To characterize blocking events, we compute several metrics from the identified blocking fields. Frequency is defined as the number of blocked days per season or year at each grid point, providing spatial and temporal information on blocking occurrence. Duration refers to the total length of each blocking episode, while size denotes the spatial extent of the blocking pattern, calculated as the physical area (in km²) of all contiguous grid points meeting the blocking criteria, with variations in grid-box size by latitude accounted for. These metrics allow for a comprehensive comparison of blocking characteristics across models and scenarios. Analyses are presented for the winter and summer seasons: December to February (DJF) and June to August (JJA). Blocking statistics are evaluated over two broad longitudinal sectors of the Northern Hemisphere, referred to here as the North Atlantic (ATL) and North Pacific (PAC) regions. These domains are defined following and encompass both oceanic and adjacent continental areas, rather than being restricted strictly to the ocean basins. This choice allows for a consistent comparison with previous blocking climatologies and ensures that continental blocking regimes, such as Ural blocking, are adequately represented, particularly during boreal summer. The exact longitudinal bounds of each domain are for 50–90° N, -90 to 90° E (ATL) and 40–90° N, 120–240° E (PAC).

To quantify synoptic-scale weather variability, we apply a Lanczos bandpass filter (2–6 d) to daily geopotential height at 500 hPa . The resulting bandpass-filtered fields are denoted by a prime (^′). Storm-tracks are identified based on the standard deviation of the filtered geopotential height (Z500′), which serves as a proxy for the intensity, frequency, and location of transient eddies.

Storm-track intensity is computed seasonally for boreal winter (DJF) and summer (JJA), providing insight into the spatial and temporal variability of synoptic activity across different models and time periods.

In this section, we compare blocking frequency, duration, and size in multidecadal storm-resolving simulations (IFS AMIP, IFS hist, and ICON hist) with the CMIP6 multi-model mean and the ERA5 reanalysis. Our focus is on Northern Hemisphere winter (DJF), using the ANOM index to highlight robust and spatially coherent biases. Additional results for the ABS index are provided in the Supplement (Figs. S1 and S2). The Northern Hemisphere summer season (JJA) is discussed in the next section.

Figure  shows DJF blocking frequency biases relative to ERA5. IFS simulations reduce the underestimation of blocking in the North Atlantic found in CMIP6 by about 10 % (Fig. a, b, d). However, this improvement comes at the cost of an overestimation in other regions, including the Pacific and continental sectors. The RMSE values in Table  confirm that IFS AMIP and IFS hist outperform the CMIP6 ensemble in the North Atlantic. Note that comparing IFS AMIP with CMIP6 is not fully fair, since the AMIP run uses observed SSTs, whereas the CMIP6 results are taken from coupled simulations that include an ocean model.

In contrast, ICON exhibits a pronounced underestimation of blocking frequency over the North Atlantic and an eastward displacement of blocking maxima toward Eurasia (Fig. c). Its North Atlantic RMSE (3.67) is higher than that of both IFS hist (2.34) and the CMIP6 ensemble (2.43). Among the IFS simulations, the AMIP run performs best, suggesting that biases of blocking frequency in the coupled version are partly driven by sea surface temperature (SST) errors. When observed SSTs are prescribed, the Atlantic blocking bias is substantially reduced (Fig. b), an improvement even more evident when using the ABS index (Fig. S1a).

In the North Pacific, IFS hist tends to overestimate blocking frequency, while AMIP simulations again show reduced biases. ICON hist underestimates Pacific blocking but overestimates activity over California, consistent with IFS AMIP.

To assess the effect of ensemble averaging, we also compute the mean bias of ICON and IFS hist (Fig. e). The resulting field appears spatially smoother and exhibits RMSE values comparable to the CMIP6 ensemble mean (Table ). However, ICON and IFS display biases of the same sign over key regions (both underestimate blocking over the North Atlantic and overestimate it over Eastern Europe), indicating that the apparent smoothing does not reflect a compensation of biases in these regions but instead results from spatial averaging of similar model errors.

