Main

Marine heatwaves (MHWs) are prolonged periods of anomalously high sea surface temperatures (SSTs) that substantially impact marine ecosystems and climate1. The Mediterranean Sea is particularly susceptible to these events due to the interplay of air–sea heat fluxes and local oceanographic processes2,3,4, with mass-mortality events strongly linked to their occurrence5. Both long-term warming and interannual variability, exacerbated by climate change, contribute to the increasing frequency, intensity and duration of MHWs in the region6. This trend is consistent with global observations, as MHWs have become more frequent, intense and long-lasting worldwide7. The Mediterranean Sea, identified as a climate change hotspot8, is experiencing warming rates exceeding global averages9, making it particularly vulnerable to extreme events such as MHWs. Notable Mediterranean Sea MHWs include those of 200310,11, 200712 and 202013,14 and the exceptional 2022 event15,16. These events have had severe ecological and economic consequences, including mass mortality of marine species17, shifts in species distributions18 and impacts on fisheries and aquaculture19. The cascading effects of MHWs can persist long after the thermal anomalies have dissipated, highlighting the need for improved understanding and prediction of these events.

An understanding of the synoptic large-scale drivers of Mediterranean MHWs, a necessary component to evaluating event predictability, is still lacking since previous studies have examined only atmospheric conditions associated with individual MHWs in the Mediterranean Sea10,13,16. Denaxa et al.2 identified air–sea heat fluxes as the primary driver of MHWs in the Mediterranean, finding the latent heat flux to be the major heat-flux component. Such anomalies in the Mediterranean are probably linked to regional synoptic atmospheric circulation patterns that govern the onset of MHWs. Garfinkel et al.20 identified robust precursors of atmospheric heatwaves in the eastern Mediterranean present 7–10 days before onset, demonstrating that horizontal temperature advection in the days before the extreme is the cause of the heatwave, particularly through a weakening of the Etesian winds that would otherwise advect relatively cool maritime air inland. High-pressure systems and anticyclonic conditions enhance solar radiation and minimize heat loss from the Mediterranean Sea to the atmosphere16. Common drivers of atmospheric heatwaves, however, have been studied in more detail. For example, atmospheric blocking, the manifestation of a persistent high-pressure system, is a known driver of atmospheric heatwaves over central to northern Europe21. These high-pressure systems sometimes initially appear as anticyclonic intrusions from the subtropics22,23, formally known as subtropical ridges (STRs). STRs have also been linked to the occurrence of atmospheric heatwaves over Mediterranean countries, further south than blocking-related events20,24,25,26. While STRs are specifically extensions of the subtropical high-pressure belt that protrude northward, atmospheric blocking displays closed circulation patterns27. As STRs are also known to influence MHWs around Australia28, in this study we examine their role in the formation of these events in the Mediterranean.

Our work uses a comprehensive multi-decadal dataset of multiple Mediterranean MHW macroevents29, built from the European Space Agency (ESA) Climate Change Initiative (CCI) SST satellite dataset30, to identify common atmospheric conditions and circulation patterns across the basin conducive to MHWs, as opposed to previous studies, which have examined atmospheric conditions associated with individual events in the Mediterranean Sea10,13,16. We analyse composite anomalies of wind, geopotential height and atmospheric heat fluxes, as provided by the European Centre for Medium-Range Weather Forecasts Reanalysis v5 (ERA5) atmospheric reanalysis31, to identify conditions favouring MHW development in three representative sub-basins (west, central and east). Our results explicitly identify the role of persistent anticyclonic STRs in Mediterranean MHW, using a multi-event dataset. The identification of the source of predictability as one with a short-term window has crucial implications for improving the accuracy of forecasts and enhancing the effectiveness of impact mitigation strategies, in this region and others potentially subject to MHWs determined by STRs.

