Future increased risk from extratropical windstorms in northern Europe

Extratropical cyclones dominate European weather and are a major driver of insured losses, with single events exceeding €6 billion. Intense cyclones (windstorms) cause widespread damage and disrupt critical sectors. Historic examples include Lothar and Martin (1999) and Kyril (2007), each causing multi-billion USD losses. Winter storm frequency shows substantial interannual variability linked to the North Atlantic Oscillation, with peaks in the early 1990s and declines through 2014. Projections of future storm tracks and damaging winds contain uncertainties, especially across the North Atlantic and western Europe, though consistent signals include decreases over southern Europe and increases over the British Isles and parts of northwestern Europe. Studies suggest that in Northern Hemisphere winter, wind intensity of extreme storms may increase despite declines in overall storm numbers, with strong regional variations. Winds associated with individual cyclones could strengthen, and mesoscale features such as sting jets may intensify, consistent with robust projections of increased precipitation and latent heating in European extratropical cyclones. Prior assessments of potential future storm losses generally indicate increases over northwestern Europe, with a synthesis suggesting that a 2.5 °C warming could yield a 23% increase in European windstorm losses. Framing risk as a function of hazard, exposure, and vulnerability, the hazard depends on windstorm frequency and wind intensity, exposure on storm tracks/locations and population (as a proxy for insured assets), and vulnerability on adaptation to extreme wind speeds and other factors. This study investigates how these components evolve under future climate and socioeconomic scenarios to quantify changes in storm severity and risk across Europe.

Previous work documents high storm-related losses in Europe and strong interannual variability tied to large-scale climate modes like the NAO. Multi-model projections have shown regional patterns with decreased storminess in southern Europe and increased activity over the British Isles and northwestern Europe. Some studies find increasing intensities of extreme storms in winter despite fewer storms overall, while others report fewer cyclones associated with strong winds across the Northern Hemisphere but increases over NW Europe, highlighting strong regional dependence. Mesoscale wind extremes (e.g., sting jets) may intensify with warming, likely linked to increased cyclone precipitation and latent heating. Loss modeling studies using earlier-generation models or single-model approaches generally project increased windstorm losses in NW Europe; a synthesis estimates about 23% higher European windstorm losses for 2.5 °C warming. However, uncertainties remain due to model resolution, internal variability, scenario differences, and differing climate sensitivities across model generations.

Design: The study assesses changes between a historical baseline (1980–2010) and late-century future (2070–2100) winters (DJF) using eight CMIP6 GCMs under SSP2-4.5 and SSP5-8.5. A multi-model ensemble enables estimation of robustness and uncertainty. Analyses focus on an All-Europe domain (23.5°W–31°E, 35°N–71°N) and three subregions: NW Europe (45°N–71°N, 23.5°W–13°E), NE Europe (45°N–71°N, 13°E–31°E), and S Europe (35°N–45°N, 23.5°W–31°E). All data are regridded to 1.875° × 1.875° (the coarsest model resolution). Grid points with population density <1 person per km² are masked to focus on inhabited land.

Data: ERA5 reanalysis (six-hourly, 1980–2010) provides reference fields. CMIP6 models (ACCESS-CM2, BCC-CSM2-MR, EC-Earth3, KIOST-ESM, MIROC6, MPI-ESM1.2-HR, MPI-ESM1.2-LR, MRI-ESM2-0) supply historical and future (SSP2-4.5, SSP5-8.5) six-hourly 850-hPa winds and 10-m winds; the first available ensemble member is used.

Cyclone identification: A Lagrangian objective tracking algorithm (Hodges method) is applied to 850-hPa relative vorticity, smoothed to T42 to isolate synoptic scales. Tracks must satisfy lifetime >48 h, displacement >1000 km from genesis, and include at least one point within the All-Europe domain.

Storm footprints: For each tracked cyclone, near-surface (10 m) wind speeds within 5° of track positions and within a ±12 h window are attributed to the cyclone. At each grid point, the maximum cyclonic wind in the 24-h window defines the footprint value. This captures the area and timing where strongest winds typically occur.

Storm severity indices: Damage is assumed to occur when winds exceed a local threshold defined as the winter (DJF) 98th percentile of six-hourly 10 m winds; all local 98th percentiles below 9 m/s are raised to 9 m/s to avoid counting weak winds. Two indices are computed:

• METSSI (meteorological storm severity index): At each grid cell, METSSI_ij = (u_max/u_98 − 1)^3 for cyclones where u_max exceeds u_98. The total METSSI is the sum over grid cells and storms, accumulated over 30 winters, representing an annual exceedance-like aggregate.
• SOCSSI (socio-economic storm severity index): SOCSSI_ij = METSSI_ij × pop_ij, where pop_ij is the human population at the grid cell. Total SOCSSI is the domain sum; for plotting, indices are scaled to comparable magnitudes.

Adaptation vs no-adaptation thresholds: Two sub-scenarios are considered for future climates. No-adaptation (NAD): the historical u_98 threshold is used for future periods. Adaptation (AD): if a grid cell’s future u_98 exceeds the historical value, the threshold is raised to the future u_98, representing idealized adaptation (e.g., enhanced building codes); if the future u_98 is lower, the historical threshold is retained (no de-adaptation). This isolates the effect of evolving wind climatology on exceedances. Population exposure is varied using SSP-consistent gridded population projections (SEDAC) with base year 2000 and projection year 2090, approximately midpoints of the analyzed periods. The SOCSSI’s validity is checked by correlating top-storm SOCSSI with observed losses (R² = 0.46).