Blocking frequency biases reflect not only differences in the number of blocking events, but also how blocking frequency is accumulated by the blocking index through event duration and spatial extent. In the applied methodology, longer-lived or spatially larger blocking events contribute more blocked days and grid points, thereby increasing blocking frequency even if the total number of events remains unchanged. In addition, short-lived or spatially small events may fail to satisfy persistence or size criteria.

Figure a and d show the number of blocking events per year in each basin. In the North Atlantic, IFS AMIP and the CMIP6 ensemble match ERA5 reasonably well, while ICON and IFS hist underestimate the number of events. In the North Pacific, IFS hist produces fewer blocking events than ERA5, whereas ICON generates a larger number of events.

Blocking duration statistics show that IFS simulations tend to overestimate mean blocking duration in the North Atlantic, while IFS AMIP captures both the mean and the 95th percentile relatively well. ICON hist exhibits mean durations closer to ERA5 and comparable to the CMIP6 ensemble. In the North Pacific, IFS hist produces fewer but longer-lasting blocking events, whereas IFS AMIP yields slightly shorter events, and ICON generates more frequent but shorter-lived blocks (Figs. b, e).

Blocking size statistics show that IFS simulations overestimate block size in the North Atlantic, while ICON simulates systematically smaller blocks. In the North Pacific, IFS hist tends to overestimate block size and ICON to underestimate it, whereas the CMIP6 ensemble closely matches ERA5 (Figs. c, f).

The discussed blocking characteristics illustrate that blocking frequency biases emerge from a combination of event counts and methodological sensitivities to duration and size. In the North Pacific, the overestimation of the blocking frequency in IFS hist is primarily linked to persistent and spatially extensive blocks, whereas ICON’s biases are associated with a larger number of shorter-lived and smaller events.

In summary, we find that:

IFS generally outperforms ICON, especially in the North Atlantic.

These findings suggest that, while higher resolution alone does not guarantee improved blocking representation, storm-resolving models with realistic boundary conditions (as in AMIP) provide a clearer pathway toward better simulation of blocking. This result also emphasizes the need to better understand the interplay between SST biases, background flow, and moist processes in shaping blocking characteristics.

We now turn to Northern Hemisphere summer (JJA) to assess the most robust and significant blocking biases across models. Note that summer biases (Fig. ) are generally smaller than winter biases (Fig. ).

In the North Atlantic, the IFS historical simulation overestimates blocking frequency, particularly south of Greenland, despite having an RMSE (1.24) similar to the CMIP6 ensemble mean (Table ). This highlights a limitation of basin-averaged RMSE: it can mask strong localized anomalies. The pronounced overestimation of Greenland blocking in IFS hist (exceeding 80 %) is largely absent in IFS AMIP, consistent with improvements also seen in the ABS index (Fig. S2b). The atmosphere-only configuration (IFS AMIP) achieves a lower RMSE (0.85) than IFS hist. ICON hist exhibits the smallest basin-averaged blocking frequency bias over the North Atlantic, with an RMSE of 0.69 compared to 1.21 for the CMIP6 ensemble mean. This indicates good overall agreement with ERA5 at the basin scale, although ICON still shows weak regional underestimations along the eastern flank of the domain.

The number of blocking events per year (Fig. ) shows that IFS hist produces mean and median values close to ERA5. IFS AMIP slightly underestimates the number of blocks. ICON simulations underestimate the number of blocking events, consistent with the localized regions of reduced blocking frequency. Blocking duration statistics (Fig. b) indicate that all models capture JJA duration reasonably well. IFS simulations show a slight positive bias in median duration, while ICON reproduces ERA5 durations closely. Blocking size (Fig. c) reveals that IFS AMIP produces larger blocks, whereas ICON tends to simulate smaller blocks than ERA5. Thus, ICON's low basin-averaged RMSE indicates that its gridded blocking frequency field is close to ERA5 at most locations. However, this good agreement in the frequency field does not imply a correct representation of blocking lifecycles: ICON underestimates the number of blocking events and tends to produce smaller blocks than ERA5. In contrast, IFS hist's frequency overestimation is associated with overly persistent and spatially extensive blocking south of Greenland.

