This section presents the mean spatial fields and the dominant modes of spatio-temporal variability, explored through both univariate and multivariate EOF analyses, derived from 27 years of satellite observations of biological and physical variables.
The average surface mixed layer thickness (Fig. 1a) is generally thinner than 50 m, apart from the Aegean Sea, the south Adriatic and in particular the Gulf of Lion where it presents values twice as large as the rest of the basin. MLD pattern confirms the dynamic nature of the cold-water spot in the Gulf of Lion where it exceeds 125 m and represents one of the most energetic water mass formation sites in the Mediterranean (Marshall and Schott, 1999). The MLD signature in the Gulf of Lion leaves a clear imprint on phytoplankton abundance and productivity, as discussed below.
Mean time averages (1998–2025) of MLD
The average SST values (Fig. 1b) show a NW-SE gradient in all sub-basins, with the Alboran Sea, the Gulf of Lion, the North Adriatic and Aegean seas to be the cold end of the range of variability. This thermal structure is consistent with the MLD pattern described above and is likely driven by dynamical factors such as river outflows (Struglia et al., 2004), water mass formation (Marshall and Schott, 1999) or wind forcing (Millot, 1999) rather than due to insolation, only.
Figure 1c shows the basin-scale average PAR field. The southeastern basin receives approximately double the solar radiation compared to the northern part, reflecting both the expected meridional gradient and the longitudinal asymmetry visible in the image. This pattern also reflects the spatial distribution of cloudiness across the basin (Enriquez-Alonso et al., 2016).
Figure 1d shows the average CHL concentration over the entire period, revealing the well-known west-east gradient characteristic of the Mediterranean, with higher phytoplankton biomass in the western basin. The most phytoplankton-rich areas are the bloom region of the Gulf of Lion (Marty et al., 2002; D'Ortenzio and Ribera d'Alcalà, 2009) and the African coast, where Atlantic surface waters entering through the Gibraltar Strait drive the Algerian current (Millot, 1999; Bosc et al., 2004) and, together with Mediterranean intermediate waters, give rise to the western Alboran Gyre (Gascard and Richez, 1985), whose signature is clearly visible across all fields. Coastal areas are on average more enriched in phytoplankton than open waters, likely due to riverine nutrient inputs and coastal upwelling.
The average PP (Fig. 1e) spans approximately one order of magnitude, ranging from roughly 0.25 to 1.0 g m−2 d−1. Although PP broadly mirrors the spatial patterns seen in CHL (Fig. 1d), with higher values in the western basin and along coastal areas, its basin-scale gradients are comparatively less pronounced.
Not surprisingly, CHL and SST mean fields (Fig. 1d–b) here computed with 27 years of observations appear at first sight very similar to those previously computed with only nine years of observations as shown in Volpe et al. (2012, Fig. 3). However, a closer look at the data in the two reference periods reveals a general warming (Pisano et al., 2020; EU Copernicus Marine Service Product, 2024b) consistent with a basin-scale tendency toward upper-layer oligotrophication and reduced phytoplankton abundance (Colella et al., 2016; EU Copernicus Marine Service Product, 2024c). These differences can locally account for as much as 5 % increase in SST and 20 % decrease in CHL (Fig. B1). This SST-CHL inverse relationship can be broken locally where river inputs (north Adriatic Sea), coastal dynamics (north Aegean Sea) or permanent mesoscale features (Rhodes Gyre) play a major role in regulating both the thermal and phytoplankton fields.
Table 2 shows the outcome of the correlation analysis between the mean fields of Fig. 1. High correlation is here reported for SST-PAR, SST-CHL, PAR-CHL and CHL-PP. The other spatial correlation can be considered negligible.
Pearson spatial correlation coefficient computed between the various variables.
