Probabilistic loss modelling supports flood risk assessment under changing conditions

Código

I-EBHE0111

Autores

Heidi Kreibich, Ravikumar Guntu, Kasra Rafiezadeh Shahi, Nivedita Sairam

Tema

WG 1.04: From local to large scale human-water dynamics

Resumo

Flood risk analyses are an important decision-making basis for flood risk management and climate adaptation. However, such analyses are associated with significant uncertainty, even more since changes in risk due to global change are expected. Novel probabilistic flood loss models are developed with a focus on capturing changes in vulnerability. Such models are developed on the basis of machine learning (ML)-based frameworks (e.g., Bayesian Networks) and empirical flood loss databases. This contribution will present how data from standardized surveys on flood affected private households and companies in Germany is used to gain more knowledge about damage processes and how they can be altered by private precaution and emergency response. For instance, ML-based findings show that in extreme flash flood scenarios (i.e., high water depth), successful emergency measures can reduce building losses by up to 75% for large companies. Another analysis shows that the implementation of private precaution reduced average residential building loss by 11 to 15 thousand EUR per object, confirming the significant effectiveness of precautionary measures during riverine floods. Furthermore, probabilistic flood loss models (e.g., BN-FLEMOps) are developed, which have the significant advantage to capture changes in vulnerability and inherently provide quantitative information about the uncertainty of the prediction. Applications of probabilistic flood loss models from local flash floods to large scale riverine floods are shown, demonstrating that the models are able to describe temporal changes in flood risk.