The occurrence of persistent STRs and weakened winds in relation to MHWs

In this Article, we use hierarchical clustering analysis to group summer (May–June–July–August) MHWs that extend beyond 100,000 km² for the period of 1982–2022 (Methods). These extensive MHWs have occurred in the western, central and eastern sub-basins of the Mediterranean Sea (WMed, CMed and EMed, respectively), with SST anomalies reaching up to 3 °C during the onset period (Fig. 1a–c, respectively). The MHWs are associated with anticyclonic (clockwise) atmospheric circulation depicted by positive anomalies of geopotential height at 500 hPa (z500 anomalies; red shading in Fig. 1d–f). This suggests the presence of stable, high-pressure systems that probably enhance the persistence and intensity of MHWs. Cyclonic (anticlockwise) centres represented by negative z500 anomalies (blue shading in Fig. 1d–f) associated with low-pressure centres surround the anticyclonic centres, suggesting the presence of a coherent wave-train pattern. Thus, we conclude that atmospheric conditions conducive to MHWs are part of a broader-scale atmospheric circulation dynamic, where alternating high- and low-pressure systems propagate throughout the Mediterranean basin.

Fig. 1: MHW cluster patterns and atmospheric circulation patterns during the MHW onset.
figure 1

a–c, MHW cluster intensity from ESA CCI SST for the three Mediterranean Sea clusters: WMed (a), CMed (b) and EMed (c). d–f, Corresponding ERA5 z500 anomaly composites during the onset period for the three Mediterranean Sea clusters: WMed (d), CMed (e) and EMed (f). Black points in d–f denote statistically significant anomalies (Methods). The anomalies are calculated against the full study period (1982–2022).

Full size image

Atmospheric circulation patterns associated with atmospheric and marine heatwaves32, similar to those identified in Fig. 1, are often attributed to blocking, although sometimes erroneously21. Moreover, previous studies have identified z500 anomalies during individual MHWs6,10,13,16,33, without either attributing them to specific processes or using dedicated indices for blocking and ridges. While STRs are common over the Mediterranean, with climatological frequencies between 0.5 (that is, one every 2 days in the summer) and 0.8 for all the three clusters, blocking is extremely rare, with frequencies below 0.02 (Extended Data Fig. 1). Here we explicitly show that anticyclonic centres linked to MHWs in the Mediterranean Sea are associated with persistent STRs (Fig. 2). The occurrence of STRs is significantly more frequent, with respect to their climatology (Extended Data Fig. 1), over the regions undergoing MHWs during their onset period (Fig. 2a–c). The rate of occurrence rises to one every day, implying the presence of a persistent ridge. Positive anomalies of STR occurrence are flanked east and west by negative anomalies, indicative of a wavy flow pattern persisting over the MHW regions6. This result is consistent across all clusters, with STR patterns simply shifting longitudinally (eastwards) and aligning with the cluster locations.

Fig. 2: STR frequencies and wind-speed anomalies during MHW onset.
figure 2

a–f, Composites of anomalies of STR (a–c) and blocking (d–f) frequency indices, derived from ERA5 z500 data, for the WMed (a,d), CMed (b,e) and EMed (c,f) clusters (Fig. 1a). Black points denote statistically significant anomalies (Methods). The anomalies are calculated against the full study period (1982–2022).

Full size image

Moreover, the critical factor we identify is not merely the presence of STRs but their persistence and co-occurrence with reduced wind conditions (Fig. 2d–f). To quantify this relationship, we calculated likelihood ratios (LRs), which provide a metric for evaluating the strength of association between atmospheric conditions and MHW occurrence (Table 1). Our analysis reveals evidence for the combined impact of STRs and reduced wind speed. When persistent STRs and negative wind anomalies (NegWind) co-occur, the likelihood ratio for MHW occurrence rises to 4.35, 4.29 and 4.81 for the WMed, CMed and EMed, respectively, with respect to the background probability (Table 1). This means that MHWs are approximately 4–5 times more likely to form when both conditions are present simultaneously compared with the background probability. These associations are highly statistically significant (P < 1 × 10−25 for all clusters). Importantly, the likelihood ratios for STR only (without wind reduction) and wind reduction only (without STRs) are much lower—often below or only slightly above 1.0 (Table 1). This indicates that neither condition alone substantially increases MHW probability. A substantial portion of Mediterranean MHWs occur during periods with both STRs and reduced wind conditions: 41.3%, 46.4% and 63.3% for the EMed, CMed and WMed, respectively (Table 1). This is particularly striking considering that the STR + NegWind condition itself occurs during only 8.6%, 10.8% and 14.6% of all days in the EMed, CMed and WMed, respectively.

