Heat extremes in Western Europe increasing faster than simulated due to atmospheric circulation trends

The study addresses why Western Europe has experienced a much faster increase in summer heat and heat extremes than simulated by state-of-the-art climate models over recent decades. Despite global increases in extreme heat, Western Europe’s recent heatwaves (e.g., 2003, 2018, 2019, 2022) have been unprecedented in magnitude and frequency. Climate models generally do not capture the pace of increase in extreme temperatures (TXx) and mean summer temperatures, suggesting a mismatch between observations and simulations. The research aims to quantify the contribution of atmospheric circulation trends—particularly more frequent southerly flows—to observed warming of summer heat extremes, and to evaluate whether CMIP6 models reproduce these circulation-induced trends. Understanding this mismatch is critical for reliable projections and for informing adaptation to European heatwaves.

Multiple mechanisms have been proposed to explain amplified summer warming in Western Europe relative to global change: shifts in mean atmospheric circulation, aerosol forcing changes, and soil moisture depletion with related land–atmosphere feedbacks. For extreme heat, increases in the frequency and persistence of split midlatitude jet states and a weakened mean summer zonal circulation have been linked to about one-third of the amplified trend in heatwave intensity. Circulation patterns that favor heat—a dipole with low pressure over the Eastern Atlantic and high pressure over the Mediterranean extending into central Europe—have been highlighted. However, no robust increasing trend has been found in Scandinavian blocking associated with the 2018 heatwave, and reported changes in Rossby waves are sensitive to definitions. Summer temperature variability in Central Europe is large, complicating attribution. Overall, past work suggested a role for dynamics but with mixed evidence and uncertainties, motivating this study’s quantitative assessment of circulation impacts.

Data: ERA5 reanalysis and E-OBS observations for Europe; 273 CMIP6 simulations (with a subset of 170 simulations from 32 models providing necessary circulation fields for analogue analyses). Region of focus: Western Europe land area [-5°W–15°E; 45°N–55°N]. Metrics: summer (JJA) maximum of daily maximum temperature (TXx) and mean of daily maximum temperature (TXm). Trends expressed per global warming degree (GWD), where GWD is a centered 5-year mean global temperature anomaly referenced to 2022 (GWD=0 in 2022; negative prior to 2022). Primary method—Circulation analogue approach: Atmospheric circulation is characterized by 500 hPa streamfunction anomalies over a North Atlantic–European domain (approximately -30° to 20°E; 30°–60°N). For each summer day (1950–2022), analogue days with similar circulation are identified using anomaly correlation coefficients (ACCs), ensuring analogues are at least 6 days apart and excluding self-analogues. To isolate the dynamical contribution to temperature trends, daily temperatures are replaced by those from analogue days, which preserves circulation but shuffles thermodynamic timing. A thermodynamic correction is applied to account for non-stationary warming: for each circulation state X, a circulation-conditioned warming rate b(X) is estimated by regressing temperatures from the best ~1% summer analogues (~67 days, spacing ≥6 days; 99% have ACC>0.5, 65% ACC>0.7) against GWD. Temperatures are scaled to a reference warming level (2022) using T_s = T − b(X)·(GWD − GWD_ref). From analogue time series (constructed using up to the three best analogues), annual TXx and TXm are recomputed and regressed on GWD to estimate the circulation-induced (dynamical) trend and its confidence interval (±2 standard errors). Thermodynamic trends are obtained as residuals (total minus dynamical). Sensitivity tests reduced the analogue domain by 10° longitude and 5° latitude; dynamical trends generally remained significant (≈0.5–0.9 °C/GWD), except for reductions toward the NE corner, where they declined (~0.35 °C/GWD). Secondary method—Dynamical adjustment via ridge regression: Using z500 spatial fields over the same domain, a regularized linear model (ridge regression) predicts the circulation-driven component of TX variability. The model is trained on the 15 warmest days each summer during 1950–2021 at each grid cell (≈180 observations), using z500 anomalies with the global mean removed to reduce thermal expansion effects. Regional circulation-induced TX components are then averaged over Western Europe. This provides an independent estimate of circulation-driven trends, found consistent with the analogue method albeit slightly weaker. Model–observation comparison and significance: Trends were also estimated relative to time and to model ensemble means for sensitivity. A False Discovery Rate (FDR) multiple testing procedure assessed field significance of model–observation mismatches. Large initial-condition ensembles (e.g., ACCESS-ESM1-5, CanESM5, IPSL-CM6A-LR, MIROC6, MPI-ESM-1-2-R) were examined. The analogue technique was applied to 170 simulations with required circulation fields. Southerly Flow (SF) patterns associated with Western Europe extremes were identified and their frequency trends computed. Implementation details: Land masking at 0.5°×0.5° (E-OBS) was applied. Annual maxima/means were recomputed from analogue series. Confidence intervals used a 2-sigma rule for regression coefficients, accounting for the number of well-separated analogues.

