Item

Abstract

Flooding events are commonly categorised per “pathway”, such as levee breaches, riverine floods, or surface water floods. The definition of these categories and their distinction are however multiple and ambiguous. Moreover, floods are frequently compound events: when multiple hazard dynamics take place simultaneously and/or in sequence. Flood losses documentation, estimation, and modelling likewise start with the assignment of each flooding event to one single flood pathway. Ignoring potential synergistic effects of the co-occurrence of different pathways can be an important source of bias in flood loss models. The investigation of compound floods (monetary) impacts is furthermore still in its infancy. Therefore, we explore how compound flood cases could be represented in and integrated into flood loss models by developing Bayesian models, an approach that can deal with different sample sizes, variable types, monotonic effects, and intrinsically address uncertainty. With a loss database of individual residential buildings affected by flooding events in Germany, we developed single- and multi-level models for the most frequent flood pathways and compound cases at hand. The models were either trained with a dataset of buildings solely affected by single flood pathways (i.e., riverine floods or surface water floods), compound events of riverine and surface water floods, or a multi-level model that learns all three cases. In the multi-level model, each data group informs the others in learning the model parameters simultaneously. Obstacles to such development, including data features, as well as comparison of parameters and loss estimates from the different models are presented.