An analysis of return levels of Valencia's 2024 extreme rainfall

The flooding in the Valencia region on October 29 was unprecedented. It caused over 200 deaths and impacted an area far beyond the boundaries of the official 500-year flood maps. Its vast scale has raised many questions about the predictability of such events. In many ways, this disaster is reminiscent of the 1999 Vargas tragedy in Venezuela[1].

In this study, we analyze the return levels of rainfall recorded by several pluviometers in the affected area to assess the likelihood of such an event occurring and how this probability changes after the event has been observed. We apply multiple techniques to evaluate their stability and robustness in estimating return levels. These techniques include: maximum likelihood fitting of extreme value distributions, regional frequency analysis, L-moments distribution fitting, Bayesian techniques for station data, Bayesian hierarchical models for the region, and stochastic generation.

Our preliminary results indicate that the magnitude of the event far exceeded the usual design values, which may explain the extent of the destruction in the affected area. However, the stability of predictions varies significantly across methods. Our findings highlight that representing design values as distributions rather than single values provides a clearer understanding of the uncertainties inherent in extreme value modeling.

Additionally, some of our results suggest that such an event alters expectations for future extreme events, necessitating a reassessment of risk levels for infrastructure across the entire region.

References:

[1] Coles, S., Pericchi, L., 2003. Anticipating catastrophes through extreme value modelling. Journal of the Royal Statistical Society Series C-Applied Statistics 52, 405–416.