Severe Convective Storms in a Changing Climate: Uncertainties in risk views across North America and Europe

Abstract

Severe convective storms (SCS) have become a major, recurrent source of insured catastrophe loss, particularly in North America and, increasingly, Europe. Observed loss growth is driven primarily by socio-economic trends, including growth and concentration of insured exposure, changes in vulnerability and loss-cost inflation; a distinct climate-change signal has not yet been robustly isolated in historical insured-loss records. At the same time, anthropogenic warming is altering the physical ingredients that support SCS, including atmospheric moisture, instability, heavy precipitation intensity and, in some regions, environments favourable to organised convection. Robust assessment remains constrained by short and inhomogeneous hazard records, reporting and detection biases, changes in observing systems and claims practices, and limited event-level observations of hail, tornadoes and damaging convective winds.

We synthesise current evidence on historical trends and future changes in the principal SCS sub-perils relevant for catastrophe modelling: hail, tornadoes, damaging straight-line winds including derechos, and SCS-related extreme precipitation. In addition to observed hazard and loss records, we evaluate reanalysis-derived environmental proxies and convection-permitting simulations commonly used to infer climate-conditioned SCS risk. Proxy metrics are useful for identifying regions and seasons where SCS-favourable environments may change, but they remain imperfect substitutes for loss-relevant event footprints, as they do not reliably resolve storm mode, intensity distribution, spatial dependence, temporal clustering, or sub-peril co-occurrence.

Looking ahead, projections suggest increases in the frequency of favourable environments and growing evidence that the most intense SCS events are likely to become more severe under continued warming, even where frequency changes remain uncertain. Compound hazards, such as hail, tornado and other damaging winds, and extreme rainfall co-occurring within single convective systems, further underscore the need for integrated hazard modelling. While current catastrophe models increasingly combine hail, tornado, and straight-line wind hazards in data-rich regions such as the United States, important methodological and regional limitations remain, including in Europe where observations are more heterogeneous. Incorporating pluvial flooding associated with SCS adds further complexity, as translating intense precipitation into losses requires coupling with hydrological approaches that estimate runoff, water depth, and flow characteristics.

Advancing SCS risk assessment will require improved observations, more consistent event-level data, better representation of compound hazard behaviour, and continued development of integrated, decision-useful modelling frameworks. These advances are critical for model validation, risk pricing, underwriting, infrastructure adaptation, and climate-resilient planning.

Key points

1. Rising Losses, Exposure, and Vulnerability: Insured losses from severe convective storms have increased significantly over the past decade, driven primarily by socio-economic development, exposure growth, vulnerability, and loss-cost inflation, with climate change acting as an additional risk driver.
2. Multi-Hazard and Compound Risk: SCS involve multiple interconnected hazards, including hail, tornadoes, damaging winds (i.e. straight-line winds and derechos), extreme precipitation, and resulting pluvial flooding, whose co-occurrence, clustering, cascading impacts, and dependencies remain challenging to represent in catastrophe models.
3. Observational and Modelling Uncertainty: Sparse and inconsistent observational data, combined with limitations in numerical models, proxy-based approaches, stochastic event sets, and synthetic event generation, remain major obstacles to robust trend detection, model validation, and forward-looking risk assessment.

1. Introduction

Severe convective storms (SCS), including hail, tornadoes, damaging extreme winds (i.e. straight-line winds and derechos), and extreme precipitation, have become some of the most financially disruptive weather perils worldwide, accounting for tens of billions of dollars in annual insured property losses (Swiss Re, 2026). In the insurance industry, SCS are often regarded as high-frequency, low-impact events, although in recent years SCS have emerged as a dominant driver of insured catastrophe losses, particularly in North America, where they currently account for the majority of such losses. As shown in Figure 1, annual insured losses from SCS have increased markedly over the past three decades in North America and Europe, with Europe showing a clear rise in recent years. Recent SCS events in France (2022) and Italy (2023) alone resulted in claims reaching the high single-digit billions of euros (Swiss Re, 2023; Münch, et al., 2024). In the U.S., annual SCS losses have climbed beyond US$ 10 billion every year since 2010 (Aon, 2023). Notably, hailstorms account for 50–80% of the insured losses attributed to SCS, highlighting their significant role within this peril (Swiss Re, 2023; Banerjee, et al., 2024).

Figure 1: Annual insured property losses from severe convective storms in North America and Europe (1994–2025), shown in US$ billions and adjusted to 2025 values for inflation. Data sourced from Swiss Re.

While climate change has clearly intensified many weather-related hazards, its influence on SCS is more complex and still under investigation (IPCC, 2021). To date, the sharp rise in SCS-related losses has been driven primarily by socio-economic factors, including increasing property values, higher repair costs, and continued development in areas exposed to convective storms (Banerjee, et al., 2024; Münch, et al., 2024; Swiss Re, 2026). Nevertheless, the upward trend is being amplified by a warming climate, which is intensifying the underlying thermodynamic conditions that fuel convective activity (Prein, et al., 2017; Allen, 2018; Rädler, et al., 2018). A key challenge for the insurance industry is that individual SCS events rarely trigger large-scale economic or insured losses, leading to potential underestimation of their long-term financial impact. However, the cumulative effect of frequent, smaller events, coupled with inflation in asset values and repair expenses, can erode profitability over time if not properly accounted for.

Moreover, conventional risk assessments have largely considered SCS sub-hazards in isolation. This reflects a broader current paradigm in which SCS risk is often assessed through separate or only partially integrated hazard views, where hazard occurrence, exposure, vulnerability, and climate change are treated as distinct modelling components (Mitchell-Wallace, et al., 2017; van Ederen, et al., 2025). However, SCS comprise multiple interacting hazards with distinct physical drivers and impacts (Table 1), and there is growing recognition within the insurance and resilience community that these events are often compound in nature rather than driven by a single hazard. Hail, tornado, damaging winds, and intense rainfall frequently co-occur within organised storm systems, producing combined impacts over contiguous regions. In data-rich regions such as the United States, where the observational basis and scientific understanding of hail, tornadoes, and damaging winds are comparatively advanced, catastrophe models increasingly represent these sub-perils in combination, including aspects of their interdependence. In other regions, including parts of Europe, such integrated modelling is less mature, reflecting more limited and heterogeneous observational records and greater uncertainty in the underlying hazard representation.

