Targeted weather regimes identify circulation patterns behind Western European summer heat extremes and trends

Julianna Carvalho Oliveira, Fiona Spuler, Marlene Kretschmer

Western European heat extremes have intensified in recent decades, with their rate of warming
outpacing the global mean. Against this general human-induced warming trend, understanding the
circulation patterns that drive such heat extremes is crucial. Weather-regime (WR) approaches
have been widely used to characterise large-scale circulation variability; however, conventional
classifications are not optimised to identify the dynamical drivers of extremes. Here we apply a
novel targeted machine learning-based approach, the regression mixture-model variational
autoencoder (RMM-VAE), to characterise summer weather regimes relevant for Western European
temperatures. We compare its performance against the standard, non-targeted k-means approach
and find RMM-VAE to yield dynamically coherent regimes while being more informative of
(extreme) temperatures in the target region.
Our analysis identifies a southerly-flow regime that accounts for the vast majority of heatwave
days, while k-means disperses them across multiple regimes. Moreover, the seasonal frequency of
this impact-relevant regime, combined with global mean temperature, explains a large fraction of
interannual variability in both mean (R² = 0.84) and extreme summer temperatures (90th
percentile; R² = 0.65), with predictive skill persisting out-of-sample tests. Finally, this simple
regression model allows us to attribute 34% of total summer warming in Western Europe and
about 70% of the observed “excess” warming (relative to the global mean) to an observed increase
in the identified southerly flow circulation patter, which we quantify to be largely forced.
Our results demonstrate that targeted weather regime approaches can sharpen the link between
circulation and surface extremes, offering attribution of regional warming. Furthermore, the
identified regimes provide interpretable predictors with potential for improving seasonal forecasts
and climate risk assessments.