The 2018 summer heatwaves over northwestern Europe and its extended-range prediction

Prolonged and severe heatwave and drought conditions occurred across northwestern Europe in summer 2018, with major socio-economic and environmental impacts. While anthropogenic warming contributed to the record-breaking heat and precipitation deficits, natural variability also played a key role. Prior work proposed several drivers for the 2018 European heat and drought, including persistent atmospheric blocking, subtropical ridging or tropical continental air intrusions, a strong stationary Rossby Wave-7 pattern, a predominantly positive summer NAO, Atlantic SST tripole patterns, and local SST and soil moisture anomalies. Although improvements have been shown for seasonal and short-term forecasts when these factors are represented, extended-range (2–4 weeks) prediction of the specific 2018 European heatwaves had not been addressed. Skillful sub-seasonal predictions of heatwaves are crucial for mitigation and preparedness, yet forecasting the onset, duration, and amplitude of specific events 3–4 weeks ahead remains challenging and likely depends on the predictability of slow-varying drivers such as the Madden–Julian Oscillation, persistent midlatitude regimes (e.g., blocking), and land–atmosphere interactions. This study aims to identify the dominant circulation drivers of the mid-July to early August 2018 heatwaves over Scandinavia (SC) and Western Europe (WE) and to assess their extended-range predictability using real-time S2S multi-model ensemble forecasts.

Most European summer heatwaves are associated with atmospheric blocking, and other persistent regimes (e.g., subtropical ridge/Atlantic Low) have been linked to Western European heat extremes. The positive phase of the summer NAO, reflecting jet stream and storm track variability, has been connected to warm and dry conditions over northwest Europe and exhibits spatial similarity to a European summer blocking regime. Variations of the Azores High and Icelandic Low, and North Atlantic SST patterns (including the SST tripole) as well as tropical Atlantic convection, have been implicated in European heatwave development, offering potential predictability sources. For 2018, studies reported contributions from persistent blocking, subtropical ridge/tropical air intrusions, a strong hemispheric wave-7 pattern, a positive summer NAO, the Atlantic SST tripole, and local SST/soil moisture. Seasonal and short-term forecast improvements were achieved when these oceanic and land surface anomalies were represented. However, explicit extended-range prediction of the 2018 European heatwaves and the dynamic roles of blocking and related regimes at daily to weekly timescales had been lacking.

Data: ERA-Interim (ERA-I) reanalysis (6-hourly, daily means) of 2-m temperature (T2) and 500-hPa geopotential height were used, with anomalies relative to 1999–2010. Teleconnection indices NAO and SCA were obtained from NCEP/CPC (rotated EOF of 500 hPa height anomalies). Real-time subseasonal-to-seasonal (S2S) forecasts from 11 models (ensemble sizes 4–51; total 295 members) formed a multi-model ensemble (MME). Model daily anomalies used each model’s July–August reforecast climatology (1999–2010). Forecasts initialized on nine dates (14, 21, 28 June; 5, 12, 19, 26 July; 2, 9 August 2018). Lead weeks defined as days 1–7 (LW1), 8–14 (LW2), 15–21 (LW3), 22–28 (LW4). Target analysis window: early July to mid-August 2018. Heatwave metrics: Using daily T2, computed (i) daily anomaly aT2 (relative to monthly climatology), (ii) deviation from 90th percentile (T2-P90; monthly 90th percentile based on 1999–2010), and (iii) Excess Heat Factor (EHF) combining heat severity (EHI_sig: 3-day mean minus reference) and acclimatization (EHI_acc: 3-day mean minus preceding-30-day mean or monthly climatology). EHF = EHI_sig × max(1, EHI_acc). Here daily mean T2 was used; monthly climatology used in place of 30-day mean (no practical difference observed). aT2, T2-P90, and EHF represent warm-spell duration, heatwave occurrence, and heatwave severity, respectively. Circulation indices: - Atmospheric blocking (BL): Objective index based on meridional gradients of 500-hPa height (GHGS and GHGN) computed at latitudes φ_s=40°N, φ_0=60°N, φ_N=79.5°N with small ±Δ latitudinal offsets. A longitude is locally blocked if GHGS>0 and GHGN<−5 m/°N for at least one Δ; the European sector (0°–40°E) is blocked if ≥3 adjacent longitudes are blocked. GHGS strength (denoted BL) sufficed to represent summer 2018 blocking activity. - Atlantic Low (AL): Center-of-action index following Cassou et al. (2005): daily difference of normalized 500-hPa height anomalies between 3°–9°E and 18°–24°W within 48°–54°N. - Azores High (AZH): Regional mean normalized 500-hPa height anomalies over 24°–54°W, 30°–45°N (adapted to summer AZH center variability). - NAO: Meridional gradient method: zonal-mean difference of normalized 500-hPa height between 36°–45°N and 66°–75°N over 70°W–10°W; chosen to avoid station noise and nonstationary EOF patterns. Statistical analyses: For reanalysis (July–August 2018): bivariate and partial correlations among indices and with T2-P90 over SC and WE; multiple linear regression (MLR) of regional T2-P90 onto standardized BL, AL, AZH, NAO to assess contributions (reporting coefficients, beta weights, adjusted R², significance). For forecasts: Weekly distributions analyzed using box–whisker quartiles for each lead week and region; counts of models with negative Q1 or Q2. Forecast members stratified by sign combinations of SC/WE metrics: SpWp (SC+, WE+), SpWn (SC+, WE−), SnWp (SC−, WE+), SnWn (SC−, WE−). Proportions computed per week and lead. Composite circulation/temperature anomaly patterns computed for each group and lead. Additional data-binning: forecasts stratified into quartiles by magnitude of aT2, T2-P90, or EHF for SC-based and WE-based binning; corresponding distributions of BL, AL, AZH, NAO examined to assess monotonic relationships with heat metrics.

