MCs are here detected and tracked based on a composite cyclone detection algorithm presented by Flaounas et al. (2023) and used by Givon et al. (2024a) for the dynamic classification of MCs. With this approach, 10 different tracking methods are applied to 1-hourly ERA5 reanalysis of ECMWF (Hersbach et al., 2020). Confidence level 5 is chosen, denoting the agreement of at least 5 detection methods on every MC track. Each composite cyclone track contains the location of its center throughout its lifetime. Sea level pressure at the cyclone center is also reported, and its minimum along the track is used to define MC peak time for classification. Some cyclone tracks that do not cross east of longitude 5° E are removed from the analysis, see more details in Givon et al. (2024a). Overall, 3190 MC tracks are captured throughout the period, with a minimum lifetime of 2 d.

Here, we utilize the cluster separation presented in Givon et al. (2024a). Specifically, for each MC at peak intensity, the surrounding upper-level isentropic potential vorticity field (PV, vertically averaged between the 320–340 K isentropic levels) is extracted around the cyclone center. The PV field is considered within a domain extending 20° east and west of the MC center, 40° to the north and 20° to the south. The PV field is classified using a self-organizing map algorithm (SOM), producing 9 robust clusters with distinct Rossby wave patterns, unique surface impacts, and inherent characteristics. Here, we attribute the cluster number originally determined for the cyclone peak time throughout the cyclone's lifetime.

We integrate the hourly P and E accumulated along each MC track to derive MC-induced freshwater fluxes. To attribute the fluxes to the MC track and allow an objective comparison of the different MC types we define a constant 10° radius impact area around the cyclone center, within which the fluxes are accumulated in 1-hourly steps along the track. This ensures that no cluster is favored or penalized by impact area differences. While a 10° radius may be considered excessive to associate to MCs (Flaounas et al., 2016), it has been adopted by recent studies on MC impacts (Rousseau-Rizzi et al., 2024; Portal et al., 2024). Results show little sensitivity to the MC impact area in the present framework: with a 5° radius, the spatially accumulated cyclone-induced fluxes naturally decrease, but the fundamental differences among the MC types and their temporal variability are unchanged. Area-weighted averages are calculated across the MC impact area and accumulated annually to quantify total MC-induced fluxes.

To investigate interannual trends in MC-induced evaporation and precipitation, we accumulate freshwater fluxes spatially and temporally across each year, integrating along the MC tracks. We thus obtain a Lagrangian framework used to unveil the cyclone-driven component of the MHC under a process-based prism, resulting in a novel hydrological-budget analysis of differently driven MCs.

As noted by Zappa et al. (2015), changes in MC-induced P are influenced by both changes in MC frequency and intensity. A similar argument can be made for MC-induced E. To address both sources of variance, we complement the accumulated MHC contributions with their normalized ones. To eliminate variations in frequency and obtain the fluxes per cyclone, normalized P and E are evaluated as follows: 1Pnorm=F‾MC∑yPyFy2Enorm=F‾MC∑yEyFy Where Py, Ey stands for annual cyclone-induced precipitation and evaporation, respectively, Fy denotes the corresponding annual frequency of MCs, and F‾MC denotes the mean MC frequency across all years. The annual frequency accounts for the co-occurrence of multiple cyclones in the domain by counting the number of MC centers at every time step. Although this approach leads to large annual frequencies (up to 80 %, as also recognized by Flaounas et al., 2016), it ensures the overall conservation of mass. We similarly normalize the contributions from each cluster to reveal MC types that are more susceptible to these elusive changes in intensity.

Ocean heat content (OHC) in the layer 0–300 m is computed using the Euro-Mediterranean Center on Climate Change (CMCC-Foundation) eddy-permitting global ocean reanalysis, C-GLORS v7. The model data is provided at daily frequency from 1993 up to 2019 and 1/4° horizontal resolution with 50 vertical levels. Details on the model and improvements from previous versions are presented by Storto and Masina (2016). While suffering from low temporal resolution and a shorter temporal extent, the model proves valuable for process-based research on heat exchanges, in agreement with comparable independent estimates.