- Observed trends: ERA5 and E-OBS show Western Europe TXx trends of about 3.4 °C per global warming degree (GWD), with local maxima >5 °C/GWD in northern France and Benelux. TXm trends are lower: ~2.4 °C/GWD (ERA5) and ~2.6 °C/GWD (E-OBS). The 20° longitude by 10° latitude Western Europe region has the highest global TXx trend for regions of comparable size between 75°S and 75°N. - Circulation-induced contribution: Using circulation analogues, the dynamical trend is generally positive and reaches ~1.5 °C/GWD in several areas. Averaged over Western Europe, dynamical trends are ~0.74 °C/GWD (0.26–1.21) for TXm and ~0.79 °C/GWD (0.24–1.35) for TRx; an alternative dynamical adjustment method yields a slightly weaker dynamical contribution of ~0.56 °C/GWD. The abstract summarizes that about 0.8 °C (0.2–1.4) of the ~3.4 °C/GWD TXx trend is attributable to circulation changes, linked to more frequent southerly flows. - Model–observation mismatch: Among 273 CMIP6 simulations, only 4 simulations from 3 models (ACCESS-ESM1, NorESM2-LM, KIOST-ESM) exhibit TXx trends larger than observed when averaged over Western Europe; observed TXx trends lie near the extreme upper tail (~98.9th percentile) of the CMIP6 distribution. In large ensembles, only ACCESS-ESM1-5 has a few members matching observed TXx trends, but it strongly overestimates TXm trends. - Dynamical trends in models: None of the 170 simulations analyzed with the analogue method reproduce the observed dynamical TXx trend over Western Europe; all members of three large ensembles (ACCESS-ESM1-5, IPSL-CM6A-LR, MPI-ESM-1-2-R) fall below the observed dynamical TXx value. The chance that the observed dynamical TXx estimate is drawn from the model population is <1%. - Thermodynamic contribution: Models broadly match observed thermodynamic contributions for TXm and tend to slightly underestimate TXx thermodynamic trends; thus, the mismatch largely arises from dynamics. - Southerly Flow (SF): Models simulate realistic mean SF frequencies (12.5–18%), but the observed rapid increase in SF frequency (+43% per GWD; 10–76% range) is only roughly captured by one model (NorESM2-LM) and is weaker in others. - Significance: An FDR multiple-testing framework indicates significant model–observation mismatch for TXx over Western Europe; no significant mismatch is found for TXm in that region.

The findings demonstrate that the accelerated warming of Western European summer heat extremes relative to global warming is strongly linked to changes in atmospheric circulation—particularly an increased frequency of southerly flow patterns that advect warm air and promote subsidence and radiative heating. CMIP6 models generally underestimate these circulation-induced changes, explaining much of the discrepancy between observed and simulated TXx trends. Thermodynamic warming is comparatively well captured. These results imply important implications for projections: if the mismatch stems from an underestimated forced circulation response to greenhouse gases, future model-based projections of Western European heat extremes may be too conservative. If the mismatch is due to underrepresented low-frequency internal variability, then the trajectory of future extreme heat remains deeply uncertain over years to decades. Either way, caution is warranted when relying on current model projections for heatwave risk assessment and adaptation planning in Western Europe.

This study quantifies the circulation-driven component of Western Europe’s rapidly increasing summer heat extremes and shows that CMIP6 models systematically underestimate the observed dynamical contribution, while thermodynamic components are broadly consistent. A substantial fraction (~0.8 °C per GWD) of the observed TXx trend (~3.4 °C/GWD) is attributed to more frequent southerly circulation patterns. As a result, most simulations fail to reproduce the observed TXx trends, and none replicate the observed dynamical TXx trends. These findings underscore the need to better understand and represent circulation responses and variability in climate models. Future research should clarify whether the mismatch arises from underestimated forced responses, insufficient low-frequency variability, or other systematic uncertainties (e.g., forcings, reanalysis homogeneity), and should improve methods to constrain and project regional circulation changes to enhance the reliability of heatwave projections and inform adaptation.

- Potential inhomogeneities and uncertainties in reanalysis products (e.g., circulation fields) could affect analogue selection and trend estimates. - Possible inaccuracies in external forcings (e.g., aerosol and land-use changes) may induce systematic model–observation mismatches. - The analogue method’s sensitivity to domain selection and analogue quality (ACC thresholds, spacing) can influence dynamical trend estimates, although sensitivity tests suggest robustness for most domain changes. - Ambiguity remains regarding the relative roles of forced circulation responses versus low-frequency internal variability in explaining the mismatch, leaving future projections uncertain. - Some textual inconsistencies/typos (e.g., variable labels) in reported metrics reflect source material; however, primary quantitative conclusions are robust.