A particular challenge is the treatment of SCS-related extreme precipitation. Unlike hail, tornadoes, and damaging winds, intense rainfall cannot be translated directly into losses without hydrological or pluvial flood modelling to estimate runoff, water depth, and flow characteristics. Integrating precipitation-driven or “wet” losses into SCS frameworks therefore requires a different modelling chain and introduces additional uncertainty. In an insurance context, this complicates the attribution of losses to a specific hazard, for example, whether a claim should be classified as hail damage, where hail compromises the roof, or as water damage resulting from rainfall entering the damaged structure.

At the same time, the highly localised and episodic nature of SCS, including clustering of multiple events within short time periods, is being addressed through rapidly evolving modelling approaches, but these capabilities remain relatively new and are still being refined and evaluated across the industry. Modern catastrophe models already incorporate high spatial resolution, large stochastic event sets, and methods to capture temporal clustering; however, important challenges remain. These include the consistent definition of events in relation to policy terms and conditions, the representation of highly localised phenomena such as downbursts and derechos, the accurate modelling of strong spatial gradients in hazard intensity, and the integration of precipitation-driven (wet) losses.

This underscores the need for a more holistic approach to SCS risk, building on existing modelling advances by further improving the representation of individual sub-perils, their interactions, and their dependence structures. Further progress will require integrated modelling frameworks, including large stochastic event sets and synthetic event-generation approaches, that can better capture co-occurrence, temporal clustering, cascading impacts, and exposure–vulnerability interactions. Such developments are required to support more accurate underwriting and robust risk management in the context of evolving socio-economic and climate conditions.

In the U.S., SCS are well-studied and integrated into underwriting practices, particularly across the Great Plains where tornado outbreaks and severe hail events are frequent and public awareness is high. However, recent geographic shifts in storm frequency and intensity, such as eastward expansions of tornado activity and growing hail risk in non-traditional regions, are beginning to challenge these established assumptions (Agee, et al., 2016; Gensini & Brooks, 2018). In Europe, the historical underestimation of hail and thunderstorm risk has left markets underprepared for the rising severity and financial impact of these events (Swiss Re, 2022; Münch, et al., 2024).

While future projections remain uncertain due to the small-scale nature and large interannual variability of SCS and limitations in current modelling tools, the scientific consensus is strengthening: the atmospheric environment is becoming more favourable to high-impact convective events except in regions where moisture limitations become dominant (Allen, 2018). At regional scales, however, these trends vary, with increases across the central and eastern United States, potential decreases along the lee of the Rocky Mountains, and broadly more favourable conditions across Europe (Tang, et al., 2019; Taszarek, et al., 2020b; Battaglioli, et al., 2023; Woods, et al., 2023; Thurnherr, et al., 2025; Battaglioli, et al., 2026). This evolving risk profile, combined with increasing exposure in urban, industrial, and infrastructure-dense areas, underscores the urgent need for targeted research, updated cat modelling frameworks, proactive adaptation strategies, and enhanced data collection to inform underwriting, pricing, and loss mitigation. These challenges are compounded by sparse and heterogeneous observational records, which limit trend detection, model validation, and the robust estimation of rare but high-impact events. Here, we provide a review and synthesis of observed and projected changes in SCS-related hazards across North America and Europe, with a focus on how observational limitations, modelling uncertainty, compound hazard representation, and regional climate signals constrain present and future risk assessment.

2. Observational Limitations and Modelling Uncertainties

High-quality observations of hail, tornadoes, damaging winds (i.e. straight-line winds and derechos) are essential for validating and improving numerical weather prediction (NWP), natural catastrophe, and climate models. However, the short-lived and highly localised nature of these hazards makes them difficult to observe and reliably capture in observational datasets. This challenge is particularly pronounced for hail, as hailstones may melt before being reported and consistent measurements of their size are often lacking. While radar-derived hail estimates can partially address these gaps (Wendt & Jirak, 2021), they remain uncertain and require careful interpretation, as derived hail-size estimates can be affected by horizontal advection during fall, variability in hail-size distributions, complex scattering and attenuation effects, and the presence of mixed hydrometeor types (Ackermann, et al., 2024), particularly given limitations in radar coverage. In contrast, hazards such as tornadoes and derechos are often better documented, as they can be identified through damage signatures, such as impacts on vegetation and infrastructure, which help delineate their spatial extent. However, this reliance on indirect detection introduces additional uncertainties, particularly in linking observed damage to the underlying hazard intensity, compared to more directly observed variables such as precipitation (Goff, et al., 2021; Butt, et al., 2025). For large-scale wind patterns such as straight-line winds, observational records exhibit substantial inhomogeneities, including reporting biases and a tendency for human observers to overestimate wind intensity, limiting their usefulness for robust model evaluation and trend analysis (Edwards, et al., 2018; Taszarek, et al., 2020). In contrast, extreme precipitation associated with SCS is generally better captured by rain gauge networks and radar observations, but remains subject to uncertainties related to spatial representativeness, gauge density, and the difficulty of resolving short-duration, high-intensity convective bursts.

As a result, observational datasets are incomplete and affected by systematic biases. Apparent increases in reported events may reflect improved detection capabilities and reporting practices, particularly with the expansion of weather radar, rather than true physical trends. In addition, spatial biases linked to population density mean that events are more likely to be observed and reported in densely populated areas, which particularly holds for hail (Allen & Tippett, 2015c; Kunz, et al., 2024).

These limitations persist even in regions with relatively advanced reporting systems, such as the United States. While the Storm Prediction Center (SPC) provides a centralised database of severe weather reports, these records remain influenced by reporting biases, evolving observation practices, and technological changes. By comparison, Europe faces an even larger observational gap in hail and tornado reporting. Although the European Severe Storms Laboratory (ESSL) has made significant progress in expanding its database, report numbers continue to increase rapidly, and spatial inconsistencies, often aligned with national borders, remain evident. Historically fragmented data collection across countries has further limited the development of homogeneous long-term records, although recent efforts have improved data sharing and integration (Dotzek, et al., 2009; ESSL, 2025).