Percent dominance of PSC size classes (Micro, Nano and Pico in panels
As expected from their functional forms and from being derived as a direct function of CHL (Di Cicco et al., 2017, 2025), the mean fields of PSC (Fig. B2) spatially show patterns very close to CHL. By design and because of the known co-variability between phytoplankton accessory pigments associated with each fraction and total CHL concentration (Chisholm, 1992; Hirata et al., 2011), the three size classes individually highly correlate with CHL:
To achieve the percent PSC pixel dominance (Fig. 2), we counted, for each pixel, the number of months in which a given PSC showed a higher value than all the others. The resulting map was then normalized by the total number of months in the time series. Figure 2a, d show that microphytoplankton is the most abundant class only over 1 % of the basin in correspondence with coastal waters and the main river outflows. Thus, the terrigenous origin of the high nutrient levels in these areas sustains the development of larger phytoplankton cells throughout the time series, a condition otherwise rarely observed in the basin except in the deep-water formation region associated with the intense spring bloom in the northwestern Mediterranean (Marty et al., 2002; Uitz et al., 2006; D'Ortenzio and Ribera d'Alcalà, 2009; Siokou-Frangou et al., 2010; Volpe et al., 2012). Nanophytoplankton predominance is more widespread in both space and time (Fig. 2b), significantly contributing to the size distribution of the entire western basin, in the Adriatic and in the north Aegean Sea. The relevance of its contribution reflects the trophism of the areas, as clearly shown by the yellow areas in Fig. 2d where nano class is shown to be dominant. Figure 2c shows that the entire eastern basin along with the Tyrrhenian and the deep-bottom open waters of the Adriatic Sea are nearly always dominated by pico phytoplankton, consistently with in situ observations (Marty et al., 2002; Siokou-Frangou et al., 2010). The occurrence of this size class remains relevant also over the western basin where it dominates the size spectrum 75 % of the time. The dominance of the smallest size class is once more the fingerprint of the oligotrophic nature of the surface Mediterranean open waters.
Figures 3 and 4 show the spatial pattern and temporal amplitudes of the first EOF modes computed over MLD, SST, CHL, PAR and PP.
Spatial patterns of the first mode of EOF for MLD
First mode of EOF MLD
The first MLD mode accounts for 61.7 % of the total variance and is dominated by the Gulf of Lion, where values are two to three times larger than the rest of the basin (Fig. 3a), consistent with its role as the primary deep-water formation site in the Mediterranean (MEDOC Group, 1970; Marshall and Schott, 1999). Large variability is also found in the south Adriatic and in the Aegean Sea, both of which are subject to recurrent deep convection events. On the other hand, reduced variability is found in correspondence of permanent current systems such as the Algerian current in the western basin which splits into two branches: one in the Tyrrhenian and the other following the Atlantic Ionian Stream (AIS) towards the Levantine Basin. The temporal evolution of this mode (Fig. 4a) shows a pulsed sinusoidal signal with sharp maxima during winter mixing, much more pronounced than the low background value characteristic of the rest of the stratified period. Unlike the temporal evolutions of the first mode of the other variables (Fig. 4b–e), the one associated with the first mode of MLD shows a pronounced interannual variability with values spanning over one order of magnitude and with lower amplitudes during the last decade. The combination of the spatial pattern and temporal evolution indicates that the first MLD mode primarily represents the dominant basin-scale seasonal cycle of winter mixed-layer deepening, with the largest amplitude in the Gulf of Lion.
Figure 3b shows the SST first mode (97 % of the total variance) with three main patterns of homogeneous variability. The highest pattern of variability is associated with coastal areas of the western Adriatic Sea. The lowest SST variability is found at Gibraltar, in the Gulf of Lion, in the Aegean Sea and south of Crete Island, and to a lesser extent, in the Sicily Channel. These areas are affected by distinct physical mechanisms that attenuate the amplitude of the seasonal warming and cooling cycle driven by solar shortwave radiation. At Gibraltar, the continuous inflow of cooler Atlantic waters mitigates surface warming and reduces the seasonal thermal range (Harzallah et al., 2014). In the Sicily Channel, the AIS-driven upwelling system brings colder waters to the surface preferentially during summer (Bosc et al., 2004; Bonanno et al., 2014) dampening the seasonal SST signal. In the Aegean Sea and the Gulf of Lion, the dominant seasonal wind regimes (the Meltemi and the Mistral, respectively) drive enhanced turbulent heat loss and vertical mixing that limit surface ocean warming (Ziv et al., 2004; Millot, 1999; Herrmann and Somot, 2008). The temporal evolution of the SST mode (Fig. 4b) is a smooth, nearly symmetric sinusoid with positive values in summer (August) and negative values in winter (January–February), consistent with the annual cycle of solar heating.
