Dense Water Formation Variability in the Aegean Sea from 1947 to 2023

Abstract

The formation of dense water in the Aegean Sea is important as it affects the deep circulation and the hydrography of the Eastern Mediterranean Sea. In this study, the variability of dense water formation is investigated in relation to forcing mechanisms from 1947 to 2023 in the subbasins of the Aegean Sea, utilising in situ observations from various sources, which have been analysed in combination with satellite altimetry and reanalyses products. The analysis reveals that the Aegean Sea has been in a state of increased dense water formation since 2017 due to the combination of increased surface buoyancy loss and reduced Black Sea water inflow. Extremely high salinity has been recorded in the intermediate layers of the Aegean Sea since 2019. The anticyclonic circulation of the North Ionian gyre during 2017 and 2018 probably also contributed to the rapid transport of highly saline waters in the intermediate and, through dense water formation, the deep layers of the Aegean Sea in 2019. Until 2022, the dense waters formed during the peak of the Eastern Mediterranean Transient still occupied the bottom layers of some deep subbasins of the North and South Aegean; however, the $29.4 \mathrm{k}\mathrm{g} {\mathrm{m}}^{-3}$ isopycnal in the North Aegean and the $29.3 \mathrm{k}\mathrm{g} {\mathrm{m}}^{-3}$ isopycnal in the Southeastern Aegean have gradually deepened by 800 m, permitting the waters forming in the last ten years in the Aegean Sea to settle at ever greater depths. Temperature controls the density variability of the Cretan intermediate water up to the decadal time scale. Increased data availability since 2010 was sufficient to clarify that intrusions of dense water from the North–Central Aegean Sea contributed to the erosion of the Eastern Mediterranean transitional waters in the South Aegean Sea after 2017, as well as to raising the intermediate water masses of the South Aegean to shallower depths. The erosion of the transitional Mediterranean waters in the South Aegean Sea between 1947 and 1955 and 1973 and 1980 coincided with increased dense water formation in the North–Central Aegean Sea. During the peak of the Eastern Mediterranean Transient, the North Ionian circulation, the Black Sea water inflow, the Atlantic Multidecadal Oscillation, and the surface buoyancy fluxes favoured dense water formation in the Aegean Sea.

1. Introduction

1.1. The Aegean Sea and the Eastern Mediterranean Transient Event

The Aegean Sea (Figure 1) is among the Mediterranean Sea subbasins where dense water formation (DWF) occurs regularly [1,2]. Locally formed water fills the deep subbasins of the region, while part of it spreads into the intermediate and deep horizons of the Eastern Mediterranean Sea. The extremely dense water produced in the Aegean Sea about 30 years ago during the Eastern Mediterranean Transient (EMT) reached the bottom of the Eastern Mediterranean Sea, which was formerly filled with water from the Adriatic Sea [3,4,5].

The EMT was a climatic shift during which the Aegean Sea replaced the Adriatic Sea as the main source of deep water of the Eastern Mediterranean Sea [4,6], and produced a new configuration of its deep waters due to a net transport of salt from the near-surface to the bottom layers of the basin [7,8]. It took place roughly from 1987 to 1997 and it is the most intense thermohaline circulation change recorded in the last 120 years for which observations are available [9]. Model results and sediment core analyses indicate that similar events have taken place between 1500 and 1900, with some regularity around centennial time scales [10,11]. The main driver of the EMT was the increased winter heat losses in 1987, 1992 and 1993, reinforced by other factors, such as the long-term decrease in the surface water temperature of the Eastern Mediterranean Sea, and the increase in salinity in the Levantine Sea due to changes in large-scale circulation [6,12,13]. In the Aegean Sea, the reduced inflow of buoyant Black Sea water (BSW), which also insulates the North Aegean Sea by reducing winter heat loss through decreased air–sea temperature difference, was observed [14]. Additionally, increased DWF in the North Aegean Sea intensified its exchanges with the South and Central Aegean Sea, in a positive feedback mechanism bringing saltier water to the north [14].

1.2. Intermediate Water Formation in the Aegean

The formation of intermediate water presumably takes place on an almost annual basis in the Aegean Sea. An intermediate water mass is formed in the Cretan Sea, the Cretan Intermediate Water (CIW), which settles at $\sim 300 \mathrm{m}$ in the Aegean Sea and at $\sim 400$–700 $\mathrm{m}$ in the Eastern Mediterranean Sea [1,15,16,17]. Its properties are similar to those of the Levantine Intermediate Water (LIW), but slightly colder and saltier, and present significant interannual variability [18].

1.3. Influence of the BIOS Mechanism

Significant variability of DWF in the Aegean Sea is also observed at the decadal time scale [2,19,20]. The possible thermohaline interdependence of the Aegean and the Adriatic Seas, by means of compensatory flows of intermediate water masses that redistributed salt and could result in the restoration of the Adriatic Sea as the main source of dense water in the Eastern Mediterranean Sea, was noted shortly after the EMT [6,7,21]. More recently, the relation of the quasi-decadal inversions of the surface circulation in the Ionian Sea [22] with the DWF in the Adriatic Sea [23] (named the Adriatic-Ionian Bimodal Oscillating System, BiOS) and the salinity distribution in the eastern Mediterranean basin corroborated the connection of the Adriatic and Aegean Seas [24]. Evidence for the out-of-phase variability of salinity in the Levantine (Aegean) and Ionian (Adriatic) Seas due to the circulation reversals of the North Ionian Gyre and the importance of considering them as one system, at least over decadal and longer time scales, is piling up [3,19,25,26,27,28]. Various mechanisms, which favour either internal/intrinsic [18,20,25,29,30] or forced/atmospheric causes [31,32,33,34], have been proposed to explain the observed quasi-decadal reversals of the Ionian Sea circulation and the DWF variability in the Aegean and Adriatic Seas. Although the sensitivity of the basin to climatic forcing is generally accepted [25,35,36,37], the discussion about the relative importance of the proposed mechanisms is still open.

Interestingly, a clear decadal or quasi-decadal relation between the near-surface salinity variability of the Levantine and Ionian Seas with the near-surface salinity variability of the Aegean Sea has not been demonstrated yet through field observations; the salinity relation emerges only when the intermediate and deep layers of the Cretan Sea are considered, probably due to the rapid salt transport from the surface to the intermediate and deep layers of the Aegean Sea through DWF [19,24]. Utilising a sea surface height (SSH) index to describe the dynamics of the Ionian Sea, Reale et al. [28] conducted a multi-model analysis and found zero-lag correlation between the near-surface salinity of the Central Aegean Sea and the circulation changes in the Ionian Sea.