To complement these limitations, targeted field campaigns such as RELAMPAGO (Remote sensing of Electrification, Lightning, and mesoscale/microscale Processes with Adaptive Ground Observations; (Nesbitt, et al., 2021)) and ICECHIP (In-situ Collaborative Experiment for the Collection of Hail In the Plains; (Witze, 2025)) provide high-resolution, process-level observations that are critical for understanding storm dynamics, hail formation, and microphysical processes. These campaigns offer insights that cannot be derived from sparse and heterogeneous long-term datasets alone. However, sustaining such efforts remains challenging. Budget constraints affecting both operational weather services and atmospheric research programmes, particularly in the United States, are reducing the continuity and quality of atmospheric observations. Declines in routine monitoring, such as upper-air measurements from weather balloon launches, alongside reduced support for research infrastructure, limit the availability of high-quality data needed to characterise convective environments and evaluate models. While field campaigns can temporarily supplement missing data streams, they cannot replace continuous long-term observations.

These observational limitations directly translate into challenges for numerical modelling of severe convective storm hazards. At coarse resolution, convective processes are parameterised, precluding the explicit simulation of hazards such as hail and tornadoes (Maher, et al., 2018; Berthou, et al., 2020). Even in convection-permitting models, while storm structures such as supercells can be resolved, tornadoes themselves are typically not explicitly simulated and are instead inferred from favourable environmental conditions or storm-scale proxies (Cui, et al., 2026; Wang, et al., 2026). In the case of hail, representation depends on microphysical parameterisations or diagnostic tools, and without robust ground-truth data, such as impact sensor measurements or verified hailstone size distributions, both observational records and model outputs remain uncertain and poorly constrained (Adams-Selin & Ziegler, 2016; Jensen, et al., 2023). Similarly, straight-line winds and derechos remain challenging to represent accurately, as their intensity and spatial extent depend on storm-scale dynamics and cold-pool processes that are sensitive to model resolution and physical parameterisations (Gallus Jr. & Harrold, 2023; Prein, 2023).

To address these limitations, environmental proxies are widely used to infer the likelihood of severe convective hazards (Brooks, et al., 2003; Prein & Holland, 2018; Martius, et al., 2018; Battaglioli, et al., 2023; Straka, et al., 2025). These proxies employ indirect atmospheric indicators, such as thermodynamic instability (e.g. Convective Available Potential Energy, CAPE, and convective inhibition, CIN), freezing level height, and wind shear, to infer hail or tornado activity and characteristics. These proxies help bridge data gaps and can improve the accuracy of risk assessments and refine risk modelling. While environmental proxies enable the development of algorithms to estimate hazard probabilities across space and time, they are not without limitations. A high-probability environment does not necessarily result in severe convective storms, and uncertainties in historical atmospheric data can introduce artificial signals or trends. In addition, proxy-based estimates exhibit region- and climate-dependent biases, particularly in areas with complex topography or under atypical atmospheric conditions, and differentiating between specific hazard types remains challenging (Prein & Holland, 2018; Battaglioli, et al., 2023).

Proxy-based climatologies should therefore be interpreted as representations of hazard-favourable environments rather than direct records of hazard occurrence. They are particularly valuable for identifying large-scale spatial patterns, seasonal variability, and changes in environmental potential (Allen, et al., 2015), but they do not explicitly resolve key processes such as storm initiation, storm mode, microphysics, or local terrain effects that ultimately determine whether a specific hail, tornado, or wind event occurs. In contrast, explicit convection-permitting modelling can represent some of these processes more directly, but remains sensitive to model physics, resolution, and event-sampling limitations.

3. Compound Hazards: Challenges in the Insurance Industry and Resilience Applications

The limitations in observations and modelling of SCS directly translate into challenges for risk assessment, where hazard interactions and loss aggregation must be reliably represented. Historically, risk assessments, particularly in the insurance sector, often focused on individual sub-hazards of SCS, such as hail, tornado, damaging winds (i.e. straight-line winds and derechos), and heavy rainfall in isolation. In practice, however, SCS frequently generate multiple hazards within the same event or in close temporal succession (Mohr, et al., 2020; Berzina, et al., 2023; Mohamed, et al., 2024). For example, mesoscale convective systems (MCS), large, organised storm complexes that can span hundreds of kilometres, produce contiguous precipitation and strong winds over several hours to days. Particularly during their initiation, these systems can produce hail and tornadoes (Wang, et al., 2023). Here, “compound hazard” encompasses both the simultaneous occurrence of multiple hazards within a single convective system and the temporal clustering of events over periods of days. This distinction is especially important for insurance applications, where losses are frequently aggregated across defined time windows (e.g. 72-hour clauses) and attributed to a single event.

Despite growing recognition that compound hazards can amplify losses and produce more complex damage patterns, their representation in catastrophe models remains limited. While most current models in data-rich markets increasingly simulate hail and tornadoes, and in some cases straight-line winds, in combination, these capabilities are not equally mature across regions or modelling platforms. In particular, extreme rainfall is typically modelled separately within hydrological or flood-specific frameworks, because translating intense precipitation into damage requires a different modelling chain, including runoff generation and estimates of water depth and flow characteristics. This limits the ability to capture interactions between pluvial flooding and other convective hazards.

Current modelling approaches do not yet consistently represent the full chain of cascading impacts observed in real-world events, introducing structural uncertainty in loss estimation, pricing, capital assessment, and adaptation strategies. This uncertainty can manifest in both under- and over-estimation of losses. Models may underestimate risk by missing large compound events involving multiple sub-perils, but they may also overestimate losses if impacts are double-counted at the same location, for example by assigning separate full losses from flood and SCS within a single event even though total damage cannot exceed a full loss.

These limitations are compounded by the fundamental characteristics of SCS. Their highly localised footprints and short lifetimes place strong demands on model resolution and event representation. Small spatial shifts in simulated storm tracks can lead to disproportionately large differences in loss outcomes, particularly in densely exposed regions, making it difficult to robustly estimate the spatial distribution of damage. Although small spatial scales can in principle be sampled through sufficiently large stochastic event sets, this is computationally demanding and does not fully solve the problem. A key challenge is representing the hazard field in a way that is both sufficiently smooth for stable loss estimation and capable of capturing sharp local gradients, for example those induced by orography such as the Schwarzwald, the foothills of the Rockies, or the Appalachians. This “hit-or-miss” behaviour, illustrated in Figure 2, reflects the challenge of accurately aligning hazard footprints with exposure at the scales relevant for insurance applications, a problem that is particularly acute for SCS compared with larger-scale windstorms.

