Can we forecast the most extreme daily temperature within a season using the forecast seasonal mean temperature?

Extreme daily temperatures, whether hot days in summer or cold days in winter, have important societal impacts.  Such impacts can in part be mitigated with advance warning. However, current seasonal forecasts do not typically give information on the likelihood of daily extremes within the coming season, instead focussing on the seasonal mean climate. The first step in improving forecasts of the likelihood of extreme days is to understand the relationship between the seasonal mean and daily extremes. We examine reanalysis fields to show that for many regions of the Northern Hemisphere winter, the seasonal mean temperature and coldest day are more strongly correlated than random subsampling alone would suggest. We investigate the spatial variability of the seasonal mean/ extreme daily temperature correlation.

We show that in winter regions of high correlation overlap regions where known dynamical drivers of winter climate act on both the mean and extremes of the daily mean temperature distribution. In summer, however, the mean-extreme relationship is typically weaker than in winter and much less driven by large scale dynamical drivers.

Given this strong mean-extreme relationship in reanalysis, we examine whether the temperature of the most extreme day within a season can be inferred using a prediction of the seasonal mean climate from the Met Office Decadal Prediction System. We find significant skill over many ocean regions but more limited skill over land regions. However, in regions where skill is significant, forecasts primarily predict the long-term warming trend, rather than interannual variability. Our work offers hope for progress in future, since the dependence of the relationship between mean and extreme on underlying dynamical drivers in reanalysis suggests that should seasonal mean predictions improve, predictability of extremes would also improve in many regions.