1.4. Overturning Processes

Understanding the overturning processes in the Aegean is a demanding task, due to the complex coastline and seabed morphology of the basin and its interaction with the Black Sea. Despite the fact that most Mediterranean-scale analyses identify the Cretan (or South Aegean) Sea as the source of Cretan Deep Waters (CDW) [34], we now know that winter-time convection in the Cretan Sea rarely reaches $\sim 400 \mathrm{m}$ depth, and that the upper threshold of CIW in the post-EMT period is $\sim 29.1 \mathrm{k}\mathrm{g} {\mathrm{m}}^{-3}$ [5,38,39]. Consequently, the source of the saline and denser than $29.1 \mathrm{k}\mathrm{g} {\mathrm{m}}^{-3}$ waters that appear in the South Aegean Sea is the North–Central Aegean Sea, which exports dense waters stored in the latter basins’ depressions during DWF events [5,14,40]. The rapid transport of dense and saline water from the North–Central Aegean Sea to the South Aegean and the consequent increase in salinity in the intermediate and deep layers of the latter implies that intrusions of dense water could play an important role in the erosion of the recurrently appearing transitional Mediterranean waters (TMW) in the Cretan Sea, but this has not been shown yet from field observations. According to Velaoras et al. [15], the erosion of TMW from 2006 to 2014 was probably caused by vertical diffusion, whilst Theocharis et al. attributed the 1998–1999 erosion to intrusions from the North–Central Aegean Sea [15,41]. Also, whether the interannual density changes of the CIW depend more on salinity or temperature remains an open question [5].

Dense-water formation processes clearly reflect current climate conditions, determining the location of maximum buoyancy loss and ocean ventilation areas. However, they also constitute the local “engines” of overturning circulation, redistributing heat, salt and nutrients throughout the basins, and thus affecting the local atmospheric conditions and the climate in the broader regions

1.5. Variability of Hydrographic Characteristics

Mean trends show that the Mediterranean Sea, which has been identified as a climate change “hot spot”, is getting warmer and saltier since the early eighties [42,43,44,45], and its sensitivity to climate change has raised the question regarding the evolution of its thermohaline circulation [25,35]. Observational uncertainty became lower than the full-depth interannual variability of the Mediterranean Sea around 1980 for temperature and around 2000 for salinity [46]. In the last thirty years, positive long-term trends of temperature and salinity are evident almost everywhere and throughout the water column whenever observations’ availability and accuracy permit calculation of long-term changes, with a few notable exceptions, such as the Adriatic Sea [29,35,47,48,49,50,51,52,53]. Compared to the deeper layers, the climatic variability of the surface and intermediate layers is better studied due to the higher signal-to-noise ratio and the higher data availability, while their relation to large-scale oceanic and atmospheric modes has been pointed out [31,37,54,55,56].

1.5.1. Temperature Variability

The well-correlated temporal variability of heat content and net heat fluxes through the sea surface of the Mediterranean, as evidenced in a set of forced and coupled simulations, reveal that atmospheric forcing is more important than lateral fluxes through the Straits in determining the heat budget of the Mediterranean Sea over interannual time scales [57]. This is also evidenced by the sea surface temperature (SST) multidecadal variability driven by large-scale atmospheric circulation modes, such as the North Atlantic Oscillation (NAO), and the East Atlantic/West Russian mode (EAWR, also referred to as the North Caspian Pattern) [37,55,58,59,60,61].

The Mediterranean SST multiannual variability stopped following the Atlantic Multidecadal Oscillation (AMO) in 2007, a development attributed to changes in local circulation and surface fluxes [56]. During the satellite era (early 1980s onwards), a consistent, summer-driven mean warming trend of SST is observed throughout the Mediterranean Sea [62,63,64]. Widespread summer and autumn increase in the mixed layer temperature supports those findings [65]. Locally, these SST warming trends can be modified by oceanic features, as can be seen by lower trends at the Strait of Gibraltar, the upwelling region in the Eastern Aegean Sea, and the two pathways followed by the Atlantic Water after entering the Eastern Mediterranean Sea [64].

1.5.2. Salinity Variability

The Mediterranean Sea has been getting saltier in the last decades; this increase is attributed to increased evaporation over its surface rather than changes in salt fluxes through its straits [66]. The spread of salinity variability is larger in regions of brackish water inflow, such as the Dardanelles Strait and Po River [43]. Salinification of the surface layer is stronger and spatially more coherent in the Eastern Mediterranean Sea compared to the Western Mediterranean Sea, with maximum long-term salinity trends observed in the Ionian Sea and the Cretan Passage [43,67]. The basin-wide peak of salinity occurred around 2018 and coincided with the most recent (2017–2018) reversal of the North Ionian Gyre (hereafter: NIG) [59,68]. Salinity in the Levantine Sea increased rapidly and extremely high values were observed along its periphery [69,70]. Since the magnitude of the NIG reversals is not proportional to the recent salinity changes in the Levantine Sea, it has been suggested that local atmospheric forcing contributed to the observed increase in salinity [59,70]. In particular, the strengthening of winds due to the enhanced southern pole of the NAO atmospheric system has been linked with increased evaporation and the weakening of the cyclonic circulation in the eastern part of the basin [59].

Salinity variability in the South Adriatic Pit has been in line with the NIG reversals until 2016. Despite the 2017–2018 NIG anticyclonic reversal, the salinity in the Southern Adriatic Pit has been increasing continuously since 2017, reaching historically high values in 2021 [68]. A combination of local and large-scale factors has been proposed to explain this salinity increase, including the contribution of saltier intermediate waters from the Aegean and/or Levantine Seas [68,71], which, in turn, implies that the Aegean Sea could be in a state of increased DWF during that period.

1.6. Purpose and Structure of the Work

Knowledge of the interannual variability of dense-water formation in the Aegean Sea and the corresponding export of Aegean waters into the Levantine and Ionian basins is crucial in understanding the observed variability in the whole Eastern Mediterranean basin. However, the determination of the timing and the strength of overturning processes in the Aegean Sea is tricky due to its complex topography, the high temporal variability, the strong thermohaline gradients in the North and Central Aegean Sea, and the different residence times between its depressions. To our knowledge, a long-term analysis of DWF encompassing all open-sea/deep regions of the Aegean Sea does not exist in the literature. The purpose of this study is to describe the hydrographic and DWF variability of the Aegean Sea from 1947 to 2023 and its relation to proposed DWF mechanisms.

The paper is structured as follows: The methodology along with the description of the data is given in Section 2. The results are presented in Section 3. To our knowledge, no studies regarding the DWF in the North Aegean Sea exist between 2017 and 2022; thus, we first show the recent hydrographic variability in the Athos Basin. Then, we present the hydrographic and DWF variability of all the subbasins of the Aegean Sea during the good-quality period (i.e., 1985 onwards) in order to look for a possible relation of DWF between the subbasins of the North–Central Aegean Sea and the subbasins of the South Aegean Sea. We then extend our analysis backwards to investigate the density variability of an important water mass, the CIW, which forms in the South Aegean Sea. Finally, having established that the DWF incidents of the North and Central Aegean can be traced to the South Aegean, we proceed to examine the hydrographic and DWF variability of the Aegean Sea during the data-poor period by taking advantage of the above finding to increase our confidence in the observations of the period. In Section 4, we discuss our findings and conclude in Section 5 with a summary of the most important results.

