Forecast-based attribution of the role of stratospheric variability in weather extremes

Abstract. Variability of the stratospheric polar vortex, particularly its dramatic breakdown during sudden stratospheric warming (SSW) events, has been linked to a number of surface weather extremes. However, attributing the role of stratospheric variability in a specific observed weather extreme, rather than an abstracted class of extremes, has proved highly challenging. Here we use an ensemble of subseasonal forecast simulations from 7 forecast systems participating in the Stratospheric Nudging and Predictable Surface Impacts (SNAPSI) project to carry out this task. By comparing the likelihood of extreme events in free-running forecasts to those with the zonal-mean stratospheric state nudged towards its observed or climatological evolution (while the troposphere is freely-evolving), we are able to calculate the changes in the risk and severity of extremes due to the occurrence, or non-occurrence, of an SSW. We focus on three case-study events: (i) the 2018 boreal SSW and subsequent Eurasian cold air outbreak and snowfall, (ii) the 2019 boreal SSW and subsequent North American cold air outbreak, and (iii) the 2019 austral near-SSW and subsequent Australian heat wave. Through an extreme value statistical analysis, we find in all three cases a significant stratospheric contribution to the risk of relevant weather extremes. In case (i), improving the SSW prediction by nudging as much as doubles the forecast risk of extreme Eurasian cold and UK snow. The differences in risk and severity between experiments nudged to the SSW and to climatology are relatively insensitive to the lead time before the cold air outbreak of case (i). By contrast, in case (ii) this difference only emerges at short lead times before the event, indicating a stratospheric influence on this event that is dependent on the tropospheric state. For case (iii) we find a stronger and more robust stratospheric impact on the severity of the Australian heat wave than on its risk, with the latter being highly sensitive to model bias. The methodology outlined here, including both the experimental design and the semi-parametric approaches for calculating risks, can be applied to attribute several other internal climate system drivers of extreme event risk.