Event attribution aims to estimate the role of an external driver after the occurrence of an extreme weather and climate event by comparing the probability that the event occurs in two counterfactual worlds. These probabilities are typically computed using ensembles of climate simulations whose simulated probabilities are known to be imperfect. The implications of using imperfect models in this context are largely unknown, limited by the number of observed extreme events in the past to conduct a robust evaluation. Using an idealized framework, this model limitation is studied by generating large number of simulations with variable reliability in simulated probability. The framework illustrates that unreliable climate simulations are prone to overestimate the attributable risk to climate change. Climate model ensembles tend to be overconfident in their representation of the climate variability which leads to systematic increase in the attributable risk to an extreme event. Our results suggest that event attribution approaches comprising of a single climate model would benefit from ensemble calibration in order to account for model inadequacies similarly as operational forecasting systems.

We would like to acknowledge valuable discussions and feedback received from François Massonnet, Nathalie Schaller, Chloé Prodhomme, and Fraser Lott. This work was supported by the EUropean CLimate and weather Events: Interpretation and Attribution (EUCLEIA), funded by the European Union’s Seventh Framework Programme [FP7/2007-2013] under grant agreement 607085 and the ESA Living Planet Fellowship Programme under the project VERITAS-CCI. We are further indebted to the s2dverification (http://cran.r-project.org/web/packages/ SpecsVerification/index.html) and specsverification (http://cran.r-project.org/ web/packages/s2dverification/index. html) with which the calculations have been carried out. The synthetic hindcast generator has been implemented into s2dverfiction. No further data were used in producing this manuscript.

Peer Reviewed