2. Materials and Methods

2.1. Hydrography Data

The temperature and salinity dataset used in this study was constructed by merging five curated datasets available online. The Mediterranean Sea–Temperature and Salinity Historical Data Collection SeaDataCloud V2 [72] was downloaded from the Alfred Wegener Institute, the World Ocean Database 2018 [73] was downloaded from the National Centers for Environmental Information, the MEDATLAS Hydrographic and Bio-Chemical Data of the Mediterranean and Black Sea [74] was downloaded from the Alfred Wegener Institute, and the CORA-GLOBAL [75] was downloaded from the Copernicus Marine Service. Argo float profiles [76] were downloaded from the Coriolis Project (http://www.coriolis.eu.org, accessed on 21 August 2024). Because the update frequency of delayed-mode datasets is one to two years, recent Argo profiles that have not undergone delayed mode quality control have also been incorporated in our merged dataset. Only good quality measurements, according to the quality flags of the downloaded datasets, were used in the analysis. After removing duplicate measurements, the final dataset was inspected visually for any remaining obviously erroneous data which were excluded manually. We have used only the good quality data of the individual curated datasets, some of which have been reviewed by experts in the study area, but we have not taken any further steps to correct for potential biases in the historical temperature and salinity profiles. The Ocean Data View v.5.6.3 (ODV) [77] software was used to construct the merged temperature and salinity dataset.

Directly observing the convective phase of DWF is challenging due to its intermittent nature and because it takes place under harsh weather conditions. The bulk of observational information regarding the spatiotemporal variability of DWF in the Aegean Sea comes from the study of its lateral exchange and spreading phase, i.e., through the doming of isopycnals and the thickness of layers defined by different density thresholds, as well as $\theta /S$ analysis of deep and intermediate water masses. The regular deployment of a sufficient number of Argo floats in the Aegean Sea since 2014 enabled the continuous monitoring of winter convection in some cases [38,39,78]. Here, we have divided the Aegean Sea in nine subbasins and analysed their hydrographic variability through time–depth plots (Hovmöller diagrams) of potential temperature, salinity and potential density using the profiles of the merged temperature and salinity dataset. Time–depth plots were created with the ODV software (v.5.6.3) using the weighted-average gridding method. The data are unevenly spaced and significant gaps exist in all subbasins. Inevitably, temporal accuracy has been compromised during the gridding procedure in our time–depth plots. To ameliorate the problem, we have included the distribution of the data for each subbasin in the relevant figure and we will refer to the dates of available profiles in the text when necessary.

To investigate the effect of temperature and salinity variability on the density variability of the CIW, time series were calculated in the 20– 250 $\mathrm{m}$ layer from the available profiles in the Cretan Sea. First, the profiles with minimum sampled depth shallower than 20 $\mathrm{m}$ and maximum sampled depth deeper than 250 $\mathrm{m}$ were selected. The selected profiles were linearly interpolated in the vertical direction, and then depth averages were computed. Finally, the casts belonging to the same month were averaged, and whenever more than three casts were available during the same month, the standard deviation was also computed. Only the profiles belonging to January, February, March and April were used for the analysis.

To complement the analysis which was based on the profiles, the in situ observations from the fixed point observatory E1-M3A in the Cretan Sea, which are part of the Mediterranean Sea In Situ Near Real Time Observations product [79,80], were downloaded from the Copernicus Marine Monitoring Service. Monthly averaged time series were calculated from the three-hourly measurements of the observatory. First, a low-pass filter with a temporal window of one week was applied to the initial data. Then, to account for the uneven observation depths (20, 50, 100 and 250 $\mathrm{m}$), the low-passed time series were linearly interpolated in the vertical direction at regular depth intervals. Finally, the monthly averages and standard deviations were calculated only when the temporal data return was $>50\%$, and observations were available at least at three depths.

The CORA Temperature and Salinity Analysis dataset [75] was downloaded from the Copernicus Marine Service. The gridded temperature and salinity fields are produced by optimal interpolation [81] as implemented in the In Situ Analysis System (ISAS) [82]. Available hydrographic observations from various data aggregators are used to produce monthly fields on a $0.5°$ grid at the equator with varying latitude resolution ($\sim 0.4°$ in the Mediterranean), and 152 levels from 0 to 2000 $\mathrm{m}$. The CORA dataset is available since 1960. The coarser EN4 objective analysis [83] with 1$°$ spatial resolution, 42 levels, and varying vertical resolution was downloaded from the Met Office Hadley Centre. The surface salinity fields of the two datasets were used for the calculation of buoyancy fluxes as explained in Section 2.4.

The Aegean Sea is usually divided into north (Lemnos and Athos Basins), central (Skyros, Chios, Ikaria Basins and Cyclades Plateau) and south (Myrtoan and Cretan Basins) parts [13,84]. Our hydrographic analysis will follow this convention.

2.2. Sea Level and Geostrophic Circulation

The European Seas Gridded L4 Sea Surface Heights And Derived Variables Reprocessed 1993 Ongoing dataset (satellite SSH hereafter) was downloaded from the Copernicus Marine Service for the 1993–2020 period. Its near-real-time counterpart was downloaded for 2021 and 2022. These products, which provide a consistent observational dataset of SSH over the European Seas and its subbasins, are provided on a $0.125°\times 0.125°$ grid with 1 d temporal resolution. In this study, we used the Absolute Dynamic Topography (ADT) to create an index of interannual BSW flow into the Aegean Sea (see Section 2.5) and to investigate the variability of the geostrophic circulation of the Eastern Mediterranean Sea.

Empirical Orthogonal Function (EOF) analysis [85] of the ADT field was conducted east of 15 °E, excluding the Adriatic Sea. The field-mean of daily ADT was removed prior to the application of the EOF. With those analysis choices, the variability of North Ionian Reversals appears in the first mode, the seasonal and long-term variability of the mean sea level has been removed, and our results are directly comparable with those of Gacic et al. [24].

2.3. Hydrodynamic Model Data

Monthly averaged SSH fields of the Mediterranean Sea Physics Analysis and Forecast (MFC hereafter) with a spatial resolution of $0.042°$ were downloaded from the Copernicus Marine Service for the 1987–2022 period [86]. The same EOF analysis which was applied to the satellite data was also conducted on MFC SSH.