To investigate the impact of the various MC drivers on the OHC, we use a 6 d window centered at the location of each MC at peak intensity times and map the net difference in OHC before and after an MC passage. These differences are accumulated under the impact area at MC peak time on daily resolution and separated by cluster. The OHC is a measure of stored energy. As such, temporal changes in it can be interpreted as energy fluxes and are indeed affected by atmospheric fluxes of both sensible and latent heat. However, the OHC is affected by other processes as well, such as radiative and advective fluxes. Nevertheless, since the presence of MCs potentially influences all of these, and with the presence of cyclones being the only common feature of the observed data, the OHC response is interpreted as MC-driven fluxes for consistency. We thus convert the OHC difference (originally measured in Joules) to units of virtual evaporation rates (Ev, mm yr⁻¹) as follows: 3Ev=365×F‾MCd(OHCt0+3-OHCt0-3)6[days]×Lv Where F‾MCd stands for the annual-mean frequency of MCs on daily scale (either as a whole or per cluster) considering only the peak-time (t0) of each track, and Lv is the latent heat of evaporation. This way, the MC impact on the OHC is comparable to the atmospheric freshwater fluxes.

We note that this analysis does not fully disentangle the impact of the OHC on MCs, that tend to intensify with response to large OHC values. Nevertheless, the temporal evolution of the difference in OHC (not shown) suggests that the OHC response is temporally centered around the passage of MCs and provides a direct measure of the MC-related ocean-heat loss. We interpret the results as the impact of MCs on OHC, while admitting that MC intensity is partially impacted by initially high OHC values. The agreement between the MC-driven E and OHC response to MC supports this interpretation. In the future, we aim to further address this two-way coupling between MCs and the OHC in a dedicated study.

We begin with a climatological overview of the MHC (Fig. 1a and b), followed by the MC contribution to mean annual P and E. Overall, P is shown to mainly affect coastal areas and mountain ridges, reaching values up to 1500 mm yr⁻¹, whereas E dominates over the maritime regions and is minimal over land. In agreement with previous studies (Flaounas et al., 2016; Tootoonchi et al., 2025), the results highlight the prominence of MCs on the MHC, driving up to 70 % of total precipitation and up to 50 % of total surface evaporation in various regions (Fig. 1c and d). While the contribution of MCs to E overlaps well with cyclone density, the peaks of MC contribution to P are shifted eastwards. This is in line with the findings of Saaroni et al. (2010), highlighting the strong dependance of eastern Mediterranean rainfall on MCs. We note that the contribution of MCs to total surface evaporation is about twice their mean frequency, while their contribution to precipitation is about three times higher, indicating that MC-induced freshwater fluxes are strongly concentrated within the MC impact area rather than resulting from random spatial overlaps.

When examining the area-weighted accumulated fluxes in Fig. 2, grid points affected by MCs show higher P and E mean rates compared to the year-round average regardless of MC presence in the domain (compare Fig. 2a and b). While both P and E of MCs (Fig. 2b) are slightly increasing with time, the positive residual (P–E) associated with MCs is being gradually eroded. The normalized fluxes (Fig. 2c) reveal a decrease in P and an increase in E per cyclone. Interestingly, the covariance between MC-induced P and E is solely due to their shared MC frequencies, dropping to zero for the normalized E and P (Fig. 2c). This result suggests that instantaneous E and P rates are unconstrained when observed per cyclone, as MCs converge moisture from other sources in addition to their self-induced E to generate heavy P rates. To better understand the sources of these long-term variabilities, we examine the results under the nine dominant MC drivers.

Under the cyclone-centered PV classification, the complexity and competing effects of the various MC drivers on the MHC is clarified. Figure 3 highlights the variability between the different MC drivers at peak cyclone intensity. Cluster 4, representing fully developed baroclinic MCs, shows the most intense instantaneous E and P fluxes while cluster 6, typically impacting North Africa, is associated with much weaker freshwater exchanges. However, sharp differences are also evident between clusters that share similar seasonal and geographical amplitudes, such as clusters 1, 2, and 4, and clusters 3, 5, and 7. Cluster 8, associated with narrow, cyclonically curving PV streamers, and recently found to be the most favorable for deep convection (Portal et al., 2025), notably exhibits the most intense fluxes among the off-winter clusters.