Table 1 LRs and co-occurrence statistics for MHW formation
Full size table

MHWs caused by the anomalous surface heat fluxes associated with STRs

Increased STR occurrence corresponds with significant decreases in wind speed and reduction of latent heat loss, which are primary candidates for heat-flux increase into the ocean, during the MHW onset (Fig. 3 for WMed; Extended Data Figs. 2 and 3 for CMed and EMed, respectively). The weakening or switching-off of the Mistrals (strong northwesterlies over the Gulf of Lion; Fig. 2; Extended Data Fig. 4 for wind climatologies) corresponds to WMed events, and a weakening of the Etesians (northerlies over the Aegean Sea; Extended Data Fig. 3b) corresponds to EMed events. Reduction of wind speeds (up to −4 m s−1 in the Gulf of Lion; Fig. 3b) inhibits evaporation, which leads to reduced latent heat loss by the ocean (positive anomalies up to 100 W m−2 in the Gulf of Lion; Fig. 3a). The latent heat signals are also confirmed by observational data (from Objectively Analyzed Air–Sea Fluxes for the Global Oceans34; Methods and Extended Data Fig. 5).

Fig. 3: Atmospheric conditions inducing MHW onset.
figure 3

a, ERA5 latent heat anomaly composites for WMed onset. b, Respective daily anomalies of ERA5 latent heat and wind speed during the onset period. Values are area averages inside the green frame shown in a and exhibit a linear correlation equal to –0.71. Black points in a denote statistically significant anomalies (Methods). The anomalies are calculated against the full study period (1982–2022). Latent heat is positive towards the ocean.

Full size image

Our analysis reveals a strong anti-correlation between wind-speed anomalies and latent heat anomalies during the onset period, with regression coefficients (r) of −0.71, −0.68 and −0.73 for the WMed, CMed and EMed clusters, respectively (Fig. 3b and Extended Data Figs. 2b and 3b). This mechanistic relationship is consistent with our statistical findings in Table 1, which show that neither STRs nor wind reduction alone significantly increase MHW probability. The combined effect of atmospheric circulation (STRs) and its surface manifestation (reduced winds) is essential for MHW formation, with the physical connection being the reduced evaporative cooling that allows ocean warming. In the areas of maximum MHW intensity, the spatial correspondence between latent heat anomalies and wind anomalies is particularly strong (Fig. 3 and Extended Data Figs. 2 and 3).

The spatial patterns of SST tendency (Fig. 4a) before the MHW peak and the equivalent tendency derived from the atmosphere–sea heat fluxes (Fig. 4b) largely coincide: for WMed, maximum values are located across the northwest part of the basin and in the Adriatic Sea. Crucially, the magnitudes (for example, up to 0.3 °C d−1 in the Gulf of Lion) are also nearly identical. Indeed, more than 70% of the SST tendency is explained (Fig. 4b hatching) by the total heat-flux tendencies. Therefore, the non-atmosphere processes roughly account for the remaining 30%. Latent heat is found to be the major contributor to the flux-related heat transport, accounting for over 80% of the SST tendency in the areas of interest (Fig. 4c–f and Extended Data Figs. 6 and 7). This confirms that the reduction in wind speed is the primary physical mechanism influencing SST through decreased evaporative cooling. Anomalies of sensible heat, net short-wave and long-wave radiation, are in places significantly different from average conditions (Extended Data Fig. 8), but they generally provide a minor contribution to the SST tendency (Fig. 4d–f). However, there are exceptions, for example, long-wave radiation in the central part of the basin during WMed events and the short-wave radiation over the Gulf of Lion. By contrast, the contribution of sensible heat is largely negligible across the basin. Similar results are found for the CMed (Extended Data Fig. 6) and EMed (Extended Data Fig. 7) clusters.

Fig. 4: SST tendency and atmosphere–sea heat-flux components during MHW onset.
figure 4

a, Composites of ESA CCI SST tendency for WMed. b, Equivalent SST tendency of total fluxes for WMed. c–f, Relative importance with respect to the SST tendency (a) of the four flux components from ERA5: latent (c), sensible (d), net long-wave radiation (e) and net short-wave radiation (f). Hatching in b identifies where the equivalent flux tendency explains more than 70% of the SST tendency in a. The anomalies are calculated against the full study period (1982–2022).