The first PAR mode explains 98 % of its total variance (Fig. 3c). The highest variability is found in the northern and central parts of the basin, with the Aegean Sea standing out as a clear maximum. This is physically consistent with the larger seasonal swing of day length and solar zenith angle at higher latitudes, which drives a pronounced contrast between low winter PAR and high summer PAR values. The Aegean is further enhanced by the persistent clear-sky conditions maintained by the Meltemi wind regime (Tyrlis and Lelieveld, 2013), which removes clouds and aerosols and thus could maximise the amplitude of the seasonal PAR signal. Analogously, the African coast is characterized by lower variability. A band of lower variability is also visible along the northern coastlines, the Gulf of Lion, the Ligurian Sea and the northern Adriatic, where high frequency in cloud cover due to the orography of the region (Houze, 2012) could play an important role to reduce insolation and to moderate seasonal PAR variations. The temporal amplitude (Fig. 4c) is a near-perfect sinusoid with minima in December–January and maxima in June–July, similar to the SST first mode (Fig. 4b). This tight phase coherence reflects the well-established coupling between sea surface heating and incoming solar irradiance at the annual timescale, with PAR driving SST with only a slight lag (1–2 months) due to the ocean's thermal inertia.
Figure 3d shows the CHL first mode (77.1 % of the total variance). The highest CHL variability is found in the western basin, with special reference to the Balearic front area and the Algerian Current system. Variability decreases progressively eastward, reaching minima near river mouths and straits. A notable feature is the tongue of elevated variability extending from the Algerian Current toward the Tyrrhenian Sea and further into the eastern basin, closely resembling the corresponding signature in the MLD field (Fig. 3a). This spatial analogy is not coincidental: the winter reinforcement of the Algerian Current drives mesoscale instabilities that promote vertical mixing, leading to MLD deepening and the subsequent nutrient enrichment of the surface layer, which in turn enhances phytoplankton biomass and the associated chlorophyll signal. Furthermore, maximum positive values generally occur in February when cold nutrient-rich waters reach the surface through a deeper mixed layer, stimulating phytoplankton growth. CHL minima are associated with the summer stratification period (July–August, Fig. 4d). However, photo-acclimation has been claimed to explain this CHL annual cycle which is dominated by low light conditions and nutrient enhancement during winter and the opposite during summer (Behrenfeld et al., 2005; Bellacicco et al., 2016).