In the North Pacific, the IFS historical simulation underestimates the blocking frequency, while IFS AMIP shows a modest reduction in the spatial extent of the bias relative to IFS hist and substantial differences with ERA5 remain. ICON hist exhibits a northward shift in blocking activity, leading to an overestimation at higher latitudes. The CMIP6 ensemble mean shows a different bias pattern, with a westward displacement of blocking maxima and an overall smaller magnitude than that seen in ICON.

Event counts show that IFS hist underestimates the number of blocking events (Fig. d), whereas IFS AMIP reproduces the ERA5 mean and median well but exhibits an overestimation in the upper tail of the distribution (95th percentile). ICON hist overestimates the number of blocking events across the distribution, while the CMIP6 ensemble underestimates it. Blocking duration (Fig. e) is well captured by the IFS models, whereas ICON exhibits longer upper-percentile durations. Blocking size (Fig. f) is systematically underestimated in IFS AMIP relative to ERA5, with the entire distribution shifted toward smaller block sizes, while IFS hist reproduces the observed distribution more closely. In ICON, the Pacific blocking frequency overestimation therefore primarily arises from an excessive number of events rather than from overly large or long-lived blocks. In contrast, IFS hist biases over the Pacific are mainly associated with a deficit in the number of blocking events (blocking initiation) rather than block size.

The key findings are:

ICON shows realistic blocking durations but underestimates blocking frequency in the North Atlantic and overestimates it in the North Pacific. ICON tends to simulate smaller blocks than ERA5 in the North Atlantic but larger blocks than ERA5 in the North Pacific.

Taken together, these results show that storm-resolving simulations can improve aspects of summer blocking, especially when coupled with realistic SST forcing. However, their performance is not consistently better than the coarse-resolution CMIP models.

After evaluating present-day blocking biases, we next examine projected changes under future climate conditions. Specifically, we analyze the IFS SSP3-7.0 simulations to explore how blocking frequency and characteristics may evolve in response to anthropogenic forcing, with attention to seasonal and regional differences. These results should be interpreted cautiously, given the substantial biases in the historical simulations – particularly regarding jet structure, SST patterns, and blocking persistence.

The ICON scenario runs were also analyzed but are shown only in the Supplement. In the ICON simulation, the absence of SST warming implies that blocking frequencies mainly reflect internal variability, although they still provide a useful illustration of how storm-resolving models simulate blocking. In the ICON DestinE simulation, the historical (10 km atmosphere/5 km ocean–ice) and future (5 km across components) runs differ in resolution, making it difficult to disentangle scenario signals from resolution effects. Nevertheless, we document these results in the Supplement, as they still provide insight into ICON’s representation of blocking (see Figs. S5 and S6).

Figure  presents projected changes in blocking frequency for (a) DJF and (b) JJA. Regions exceeding the 80 % threshold are highlighted to emphasize areas with pronounced relative changes in blocking frequency. This threshold is not intended as a formal measure of statistical significance, but rather as an indicator of strong changes.

In DJF, IFS projects a reduction in high-latitude blocking frequency, most pronounced over northern Eurasia and parts of northern Europe, consistent with previous studies . At the same time, increases in blocking frequency are simulated at lower latitudes, particularly over the North Pacific, where enhanced frequencies occur south of the present-day climatological maximum. This indicates an equatorward tendency of Pacific blocking, accompanied by a zonal redistribution toward the eastern North Pacific.

The magnitude of these projected changes is comparable to the present-day blocking frequency biases in IFS. Note that the color scale in Fig.  differs from that in Fig. .