• Storm frequency: Overall winter cyclone frequency over Europe decreases relative to historical by −4% (SSP2-4.5) and −6% (SSP5-8.5), with robust decreases over Southern Europe and Northern Scandinavia, but increases (up to 0.8 and 1.2 cyclones per month for SSP2-4.5 and SSP5-8.5, respectively) in a 50°N–60°N band, notably Poland/Eastern Europe (SSP2-4.5) and the British Isles/Denmark (SSP5-8.5).
• Extreme wind thresholds (p98 of 10 m wind): CMIP6 models overestimate historical p98 versus ERA5; future changes show decreases over much of Europe with localized increases over Germany and Baltic coasts (SSP2-4.5) and a central Europe swath (SSP5-8.5). Changes align with track density shifts.
• Maximum cyclonic winds: Future distributions show fewer mid-strength storms in NW/NE Europe and little change in tails, implying p98 changes are driven by frequency shifts of higher-wind storms rather than increased maxima.
• METSSI (no adaptation): 30-year accumulated METSSI increases in Northern Europe and decreases in Southern Europe. For all Europe, model-mean increases are +11.2% (SSP2-4.5) and +43.7% (SSP5-8.5). In NW Europe, METSSI more than doubles under SSP5-8.5 (+129%). Results exhibit large model spread and sensitivity to BCC-CSM2-MR; excluding it yields a negligible Europe-wide increase (+0.42%) but still +13.3% over NW Europe.
• METSSI with adaptation: Raising thresholds to future u_98 substantially reduces severity. Europe-wide METSSI decreases by about 44% relative to the no-adaptation case; in NW Europe reductions reach ~60%, with minimal adaptation effect in Southern Europe where the 9 m/s floor often applies.
• SOCSSI (population-weighted, includes projected population): Europe-wide SOCSSI increases by +34.1% (SSP2-4.5, no adaptation) and +74.1% (SSP5-8.5, no adaptation). In NW Europe under SSP5-8.5, SOCSSI increases exceed a tripling (+226%), with less inter-model variability than METSSI due to urban population weighting.
• SOCSSI with adaptation: Adaptation reduces increases but does not eliminate them: Europe-wide changes are +7.6% (SSP2-4.5) and +40.5% (SSP5-8.5). Largest adaptation benefits occur over parts of the UK.
• Drivers of SOCSSI change: Population change dominates where SOCSSI increases, accounting for >50% of the increase in W/NW Europe; where SOCSSI decreases (S and E Europe), reduced METSSI is the main contributor. Even without population growth, METSSI increases alone would raise SOCSSI in W/NW Europe.
• Validation: SOCSSI correlates with actual losses for the top 13 historical storms with R² = 0.46, supporting its use as a risk proxy.

The study demonstrates that, despite a slight overall decline in European winter cyclone frequency, spatial shifts in storm tracks and extreme wind thresholds lead to increased windstorm severity and risk in northern and northwestern Europe, especially under higher warming. In NW Europe, METSSI increases are tied primarily to more frequent high-wind storms rather than stronger within-storm maxima. Incorporating idealized adaptation to rising extreme wind thresholds substantially mitigates meteorological severity but leaves significant residual socio-economic risk due to growing population exposure. Under SSP5-8.5, even with adaptation, SOCSSI remains more than double historical levels across NW Europe, indicating adaptation alone is insufficient to offset increased risk. Lower-emission pathways (SSP2-4.5) markedly reduce the projected increase in risk and the required level of adaptation. Model spread is considerable, underscoring the value of ensemble analysis and the importance of tracking both hazard and exposure. The findings align with previous assessments of heightened NW European windstorm risk and suggest that small-scale wind extremes (e.g., sting jets) not fully resolved at ~100 km resolution could further increase risk beyond estimates here. These results emphasize the need to integrate socio-economic factors with physical hazard projections when planning risk management and adaptation strategies.

Using a multi-model CMIP6 ensemble with objective cyclone tracking and storm severity indices, the study provides robust evidence of increased windstorm risk potential across northern and northwestern Europe by late century, strongest under high emissions. While adaptation to higher extreme wind thresholds can significantly reduce meteorological severity, population growth sustains large increases in population-weighted risk, particularly in NW Europe. Mitigation via lower emissions scenarios substantially curbs future increases in risk and reduces the adaptation burden. The work advances loss estimation by combining hazard, exposure, and idealized adaptation, offering a baseline for insurers, planners, and policymakers. Future research should employ higher-resolution and convection-permitting models to capture mesoscale wind features, incorporate more detailed vulnerability and land-use changes, assess serial clustering impacts, and deepen attribution of model spread and nonlinear responses across scenarios.

• Model biases: CMIP6 models overestimate baseline cyclone frequency in parts of SE Europe and overestimate p98 wind speeds, especially near coasts and mountains. Results show large inter-model spread; some metrics are sensitive to specific models (e.g., BCC-CSM2-MR).
• Resolution constraints: Typical ~100 km resolution may underrepresent small-scale high-intensity wind features (e.g., sting jets), potentially underestimating risk.
• Surface wind modeling uncertainties: Prior generations showed sensitivity of surface winds to surface property changes; although no strong evidence of this affecting results here, uncertainty remains.
• Idealized adaptation: Assumes perfect adaptation to increased wind thresholds and no de-adaptation when thresholds decrease; real-world adaptation is heterogeneous and path-dependent.
• Exposure and vulnerability simplifications: Population is used as a proxy for insured assets; does not account for differing national/regional vulnerabilities, building standards, land-use change, or asset values.
• Storm clustering and temporal dependence: Potential impacts of serial cyclone clustering on seasonal losses are uncertain and not explicitly modeled.
• Scenario and internal variability: Projections depend on SSPs, population scenarios, internal climate variability, and model climate sensitivity; nonlinear responses across warming levels may occur.