The first PP mode explains 76.2 % of its total variance (Fig. 3e), with areas of distinct variability. The highest variability is found in the coastal northern Adriatic Sea, where allochthonous nutrient inputs from rivers are well known to sustain elevated coastal productivity (Penna et al., 2004; Nixon, 2003; Cossarini et al., 2020). However, the PP signal in this region should be regarded with care because this area is characterized by optically complex waters and the PP model bio-optical parameterization derives from open-ocean conditions. Other coastal areas of significant variability are the straits of Gibraltar and the Dardanelles, where lateral advection of nutrient-enriched waters drives a clear seasonal signal, and with the open waters of the Ligurian Sea and the Gulf of Lion, reflecting the influence of winter convective mixing and the ensuing nutrient supply to the euphotic zone. This interpretation is consistent with Mediterranean biogeochemical studies showing that nutrient availability and primary production patterns are strongly shaped by hydrodynamic forcing, riverine inputs, and mixed-layer dynamics in key productive and convective regions of the basin (Cossarini et al., 2020; Reale et al., 2020). The remaining basin presents comparatively low annual PP variability. This pattern is modulated by a marked temporal amplitude, oscillating sinusoidally with positive values during summer, reaching maxima generally in July, and minima in winter, typically in December or January (Fig. 4e). This seasonal phasing perfectly matches PAR annual cycle (Fig. 4c) and SST (Fig. 4b), but in striking opposition of phase with CHL (Fig. 4d), whose maxima occur in late winter. This apparent paradox, maximum primary production when chlorophyll concentration is at its annual minimum, and vice versa, is a well-documented characteristic of the Mediterranean Sea and of oligotrophic gyres (Behrenfeld et al., 2005; Mignot et al., 2014; Bellacicco et al., 2016). The mechanistic explanation for this decoupling lies in the contrasting seasonal forcings that govern CHL and PP. In winter, deep convective mixing entrains nutrient-rich subsurface waters into the euphotic zone, stimulating phytoplankton growth and elevating bulk CHL concentrations. However, the concurrent low irradiance, as captured by the PAR minimum, limits the rate of photosynthetic carbon fixation, keeping PP suppressed despite high CHL values. Conversely, in summer, the development of strong thermal stratification reduces the vertical supply of nutrients to the euphotic layer, in agreement with Mediterranean biogeochemical studies linking nutrient availability to mixed-layer dynamics, and CHL declines to its seasonal minimum (Reale et al., 2020). Yet the high irradiance characteristic of the Mediterranean summer (PAR maximum) drives intense photosynthetic activity per unit biomass, sustaining elevated PP under otherwise oligotrophic conditions. This light-driven enhancement of summer PP is compatible with the process of cellular photo-acclimation: under high irradiance, phytoplankton reduce their cellular chlorophyll-to-carbon ratio, so that bulk CHL decreases even if photosynthetic rates remain elevated (Behrenfeld et al., 2005; Bellacicco et al., 2016). The first EOF mode of PP therefore captures a predominantly light-driven signal rather than one driven by nutrient or biomass dynamics, with PAR emerging as the primary modulator of the annual PP cycle at the Mediterranean basin scale. The alignment between the PP and PAR temporal amplitudes (Fig. 4e and c) directly reflects this dominant physical control.
It is worth mentioning that nearly the total variability of light (98 %) and temperature (97 %) is linked to the annual cycle while there is still a substantial part of variability to be explained for MLD, CHL and PP.
The first mode of phytoplankton CHL (Fig. 4d) presents high correlation values at lag zero with both SST (Fig. 4b,
Phytoplankton size diversity (PSC) is perfectly in line with CHL (Figs. 3d and 4d), as shown in Figs. B3a, b, c and B4a, b, c, with maxima in winter and minima in summer. We did use the same colorbar range in Fig. B3, just to show that Micro, being the less dominant and frequent (Fig. 2), shows the most pronounced variability. Variability reduces with the dominance of the cell size with Pico being the most stable, still with the very same space-time pattern as the others, including total CHL.
The second EOF mode computed over the MLD (Fig. 5a) explains 14 % of its total variability and presents two spatially distinct regions in opposition of phase: the Gulf of Lion on one side and the rest of the basin on the other. This spatial opposition suggests that this mode captures localised deep convection dynamics rather than the basin-scale annual signal dominated by the first EOF mode. The temporal amplitude (Fig. 6a) shows recurrent negative pulses during winter and positive values throughout the rest of the year, although occasional winter maxima can be observed in 2005, 2006, 2009, 2010, 2012, and 2013. These anomalous years have already been correlated with positive-to-negative North Atlantic Oscillation (NAO) transition (Basterretxea et al., 2018) which intensify northerly winds over the Gulf of Lion and favour deep-water formation events. Taken together, the spatial and temporal signals of this mode can be interpreted as representing sub-basin scale deep-water formation, with the Gulf of Lion as the primary driver, largely decoupled from the smoother seasonal pattern described by the first mode.