In JJA, IFS simulates a decrease in blocking frequency across the mid-latitudes most prominently over the central North Atlantic and northern Europe. IFS simulates an increase in blocking frequency at higher latitudes, particularly northern Greenland. The reduced blocking south of Greenland is spatially co-located with SST anomalies in the North Atlantic (Fig. d), including the region of the North Atlantic warming hole.

Although the IFS response broadly follows that of CMIP6, some discrepancies remain. In winter, CMIP6 projects an increase in blocking frequency over eastern Europe and western Asia that is not found in the IFS projections. In summer, IFS likewise does not project the decrease in blocking frequency over the North Pacific seen in CMIP6 .

Beyond frequency, we also assess projected changes in blocking characteristics such as duration, size, and event count (Figs. ). Overall, these properties do not differ significantly from historical values across most regions, but contrasting seasonal tendencies emerge. In summer, IFS projects slightly larger and longer-lived blocks, particularly over the Atlantic, accompanied by a decrease in the number of blocking events in this basin. In winter, blocking events tend to be smaller and shorter-lived over the Atlantic, while blocking counts remain broadly similar to historical values. Over the North Pacific, blocking counts show little change in winter but a modest increase in summer, particularly in the upper percentiles, alongside relatively small changes in block size and duration.

Projected changes in jet and storm-track structure under SSP3-7.0 are small compared to the present-day model biases (Fig. ). In general, the winter jet becomes slightly more zonal, while in summer it shifts poleward, consistent with previous studies e.g.,. In summer, over the northwestern Atlantic, IFS indicates a modest strengthening of the jet. Downstream of this region, south of Greenland, IFS projects a enhancement of storm-track activity that is spatially co-located with the simulated decrease in blocking frequency. In winter, the largest circulation response occurs in the Pacific, with intensified midlatitude westerlies and storm-tracks in regions where blocking frequency increases. However, the magnitude of these circulation changes remains small, making it difficult to disentangle the relative roles of individual processes.

The projected circulation changes occur alongside distinct SST patterns (Fig. c, d). In the North Atlantic, a warming hole south of Greenland contrasts with pronounced warming along the Gulf Stream. These SST patterns spatially coincide with changes in jet structure and reduced summer blocking over the Atlantic, although a direct causal link cannot be established given the magnitude of the model biases. In the North Pacific, SSTs warm more uniformly, with stronger warming at midlatitudes during winter, potentially modifying meridional temperature gradients and the background flow.

Overall, the SSP3-7.0 simulations point to a net reduction in Atlantic blocking in summer and a net increase in North Pacific blocking in winter. However, these signals should not be interpreted as robust projections, as they are derived from a single storm-resolving model and may be influenced by biases in the mean state and circulation identified in the historical simulations. Instead, they provide an indication of the sensitivity of blocking to future forcing within this modelling framework.

These results highlight the need for process-based analyses to disentangle the interplay between SST changes, jet dynamics, and blocking, particularly in light of the differences between AMIP and coupled simulations, which suggest an important role of SST biases.

Our results show that increasing horizontal resolution to the storm-resolving scale can improve certain aspects of atmospheric blocking representation, but does not systematically eliminate long-standing biases. Based on multi-decadal simulations, the storm-resolving IFS and ICON configurations reproduce several regional blocking characteristics more realistically than the CMIP6 ensemble mean, particularly when realistic sea surface temperatures (SSTs) are prescribed. For example, IFS AMIP better captures blocking duration and size during boreal winter, consistent with earlier work from idealized experiments and high-resolution CMIP6 simulations showing that higher resolution and improved forcing enhance the persistence of large-scale circulation regimes. However, these improvements are spatially and seasonally dependent and are not uniform across models.