Liu et al. 2021 [33] simulated the Mediterranean Sea from 1901 to 2010 to investigate the drivers of the North Ionian Sea reversals, and created an index of NIG cyclonic and anticyclonic circulation based on the near-surface zonal velocity of the North Ionian Sea. The simulation compared well with the known North Ionian Sea reversals during the satellite period. We have digitised the index from their Figure 4 and calculated its year-mean to complement our EOF analyses and extend it backwards to 1940, as satellite altimetry data are available since 1993 and MFC data since 1987.

2.4. Atmospheric Forcing

The ERA5 reanalysis [87] monthly averaged longwave, shortwave, sensible and latent heat fluxes, and precipitation and evaporation from 1940 to 2022 were downloaded from the Climate Data Store and were used to calculate the surface buoyancy fluxes over the study area. The surface buoyancy fluxes were calculated as [88]:

$$B_o=Bq+Bp=-g\alpha {\displaystyle \frac{Q_o}{\rho C_p}}+g\beta (E-P)S_s,$$

(1)

where $B_q$ are the thermal and $B_p$ the haline components, respectively, $Q_o$ is the net thermal flux, g is the acceleration of gravity, $\alpha$ and $\beta$ are the thermal expansion and haline contraction coefficients of seawater, respectively, $C_p$ is the heat capacity of seawater, E is the evaporation, P the precipitation, and $S_s$ the sea surface salinity. The buoyancy fluxes were calculated twice, using the $S_s$ from the CORA and EN4 datasets. The monthly mean $S_s$ fields of CORA and EN4 were bilinearly interpolated on the ERA5 grid. Only the ERA5 cells with proportion of water ≥80% were used for the calculation of buoyancy fluxes, in order to reduce extrapolation errors from the coarser CORA and EN4 grids and the contamination from land fluxes. The differences in the buoyancy fluxes calculated with the CORA and EN4 datasets were negligible for the purposes of our study; thus, only the buoyancy fluxes calculated using EN4, which extend back to 1940, are shown. Winter mean fluxes (November–March) were calculated from the monthly ERA5 fields.

2.5. BSW Inflow

The Mediterranean and Black Seas exchange water through the Turkish Straits System (TSS), where a two-layer flow occurs due to the density difference between the two basins. The cold and brackish upper layer flows into the Aegean Sea, while the salty and warm lower layer flows into the Black Sea. The net flow at interannual time scales is from the Black Sea to the Aegean Sea due to the sea level difference between the two basins. To date, no information on the long-term variability of BSW flow into the Aegean Sea is available for the whole period covered in this study. To assess the interannual variability of the BSW inflow, we used the upper-layer volume flow Q and buoyancy flux F from the Dardanelles to the Aegean Sea from Maderich et al. 2015 [89] that spans the 1970–2009 period, and the net water exchange anomaly $\langle N\rangle$ between the Mediterranean and Black Seas from Garcia-Garcia et al. 2022 [90] that spans the 2003–2016/2019–2020 period. The daily volume rate and buoyancy flux of the upper layer from the Sea of Marmara to the Aegean Sea, kindly provided by V. Maderich to our laboratory, were calculated by a chain of box models for the seas and hydraulically controlled strait models based on the sea level gradient along the TSS and the density difference between the upper and lower layers. The net exchange between the Mediterranean and Black Seas, which we digitised from Figure 7 of Garcia-Garcia et al. [90], was calculated through a water-budget analysis based on satellite gravity observations and model output of evaporation and precipitation over the two basins and their catchment areas. Since net water flow at interannual scales is from the Black Sea to the Aegean Sea [89], the long-term variability of net exchange could also serve as an indicator of BSW inflow to the Aegean Sea.

The daily Absolute Dynamic Topography (ADT) was spatially averaged in two boxes with area $\sim 4.5\times {10}^3 {\mathrm{k}\mathrm{m}}^2$ at the exits of the Dardanelles and the Bosporus Straits, and then the difference $\Delta \zeta$ between the two boxes was computed. $\Delta \zeta$ was shifted in time by one month to take into account the lag between the SSH difference along the Turkish Strait System and the upper layer flow at the Dardanelles [89,91]. The adequacy of $\Delta \zeta$ as an indicator of the interannual BSW inflow was assessed by comparing it with Q, F and $\langle N\rangle$. The correlation coefficient r was calculated for monthly averaged, low-passed monthly averaged and annually averaged $\Delta \zeta$, Q and F. Because only non-seasonal exchange is shown in Figure 7 of Garcia-Garcia et al. [90], r was calculated only for the low-passed monthly averaged and annually averaged $\Delta \zeta$ and $\langle N\rangle$.

The results of the assessment of the suitability of $\Delta \zeta$ as an index of BSW flow into the Aegean Sea are shown in Figure S1. The correlation for the low-passed and annually averaged $\Delta \zeta /Q$, $\Delta \zeta /F$ and $\Delta \zeta /\langle N\rangle$ was roughly between ∼0.7 and ∼$0.75$. The monthly averaged $\Delta \zeta /Q$ and $\Delta \zeta /F$ correlations were ∼$0.65$ and ∼$0.5$, respectively. The high correlation of $\Delta \zeta /F$ (∼$0.7$) implies that lateral buoyancy flux into the Aegean Sea is mostly due to variations in volume flow changes. Lateral buoyancy input from the Black Sea is important for DWF in the Aegean Sea as it almost counteracts the surface freshwater buoyancy flux over the Aegean Sea [92]. Since we are interested in the interannual variability of the flow rate, we conclude that the $\Delta \zeta$ suffices as an index of BSW inflow for the requirements of this study, although the $\Delta \zeta$ index will fail to detect events of strong baroclinic inflow at the Dardanelles Strait that do not affect SSH at the TSS. In the following sections, we will refer collectively to the year-mean of the three variables ($\Delta \zeta$, Q, $\langle N\rangle$) as BSW inflow.

3. Results

3.1. Variability in the Athos Basin from 2016 to 2023

The hydrographic and DWF variability in the Athos Basin from 2016 to 2023 is shown in Figure 2. From the doming of the isopycnals, it can be seen that dense water with ${\sigma }_{\theta }>29.3 \mathrm{k}\mathrm{g} {\mathrm{m}}^{-3}$ was produced in 2017, 2019 and 2022, while missing observations in 2018 and 2021 hinder further investigation during these years. DWF occurred also in 2020, but it was much weaker as convection reached down to $\sim 250 \mathrm{m}$ and only the isopycnals $<29.2 \mathrm{k}\mathrm{g} {\mathrm{m}}^{-3}$ shoaled. In 2016, the water column presented a typical structure, with cold and saline waters below $\sim 500 \mathrm{m}$, warmer and less saline waters from $500 \mathrm{m}$ to $150 \mathrm{m}$, and very low salinity waters shallower than $150 \mathrm{m}$. This typical structure of the column changed after the formation of 2017, which was initiated by large surface heat loss during early winter [36]. The $29.1 \mathrm{k}\mathrm{g} {\mathrm{m}}^{-3}$ isopycnal rose by more than $200 \mathrm{m}$ and the $100$–$300 \mathrm{m}$ layer was filled with very cold and low salinity waters, probably due to the penetration of the thin surface BSW layer and the transport of brackish waters from the shelves of the North Aegean Sea. By mid 2018, the salinity has increased again in the intermediate layer and a low-salinity core, probably the remnants of the 2017 formation, was present at the sill depth around $400\mathrm{m}$. The 2017 formation affected only the intermediate layers of the Athos Basin down to $\sim 700 \mathrm{m}$ [36], and the deep layers remained stagnant.

