A systematic exploration of the relationship between synoptic dynamics and Euro-Mediterranean extreme rainfall in a changing climate

Few classes of weather events have more acute or dramatic impacts on Europe than extreme rainfall. In a warmer world, both thermodynamical and dynamical factors will interact to alter European rainfall; both its average and extreme characteristics, and at both seasonal-mean and synoptic timescales. Understanding these changes is societally vital and yet complex, as we are faced by the challenge that CMIP6 class models are not able to resolve rainfall directly. Meanwhile, kilometre scale simulations are either limited in geographical extent, simulation length or both, and have not yet been extensively validated from a physical perspective.  Indeed, developing a coherent, continent-wide perspective on the physics of rainfall is complicated by the fact that the dynamics of rainfall vary tremendously between seasons and between regions. As just a few examples of this dynamical richness, Atlantic Rossby wave packets favour downstream cyclogenesis in the lee of the Alps and so bringing storms to Italy and the Aegean, southern deflections of the wintertime jet stream can direct atmospheric rivers to Iberia — and cutoff-lows can bring the same region torrential rain in Summer. 

 

In this presentation we adopt a 'dynamics-first' approach to understanding rainfall in CMIP6-class climate models, focusing on their flow-dependent biases in rainfall in order to understand their errors and assess the physical plausibility of their projections. To practically handle the dynamical richness of Euro-Mediterranean rainfall dynamics we use a flow-precursor approach, developed for weather forecasting applications, in order to systematically identify the circulation patterns that drive extreme precipitation across Europe and reduce them to scalar metrics. By doing so, we are able to distill the multi-faceted synoptic dynamics into a manageable, low-dimensional space. Using this novel approach, we explore the potential of bias correcting climate model rainfall using dynamically-aware AI methods and, additionally, compare the calibrated results to those obtained from convection-permitting regional simulations carried out over the Alps.