- Observed drivers: Persistent blocking dominated July–early August 2018 over Europe with positive 500-hPa height anomalies centered over northern Scandinavia. Regional warm spells from 12 July to 8 August showed distinct evolutions over SC (double-peaked) and WE (slower rise with persistence beyond 8 August). Heatwave durations based on exceedance of 90th percentile thresholds were 21 days (SC) and 17 days (WE). - Regression patterns: Standardized BL and AL indices yielded distinct patterns resembling SC and WE warm anomalies, respectively. BL and AZH were associated with 1.5–3 °C per 1σ local positive T2 anomaly over SC; AL with >2 °C per 1σ over WE. Standardized NAO regressed onto daily anomalies showed widespread cool anomalies over Europe in this period. - Interdependence among indices (Table 1): Significant positive bivariate correlations R_BL-AL and R_BL-AZH (>0.4). When controlling for other indices, AL and AZH exhibited a significant negative partial correlation R_AL-AZH(BL,NAO). NAO showed no direct effect on relationships among BL, AL, AZH. - Relationships with heat metrics: Over SC, bivariate correlations with T2-P90 were strong and positive for BL (0.803), AL (0.497), and AZH (0.392); NAO weakly negative (−0.221). Third-order partial correlations controlling other indices remained significant and positive for BL (0.664) and AL (0.352); AZH became insignificant, and NAO showed a weak significant negative relationship (−0.390). Over WE, T2-P90 correlated strongly with AL (0.756) and weakly with BL (0.212); partial correlation remained strong for AL (0.717) while others were insignificant. - Multiple linear regression of T2-P90 (July–August 2018): Adjusted R² = 0.72 (SC) and 0.57 (WE), both p<0.001. Significant predictors: SC—BL (+), AL (+), NAO (−); WE—AL (+). Per 1σ increase: T2-P90 rose by ~1.6 °C for BL over SC and ~1.8 °C for AL over WE. Using only BL (SC) or AL (WE) explained ~64% and 57% of variance, respectively. - Causal interpretation: Positive BL and AL phases produced anomalous ridging over SC and WE, respectively, consistent with local warming. AZH had no direct linear contribution to SC heat anomalies once BL effects were controlled; its apparent influence arose from covariance with BL. NAO’s daily variability exerted a weak negative effect on SC when other regimes were held constant. - Forecast performance (S2S MME): - aT2 (warm-spell) evolution over SC and WE captured up to ~3 weeks lead; lead-time increase reduced amplitude skill. Lead week 4 still showed elevated warm-week probabilities but poor week-to-week discrimination. - Heatwave occurrence (T2-P90) and significance (EHF) poorly predicted; valid signals mainly in peak weeks with 1–2 week lead (SC ~2 weeks; WE ~1 week). - Circulation indices: BL predicted 1–2 weeks ahead during its mature phase (19 Jul–1 Aug); AL predicted 2–3 weeks ahead near its mature phase (26 Jul–8 Aug). AZH persistence well captured; NAO positive persistence captured but contributions to warm spells unclear. - Stratified composites: SpWp forecasts associated with broad-scale blocking and regional warmth; SpWn linked to blocking plus upstream trough (SC warmth only); SnWp associated with subtropical ridge (WE warmth only); SnWn with troughing (cool). For EHF and T2-P90, positive BL and AL phases were preferential for SC and WE heat, respectively. - Quartile-based binning: SC-based bins showed monotonic increase of BL with higher aT2/EHF across lead weeks; WE-based bins showed monotonic increase of AL with higher aT2/EHF. AZH showed no direct monotonic relation for SC; a negative relation with WE heat metrics. NAO showed no clear monotonic relation in forecasts. - Overall: BL was the dominant driver for SC heatwaves; AL the dominant driver for WE; NAO and AL were secondary over SC with opposite signs; AZH showed no direct linear linkage. MME captured warm-spell evolution but struggled with event occurrence/severity at longer leads.

