Projecting frequencies of extreme rainfall compound events under climate change using bivariate extreme value modeling and multivariate bias corrections

Extreme compound events, defined as the “combination of multiple drivers and/or hazards that contributes to societal or environmental risk”, present a
growing concern for the scientists and the civil society (Zscheischler et al. 2020). Climate models provide physical simulations of the climate until 2100, which permits to better understand the evolution of the extreme events under climate change. This study proposes a novel modeling of bivariate extreme events using bivariate Generalized Pareto Distributions (biGPD), with Extended GPD for the univariate part (EGPD). This novel semi-parametric modeling is applied to an extreme event: the flooding of the Seine and the Loire watersheds in June 2016. This event is a spatially compound event between the accumulated precipitations over the two watersheds. The accumulation of rain over several days is approximated by the Antecedent Precipitation Index (API), and high values of API are considered to lead to flooding. This approach is compared to a more classic copula modeling over simulations in Jacquemin et al. (2025, submitted). As climate simulations often have statistical biases, they must be corrected using bias correction algorithms. This is also the case for their simulations of compound events. This study compares several multivariate bias correction algorithms (CDF-t, dOTC and R2D2) on this event. dOTC seems to perform better than R2D2 for extreme values. The proposed methodology illustrates how compound events can be analyzed, and their evolution in frequency projected. As a perspective, this method can be applied to more diverse compound events, and it could be generalized to events in higher dimensions.

Bibliography:
Zscheischler, J., Martius, O., Westra, S., Bevacqua, E., Raymond, C., Horton, R. M., van den Hurk, B., AghaKouchak, A., Jézéquel, A., Mahecha, M.
D., et al.: A typology of compound weather and climate events, Nature reviews earth & environment, 1, 333–347, 2020.
Jacquemin, G., Vrac, M., Allard, D., and Freulon, X.: Estimating the return period of climate compound events using a non parametric bivariate Generalized Pareto representation. Submitted