For atmospheric and surface data, we use monthly and daily data from the European Center for Medium-Range Weather Forecasts (ECMWF) ERA5 atmospheric reanalysis at 0.25° × 0.25° grid-spacing, covering the period 1979–2023 i.e., post-satellite era; . ERA5 data has been shown in previous studies to provide reliable estimates of in situ observations of the hydrological cycle . We nevertheless validate our results using in situ data from rain gauges distributed throughout Israel provided by the Israel Meteorological Service (IMS, https://ims.gov.il, last access: 7 January 2025). In accordance with ERA5, the reference observed sea surface temperature (SST) data are taken from the HadISST2 dataset up to September 2007 and from the Operational SST and Ice Analysis (OSTIA) dataset thereafter. For precipitation over land, we use ERA5-land at approximately 9 km resolution, which provides an improved representation of land processes.

Upper-ocean heat content, given by the product of sea-water heat capacity cp, ocean mixed-layer depth (hml, MLD), and SST (T), has been shown in previous work to be a contributing factor to processes affecting Levant precipitation over land, such as ocean-atmosphere heat and moisture exchange and land-ocean temperature contrasts . In particular, and demonstrated the potential of utilizing the ocean's upper-layer heat content during fall months to enhance predictions of winter precipitation in the Levant. Building on these results, we hypothesize that changes in Mediterranean SST and ocean heat uptake are linked to precipitation changes in the Levant. However, since estimates of MLD can vary significantly across datasets and methodologies , we avoid using MLD as a predictor.

Specifically, the energy balance equation of the ocean mixed layer can be written as 1cphmlT˙+cp∇⋅(ũT)=Qf+Kz where T˙ is SST tendency, ũ is the vertical-mean horizontal flow in the mixed layer, Qf is the net downward heat flux at the ocean surface, and Kz is the input of heat to the mixed layer from below, associated primarily with small-scale vertical mixing. Here positive Qf values indicate heating of the upper ocean and negative Qf values indicate release of heat from the ocean surface to the atmosphere. Also, for sufficiently small temporal variations of the ocean mixed layer depth, SST tendency is approximately proportional to Qf.

The net surface energy input into the ocean mixed layer (Qf) consists of the net downward shortwave (SW) and longwave (LW) surface radiation, and the downward sensible (SH) and latent (LH) surface heat fluxes, minus the fraction of the shortwave radiation penetrating below the mixed layer (QP) 2Qf=SW+LW+SH+LH-QP where QP is calculated as a 25-m e folding decay of shortwave radiation, given by the equation QP=0.45 SW e-γhml, with γ being a decay rate constant equal to 0.04 m⁻¹ .

To objectively capture the spatial patterns of Mediterranean Sea variability, we employ the Empirical Orthogonal Function (EOF) analysis , as well as Self-Organizing Map (SOM) unsupervised neural network analysis for clustering and visualizing high-dimensional data . The two methods yield very similar spatial patterns and temporal behavior. We therefore focus in our analysis on the EOF patterns, but provide additional results for the SOM analysis in the Supplement (Sect. S1).

We produce EOF patterns of SST and ocean heat uptake (Qf) in the Mediterranean Sea, using detrended monthly anomalies from climatology. The EOFs are computed as the eigenvectors of the covariance matrices of SST and Qf, representing the primary orthogonal modes of variability in the SST and Qf data. The variance explained by each EOF mode is calculated as each eigenvalue of the covariance matrix of SST and Qf normalized by the sum of all eigenvalues. The corresponding time series of the relative amplitude of each EOF pattern (the principal component) is used to quantify how the dominance of these patterns varies over time. Subsequently, we calculate the lagged Pearson correlation coefficient between the monthly amplitude of each pattern and winter (December–February) mean land precipitation in the Levant (Fig. ).

The steady moisture equation of the atmosphere can be written as 3P‾=E‾-〈∇⋅(u‾q‾)〉-〈∇⋅(u′q′‾)〉-qsus‾⋅∇p‾s where P and E represent precipitation and evaporation, respectively, u is the wind vector, q denotes specific humidity, p denotes pressure, and the subscript (⋅)s denotes surface values. Angled brackets denote mass-weighted vertical integrals, 〈⋅〉≡1ρwg∫0ps(⋅)dp where ρw is the density of water and g is Earth's gravitational acceleration; overbars and primes denote monthly temporal means and deviations thereof, respectively.

Following the methodology and terminology described in , it follows from Eq. () that changes in precipitation can be decomposed into those involving changes in evaporation, the mean wind field (dynamic changes), the mean moisture field (thermodynamic changes), surface moisture transport, and transient eddies .

