirical damage data from six events in the Elbe, Danube, Rhine and Oder catchments in Germany [ABSTRACT FROM AUTHOR]