Full size image

Implications for predictability and modelling

This work has identified the combined effect of STRs and associated wind reduction as the key mechanism leading to the development of summer MHWs across the Mediterranean basin. Here we use an event-based dataset covering the Mediterranean Sea over the satellite data period (1982–2022) to evaluate the role of this synergistic relationship.

Our analysis demonstrates that STRs correspond to the wind and heat-flux conditions necessary to induce MHWs, with wind reduction under STRs causing decreased evaporative cooling as the primary physical mechanism, accompanied by a more moderate and less-widespread increase in short-wave radiation. The study by Garfinkel et al.20 found that atmospheric heatwaves over land areas in the eastern Mediterranean co-occur with warm SSTs and are preceded by warm SSTs farther west. Their analysis revealed that horizontal temperature advection due to weakened Etesian winds is the proximate cause of atmospheric heat extremes, which aligns with our findings that reduced wind speeds under STRs drive MHWs through decreased evaporative cooling. Similarly, Denaxa et al.2 investigated the role of air–sea heat exchange during MHWs in the Mediterranean and found that latent heat flux emerges as the major heat-flux component in the SST evolution across all seasons, with air–sea heat flux being the major driver of MHW onset phases. Our results complement and extend these findings by identifying the specific atmospheric circulation pattern (persistent STRs) that drives the reduction in wind speeds and subsequently the anomalous latent heat flux that generates MHWs. Unlike in other regions (for example, the South Atlantic Ocean35), atmospheric blocking does not play a role in Mediterranean summer MHW onset due to its rarity in this region. Instead, the Mediterranean is characterized by frequent STRs, and it is specifically their persistence and co-occurrence with reduced wind conditions that creates ideal conditions for MHW formation.

The combined effect of STRs and reduced winds therefore represents a potential source of predictability for MHWs. Persistent z500 anomalies extending past the subtropical belt are related to MHWs. The persistent STR index used here can act as an indicator for MHWs when combined with information on surface wind anomalies. Previous studies36,37 have also proposed indicators for local temperature extremes based on atmospheric circulation. Indicators of potential predictors of extreme events are applied in forecasting, especially in the emerging field of machine-learning-based forecasting. Machine-learning forecasts of MHWs in particular require information on atmospheric conditions38,39.Therefore, we propose the combined STR–wind-reduction pattern as an indicator of MHWs across the Mediterranean basin.

However, STR–wind reduction provides only a small window of predictability of approximately less than 10 days, similar to blocking40,41. The next step is to understand what drives the excitation, at subseasonal to seasonal timescales, of STRs at a given longitude to boost the window of predictability. The atmospheric circulation anomalies associated with the MHW clusters closely resemble a wave train, with depressions on either side of the high pressure35, suggesting a role of hemisphere-scale dynamics. For example, similar composite analyses have detected heightened summer 2 m temperature and SST anomalies in Europe during instances of North Atlantic depressions6,42. The genesis of these depressions/wave trains may influence the propensity of STR excitation in a season at particular longitudes, which in turn may shed light on the predictability of Mediterranean MHWs. Furthermore, current seasonal forecasts of MHWs in the Mediterranean area are often unreliable43,44, possibly due to their inability to correctly represent the occurrence of STRs. Our work highlights previously unidentified processes that are essential for accurately representing Mediterranean MHWs. These results are critical for improving forecast systems and Earth system models, representing a key step towards effective early-warning and mitigation strategies in the basin.

Methods

Datasets used

ERA5, the multi-decadal three-dimensional reanalysis of the atmosphere at 0.25° horizontal resolution at the surface31, is used to produce regional-scale composites of atmospheric conditions. The following variables are used: geopotential height at 500 hPa, potential temperature at the tropopause (−2 PV units), horizontal components of winds at 10 m, mean surface values of latent and sensible heat, and long-wave and short-wave radiation fluxes. Objectively Analyzed Air–Sea Fluxes for the Global Oceans, specifically the daily 1° satellite-derived analysis, is also used, to validate the latent heat flux of ERA5.

