Warming with Consolidation: The New Normal of Temperature Extremes

Understanding how temperature extremes evolve under climate change requires distinguishing between shifts in the typical magnitude of extreme events (location) and changes in their variability (scale). While time-varying scale has been proposed in extreme value analyses, estimating it reliably remains challenging: station-level approaches yield high uncertainty, and changes in scale are easily confounded with location shifts or absorbed by other model components.
We address this through a Bayesian hierarchical framework that pools information across a dense observational network while preserving spatial flexibility. Spatial structure is captured via a Matérn field implemented through the SPDE approach and through covariate effects, while temporal dynamics enter through random walks for location and flexible parametric structures for scale. This hierarchical sharing of information reduces uncertainty in scale estimation compared to single-station analyses. Comparing models with time-constant, linear, and nonlinear scale evolution allows formal testing of whether observed changes arise from location shifts alone. The tail shape parameter was explored but found consistently indistinguishable from zero, indicating that changes are governed by location and scale rather than by increasing tail heaviness.
Application to a century-long record (1916–2024) of annual temperature maxima from a dense observational network reveals a pronounced acceleration in location since the mid-1970s, accompanied by a systematic contraction in scale—a pattern we term "warming with consolidation." This defines a new normal for temperature extremes: summers that once would have been exceptional are now routine, occurring year after year with diminishing contrast between hot years and moderate ones. Meanwhile, reduced spatial variability means that extreme heat increasingly affects entire regions simultaneously rather than isolated areas. Paradoxically, while moderate extremes have become pervasive, the most exceptional events—those with high return periods—grow less steeply than location-only models would predict. The result is a climate where extremes are less surprising but more inescapable.
The framework is transferable to other regions and variables, providing a principled tool for characterising non-stationary extremes and informing climate adaptation.