The most drastic thermohaline change in the water column occurred during the formation of 2019. Salinity increased by more than $0.1$ from the surface down to 600 $\mathrm{m}$ after May 2019. From 2016 to 2019, the deep layers were stagnant and progressively gained buoyancy through vertical mixing; hence, the $29.4 \mathrm{k}\mathrm{g} {\mathrm{m}}^{-3}$ isopycnal gradually deepened. In late 2019, a saline and warm intrusion caused the $29.42 \mathrm{k}\mathrm{g} {\mathrm{m}}^{-3}$ isopycnal to shoal by >$150 \mathrm{m}$, ending the stagnation period. The formation of 2022 was strong enough to affect the $29.35 \mathrm{k}\mathrm{g} {\mathrm{m}}^{-3}$ density horizons and resulted in the drop of salinity and temperature, especially in the first $600 \mathrm{m}$ of the water column. The decrease in salinity and temperature were higher in the first $\sim 400 \mathrm{m}$ due to open-sea convection. In general, open-sea convection in the Athos Basin, which causes the doming of isopycnals in the first $\sim 300 \mathrm{m}$, is accompanied by a decrease in salinity due to the presence of modified BSW in the surface layers. This is quite evident in 2017, 2019, 2020 and 2022. The increase in density in the deeper layers, which is accompanied by an increase in salinity, is attributed to intrusions of dense waters from the southeastern region of the North–Central Aegean Sea [93], which accumulate in the Athos Basin due to the general cyclonic circulation of the Aegean Sea [94].

3.2. Variability in the Aegean Sea since 1985

The hydrographic and DWF variability in the nine subbasins of the Aegean Sea since 1985 is shown in Figure 3. The change in the thermohaline properties was not uniform in the subbasins of the North–Central Aegean Sea during the formation of 1987. In the Athos and Lemnos Basin, temperature and salinity dropped, in the Chios and Ikaria Basins, temperature dropped and salinity increased, whilst in the Skyros Basin, temperature remained almost constant and salinity increased. In the South Aegean, temperature dropped and salinity increased in the Myrtoan Basin, while temperature remained unchanged and salinity increased in the subbasins of the Cretan Sea. Density increased in all the subbasins of the North–Central Aegean Sea. The highest densities ($>29.4 \mathrm{k}\mathrm{g} {\mathrm{m}}^{-3}$) have been recorded in the basins of Lemnos and Skyros. In the Athos Basin, the observed densities were $>29.35 \mathrm{k}\mathrm{g} {\mathrm{m}}^{-3}$, and in the Chios and Ikaria Basin $>29.3 \mathrm{k}\mathrm{g} {\mathrm{m}}^{-3}$. The response to the formation of 1987 has not been the same in all the subbasins of the South Aegean Sea. Waters with density $>29.2 \mathrm{k}\mathrm{g} {\mathrm{m}}^{-3}$ appeared in the Myrtoan and West Cretan Basins already by April 1987. The $29.2 \mathrm{k}\mathrm{g} {\mathrm{m}}^{-3}$ isopycnal rose by $>500 \mathrm{m}$ in the Myrtoan Basin and by $>300 \mathrm{m}$ in the West Cretan Basin. The $29.2 \mathrm{k}\mathrm{g} {\mathrm{m}}^{-3}$ isopycnal started to rise with a lag of about one year in the Central Cretan Sea and two years in the East Cretan Sea. Although the increase in density in the North–Central Aegean Sea was not accompanied by an increase in salinity in all of its subbasins, the result of the 1987 formation was a net salt transport towards the deep and intermediate layers of the Cretan and Myrtoan Seas, probably due to the production of saline waters also in the Cyclades Plateau and because the brackish waters formed in the north extremities of the Aegean Sea have to traverse its central part where high-salinity intermediate water masses are found. The formation of 1987 was strong enough to affect all the subbasins of the Aegean Sea, and reinforced the formations of the following years by bringing dense waters close to the surface as well as “pulling” highly saline Levantine water into the Aegean at the intermediate layers [14].

The formation of dense water continued after 1987 and peaked in 1992 and 1993. The densest waters, which were formed near the Lemnos Plateau and Lesvos islands [95], filled the Skyros and Lemnos Basins with waters of extremely high density ($>29.6 \mathrm{k}\mathrm{g} {\mathrm{m}}^{-3}$). As a result, density increased in all the subbasins of the Aegean Sea and the $29.15 \mathrm{k}\mathrm{g} {\mathrm{m}}^{-3}$ isopycnal was brought at depths shallower than 200 $\mathrm{m}$. The modelling study of Nittis et al. [96] suggests that the $29.1 \mathrm{k}\mathrm{g} {\mathrm{m}}^{-3}$ isopycnal outcropped over all of the Aegean Sea during 1993 and over a large part of the Aegean Sea from 1989 to 1994. During the peak of the EMT, as happened also in 1987, the salinity increased in the Cretan Sea ($S>39.05)$, but this time the temperature dropped. Consequently, the $>29.3 \mathrm{k}\mathrm{g} {\mathrm{m}}^{-3}$ isopycnal was uplifted at 800 $\mathrm{m}$ in the Cretan Sea. After the peak of the EMT, the density began to drop in all the subbasins of the Aegean Sea at different rates. The Chios and Ikaria Basins gained buoyancy faster than the other subbasins of the North–Central Aegean Sea. By 2002, the maximum observed density was $<29.35 \mathrm{k}\mathrm{g} {\mathrm{m}}^{-3}$ in the Chios Basin and $<29.4 \mathrm{k}\mathrm{g} {\mathrm{m}}^{-3}$ in the Ikaria Basin. By 2006, out of the three basins which have accommodated the densest waters (Skyros, Athos and Lemnos), density exceeded $29.5 \mathrm{k}\mathrm{g} {\mathrm{m}}^{-3}$ only in the deep layers of the Skyros Basin. Waters with ${\sigma }_{\theta }>29.4 \mathrm{k}\mathrm{g} {\mathrm{m}}^{-3}$ were diffused away by the mid-2000s in the Lemnos Basin and by the mid-2010s in the Skyros Basin, whilst in the Athos Basin they could be observed at least until 2020. In the Cretan Sea, the most remarkable thermohaline change just after the peak of the EMT was an intrusion of TMW with its core depth around 350 $\mathrm{m}$ and salinity <$38.95$, which appeared no later than 1994, and became more pronounced in 1995 [41]. In the following years, from 1997 to 1998, the TMW eroded, and the minimum salinity (<$38.95$) associated with it could be found only in the East Cretan Sea at $\sim 450 \mathrm{m}$. A new intrusion of TMW appeared in the Cretan Sea after 2001, indicative of the compensatory flow due to the outflow of dense waters from the Cretan Sea to the Eastern Mediterranean Sea [18].