For our study, we used an MHW dataset, which provides a detailed record of MHW macroevents and their characteristics across the southern European and western Asian (SEWA) basins from 1981 to 2016, named the SEWA–MHW dataset as described by Bonino et al.29. This dataset, derived from the ESA CCI SST v2.1 dataset30, identifies MHWs at each 5 × 5 km grid point and applies connected component analysis to group spatiotemporally linked MHWs into larger macroevents while maintaining high-resolution data. To extend the temporal coverage, we included daily SST data from 2017 to 2021, obtained from the Copernicus Climate Data Store (last accessed 12 June 2025). This extended dataset, covering 2017 to 2022, is generated at level 4 by the Copernicus Climate Change Service and builds on the ESA CCI SST dataset. MHWs in the SST dataset are first detected relative to a time-varying baseline climatology, calculated along the full study period (see ref. 29 for more details), which accounts for both trends and seasonal variability. This approach adjusts for shifts in the climate mean state, isolating the interannual variability of the signal45. Second, connected component analysis aggregates MHWs that are spatially and temporally linked, transitioning from a grid-cell-based dataset to an event-based dataset without sacrificing the high-resolution grid-cell information. This method provides a detailed temporal evolution of MHWs at the basin scale. By using this dataset, we ensured consistency and comparability with other MHW studies, adhering to recent recommendations for a standardized MHW definition and spatial considerations46,47.

MHW macroevent detection methodology

We used hierarchical clustering to group the summer (May–June–July–August) macroevents, following the methodology of Bonino et al.29. This enabled us to effectively characterize and analyse the largest MHW macroevents, identifying typical spatial patterns of summer MHWs over the SEWA basins. From the 68,900 identified macroevents, we focused on those with the largest area extensions, retaining 123 macroevents exceeding 100,000 km². For each macroevent, we extracted daily mean intensity maps and averaged them to produce a single event map. Using an agglomerative hierarchical clustering algorithm based on cosine distances48, we identified three clusters over the Mediterranean Sea. As expected, the spatial patterns closely match those in ref. 29 for the Mediterranean Sea, except for the number of events in each cluster. Cluster 1, spanning the western Mediterranean and Adriatic seas, contained 26 macroevents. Cluster 2 centred on the Adriatic Sea and included 18 macroevents; cluster 3, centred around the Aegean Sea, included 14 macroevents. In this study, the clusters are referred to as WMed, CMed and EMed, respectively.

To identify the peak of each macroevent, we determined the time at which the MHW experienced its first and fastest growth in terms of area. This was done by analysing the time-series data of MHW area, focusing on the period when the area expanded most rapidly. The area is defined as the sum of the pixels experiencing MHWs on each day. To account for short-term fluctuations, we smoothed the time-series data using a running mean with a time window proportional to the event duration. Then we calculated the rate of change in area to pinpoint the peak growth phase. The date corresponding to this peak was recorded for each macroevent, ensuring a consistent identification of the critical growth phase across all events. We chose the 5-day period before the peak as the onset period of MHWs.

Atmospheric circulation patterns

STRs are defined as extensions of the subtropical belt (STB) beyond its zonal mean latitude (see fig. 1 in ref. 49; Extended Data Fig. 1a). They are meanders of westerly flow that redirect atmospheric circulation over the Mediterranean region and farther poleward, and are either short-lived stand-alone phenomena or the first stage in the life cycle of blocks. To identify STRs, the following steps have been recommended49. First, a high-pressure zone is defined as a region where the z500 is greater than the hemisphere-wide average over the previous 15 days (to account for long-term trends and seasonal cycle). Second, the STB is defined as the minimum latitude of the high-pressure zone, splitting the hemisphere into two domains. The STB is therefore seasonally varying and, in the summer, reaches latitudes of 40° and thus beyond the extent of the Mediterranean Sea. Finally, STRs are defined where positive anomalies of z500 occur polewards of the STB49. Here we choose to modify the second step of the definition by imposing a fixed definition of the STB as 30° N to identify ridges that cover the Mediterranean Sea. The STR indices used here are two-dimensional, daily and impose no minimum duration.