Notably, ICON underestimates North Atlantic blocking frequency in both winter and summer, despite its explicit treatment of moist processes, while some CMIP6 models perform comparably well or better for specific blocking metrics such as duration or spatial extent. This confirms that horizontal resolution alone is not sufficient to resolve blocking biases, and that model formulation and boundary forcing remain critical. Similar conclusions have been drawn in previous studies using high-resolution CMIP and idealized model configurations, suggesting that storm-resolving models represent an important but incomplete step toward improved blocking simulation e.g.,.

Sea surface temperatures and ocean–atmosphere coupling play a central role in shaping midlatitude circulation and blocking behavior e.g.,. At storm-resolving resolution, mesoscale SST gradients and frontal structures are better resolved, enhancing air–sea coupling and increasing the sensitivity of the atmosphere to SST biases . This sensitivity is clearly reflected in the contrasting behavior of coupled and atmosphere-only simulations.

Among the IFS configurations, the AMIP simulation (forced with observed SSTs) most closely reproduces ERA5 blocking characteristics, particularly with respect to wintertime duration and North Atlantic and North Pacific blocking frequency. This highlights the importance of realistic SST boundary conditions. In contrast, IFS hist exhibits a negative SST bias in the North Atlantic, especially in summer, coinciding with regions of enhanced blocking frequency. Rather than implying a direct causal relationship, this spatial correspondence suggests that SST biases can modulate the background state in which blocking develops, consistent with earlier studies linking North Atlantic SST errors to circulation biases .

ICON simulations display positive SST biases in both the North Atlantic and North Pacific mid-latitudes. In winter, these warm anomalies coincide with intensified jets and reduced blocking, broadly consistent with theoretical expectations that stronger baroclinicity suppresses blocking. In summer, however, the relationship is less robust: ICON exhibits a poleward-displaced jet without a commensurate intensification, and the associated blocking response is weak and regionally variable. This seasonal contrast indicates that the influence of SSTs on blocking depends strongly on the background flow regime.

While SST biases in IFS hist extend across much of the North Atlantic, their potential influence on blocking and storm-track behavior is unlikely to be spatially uniform. The dynamical impact of SST anomalies depends on their position relative to the climatological SST gradients and the mean storm-track location. In particular, high-latitude regions south of Greenland coincide with strong background baroclinicity and frequent storm-track activity, making blocking potentially more sensitive to SST-related modifications of surface fluxes and lower-tropospheric stability there. In contrast, SST anomalies in other parts of the basin may have a weaker or different dynamical imprint. This spatial modulation helps explain why an apparent co-occurrence between negative SST biases and enhanced blocking emerges primarily in the high-latitude North Atlantic, while similar relationships are absent in ICON and differ in the North Pacific. Studies have even shown that Pacific SST gradients can even influence wave activity and blocking downstream over the Atlantic sector .

Overall, these results emphasize that an accurate representation of SSTs (both in magnitude and spatial gradients) is essential for simulating blocking, particularly in high-resolution coupled models. Storm-resolving configurations amplify the atmospheric response to SST errors, underscoring the need to reduce oceanic biases alongside increases in resolution.

Blocking behavior is tightly coupled to the structure and variability of the midlatitude jet and storm tracks. In storm-resolving simulations, biases in blocking frequency often co-occur with errors in jet latitude and intensity relative to ERA5. During winter, the overly zonal and poleward-displaced jets in ICON hist are associated with reduced blocking, whereas the more realistic jet configuration in IFS AMIP coincides with improved blocking representation.

Jet exit regions and areas of strong diffluent flow are particularly conducive to blocking formation . In both IFS and ICON, biases in jet position and strength align spatially with blocking biases in the North Atlantic and North Pacific. These interactions are further shaped by storm-track variability: enhanced transient eddy activity can promote blocking onset through wave breaking, while displaced storm tracks can be a consequence of blocked flow .

In summer, however, this coupling changes. Both IFS and ICON exhibit predominantly positive correlations between blocking biases and storm track biases. These findings support previous work showing that while transient eddies are crucial for blocking onset, persistence and spatial extent are strongly conditioned by the background circulation and thermodynamic state e.g.,.