These modelling challenges directly propagate into loss uncertainty. This sensitivity is further amplified by heterogeneity in exposure and vulnerability, leading to large variations in loss outcomes even under similar hazard conditions, and often dominating the uncertainty in insured losses. While exposure is typically treated as a model input, uncertainties in its spatial resolution, completeness, and representation, such as building characteristics or asset values, can influence loss estimates. Vulnerability adds another layer of uncertainty, because for a given hazard intensity the resulting damage degree can vary substantially, producing a wide range of possible losses even when the hazard itself is well defined.

Figure 2: Spatial distribution of property insurance claim frequency (in %) across Germany resulting from large-scale weather events, such as an extratropical storm (left), and more localised events, such as a severe convective storm (right). White areas represent regions unaffected by the event, whereas colours from green to red represent progressively higher claim frequencies (GDV, 2025).

In addition, a further source of uncertainty arises from the representation of spatial and temporal dependence. Severe convective storms often occur in clusters or outbreaks, with multiple events affecting the same region over hours to days. Current modelling frameworks struggle to consistently reproduce this behaviour, particularly the dependence structure between events and the resulting accumulation of losses. Simplified event definitions or independence assumptions can lead to material misrepresentation of portfolio risk, especially for reinsurance and capital applications where aggregation effects dominate. In practice, losses from such clustered events are often combined under operational definitions (e.g. 72-hour clauses), yet models may not fully capture the underlying meteorological persistence that drives these sequences. This difficulty is compounded by observational inconsistencies, which limit the ability to robustly identify and characterise clustered events in the historical record.

More broadly, the short observational record and limited sampling of extreme events constrain the ability to robustly quantify tail risk. Synthetic event generation is therefore required to extend beyond the historical record, but the representation of rare, high-impact compound events remains highly sensitive to modelling assumptions, including the choice of environmental proxies, event-generation methods, and dependence structures. This introduces uncertainty not only in hazard frequency and intensity, but also in the spatial correlation of losses across portfolios. This is particularly relevant for compound hazards, as proxy-based approaches typically represent favourable environments rather than the co-occurrence, sequencing, and interaction of multiple hazards within individual events.

Together, these factors highlight that the modelling challenge is not confined to individual hazards but lies in the integrated representation of multi-hazard processes, spatiotemporal dependence, and exposure interaction. Addressing these limitations will require continued development of integrated catastrophe modelling frameworks, including large stochastic event sets and synthetic event-generation approaches, that better represent hazard interactions, co-occurrence, temporal clustering, cascading impacts, and dependence between SCS sub-perils. In model validation the emphasis is therefore no longer only on hazard occurrence, but it typically also includes event-day statistics, spatial correlation patterns, portfolio-level loss behavior and the sensitivity of results to event definitions and financial terms.

As warming creates a more favourable environment for intense weather events (Allen, 2018; Lepore, et al., 2021; Rasmussen, et al., 2020), the likelihood of compound or sequential hazards is expected to rise, placing growing strain on early warning systems, infrastructure resilience, and emergency preparedness. However, there remains significant uncertainty regarding where and how these changes will unfold, with substantial regional variability (IPCC, 2021). Managing this uncertainty will require not only improved hazard modelling, but also adaptive risk management strategies that explicitly account for a wider range of plausible outcomes, further challenging the validity of models calibrated on historical conditions.

4. Climate Change and Severe Convective Storms

We provide a synthesis of current scientific understanding regarding trends in hail, tornadoes, extreme precipitation, and damaging winds (i.e. straight-line winds and derechos) associated with SCS. Particular emphasis is placed on the challenges and uncertainties in detecting and attributing observed changes to climate change. In addition to direct hazard trends, we also examine changes in the broader environmental conditions that support SCS development, which are often used as proxies in the absence of consistent observational data. Finally, we provide an overview of future projections for these hazards.

4.1. Climate Change Impacts on Severe Convective Storms Environments

A warming of the atmosphere enhances its capacity to retain moisture by approximately 7% per degree of warming (Clausius, 1850), which in turn increases convective available potential energy (CAPE), a fundamental driver of deep convection and severe convective storms. Basic physical considerations and modelling studies confirm rising CAPE trends globally (Chen, et al., 2020) and across the central US (Prein, et al., 2017; Allen, 2018) and parts of Europe (Rädler, et al., 2018), particularly under warming scenarios approaching +3 °C (Thurnherr, et al., 2025). Seasonal patterns are shifting as well. In the southeastern U.S., for example, springtime CAPE is increasing more rapidly than summer values, suggesting a potential earlier onset of the severe weather season (Gensini & Brooks, 2018). Although changes in vertical wind shear are more uncertain, thermodynamic changes dominate and will likely increase the frequency of environments where high CAPE coincides with strong vertical wind shear (Gensini, 2021), a combination that favours the development of SCS. Taszarek, et al., (2020) project that such environments will become significantly more frequent across mid-latitude continents as global temperatures rise. However, how these changes in large-scale environments translate into changes in occurrence and intensity of SCS is uncertain. Convective storm environments differ regionally: the U.S. experiences more frequent and intense setups with higher CAPE, moisture, and wind shear, while Europe, particularly southern regions like northern Italy, shows higher low-level lapse rates and 0–3 km CAPE despite lower overall values. CAPE is generally a more reliable proxy for SCS in Europe, whereas in the U.S., additional dynamic triggers are often needed. These regional differences highlight the need to tailor risk assessments to local convective conditions (Taszarek, et al., 2020b).

Tropospheric warming also leads to rising freezing and melting level heights, especially in mid-latitudes, affecting hailstone growth and survivability. While higher freezing levels may cause smaller hailstones to melt before reaching the ground, stronger storm updrafts and deeper storm columns foster the formation of larger hail that can survive the fall (Berthet, et al., 2011; Brimelow, et al., 2017; Prein & Heymsfield, 2020; Raupach, et al., 2021). This dynamic is reflected in climate model projections, which show stronger convective transport and thicker warm-cloud layers in future summer storms (Prein & Heymsfield, 2020; Raupach, et al., 2021).