We note that the impact area well captures the E and P features associated with MCs. The relatively strong evaporation observed about 15° to the west of the MC centers of clusters 3, 5, and 7 (i.e., outside of the impact area) mostly occurs over the Atlantic Ocean, given their preferred locations (black contours in Figs. 4 and 5 and Fig. 7 in Givon et al., 2024a in more detail). With this MC-centered view it appears that a first-order correlation exists between instantaneous MC induced E and P. However, since some of the clusters appear more frequently on land, we next evaluate the accumulated fluxes throughout the MC duration and consider their geographical areas of influence.

The independent contributions of each MC type to the overall annual E are shown in Fig. 4. Cluster 4 stands out as a major contributor, accounting for up to 24 % of the MC-induced E despite forming 20 % of the total MC frequency (see Fig. 4 titles). The ratio between the MC contribution and local frequency surpasses 3 in the eastern Mediterranean and along the north African coast, suggesting high conditional probability. Cluster 1, representing cold cyclones anchored to topography, shows significant E contributions especially in the Ligurian and Adriatic seas, as may be expected from the strong gap-winds regimes triggered at the early stages of Alpine lee-cyclogenesis and their link to evaporation (Buzzi et al., 2020; Givon et al., 2021, 2024b). On the other hand, the contributions of summer clusters 6 and 9 are weaker than their mean frequency, which is often over land, possibly due to already high evaporation rates throughout the season (Ruiz et al., 2008). Nevertheless, Sharav lows (North-African heat lows captured by cluster 6) are associated with strong evaporation hot-spots in north-western Africa, possibly affecting the great lakes and land-moisture in the region (Rieder et al., 2025). Cluster 2 MCs are denoted by a unique combination of anticyclonic and cyclonic Rossby wave breaking simultaneously deforming the same PV streamer, that extends from the anti-cyclonic shear zone of the sub-polar jet to the north to the cyclonic shear zone of the sub-tropical jet to the south (Fig. A1). The double-jet configuration denoted by cluster 2 notably contributes 15 % to total MC-induced evaporation despite its 12 % relative frequency, suggesting strong – yet apparently more local – evaporative capabilities, appearing as relatively sporadic evaporation hotspots.

The cluster-separation of contributions to P is even more pronounced (Fig. 5). Cluster 4 dominates in the central and eastern Mediterranean, overall contributing 28 % of total MC induced P, and cluster 1 in the western coasts of Italy, Greece and Turkey, and along the Adriatic, providing 22 % of total MC induced P. The precipitation induced by cluster 2 again exhibits more localized hot spots, while the contributions of the other clusters are significantly weaker. Intriguingly, precipitation along the Egyptian coast shows a tight link to cyclonic wave breaking (CWB) MCs captured in cluster 8, with ∼ 20 % of annual precipitation delivered in a mere 1 % regional cluster frequency. Moreover, this region is hardly responsive to other MC clusters. The tight connection between RWB patterns and extreme precipitation were recently highlighted by de Vries (2021), especially for semi-arid areas such as the eastern-Mediterranean.

These results illustrate well the robustness of the PV clustering even beyond the classification time, with profound implications on geographical, seasonal, and dynamical variability. Each MC driver thus plays a fundamentally different role in shaping the MHC.

To evaluate the total contribution of MCs and their drivers to the MHC, we accumulate the fluxes spatially and temporally across the years, analyzing both annual P and E distributions as well as their difference, P–E, per cluster. Figure 6 shows that the overall slightly positive P–E contribution of MCs is the result of a delicate balance between inherently imbalanced MC contributions: while winter clusters 1, 2, and 4 act as moisture sources, each adding about 30 mm yr⁻¹ per grid-point within a 10° radius impact area, these inputs are partly offset by a consistent drying effect imposed by the summer-prone clusters 5, 6, and 9. Seeing as the relative frequencies of each cluster are changing in recent decades (Givon et al., 2024a), we further analyze the long-term variability of each cluster, accounting for both variations in frequency and intensity.