The study addressed the research questions by isolating dynamic circulation regimes that modulated the 2018 northwestern European heatwaves and by evaluating their sub-seasonal predictability. Over Scandinavia, persistent blocking (BL) provided the primary ridging and subsidence conducive to prolonged warming, with AL contributing positively and NAO exerting a modest negative daily-scale influence when controlling for other regimes. Over Western Europe, the Atlantic Low (AL) pattern emerged as the principal driver, consistent with past associations between WE heat and subtropical ridge/AL regimes. The interdependence among BL, AL, and AZH—positive BL–AL and BL–AZH correlations and negative AL–AZH partial correlation—clarifies how large-scale circulation co-variability can shape regional heat anomalies and complicate attribution if indices are considered in isolation. From a prediction standpoint, results show that extended-range skill for specific heatwave events is primarily limited by the ability to predict BL and AL regime amplitudes and timing. The MME reproduced warm-spell evolution 2–3 weeks ahead but had limited skill for heatwave occurrence and severity except near peak intensity, reflecting forecast amplitude degradation and large ensemble spread. Despite the prevalent positive summer NAO on monthly timescales, daily NAO variability did not provide a positive predictor for event-scale heat; thus, relying on NAO alone is insufficient for extended-range heatwave warnings. The binning and composite analyses further demonstrated that stronger positive BL (SC) and AL (WE) are associated with higher heat severity, providing actionable indicators for regime-based sub-seasonal outlooks. Enhancing model representation of regime transitions and short-term co-variations between AZH and NAO may extend lead times for predicting BL/AL and, consequently, heat extremes over Europe.

This work identifies the key circulation drivers of the mid-July to early August 2018 heatwaves over northwestern Europe and evaluates their extended-range predictability. The persistent blocking regime (BL) was the dominant contributor over Scandinavia, with AL and NAO as secondary contributors of opposite sign, while AL was the major contributor over Western Europe. AZH showed no direct linear linkage to the heatwaves once interdependence with BL was accounted for. Multiple linear regression using four indices explained 72% (SC) and 57% (WE) of daily T2-P90 variance, and per-1σ changes in BL (SC) and AL (WE) increased T2-P90 by about 1.6–1.8 °C. The S2S multi-model ensemble captured warm-spell evolution up to about 3 weeks but struggled to predict heatwave occurrence and severity except near peak intensity. BL and AL were predictable about 1–2 weeks in advance during their mature phases, underpinning successful warm-spell forecasts; NAO and AZH variability was captured but did not directly aid warm-spell prediction. Future work should focus on improving the simulation and prediction of regime transitions and persistence (especially blocking and AL), refining model representations of short-term co-variations among AZH and NAO, and integrating oceanic and land-surface anomalies to enhance sub-seasonal predictability of European heatwaves. Expanding analyses across multiple summers and events would help generalize these findings and quantify model-dependent uncertainties.

- The attribution relies on linear statistical relationships (correlation, partial correlation, regression) and cannot fully demonstrate underlying dynamical causality; evidence for detailed dynamical processes is limited. - Forecast skill for heatwave occurrence/severity degrades rapidly with lead time; valid predictions were mainly during mature phases with short leads (1–2 weeks), indicating large model spread and uncertainty at longer leads. - Stratification by sign and quartile-based binning provides probabilistic associations but may not capture nonlinearities or compound drivers. - NAO and AZH contributions to warm spells remained unclear in forecasts despite being well captured individually, suggesting model deficiencies in representing regime co-variability. - The analysis centers on a single season/event (July–August 2018) and two regions (SC, WE), which may limit generalizability. - Anomalies and thresholds were referenced to 1999–2010 climatology; different baselines could slightly alter thresholds/magnitudes.