Atmospheric blockings, like STRs, are also interruptions/redirections of westerly flow, but the key difference from STRs is that blocks are closed anticyclonic features that trap air masses (see fig. 1 in ref. 49). STRs that develop a closed anticyclonic flow at their core and, once unstable, undergo wave-breaking, become atmospheric blocking, completely separating the anticyclonic flow from the STB. There are many competing definitions for blocks, with studies proposing definitions based on different metrics of atmospheric circulation, or identifying more stages in the life cycle of blocks. Here we use positive anomalies of potential temperature in the tropopause (2 PV units) (ref. 50) to identify anticyclonic features that have undergone wave-breaking and persist for 5 days or longer, a definition previously used for detection of MHW drivers35. We define boxes with dimensions of 15° latitude by 5° longitude to both the south and north of each central location. The potential temperature at 2 PV is averaged within these defined areas. We calculate a large-scale measure of the meridional potential temperature gradient by subtracting the average temperature in the southern box from that in the northern box. Positive differences for at least four consecutive days identify a wave-breaking event.

In this study, we adopt the terminology of STRs as defined in previous literature49. It is important to distinguish these STRs from omega blocks or other blocking patterns. While they may appear similar in some respects, STRs are specifically extensions of the subtropical high-pressure belt that protrude northward, whereas omega blocks are characterized by a distinct omega-shaped closed circulation pattern27.

Composites of atmospheric conditions are calculated by averaging the values over the onset period. Statistical significance of composite anomalies is determined with bootstrapping; 1,000 random samples, with a size equal to that of the composite samples (the number of onset days specific to each cluster), are taken from the months May to August and used to create a distribution of composites. The composites of MHW onset conditions are considered statistically significant if they fall above the 95th percentile of the distribution of random samples.

Event likelihood analysis

To robustly analyse the relationship between STRs, wind conditions and MHWs, we developed three specific indices:

  1. (1)

    Persistent STR Index (STR). We identified persistent STRs by requiring the presence of an STR system for five consecutive days within the defined box regions (shown in green in Fig. 3 and Extended Data Figs. 2 and 3) for each cluster. This persistence criterion is crucial as it filters out transient systems and focuses on the sustained atmospheric patterns capable of influencing SSTs. The index is binary (1 if the condition is met, 0 otherwise) for each day in the time series.

  2. (2)

    Persistent Wind Anomaly Index (NegWind). For each cluster, we calculated the area-averaged wind-speed anomaly within the defined box regions. Negative values indicate lower-than-normal wind speeds. The presence of a negative speed anomaly has to be persistent for five consecutive days. The index is binary (1 if the condition is met, 0 otherwise) for each day in the time series.

  3. (3)

    MHW Onset Index (MHW). Our analysis specifically targets the MHW onset period, defined as the 5 days preceding the peak intensity of each event. We computed binary time series for onset days (1 if it is an onset day, 0 otherwise).

To quantify the relationship between these indices, we calculated LRs, which provide a powerful metric for evaluating the strength of association between atmospheric conditions and MHW occurrence. The likelihood ratio is calculated as the ratio of the following probabilities (P):

LR = P(MHW|Condition)/P(MHW)

This metric directly quantifies how much more likely an MHW becomes under specific atmospheric conditions compared with its baseline probability. An LR of 1.0 indicates no association (that is, the condition does not affect MHW probability), while values greater than 1.0 indicate increased likelihood.

Heat-flux contribution analysis

The contribution of air–sea heat fluxes to the tendency in SST can be used to estimate whether MHWs are driven primarily by inflow of heat from the atmosphere. The equivalent SST change induced by the air–sea heat fluxes with time (t) can be estimated with:

$$\frac{{{\mathrm{dSST}}}}{{{\mathrm{d}}t}}=\frac{{Q}_{{\mathrm{h}}}}{{h\times\rho \times c}_{\mathrm{p}}}$$

assuming a fixed summer mixed-layer depth (h) of 5 m based on the climatological study of d’Ortenzio et al.51, a specific heat capacity of water (cp) of 3,985 J kg−1 °C−1 and a density (ρ) of 1,025 kg m−3. Qh (positive downwards) represents the atmospheric heat-flux anomaly (either total or one of the four components) computed against the full study period. This is an approximation of the true value of SST tendency, which would include contributions from ocean advection and unresolved processes. The relative importance, or contribution to the total flux, of an atmospheric heat-flux anomaly component is its value as a percentage of the total flux component, allowing for negative contributions and values above 100%.