We define δ as the difference between two composites of monthly means, 4δ≡(⋅)‾‾2-(⋅)‾‾1 where a double overbar denotes a temporal average over some period. Neglecting changes associated with surface pressure, we rewrite the moisture balance equation to express the difference between two composites of winters, 5δP≅δE-δ〈∇⋅u‾q‾〉-δ〈∇⋅u′q′‾〉. In the next section, we refer to the second and third right-hand side terms as the changes in the mean and transient components of precipitation, respectively, where the mean component is further decomposed into mean thermodynamic (-〈∇⋅(u‾‾δq‾)〉) and the mean dynamic (-〈∇⋅(δu‾q‾‾)〉) components .

Due to the critical influence of synoptic-scale conditions on precipitation in the Levant , we examine the relationship between these conditions and our calculated patterns of variability. Specifically, we classify synoptic conditions based on the semi-objective methodology proposed by , which has been used in previous works to study EM seasonal synoptic variations , as well as changes under climate change in weather patterns , extreme weather events , and their impacts .

The classification method uses four single-level atmospheric fields: geopotential height, temperature, and the zonal and meridional components of wind; all at 1000 hPa and at a grid-spacing of 2.5° × 2.5° over the EM region (defined within the coordinates 27.5–37.5° N and 30–40° E), sampled at 12:00 UTC each day. The synoptic classification of each day within the EM is therefore calculated using 100 data points (5×5 grid points for each of the four atmospheric fields), which are also standardized by subtracting the long-term mean and dividing by the standard deviation of the time series of each field. The classification of each day to its synoptic type is performed by comparing each of these daily datasets to 426 d that were manually classified by a team of expert meteorologists (335 d from 1985 and 91 d from the winter of 1991–1992). The synoptic type of each day is determined as that for which the Euclidean distance from each of the manually classified days is lowest. The reference 426 d are categorized into five synoptic groups , of which two are associated with winter precipitation: i.

Cyprus Low (CL): A Mediterranean cyclone centered near Cyprus, often responsible for significant rainfall and stormy weather in the Levant region and significant impacts ;

To better identify the driving factors impacting the synoptic change, we analyze the regional atmospheric conditions between winter composites. Accordingly, we study the changes in mean sea-level pressure conditions, the 500 hPa geopotential, and the strength and position of the subtropical jet. Additionally, to quantify the difference in the regional baroclinic conditions, we use the Eady Growth Rate, as defined in : 6σ=0.31fN-1|∂zv| Where σ is the Eady Growth Rate between the 850 and 500 hPa pressure levels, f is the Coriolis parameter, N is the Brunt-Väisälä frequency, and ∂zv) is the vertical shear of the horizontal wind.

For both SST and Qf, the first three EOF patterns explain the majority of the variance in the data (76 % for SST and 78 % for Qf, left and right columns of Fig. , respectively), indicating they represent the dominant patterns of variability in these fields. The first EOF pattern (Fig. a, e, 47 % variance explained in both fields) can be described as generally capturing a gradient between the central Mediterranean and its eastern and western parts. The second EOF pattern of SST and Qf (Fig. b, f, 24 % and 27 % variance explained, respectively) generally captures east-west gradients across the Mediterranean basin, consistent with the characteristic east-west dipole seen in atmospheric variables . Specifically, the second EOF pattern of SST and Qf alike features a surface anomaly located between the Ionian and Tyrrhenian Seas (i.e., east and west of Sicily). The third EOF pattern of SST and Qf explains 5 % and 4 % of the variance, respectively, and depicts a north-south gradient of SST and Qf anomalies over the Mediterranean.

Statistically significant lagged correlations of the principal components of the EOF patterns and Levant winter precipitation are seen in different months for the three EOFs (Fig. d, h). Of these, the second EOF pattern of Qf stands out with a correlation coefficient R=-0.56 in August and R=0.40 in October. During the peak correlation month of August, Qf variations are dominated by latent heat fluxes, with minor contributions from sensible heat fluxes and negligible contributions from radiative fluxes (Fig. S15). Since latent and sensible heat fluxes are strongly dependent on near-surface winds and atmospheric conditions (i.e., temperature and humidity), this suggests that coupled ocean-atmosphere processes are linked to the lagged correlations.

We find that mean Qf values in the Aegean Sea reproduce the lagged correlations with Levant precipitation seen for the second EOF pattern. We therefore define an Aegean Sea Qf Anomaly index (AQA) as a precursor to Levant winter land precipitation. Specifically, AQA is defined as the Qf detrended anomaly from seasonal climatology in the north-eastern region demarcated in Fig. a–b (23.5–26.5° E, 32.5–35° N), normalized by the standard deviation of the anomaly timeseries. Note, however, that based on the second Qf EOF pattern, this index does not only indicate Qf changes in the Aegean Sea, but also contrasting Qf changes in the western Mediterranean (and similarly for SST; Fig. ). The AQA index was found to have considerably stronger predictive power compared to indices based on other regions of the Mediterranean and is not sensitive to ±0.25° variations of its boundaries (Fig. S11). A similar AQA index based on SST yields similar but statistically weaker results (see Sect. S2).

