A Multi-Model Approach to Return Period Estimates: The Example of Floods

Traditionally, hazard assessments have relied on the concept of a T-value (or return period) to define the average frequency of extreme events, which is linked to an exceedance probability. However, despite significant advancements in continuous hazard modeling, the most common method for hazard definition still relies on pre-determined probabilities of exceeding a specific event. As a result, current models typically provide hazard maps for specific return periods, often resulting in limited overlap among the considered return periods.

To address this limitation, we introduce a multi-model expected return period framework for extreme events, which estimates the average frequency of climate hazards by integrating multiple open datasets. This approach offers a novel methodological perspective on return period estimation, providing a statistically robust and user-friendly tool to address model heterogeneity. We demonstrate the framework's applicability by utilizing open and accessible web data on flood hazards, specifically three spatial datasets detailing global inland fluvial flood maps: the World Resources Institute's Aqueduct Floods Project maps (Ward et al., 2020), the European Joint Research Center's (JRC) maps (Dottori et al., 2016), and the maps produced by the CIMA Research Foundation and the United Nations Environment Programme (Rudari et al., 2015).

We introduce here a multi-model expected return period value, TMM, determined by an average function based on the number of active models, i.e., the number of models that provide a valid output for a given cell within their active spatial domain. This concept shifts the perspective on extreme event frequency from evaluating hazard as event-, model-, and return-period-specific to a more integrated hazard estimation approach.

Although demonstrated using flood hazard data, the framework can be adapted to any other spatially-explicit extreme event characterized by return periods, such as drought, cold spells, and wind storms. By leveraging the differences among existing datasets, our mathematical framework adds value to open science efforts, introducing a tool to exploit high-quality, open data from distinct modeling designs. This can particularly benefit socio-economically vulnerable communities. Our work showcases the potential of heterogeneous, open data sources to improve climate knowledge and provides a robust foundation for future research in hazard modeling and climate risk assessment.