Second mode of EOF variability for MLD
Second mode of EOF MLD
Figure 5b displays the second EOF mode of SST (1.2 % of the total variance) with a clear northwest–southeast gradient, with predominantly positive anomalies in the western Mediterranean and negative values over the Levantine Basin. The northern Aegean Sea constitutes an exception within the eastern sub-basin, showing positive variability. This spatial pattern closely resembles that reported by Volpe et al. (2012), despite the longer time series used here. The corresponding temporal amplitude (Fig. 6b) shows a regular seasonal cycle, with positive values typically from January to August–September and negative values during the remaining months. Nevertheless, the amplitude of both maxima and minima varies considerably from year to year, reflecting interannual modulation likely linked to large-scale atmospheric forcing such as the NAO, consistent with the variability identified in the MLD second mode.
The second EOF mode of PAR (Fig. 5c) explains only 0.6 % of the total variance, confirming that PAR variability is largely dominated by the annual cycle. Its spatial pattern shows a weak dipole-like structure between the northwestern Mediterranean and the Levantine sub-basin. This residual signal may partly reflect non-seasonal modulation of cloud cover and surface shortwave radiation by atmospheric circulation patterns, but its physical interpretation remains uncertain and a contribution from noise cannot be excluded. Therefore, the second mode of PAR is not considered a robust component of the deep-water formation and spring bloom mechanism discussed in this section.
The second EOF mode of CHL (Fig. 5d) explains 5.6 % of the total variance and is spatially coherent with the MLD and SST second modes and aligns well with the results of Volpe et al. (2012). Pronounced positive anomalies are found in the northwestern Mediterranean, particularly in the Gulf of Lion, the Ligurian Sea, and to a lesser extent around the Bonifacio Gyre, while the rest of the basin shows limited variability. This pattern is clearly linked to spring bloom dynamics: the deep winter mixing captured by the MLD second mode in the Gulf of Lion supplies nutrients to the euphotic zone, and the subsequent restratification in spring, consistent with the SST second mode gradient, triggers intense phytoplankton growth. The temporal amplitude (Fig. 6d) confirms this interpretation, displaying recurring positive peaks during the spring months and negative values throughout winter and summer, with an interannual variability in bloom intensity.
The PP second EOF mode (Fig. 5e) explains 7.6 % of the variance and exhibits widespread positive variability throughout the Mediterranean basin, with negative values confined to the Adriatic Sea, Aegean Sea and along African coasts of the Levantine basin. This broader spatial signature, compared to the more localised chlorophyll second mode, suggests that PP integrates the regional bloom signal of the northwestern Mediterranean with a diffuse positive response across the open basin, possibly linked to the increased light availability. The temporal amplitude (Fig. 6e) follows CHL maxima in correspondence of the spring months but exhibits a faster and deeper decay during summer, with more intense negative minima than those observed in the CHL temporal signal. This asymmetry indicates that PP is more sensitive than CHL to the combined effect of nutrient depletion and reduced mixing in the post-bloom period.
Consistently with previous findings (Volpe et al., 2012), minima in the second mode of SST (Fig. 6b), that are likely associated with the pre-conditioning phase to deep-water formation in the Gulf of Lion, have the potential to drive the late-winter bloom in PP (Fig. 6e,
PSC second modes (Figs. B3d, e, f and B4d, e, f) perfectly align with the CHL second mode (Figs. 5d and 6d), exhibiting the dynamics of the spring bloom in response to the deep convection. This is consistent with the considerations expressed for both the average fields and the annual cycle variability. As in the first mode, the spatial intensity expressed by the second mode increases with the size classes, although less evidently. This indicates that during deep mixing nutrient abundance sustains the entire phytoplankton community. This is consistent with classical and trait-based phytoplankton size theory, according to which larger cells, particularly diatoms, are generally favoured under nutrient-rich and turbulent conditions, whereas smaller cells are more competitive under stratified and oligotrophic conditions (Margalef, 1978; Litchman and Klausmeier, 2008; Finkel et al., 2010). This is reflected here in the increasing extension of the blooming area with phytoplankton size.
