Over the past decade there has been a burst of studies examining how anthropogenic emissions have altered the chance of specific recent extreme weather events.  Results of these analyses are intrinsically probabilistic in nature but they are rarely considered in the global probabilistic context, leaving open the possibility that the growing literature reflects a substantial publication bias reporting on “false discoveries”.  These studies also rarely use observational data in a way that can confront overall conclusions of an anthropogenic role.  Global studies using metrics of extreme weather lend only qualitative evidence against strong selection bias in these regional conclusions, because they use global measures whose probabilistic connection to the regional studies is not quatified.

In this presentation, we tackle this global-regional disconnect using a Bayesian decision theoretic approach for multiple testing  that flexibly controls false discovery.  The method is systematically applied to unusually hot, cold, and wet 5-day events over 237 regions of approximately 500 000 km2 size covering most of the world’s land area, using output from large ensembles of climate model simulations from the C20C+ D&A project covering multiple decades under historical and natural historical forcing scenarios.  We incorporate observed trends in frequencies of these extremes clustered at larger regional scales to provide observational constraints.  The approach produces maps that combine information about the evidence for or against a human influence with information about the magnitude of that influence.  Finally, we pool the 237 regions to examine the progression of categorical emergence of a human influence on extreme weather on this spatial scale.