However, changes in the vertical temperature profile are also reshaping SCS environments. Climate change leads to enhanced warming in the mid- to upper troposphere, increasing atmospheric stability (Manabe & Wetherald, 1967; Prein, et al., 2017). At the same time, warming increases both CAPE and convective inhibition (CIN). The rise in CIN suppresses weaker convection, allowing CAPE to accumulate to larger values and favouring a shift towards fewer but more intense convective storms. As a result, weak to moderate convection is expected to decline, while strong convection becomes more frequent, particularly in regions such as the lee of the Rockies where both CAPE and CIN are projected to increase (Rasmussen, et al., 2020). Additionally, studies suggest that favourable atmospheric conditions for convection are becoming more frequent in regions that have historically seen fewer such events, both in general and during specific seasons, such as northern Central Europe, Northern Europe, and the northeastern U.S. (Glazer, et al., 2020; Koch, et al., 2021; Taszarek, et al., 2021c).

4.2. Past and Future Hail Trends

Climate change impacts on hail are still uncertain and will be regionally and size dependent. Generally, tropical and subtropical regions might see decreasing hail frequencies while some mid-latitudes regions might see decreases in small but increases in larger hail (Mahoney, et al., 2012; Dessens, et al., 2015; Brimelow, et al., 2017). Uncertainties in hail projections mainly stem from the complex interactions of multiple variables in hail-producing storms such as changes in CAPE, CIN, wind-shear, and freezing level height.

North America

In North America, hail trends are complex. Besides the challenges in observed hail events and using atmospheric proxies, the southern Great Plains and southern states of the U.S. further experience substantial year-to-year variability due to natural variability modes. Nonetheless, in parts of the northern Great Plains and localised regions of Texas and Oklahoma, analyses using different hail datasets have revealed positive trends over two decades (1994–2016), with statistically significant increases (Jeong, et al., 2020; Jeong, et al., 2021).

As noted above, due to the sparse and biased nature of hail reports, researchers often rely on favourable atmospheric conditions as proxy indicators, either directly or as inputs for models that estimate hail probabilities, to assess spatial and temporal trends. Battaglioli, et al., (2023) show, using a hail probability model, mostly modest trends in large (>2cm) and very large hail (>5cm) across North America, with isolated significant increases in the lee of the Rocky Mountains and across the Canadian plains, and some decreasing trends (not statistically significant) in the eastern Great Plains during the peak months. However, the findings contrast with studies focusing on key atmospheric ingredients for severe thunderstorms such as CAPE, Storm Relative Helicity (SRH), Large Hail Parameter (LHP) and Significant Hail Parameter (SHIP) showing mostly positive trends (statistically significant) particularly in the eastern plains (Tang, et al., 2019; Taszarek, et al., 2021; Koch, et al., 2021). In addition, differences in the reliability and physical representation of these parameters further complicate direct comparisons between studies and contribute to the overall uncertainty in trend assessments. Although seasonal differences in the trend exist, the relevant atmospheric conditions point towards more favourable conditions, particularly in the Great Plains, and potentially more and larger hail formation. Variations may also stem from different parameters and differing synoptic-scale environments, such as frontal systems, the Great Plains low-level jet, and smaller-scale convective cells, which influence hailstorm development in distinct ways, depending on the specific atmospheric parameters at play (Fan, et al., 2022). In addition to long-term trends, natural climate variability, particularly patterns such as the El Niño–Southern Oscillation (ENSO), can substantially influence the atmospheric environments that support severe convective storms, modulating their frequency and intensity from year to year. Studies have shown that hail occurrence is more likely during the cold phase of ENSO (La Niña) than during the warm phase (El Niño), primarily due to changes in the large-scale atmospheric environment (Jeong, et al., 2020; Jeong, et al., 2021; Koch, et al., 2021). Overall, while some indications point towards increasing trends over the recent decades, such as atmospheric conditions and statistical models, there is no consistent picture across studies; hence, large uncertainties remain, which was also stated by (Raupach, et al., 2021). In terms of future projections, high-resolution modelling tools such as HAILCAST project increased frequencies of very large hailstones (>5 cm) across the Great Plains and Midwest by mid-century (Brimelow, et al., 2017).

Europe

Observational evidence for hail trends across Europe is highly heterogeneous, with no consistent long-term signal emerging across regions. Punge and Kunz (2016) found no consistent trends in hail event frequency based on direct observations across Europe reviewing more than 400 articles up to 2015. Some studies indicate an increase in hail days in parts of central (Burcea, et al., 2016) and southeastern Europe (Simeonov, et al., 2009), Spain (Aran, et al., 2011), and Switzerland (Hohl, et al., 2002; Wilhelm, et al., 2024). In contrast, decreasing trends have been reported in Serbia (Ćurić & Janc, 2015) and Croatia over a 100-year time series, although the last three decades suggest a weak, non-significant positive trend (Blašković, et al., 2023), while evidence from Italy points to fewer but larger hailstones on average (Manzato, et al., 2022).

However, most of these assessments did not consider data after 2014, while hail reports have increased fivefold since 2010, mainly due to increased reporting rather than physical changes. Part of this variability is also linked to large-scale climate variability, with reduced thunderstorm activity observed during negative phases of the North Atlantic Oscillation (NAO) in France (Augenstein, et al., 2025). Furthermore, the reliance on relatively short records, often beginning in the 2000s and in regions with low event frequency, makes robust trend detection highly challenging.

Given these observational limitations, studies increasingly rely on environmental proxies to assess changes in hail risk. Despite these observational uncertainties, favourable atmospheric conditions have shown increasing trends over southern, central, and northern Europe, with decreases over eastern Europe, where increased convective inhibition leads to less favourable conditions (Taszarek, et al., 2021). Using a modelling approach simulating the last few decades (Battaglioli, et al., 2023) show increasing trends in the occurrence of hail (>2cm) and large hail (>5cm) over southern Europe, excluding Spain and Portugal, with a most pronounced trend in northern Italy. Looking ahead, projections of European hail activity remain complex and depend strongly on the modelling approach used. Projected shifts in European hailstorm activity based on environmental proxies are chiefly driven by enhanced convective instability stemming from higher moisture levels in the lower and middle troposphere (Mohr, et al., 2015; Piani, et al., 2005). However, hail environmental changes do not necessarily translate directly into changes in hail frequency, as shown by Thurnherr, et al (2025). Their simulations under +3 °C warming feature increases in CAPE over the Iberian Peninsula but an overall reduction of hail frequencies during the peak summer season in this region, likely related to increased melting and changes in storm structure, while increasing hail frequencies are found in central and eastern Europe.