The shoaling of the $29.15 \mathrm{k}\mathrm{g} {\mathrm{m}}^{-3}$ isopycnal in the Lemnos and Athos Basins marked the next period of increased DWF in the North–Central Aegean Sea, which started in 2002–2003 and intensified after 2006. Observations during that period are scarce in our dataset, in particular, in the South Aegean Sea. The more complete dataset used by Cardin et al. [3], however, confirms that dense waters produced in the Aegean Sea were accumulating in the Cretan Sea during 2002–2006. The DWF peaked in 2008–2009, bringing waters with ${\sigma }_{\theta }\sim 29.25 \mathrm{k}\mathrm{g} {\mathrm{m}}^{-3}$ shallower than the 400 $\mathrm{m}$ sill depth of the North–Central Aegean Sea. In 2008, the salinity of the intermediate layers increased abruptly by $\sim 0.1$ in the Chios and Ikaria Basins and by $\sim 0.15$ in the Skyros Basin. In the North–Central Aegean Sea, the $29.15 \mathrm{k}\mathrm{g} {\mathrm{m}}^{-3}$ isopycnal was found at depths shallower than 150 $\mathrm{m}$. Ventilation reached the deep layers (at least $800 \mathrm{m}$) of the Chios and Ikaria Basins, and the intermediate layers (at least $600 \mathrm{m}$) of the Skyros Basin. Also ventilation reached the bottom in the Lemnos Basin, as can be seen by the sharp temperature and salinity increase throughout the water column. In the Athos Basin, only the intermediate layers (down to $\sim 600 \mathrm{m}$) were ventilated. In the South Aegean Sea, the $29.15 \mathrm{k}\mathrm{g} {\mathrm{m}}^{-3}$ isopycnal rose by $\sim 200 \mathrm{m}$ during that period and the TMW in the Cretan Sea was eroded completely between 2006 and 2010.

According to Krokos et al. [19], warm and saline waters were being exported from the Aegean Sea to the Ionian Sea at least until 2011. According to our dataset, in which observations are available after 2011 for the Central Cretan Sea, the sharp shoaling of the $29.15 \mathrm{k}\mathrm{g} {\mathrm{m}}^{-3}$ isopycnal in the Central Cretan Sea in 2012 and its gradual deepening until 2016 suggests that the increased DWF which started in 2002 ceased after 2012. The strong DWF episode of 2012 is also supported by the appearance of a new intrusion of TMW in the South Aegean no later than 2012. The recurrent appearance of TMW in the South Aegean Sea has been explained previously as a compensatory inflow to the outflowing CDW in the Mediterranean Sea and consequently has been used as a tracer of DWF episodes in the Aegen Sea [5,15]. From 2012 to 2016, the density of the intermediate water masses in the North–Central Aegean Sea was relatively low and no signs of DWF could be discerned. However, injections of dense water were evident in the Central Cretan Sea, which was the best observed subbasin of the South Aegean Sea. Those intrusions, which affected only the deeper layer by uplifting mostly the $29.2 \mathrm{k}\mathrm{g} {\mathrm{m}}^{-3}$ isopycnal, but also the $29.3 \mathrm{k}\mathrm{g} {\mathrm{m}}^{-3}$ isopycnal in 2014, probably were the dense waters formed during the previous years in the North–Central Aegean Sea and which ended up in the Cretan Sea with a lag of one to two years through bottom currents.

As shown in Section 3.1, 2017 marked the beginning of a new period of increased DWF in the North–Central Aegean Sea. In the North–Central Aegean Sea, the density of the intermediate layers in the basins of Athos, Skyros and Chios increased due to the formation of 2017. The $29.35 \mathrm{k}\mathrm{g} {\mathrm{m}}^{-3}$ density horizons were uplifted in the Skyros Basin, while ventilation reached the $29.2 \mathrm{k}\mathrm{g} {\mathrm{m}}^{-3}$ density horizons in the Chios Basin. In the Myrtoan and Cetral Cretan Basins, the $29.1 \mathrm{k}\mathrm{g} {\mathrm{m}}^{-3}$ density horizons were uplifted. Velaoras et al. [36] have shown that the whole water column was ventilated in the Myrtoan Basin due to the formation of 2017 and that those waters were exported from the West Cretan Straits. It seems that a large bulk of the densest waters formed in 2017 exited the Aegean Sea through the East Cretan Straits, as the East Cretan Basin was the only basin of the South Aegean Sea in which the $29.2 \mathrm{k}\mathrm{g} {\mathrm{m}}^{-3}$ isopycnal rose by 200 $\mathrm{m}$. In the West and Central Cretan Sea, only the intermediate layers seem to have been affected by the formation of 2017, since the $29.15 \mathrm{k}\mathrm{g} {\mathrm{m}}^{-3}$ was the highest isopycnal, which was raised by $<100 \mathrm{m}$.

The formation of 2019 had the most prominent impact in the hydrography of the Aegean Sea, as an abrupt increase in the temperature and the salinity was recorded in all of the subbasins in which data were available. The salinity increased by >$0.1$ and the temperature increased by >$0.25 °\mathrm{C}$ in the Chios, Ikaria and Myrtoan Basins. The thermohaline changes were accompanied by a significant shoaling of the $29.1 \mathrm{k}\mathrm{g} {\mathrm{m}}^{-3}$ isopycnal in the Myrtoan Basin, of the $29.15 \mathrm{k}\mathrm{g} {\mathrm{m}}^{-3}$ isopycnal in the Chios Basin, and of the $29.2 \mathrm{k}\mathrm{g} {\mathrm{m}}^{-3}$ isopycnal in the Ikaria Basin. The salinity values of $\sim 39.1$ and the temperature values of $\sim 14.5 °\mathrm{C}$ were the highest ever recorded in the intermediate and deep layers of the Ikaria and Myrtoan Basins until 2019. The sharp increase in temperature and salinity in the North–Central Aegean Sea was also reflected in the South Aegean Sea by a sudden deepening of the $39.0$ isohaline and of the $14.5$ $°\mathrm{C}$ isothermal by $\sim 200\mathrm{m}$ in the West and Central Cretan Basins. The deepening of the isothermals and isohalines was smoother in the East Cretan Basin. The formation of 2019 also caused the $29.2 \mathrm{k}\mathrm{g} {\mathrm{m}}^{-3}$ isopycnal to rise by $>200 \mathrm{m}$ in the West Cretan Basin but had no such effect in the Central and East Cretan Basins. Thus, in contrast to the formation of 2017, the densest waters seemingly exited the Aegean Sea through the Western Cretan Straits.