Despite the benefits of higher spatial resolution and improved SST forcing, biases in blocking representation persist in the IFS AMIP simulations, pointing to additional relevant processes. Misrepresentation of moist diabatic processes, particularly latent heat release associated with warm conveyor belts, are increasingly recognized as important contributors to blocking bias in CMIP6 . Diabatic heating can amplify ridges, modify potential vorticity gradients, and precondition the flow for blocking development e.g.,.

IFS and ICON differ substantially in their treatment of moist processes: IFS employs parameterized convection, whereas ICON explicitly resolves deep convection. These differences likely contribute also to the divergent blocking responses across models and configurations.

A more detailed process-based evaluation of moist dynamics (including vertical heating profiles, cloud–radiative feedbacks, and their interaction with large-scale flow) is required to understand and reduce persistent blocking biases in storm-resolving models.

In addition to present-day evaluation, our analysis provides a cautious first assessment of how blocking characteristics may evolve under future warming in storm-resolving simulations. The IFS SSP3-7.0 experiments suggest a redistribution of blocking activity, with reduced summer blocking over the North Atlantic midlatitudes and a tendency toward increased winter blocking in the southern North Pacific, consistent with an equatorward shift in this sector.

These changes occur alongside modest shifts in jet and storm-track structure. In summer, a strengthening of the Atlantic jet and downstream storm tracks coincides with reduced blocking frequency, whereas in winter the largest response is found in the Pacific, where enhanced jet and storm-track activity coincides with increased blocking. However, the magnitude of these projected changes remains small relative to present-day biases, complicating attribution and interpretation.

These results are broadly consistent with previous studies using CMIP-class models, which generally report modest and regionally dependent changes in blocking under climate change, often linked to shifts in jet position and storm-track activity e.g.,. At the same time, substantial inter-model spread and sensitivity to the mean state have been identified as key sources of uncertainty in these projections.

It is important to emphasize that these projected changes are derived from a single storm-resolving model and should therefore not be interpreted as robust projections. Instead, they provide an indication of the sensitivity of blocking to future forcing within this modelling framework, conditioned by the biases identified in the historical simulations.

In particular, the differences between AMIP and coupled configurations highlight the role of SST biases in shaping blocking behaviour, consistent with previous work demonstrating the sensitivity of blocking to SST gradients and air–sea coupling e.g.,. This suggests that uncertainties in the mean state (especially in SSTs, storm tracks, and jet structure), can strongly influence the simulated response.

Projected changes in blocking duration, size, and event count are generally modest and often fall within the range of internal variability. This further underscores the need for caution when interpreting blocking trends from storm-resolving models and highlights the importance of reducing historical biases and expanding to multi-model frameworks before drawing robust conclusions about future changes.

Beyond their immediate interpretation, these results serve an important documentation purpose. They provide a baseline for evaluating future generations of storm-resolving models as larger ensembles and multi-model frameworks become available. While the projected changes are not definitive, documenting these early signals is essential for tracking the evolution of blocking responses and their sensitivity to model biases.

To build on these advances, future research should prioritize the following directions:

Development and evaluation of coupled storm-resolving Earth system models, including multi-decadal simulations to assess blocking variability, climatological trends, and model stability across timescales.

These research avenues will be essential for leveraging the full potential of storm-resolving modeling and for addressing the persistent uncertainties that affect our understanding and simulation of midlatitude atmospheric circulation in a warming climate.

Storm-resolving modeling represents an important step toward more realistic simulations of atmospheric blocking and persistent weather extremes. However, our results show that increased spatial resolution alone is not sufficient. Meaningful improvements will require a combined approach that integrates high-resolution dynamics with accurate surface boundary conditions and targeted improvements in key physical processes. Advancing these elements together is essential for improving confidence in blocking simulations and their projected response to climate change.