Similarly, recent pan-European convection-permitting simulations indicate that, despite increasing convective activity, severe hail potential may decline overall due to higher freezing levels and changes in storm dynamics, although the likelihood of very large hailstones may increase, particularly in southern Europe. Together, these findings point towards a shift in hail characteristics rather than a simple increase in frequency. This suggests a shift towards fewer but more intense hail events, consistent with a transition to “warm-type” convective storms, characterised by an increased role of warm-rain processes relative to ice-phase microphysics under future climate conditions (Kahraman, et al., 2025).

Further storm-tracking analyses indicate substantial changes in hailstorm characteristics, including more frequent very large hail events (≥50 mm), an expansion of hail swath areas by approximately 15–30% and increases in associated precipitation and wind hazards. These results also point to a higher likelihood of compound hail–precipitation extremes, with intense rainfall following large hail events projected to become more frequent under future warming conditions (Brennan, et al., 2025).

4.3. Tornado Frequency and Track Shifts

North America

Multiple recent analyses of U.S. tornado records suggest that annual counts of reliably observed tornadoes, those rated F/EF1 and above, have exhibited slight, statistically insignificant declines, while the frequency and clustering of tornado outbreaks have shown a statistically significant upward trend (Brooks, et al., 2014; Tippett, et al., 2016b). Changes in regional tornado activity have also been observed, with a notable eastward shift away from the traditional “Tornado Alley” in the western Great Plains (Agee, et al., 2016; Gensini & Brooks, 2018). Specifically, Gensini and Brooks, (2018), report decreasing frequencies of both tornado occurrences and favourable environments from 1979–2017 in Texas, Colorado, and parts of Oklahoma, alongside increasing trends in other areas of the Great Plains, particularly in spring (MAM) over the central region.

Similarly, Agee, et al., (2016), identify a decline in tornado activity across the central Plains, especially during summer, and a concurrent increase during autumn from Mississippi to Indiana, suggesting a possible climate-driven spatial redistribution of tornado risk. Complementing this, Tippett, et al., (2016b), document an increase in both the frequency and severity of tornado outbreaks across the U.S. over the past half-century (1950s–2015). More recent analyses extending the observational record from 1960 to 2022 reinforce this dichotomy, showing long-term decreases in the number of tornado days, particularly during the summer months in the Southern Great Plains, alongside a seasonal shift in peak tornado probability from mid-June to late May, and an increasing tendency in the frequency of tornado outbreaks, especially in the Southeast United States across both warm and cool seasons (Graber, et al., 2024a).

Part of this variability can be explained by natural climate variability, as tornado activity is strongly influenced by ENSO, with pronounced impacts in the southern and central U.S. ENSO modulates tornado occurrence during winter and spring by altering the large-scale atmospheric environment, typically resulting in fewer events during El Niño and more during La Niña conditions (Allen, et al., 2015b; Graber, et al., 2024b). In addition, a recent study suggests that both internal climate variability and potential anthropogenic influences may contribute to these evolving patterns, although disentangling their respective roles remains challenging due to the relatively short and observation-based nature of tornado records (Graber, et al., 2024a).

When looking at climate projections, the assessment of future tornado risk becomes particularly challenging. Although global climate models lack the spatial resolution to directly simulate tornadoes, proxy metrics like the Significant Tornado Parameter (STP) are increasingly used to assess future tornado risk. Additionally, environmental indicators suggest a shift toward more favourable environmental conditions for tornadoes, along with potential changes in their spatial distribution (Diffenbaugh, et al., 2013; Tippett & Cohen, 2016; Gensini & Brooks, 2018; Woods, et al., 2023). Complementing proxy-based approaches, recent convection-permitting climate simulations focusing on supercells, the primary producers of tornadoes, provide further insight, suggesting increases in the frequency and intensity of these storms in future climates, particularly across the eastern United States, alongside decreases in parts of the Great Plains. These simulations also indicate a seasonal redistribution, with increased supercell activity in late winter and early spring and potential declines during the peak summer months, consistent with projected changes in thermodynamic and dynamic storm environments (Ashley, et al., 2023). Together, these findings indicate a shift not only in the spatial distribution of tornado risk, but also in its temporal concentration and event structure. However, projecting future tornado trends and changes in spatial patterns or clustering remain complex and uncertain.

Europe

Violent tornadoes are relatively rare in Europe, but when they do occur, they can cause severe damage. Data from the European Severe Weather Database (ESWD) confirm that no region in Europe is entirely free from tornadoes, and strong events have been documented across the continent, although the database also shows substantial underreporting in the Mediterranean and eastern Europe (Groenemeijer & Kühne, 2014). A combined analysis of observations and atmospheric conditions shows tornado occurrence rates are highest along the Mediterranean coast, particularly in Italy and the Adriatic region, with elevated values also found in the Po Valley and parts of northwestern France. Lower rates appear over Scandinavia, the Iberian Peninsula, and mountainous areas (Grieser & Haines, 2020). While this broad spatial pattern partly reflects the distribution of favourable SCS environments, tornado occurrence differs from other SCS hazards such as hail in being more strongly controlled by local dynamical conditions (e.g. wind shear), resulting in greater regional variability and distinct seasonality, peaking in summer across central and northern Europe, shifting to autumn and winter in the Mediterranean, and typically occurring in the late afternoon to early evening. These spatial patterns are consistent with findings by (Groenemeijer & Kühne, 2014), who observed a northwest-to-southeast decline in tornado reports across France and Germany, and an increase from northeast to southwest in Poland. While differences in tornado reporting rates across countries influence observed distributions, they do not fully explain these regional gradients.

Despite these insights, the current tornado report time series across Europe remains too sparse and inconsistent to support robust trend assessments. However, studies suggest that climate change is making the atmospheric environment increasingly favourable for severe convective phenomena, including tornadoes (Taszarek, et al., 2021).

