Using a model comparison to support the interpretation of extreme event attribution

Abstract

Extreme event attribution is used to evaluate the role of anthropogenic climate change in the occurrence or characteristics of extreme events. To be most useful, event attribution should be interpreted in the proper context and one part of this is understanding when event attribution results are robust. We investigate some factors contributing to the robustness of event attribution, by evaluating the roles of model choice, spatial scales, and ensemble sizes. We compare six single-model large ensembles to determine for which regions and variables there is agreement on at least two-fold changes in likelihood. As attribution of increases in the likelihood of daily temperature extremes is widespread by a 1 °C global warming level, we focus on precipitation extremes. In some regions, low signal to noise contributes to a lack of agreement on at least two-fold increases in likelihood in the current climate, but more model agreement in a 3 °C climate. In other regions, however, model differences remain, even at larger warming levels. Understanding where and when models do or do not agree on changes in large-scale precipitation extremes can be used to inform the design and interpretation of the local-scale analysis of specific events.