In the years following 2019, data availability was reduced in the depressions of the North–Central Aegean Sea and the DWF events of 2020 and 2021, if any, could not be examined. However, the temperature and salinity kept increasing in the Cretan Sea, and by April 2021, the TMW had been eroded completely in the West Cretan Basin, while in the Central and East Basins, its signal had become much weaker and was found at greater depths. The continuation of DWF in 2020 and 2021, or at least of the continuous export of dense waters formed in the previous years in the North–Central Aegean Sea, could be identified by the remarkable shoaling of the $29.2 \mathrm{k}\mathrm{g} {\mathrm{m}}^{-3}$ isopycnal in the West Cretan Basin by $\sim 300 \mathrm{m}$. The formation of 2022, which can be seen in the Athos Basin in Figure 3, and the export of dense waters from the North–Central Aegean Sea towards the South Aegean Sea has been studied by Potiris et al. [39], and will not be discussed further here.

3.3. Erosion of the TMW in the Cretan Sea after 2017

The TMW appears recurrently in the Cretan Sea at least since 1960, when data availability allowed the variability of the basin to be studied. Whether the TMW erodes by vertical diffusion or by intrusions of dense waters from the North–Central Aegean Sea has been an open question since the early 2000s. Theocharis et al. [41] suggested that the intrusion of dense waters from the North–Central Aegean Sea caused the erosion of TMW around 1998–1999. Velaoras et al. [15] conducted a $\theta /S$ analysis and suggested that vertical diffusion was probably responsible for the TMW erosion from 2006 to 2014. Our combined observational dataset had sufficient temporal and vertical resolution from the surface down to 1000 $\mathrm{m}$ to investigate the erosion of TMW after 2016 in the Central Cretan Sea.

The thermohaline evolution of the Central Cretan Sea from 2010 to 2022, along with a $\theta /S$ diagram of selected profiles, is shown in Figure 4. From 2010 to 2012, the signal of the TMW was very weak and confined at depths greater than 800 $\mathrm{m}$, with its core at 1100 $\mathrm{m}$. In mid-2012, the lowest salinities which were observed in the water column were >$39$ and the TMW had eroded completely, at least in the first 1000 $\mathrm{m}$. The new intrusion of TMW appeared at 800 $\mathrm{m}$ in late 2012, and until 2016, it was progressively becoming stronger. In 2015, the TMW occupied roughly the 400–1200 $\mathrm{m}$ layer and its core salinity value at $\sim 700 \mathrm{m}$ was <$38.95$. The lowest salinity value of the core ($\sim 38.85$) was observed in late 2016. From late 2012 to late 2016, the $29.1 \mathrm{k}\mathrm{g} {\mathrm{m}}^{-3}$ isopycnal was gradually deepening, while in the deeper layers, intrusions of dense water with ${\sigma }_{\theta }>29.2 \mathrm{k}\mathrm{g} {\mathrm{m}}^{-3}$ at depths greater than 900 $\mathrm{m}$ could be observed in 2012, 2014 and 2015. The intrusion of 2012 was the largest of the three as it affected not only the deeper layers but also uplifted the $29.1 \mathrm{k}\mathrm{g} {\mathrm{m}}^{-3}$ isopycnal by more than 150 $\mathrm{m}$. The doming of the isopycnals due to the intrusions of 2014 and 2015, on the other hand, was confined at depths greater than 500 $\mathrm{m}$.

The progressive deepening of the TMW core, which started in 2017 and continued at least until mid-2021, was accompanied by the increase in its salinity from less than $38.85$ to more than $38.95$, and the erosion of the ceiling of the TMW could be traced by the deepening of the 39 isohaline from ∼$400 \mathrm{m}$ to ∼$900 \mathrm{m}$. The salinity increase in the TMW during that period was characterised by smaller or larger intrusions of warm and salty waters below 400 $\mathrm{m}$ and down to the maximum observed depths (around 1000 $\mathrm{m}$), as can be seen by the protrusions of the thermohaline traces of the selected casts in the $\theta /S$ diagram and in the profile diagrams. Most of the intrusions were observed close to the depth of the maximum salinity gradient, while the concurrency between the intrusions and the doming of the isopycnals suggests that the part of the TMW which was lighter than the intrusions was uplifted at shallower depths. The largest intrusion, which was recorded in 2017 between 430 $\mathrm{m}$ and 560 $\mathrm{m}$, increased salinity by 0.05, while the second largest intrusion was recorded in 2020 between 550 $\mathrm{m}$ and 650 $\mathrm{m}$. In general, the intrusions were recorded below 400 $\mathrm{m}$, i.e., below the depth of the CIW, and apparently their net effect was the increase in temperature and salinity in the 400– 1000 $\mathrm{m}$ layer. A similar but more pronounced erosion was observed in the shallower West Cretan Basin, as shown in Figure S2.

3.4. Variability of the CIW

The lack of a clear relation between the increase in surface salinity and DWF in the Cretan Sea has been noted already in 2019 by Velaoras et al. [5], who raised the question of whether the density variability of the CIW, i.e., the water which is produced in the Cretan Sea, depends more on temperature or salinity. To investigate the density variability of the CIW, we computed the thermohaline properties of the Cretan Sea in the first 250 $\mathrm{m}$ of the water column from profiles and fixed point observations from January to April (Figure 5). Although salinity measurements that meet the requirements of the analysis (see Section 2.1) are scarce before 1985, four periods of increased CIW density could be identified: from the early to late 1970s, from 1987 to 1998, from 2005 to 2009 and from 2017 to 2023. High densities were also recorded in 2012. The density in the 20–250 $\mathrm{m}$ layer of the Cretan Sea was higher than $29.1 \mathrm{k}\mathrm{g} {\mathrm{m}}^{-3}$ only during the EMT, and the highest density values (>$29.2 \mathrm{k}\mathrm{g} {\mathrm{m}}^{-3}$) were recorded during the peak of the EMT. The density of the other three periods outside of the EMT was around $29.05 \mathrm{k}\mathrm{g} {\mathrm{m}}^{-3}$, while the maximum density threshold of individual measurements was close to $29.1 \mathrm{k}\mathrm{g} {\mathrm{m}}^{-3}$.