4.4. Extreme Winds and Precipitation Changes

The scientific basis for how straight-line winds and derechos respond to climate change is very limited. However, observational evidence suggests that straight-line wind events have already intensified over recent decades, particularly during summer across the central United States (Prein, 2023). This intensification is linked to the steepening of moist adiabatic lapse rates, which enhance convective downdrafts and are primarily driven by the increased moisture-holding capacity of warmer air.

Looking ahead, modelling studies indicate that MCS-driven windstorms, including derechos, are likely to become more frequent, widespread, and intense across much of the central and eastern United States under future climate warming. These increases are projected to occur throughout the year, with particularly pronounced changes during early spring and summer, suggesting a potential extension and intensification of the MCS windstorm season (Kaminski, et al., 2024). More broadly, the response of straight-line wind events appears to depend strongly on the environmental conditions that support severe convective storms. For example, a modelling study over Australia (Brown, et al., 2024) highlights that changes in these environmental conditions may play a key role in shaping future wind hazard characteristics. As such, an increase in environments favourable for SCS is also likely to create more conducive conditions for straight-line wind and derecho events in the future.

SCS are also capable of producing extreme precipitation due to their intense updrafts, high moisture content, and organised nature (Doswell III, 2001). These storms can produce localised rainfall extremes on timescales of minutes to hours, creating the potential to exceed return levels of infrastructure design and leading to flash flooding. Extreme precipitation is frequently produced through the repeated formation of convective cells along outflow boundaries (i.e., "training" convection). Supercells, while more spatially confined, can also produce intense short-duration precipitation bursts due to efficient warm-rain processes and strong vertical motion.

There is high confidence that extreme precipitation associated with convective storms has increased over the past few decades and will continue to increase in a warming climate (IPCC, 2021). In line with physical expectations, rainfall intensity is found to increase at a rate of approximately 6–7% per degree Celsius (Clausius, 1850). Additionally, whether extreme precipitation can intensify at super-Clausius–Clapeyron rates, exceeding the increase expected from atmospheric moisture capacity alone, remains an active area of research (Da Silva & Haerter, 2025). Super-Clausius-Clapeyron scaling could occur due to changes in storm dynamics, such as updraft strength, mesoscale organisation, and storm motion, which amplify the thermodynamic effects such as increases in water vapour. The occurrence of compound extremes, such as severe winds, hail, and extreme rainfall within the same storm system, may also become more likely (Prein, et al., 2017; Fowler, et al., 2021).

5. Pathways to Better Risk Assessment and Preparedness

To better align risk assessment and preparedness with evolving SCS hazards, advances are needed across the full risk chain: observations, process-based modelling, catastrophe modelling, exposure and vulnerability assessment, building standards, and risk governance. These elements are interdependent: improved observations support model development and validation; improved models inform underwriting, planning, and adaptation; and better exposure and vulnerability data determine how hazard information translates into losses.

Enhanced Observational Data and Data Integration: Investing in denser, better integrated high-resolution observational networks, including improved radar coverage, particularly over sparsely populated areas, enhanced use of dual-polarisation radar, expanded lightning detection, and more surface-based observations, is essential for advancing our understanding of SCS. These rich, real-time datasets are critical for developing and training next-generation predictive models, building robust Nat Cat models, and validating climate projections. Equally important is the continued collection and documentation of SCS reports to build climate-length records that support long-term variability and change assessments. In this context, crowdsourced hail reports have emerged as a valuable observational tool. For example, since 2015, more than 250,000 hail reports have been submitted via the Swiss National Weather Service (MeteoSwiss) app, providing an unprecedented volume of ground-based observations. Using advanced spatiotemporal filtering methods, these reports can be refined to improve data quality and used to validate radar-based hail metrics such as the probability of hail (POH) and maximum expected severe hail size (MESHS) (Kopp, et al., 2024). While such comparisons confirm a general consistency between radar-derived estimates and ground observations, they also highlight substantial uncertainties, including relatively high false-alarm ratios and the need for recalibration of existing radar algorithms. These citizen-sourced data help bridge the gap between broad-scale radar detection and point-specific observations, strengthening both real-time monitoring and long-term climatological analysis (Barras, et al., 2019; Kopp, et al., 2024). Improved data sharing, standardisation, and interoperability between meteorological agencies, researchers, and insurers will be essential for translating observations into usable model inputs.

Convection-Permitting Simulations and Process Understanding: Models that can explicitly simulate deep convective processes and more accurately represent topography, land-surface inhomogeneities, and microphysical processes are invaluable for advancing our understanding of how severe convective storms are evolving under climate change. Computational and model advances are increasingly enabling larger regional ensembles and, in some cases, global convection-permitting simulations. Investment in model development is needed, particularly to improve the representation of convective processes, including hail microphysics. Additionally, sub-kilometre-scale grid spacings are likely needed to represent downbursts, whereas explicitly simulating tornadoes will remain out of reach except for short case study experiments. However, km-scale models can provide valuable proxy information for tornado genesis, such as low-level vorticity, which in turn can inform hazard assessments.

Catastrophe Modelling: Catastrophe models typically rely on event-based frameworks that combine hazard, exposure, and vulnerability components to estimate losses, making the representation of hazard interactions and dependencies critical. Incorporating a multi-hazard approach, supported by improved observations and convection-permitting simulations, would strengthen underwriting, pricing, and portfolio risk assessment. However, the challenge is not only to add further sub-perils, but to represent them in a way that is physically consistent, computationally efficient, and convergent for portfolio-level risk estimates.

Model convergence remains a key challenge. Balancing spatial resolution, event-set size, and the number of stochastic years requires trade-offs between computational efficiency and physical realism. Approaches range from highly explicit modelling to hybrid frameworks that approximate high-frequency attritional losses while focusing computational resources on low-frequency, high-impact tail events. Continued research into large stochastic event sets and synthetic event-generation approaches will be important for improving the representation of hazard interactions, co-occurrence, temporal clustering, and cascading impacts across multiple SCS sub-perils. Similarly, model validation is evolving and now includes event counts, footprint realism, spatial correlation, event clustering, and portfolio-level loss behaviour.