Monthly AQA values are shown in Fig. c. AQA is unitless, with positive values indicating higher ocean heat uptake in the Aegean Sea relative to climatological conditions (Fig. a, b). AQA is strongly anti-correlated with the principal component of the first and second Qf EOF patterns (R=0.60 and R=0.71 respectively, Fig. a) and can therefore be interpreted as generally proportional to the amplitude of these EOF patterns.

AQA is significantly correlated with Levant winter precipitation, in both ERA5 data and in situ IMS rain gauges, with the strongest lagged correlation in August (R=-0.60), in agreement with the Qf Patterns 2 and 3 (Fig. a). The correlation is strongest for winter (December–February), but is significant for each of the winter months (Fig. b). These results are not sensitive to the choice of December–February as the winter months, and remain significant for earlier, later, or longer winter months combinations (specifically, November–March, November–January, and January–March, shown in Fig. S9).

We now turn to analyzing the winter conditions associated with AQA variations by examining the differences between composites of winters preceded by August AQA values below and above plus and minus half of the standard deviation of August AQA values (∼13 years in each composite group; cf. Fig. ). The difference in winter SST between the two composites, shown in Fig. a, exhibits higher SST conditions in the north-eastern Mediterranean, which persist throughout winter (not shown). The increased winter SST in the north-eastern Mediterranean is indicative of increased upward surface heat and moisture fluxes, which, in turn, imply favorable conditions for cyclogenesis and storm intensification e.g.,. Accordingly, the winter Qf difference between these winter composites (Fig. b) shows decreased ocean heat uptake in the EM, driven primarily by increased upward latent and sensible heat fluxes during negative AQA composite winters.

The composite difference of hydrological changes is shown in Fig. . A significant increase in precipitation is seen in the EM (Fig. a). The majority of the precipitation increase is seen in the mean component of the moisture flux convergence (Fig. c), with minor contribution from transient eddies in the northern Levant (Fig. d), and negligible contribution from changes in evaporation (Fig. b). Further decomposition of the mean component (Fig. c) shows that the mean thermodynamic component (Fig. e) is negligible, while the mean dynamic component dominates the precipitation response (Fig. f).

We therefore conclude that the increased winter Levant precipitation associated with negative AQA anomalies during the preceding August is mediated by changes in regional mean flow patterns, creating more favorable conditions for precipitation by synoptic systems migrating eastward from the central Mediterranean to the Levant. Next, we turn to synoptic analysis to diagnose the meteorological conditions underlying the winter precipitation response to AQA.

Using the semi-objective synoptic classification , we assess the difference in synoptic conditions between winters preceded by August AQA values above and below half the standard deviation of August AQA values (Fig. a). Negative August AQA values are associated with increased prevalence of Cyprus Lows (CL, 32 vs. 25 d per winter) and decreased prevalence of Red Sea Trough systems (RST, 27 vs. 35 d per winter), with negligible changes in the other synoptic groups. Given that winter precipitation in the EM is dominated by eastward propagating Mediterranean cyclones i.e., Cyprus Lows;, and that Red Sea Trough synoptic conditions rarely lead to precipitation, these results are consistent with the winter precipitation response to AQA shown in Fig. a.

To assess whether the enhanced Cyprus Low activity results from increased number or duration of storms, we calculate the composite difference in the number of synoptic systems occurring in the region during winter (defined as the number of days minus the number of consecutive days of each synoptic type during winter; Fig. b). The number of Cyprus Low systems, as well as of Red Sea Trough and High systems, shows no significant sensitivity to AQA. This, in turn, indicates that the wetter Levant winters in response to negative AQA anomalies in the preceding August result from more persistent precipitating Cyprus Low systems during winter.

Since the hydrological decomposition pointed to changes in the mean flow as the primary driver of wetter Levant winters (Fig. f), we now asses the relation of AQA to the prevailing regional westerlies. As shown in Fig. a, negative AQA values in August are associated with a stronger subtropical jet over the EM during winter, which goes along with a low sea-level pressure anomaly over the Aegean Sea (Fig. S4) and a negative 500 hPa geopotential anomaly north of the Aegean Sea (Fig. c). This large-scale jet intensification coincides with increased upper-level divergence along EM storm tracks (Fig. S5) and a positive anomaly in the Eady Growth Rate (Fig. e), both indicative of more favorable conditions for baroclinic convective instability, consistent with the increased precipitation in the Levant (Fig. a).

In summary, our findings reveal that a negative ocean heat uptake anomaly in the Aegean Sea during August is a precursor to enhanced winter precipitation in the Levant. This link is dynamically mediated by the increased persistence of precipitating Cyprus Low systems traversing the region, driven by a strengthening of the subtropical jet over the EM and a concurrent intensification of regional baroclinicity.