Long term variability of MC occurrence and interaction with the MHC strongly depends also on MC cluster type. The relative importance of each MC cluster is shifting over recent decades, as shown in Fig. 7. While some clusters are well balanced and hardly contribute to the MC-induced P–E (e.g., clusters 5 and 9), P–E is growing larger for some (clusters 1, 3, and 8) and smaller for others (clusters 2, 4, 6, and 7). The strongest change is captured for the double-jet cluster 2, losing 0.2 mm yr⁻¹ which corresponds to ∼ 1 % erosion of its annual precipitation surplus per year, with similar trend magnitudes reported for clusters 6 and 7. The net negative contributions from clusters 2, 4, 6, and 7 differentially stem from either a reduction in precipitation (clusters 4 and 7), an increase in evaporation (6), or both (cluster 2). By contrast, cluster 8 contribution is positive and slightly increasing with time, due to enhanced P and near-constant E. As these changes can arise from either changes in frequency or intensity, we next examine the normalized fluxes and their trends, effectively eliminating frequency changes and highlighting changes in E and P intensities.

Indeed, the net contributions (P–E) of each cluster vary differently when only changes of flux intensity are considered (Fig. 8). The intensity-only loss from cluster 2 is enhanced to -0.5 mm yr⁻¹, indicating that its increasing frequency masks a drop in intensity. Interestingly, the Enorm and Pnorm associated with cluster 2 evolve in opposite directions, with decreasing Pnorm (and increasing Enorm). As suggested by previous studies, the overall drop in precipitation for cluster 4 is indeed driven solely by a reduction in frequency, while the intensity of both Enorm and Pnorm steadily increase. For cluster 6, changes in intensity exist but are well balanced, suggesting changes in frequency are dominant in determining P–E for these Saharan heat lows.

When separated by cluster, the P–E residual shows distinct long-term trends that are affected by variations in both frequency and intensity. In agreement with previous studies, winter MCs grow in intensity and decrease in frequency, with the latter being more dominant, resulting in a long-term reduction in P–E. Summer MCs, on the other hand, are primarily affected by a rise in frequency, leading to further drying.

To derive the overall impact of MCs on the MHC, net cluster frequencies and their area weighted contributions to the MHC relative to accumulated climatological values across space and time are shown in Table 1. The net frequencies consider the temporal occurrence of each MC cluster and the ratio between the number of grid-points within the MC masks to the total number of grid points in the domain. These frequencies hence describe the fraction of MC affected areas throughout the entire analysis period and across the whole domain. The overall net climatological P–E in the Mediterranean is -52 mm yr⁻¹ (Fig. 2a). The overall MC-induced P–E is, however, positive, counter balancing 24 % of the climatological negative P–E. The MC contribution is primarily driven by lee-cyclones (clusters 1 and 4), while specific MC drivers (clusters 5, 6 and 9) act to enhance the overall excess E in the Mediterranean.

The overall impact of MCs on the OHC (Fig. 9) shows that on average, MCs extract heat from the Mediterranean at a mean rate equivalent to 645 mm yr⁻¹ of E. Although other processes affect the OHC, including lateral advection and radiative effects, the contribution of MC induced E alone accounts for up to ∼ 350 mm yr⁻¹ (Fig. 2), more than 50 % of the net OHC difference with response to the passage of MCs. Notably, the OHC response to MCs is positive along the North-African and eastern Mediterranean coasts, suggesting MCs tend to add heat to these areas. While this local reversal could be the result of sensible heat fluxes triggered by heat lows that often affect these areas, further research into oceanic flows is required to definitively explain these signals.

Once more, the cluster breakdown provides a more detailed picture of the independent role of MC types (Fig. 10): MC drivers that provide excess P also appear to drain heat from the ocean (clusters 1, 2, 4, and 8), as their E is still higher than that of E-dominated MCs. The E-dominated MCs (clusters 6, 7, and 9) act as heat sources, enhancing the OHC as they evolve rather than lowering it. This may be the result of their dry formation environments, prohibiting precipitation and leading to convergence of dry, warm air towards the MC center, generating enhanced downward sensible and possibly radiative (i.e., reduced cloud cover) fluxes. Spatial variability is evident for some of the clusters, suggesting certain drivers, e.g. clusters 3 and 5, act differently in different regions. Specifically, these MCs act as heat extractors when impacting the eastern Mediterranean, and heat suppliers when impacting the western parts. The OHC response to the various MC drivers is in-line with the atmospheric perspective, suggesting that opposing contributions of MCs to the MHC also bear the opposite impact on the OHC. This highlights the importance of considering the internal variability of MCs when assessing their role in future climate.