The first MEOF mode (Fig. 7) represents 68 % of the total multivariate variance and identifies the dominant shared variability across the physical and biological variables. The temporal component (Fig. 7f) is a nearly perfect sinusoid with an annual period (summer maxima in July), stable amplitude, and no appreciable long-term trend across the entire 1998–2025 record, confirming that the dominant mode of co-variability across all five variables is the annual cycle driven by solar radiation and surface heating. Variance of this mode is however notably lower than the variance explained by the first modes of the univariate EOFs of PAR and SST (98 % and 97 %, respectively), which almost entirely capture the annual cycle of their respective fields. In the multivariate framework, a single temporal amplitude must simultaneously represent all five variables, and since MLD and CHL exhibit annual cycles that differ in phase and shape from those of PAR and SST, a fraction of the total variance is necessarily redistributed to higher MEOF modes.
First mode of MEOF of MLD
MLD and CHL spatial variabilities are uniformly negative across the basin, confirming that these two variables are in anti-phase with the seasonal solar-thermal forcing embedded in the temporal amplitude: MLD deepens and CHL peaks in winter, when the amplitude is negative. MLD shows the highest variability in the deep-water formation areas (Gulf of Lion and south Adriatic Sea) while CHL variability is larger in the western than in the Levantine basin. Notably, as in the univariate first EOF modes, the signature of the Algerian Current splitting into two branches is clearly recognizable here in both the MLD and CHL spatial patterns.
SST, PAR and PP variabilities are everywhere positive and in phase with temporal amplitude. Their spatial patterns are consistent with the dominant seasonal structures described by the univariate analyses, indicating that PP is organized around a physically coherent seasonal cascade: solar radiation heats the surface, stratifies the water column, suppresses deep mixing and nutrient supply, reduces phytoplankton biomass, yet sustains high photosynthetic rates per unit biomass, producing the summer PP maximum under oligotrophic conditions.
Second mode of MEOF of MLD
The second MEOF mode (Fig. 8) accounts for 22 % of the total multivariate variance and identifies the main secondary mode of shared variability across the physical–biological system. This mode is associated with the coupled spring-bloom response. Although several features are consistent with those found in the univariate second EOFs, the MEOF mode reflects shared covariance among variables rather than a direct correspondence with individual univariate EOF modes.
The temporal amplitude (Fig. 8f) displays a regular seasonal cycle with positive values in spring and negative values in late summer through winter, with clear interannual variability in peak shape and magnitude. This temporal structure is consistent with the springtime variability described by CHL and PP, indicating that this MEOF mode is mainly associated with the spring bloom signal and its interannual modulation.
The MLD (Fig. 8a) shows positive variability concentrated in the Gulf of Lion and Ligurian Sea but extends more broadly into the Aegean Sea and parts of the eastern Mediterranean. This spatial footprint suggests that, within the multivariate mode, MLD contributes through a broader mixing/preconditioning signal across multiple convection-prone areas of the basin, rather than through a signal confined to the Gulf of Lion.
The SST (Fig. 8b) is negative across the whole basin, with the largest anomalies in the western Mediterranean and northwestern coastal areas. This pattern is physically consistent with post-convection cooling and with the seasonal thermal preconditioning that precedes the spring bloom.
The PAR (Fig. 8c) is positive across the entire basin, with highest values in the western and central Mediterranean. The positive PAR values, in phase with the positive MLD and the negative SST, suggest that the second MEOF mode represents the transition from winter deep mixing to spring re-stratification, during which increasing irradiance acts as the trigger for phytoplankton growth.
CHL variability (Fig. 8d) shows pronounced positive anomalies in the Gulf of Lion and surrounding northwestern Mediterranean regions, indicating that spring bloom dynamics strongly contribute to this multivariate mode.
The PP variability in the second MEOF mode (Fig. 8e) shows high variability mainly in the Gulf of Lion, emphasizing the productive bloom response of the northwestern Mediterranean and its tight link with CHL, in contrast with the more stable oligotrophic conditions of the eastern basin.