Urban Planning, Building Standards, and Vulnerability Reduction: Urban planners and regulators should also continue to update building codes, drainage standards, and land-use policies to account for evolving SCS risk. In some highly exposed regions, building standards have been strengthened in response to repeated losses, but these adaptations often lag behind changing hazard characteristics and exposure patterns. This is particularly relevant in newly exposed areas and older urban centres with vulnerable infrastructure, ageing roofs, and façades, which have contributed substantially to insured losses in the U.S. over recent decades (Sorber & Farney, 2025). Conversely, where hazard levels remain relatively stable, but exposure and vulnerability are increasing due to urban expansion, asset concentration, or changes in building materials, building codes and planning practices may not keep pace with evolving loss potential.

More broadly, changes in building materials, such as improved insulation or solar installations, often reflect climate mitigation goals but can increase vulnerability to SCS impacts, particularly hail. For example, the vulnerability of solar panels to hail damage varies depending on design standards (Zaino, et. al, 2024). Mitigation measures should therefore be better aligned with physical climate risk adaptation, and material changes, especially when driven by government programmes or subsidies, should be transparently communicated to insurers. Rapid changes in building stock can significantly alter vulnerability and loss potential and should be accounted for in insurance modelling, costing, and pricing.

Public–Private Partnerships and Risk Governance: Strengthening collaboration among scientific institutions, government agencies, the insurance industry, and local communities is essential for enhancing early warning systems, including wider use of automated protective alerts, such as messages prompting households to raise or retract external blinds ahead of hailstorms, supporting resilient infrastructure investments, and advancing innovative insurance solutions, such as parametric coverage tailored to specific SCS events. Such partnerships can also support storm-resilient building codes and dynamic risk analytics, both of which are essential for managing escalating SCS risk (Banerjee, et al., 2024).

6. Conclusion

The environmental conditions that support SCS development in the U.S. and Europe are becoming more favourable and are expanding into regions that have historically experienced little such activity. This shift is driven by thermodynamic amplification, changes in wind shear patterns, and an increase in the likelihood of compound hazards involving hail, tornado, damaging winds, and extreme precipitation. Consistent across many studies, there is growing agreement that, while frequency trends remain uncertain, the most intense SCS events are likely to become more severe in a warmer climate, with regional exceptions in some tropical, subtropical, and arid environments.

However, substantial uncertainties remain in both observed trends and future projections. Historical records alone are not sufficient for understanding SCS risk, given their inherent limitations such as observational biases or insufficient spatial coverage due to their small-scale nature. In the context of a changing climate, they are becoming increasingly inadequate, undermining the reliability of traditional hazard assumptions that underpin risk assessment and strategic planning. Furthermore, these challenges are even more pronounced when projecting future risks. Kilometre-scale models offer improved realism in simulating storm-scale processes and hail-producing updrafts compared to coarser models and add value not only for climate projections but also for creating realistic SCS climatologies that are less affected by country-specific reporting biases (Cui, et al., 2025). Yet significant uncertainties remain. These include how well models capture cloud microphysics, storm electrification, and aerosol–cloud interactions, all of which influence hail development. Often, hail is not explicitly represented in model microphysics, requiring the use of diagnostic tools like HAILCAST (Adams-Selin & Ziegler, 2016) to estimate hail sizes at the surface. Moreover, future storm environments will be shaped by regional changes in temperature, moisture, and wind shear, factors that vary across climate models and emission scenarios, contributing to scenario-dependent uncertainty (Púčik, et al., 2017; Rädler, et al., 2019; Thurnherr, et al., 2025). Exploring these uncertainties with kilometre-scale models is computationally intensive, limiting the extent to which they can be sampled. Consequently, projections of future SCS risk, particularly for hazards like large hail and tornadoes, may differ widely in their timing, intensity, and geographic distribution. Importantly, forward-looking hazard insights should therefore be interpreted with caution, as current evidence is generally more robust in identifying regions where conditions may become more favourable for SCS than in resolving how the underlying events themselves, such as storm mode, intensity, clustering, and compound impacts, will evolve.

Socio-economic drivers, including increasing exposure due to urban expansion, ageing infrastructure, the growing use of rooftop solar, and rising repair costs, are currently the primary contributors to increasing losses and are likely to further amplify the impacts of SCS. While the precise influence of climate change remains uncertain, it is expected to act as an additional contributing factor to future losses, albeit unlikely to be the dominant driver.

Future SCS risk assessment should continue to improve the representation of both individual sub-hazards and their interactions. While significant progress has already been made, including the integration of multiple perils such as hail and tornadoes in many catastrophe models, important gaps remain. In particular, improving the representation of co-occurrence, cascading impacts, temporal clustering, and dependence between hazards, alongside exposure and vulnerability dynamics, will be key to further enhancing risk assessment. This will require not only refinement of existing sub-peril models, but also continued research into integrated modelling frameworks, large stochastic event sets, and synthetic event-generation approaches that can better represent interactions across multiple SCS sub-perils, including precipitation-driven losses.

Rather than a fundamental shift, continued advancement of existing approaches is needed, building on current modelling practices that already extend beyond purely historical calibration. This includes further investment in high-resolution storm modelling, refinement of sub-hazard representation and their interactions, and ongoing improvements in observational capabilities. These advances should ultimately support convergent, decision-useful loss models for underwriting, pricing, capital assessment, infrastructure adaptation, and resilience planning in a changing climate.

References

Declarations

Handling Editor: Oliver Wing (Chair), Chief Research Officer, Fathom

The Journal of Catastrophe Risk and Resilience would like to thank Oliver Wing for his role as Handling Editor throughout the peer-review process for this article. We would also like to extend our thanks to the chosen academic reviewers for sharing their expertise and time while undertaking the peer review of this article.

Received: 23rd July 2025
Accepted: 20th July 2026
Published: 19th August 2026

Rights and Permissions

Access: This article is Diamond Open Access.
Licencing: Attribution 4.0 International (CC BY 4.0)
DOI: 10.63024/dkee-9w5q
Article Number: 04.07
ISSN: 3049-7604
Copyright: Copyright remains with the author, and not with the Journal of Catastrophe Risk and Resilience.

Article Citation Details

Aellen, N., et al., 2026. Severe Convective Storms in a Changing Climate: Uncertainties in risk views across North America and Europe, Journal of Catastrophe Risk and Resilience, (2026). https://doi.org/10.63024/dkee-9w5q

Share this article: https://journalofcrr.com/research/04-07-Aellen-et-al/