In order to quantify the degree to which the temperature or salinity has affected the CIW density, the latter was estimated based only on the variability of either the temperature or salinity, keeping the other parameter equal to its temporal average within that layer. Thus, we estimated the potential density variability based only on the temperature variability as ${\sigma }_{\theta }^T={\sigma }_{\theta }(\overline{S},T)$, where overbar denotes temporal and vertical averaging over the selected depth range, and accordingly, the potential density range based only on salinity variability, as ${\sigma }_{\theta }^S={\sigma }_{\theta }(S,\overline{T})$. It is evident from Figure 5f that the density variability of the CIW depends mostly on temperature, since temperature alone can induce a density variability of $0.7 \mathrm{k}\mathrm{g} {\mathrm{m}}^{-3}$, whilst salinity only a density variability of $0.4 \mathrm{k}\mathrm{g} {\mathrm{m}}^{-3}$. Also, the dependence of density on temperature is so dominant that the correlation coefficient between salinity and density is negative. Salinity seems to play a significant role only at inter-decadal time scales. The almost continuous increase in salinity since the mid-2000s resulted in comparable CIW densities between 2005–2010 and 2019–2023, despite the 2019–2023 temperature being higher by $0.8$ $°\mathrm{C}$ compared to the 2005–2010 period.

3.5. Variability in the Aegean Sea before 1985

The literature regarding the hydrographic variability of the Aegean Sea before 1985 is significantly smaller compared to that after 1985. The main reasons are the scarcity and the imprecision of salinity measurements, which make the drawing of safe conclusions much more difficult. In this section, we examine concurrently the hydrographic variability of the nine largest depressions of the Aegean Sea before 1985 (Figure 6). Prior to 1985 salinity measurements are less precise and only major variations can be identified with confidence. However, major salinity changes in the subbasins of the Aegean Sea usually come in tandem with changes in temperature, the measurements of which are accurate enough to study the variability of the Aegean Sea.

A large increase in density in the basins of Lemnos and Athos suggests that at least one strong DWF event took place between 1947 and 1955 in the North–Central Aegean Sea. The basin of Athos, where a few more observations are available compared to the other subbasins of the North–Central Aegean Sea, enabled us to determine the timing of the peak formation to be somewhere between 1948 and 1952. The formation(s) resulted in the decrease in temperature and increase in salinity in the Athos Basin. The few observations until late 1950s in the Ikaria, Skyros, Athos and Lemnos Basins suggest that the $29.25 \mathrm{k}\mathrm{g} {\mathrm{m}}^{-3}$ isopycnal was uplifted above the ∼$400 \mathrm{m}$ sill depth, and that the subbasins of the North–Central Aegean Sea were relaxing from the DWF episode(s), which took place during the late 1940s to early 1950s. Also, it is evident that the TMW layer in the Cretan Sea was eroded between 1947 and 1955, as shown by measurements of both temperature and salinity.

A new intrusion of TMW with core salinity lower than 38.85 appeared no later than 1959 in the East Cretan Sea. The TMW signal gradually became stronger after 1963, and by 1965 it occupied the 400–1200 $\mathrm{m}$ layer in the East and Central Cretan Basins. The low salinity signal of the TMW was more prominent in the Eastern Cretan Basin and less so in the West Cretan Basin. The first signs of a new DWF in the North–Central Aegean Sea appeared in 1972–1973 as density increased, while at the same time, the TMW in the East Cretan Sea started to erode. The DWF during that period did not have an evident effect in all of the subbasins, and combined with the relatively scarce observations, safe conclusions cannot be drawn. The formation period after 1975, however, had an easily visible impact in all the subbasins of the Aegean Sea, as the density of the intermediate layers in the North–Central Aegean increased and the TMW had eroded completely by 1977, at least in the West and Central Basins of the Cretan Sea. After 1980, the TMW appeared again in the Cretan Sea. The post-1980 TMW intrusion was markedly different from the other intrusions, as it did not take place gradually, but instead appeared almost concurrently in all of the subbasins of the Cretan Sea. The TMW layer did not become thicker progressively, but instead occupied a large part of the water column in less than two years. Also, the very low salinities in the near-surface and sub-surface layers of the North–Central Aegean Sea indicate that the BSW inflow increased significantly after 1980.

The erosion of the TMW layer in the 1970s shows a similarity with the erosion during the early stages of the EMT in the sense that the TMW core was not deepening and eroding gradually. In the post-EMT period, the low-salinity core of the TMW gradually deepened and became saltier. In the pre-EMT periods, the erosion of the TMW appears more abrupt and also it seems that it was extended throughout the water column, increasing salinity at all observed depths. Of course, the different “behaviour” could be attributed to the imprecision and the scarcity of the salinity measurements; however, the concurrent increase in temperature at all depths at which measurements are available gives us some confidence in our conclusions.

3.6. Variability of Forcing Mechanisms

The circulation of the Ionian Sea from 1940 to 2022 was predominantly cyclonic with shorter or longer periods of anticyclonic circulation (Figure 7a). Four extended periods of strong and three shorter periods of weak anticyclonic circulation could be identified. The circulation was anticyclonic from 1940 to 1946, from 1953 to 1958, from 1987 to 1998, and from 2005 to 2009. The 1949–1950, 1982–1983 and 2017–2018 periods were years of weak anticyclonic circulation. The strength of the anticyclonic circulation was comparable in strength from 1940 to 1958 and from 2005 to 2009. The strongest and longest anticyclonic circulation from 1987 to 1998 almost coincided with the EMT.

The water exchange at the TSS from 1970 to 2022 is shown in Figure 7b. In the pre-satellite SSH era, a strong interannual variability of BSW inflow is visible. BSW inflow was low from 1972 to 1977, from 1983 to 1987, and from 1991 to 1993. Significantly increased BSW inflow was observed from 1994 to 2000 and from 2004 to 2006, while 2001–2003 was a period of average BSW inflow. A sudden reduction in BSW inflow was recorded from 2007 to 2009, and was followed by an increase which lasted for two years. The BSW inflow shows a clear negative trend from the early 2010s until 2022. The present-day values are close to those of 1973–1976, which have been the years of minimum BSW inflow since 1970.

The AMO, which correlates with SST in the Mediterranean and serves as an index of SST variability at sub-decadal time scales, shows much lower frequency variability (around seventy years) compared to the other forcing mechanisms (Figure 7c). It was positive from 1940 to 1962 and from 1995 to 2022, and negative from 1963 to 1994. The lowest AMO values were recorded around 1976, 1986 and 1992.

The surface winter buoyancy fluxes are the forcing mechanism with the highest frequency variability (Figure 7d). The years 1949, 2002–2003, and 2022 were those with the highest buoyancy loss. From 1940 to 1955, the buoyancy loss was mild to moderate, with the exception of 1949, and was followed by a period of above-average buoyancy loss which lasted until 1969. From 1970 to 1987, the buoyancy loss was low, with one exception in 1976. As has been discussed extensively in the literature, the peak of the EMT in 1992 and 1993 was two consecutive years of strong buoyancy loss. A remarkable decadal variability emerged after 1992–1993, with peaks in 2002–2003, 2012 and 2021–2022. Since 2015, the buoyancy loss was higher every second year, and resulted in the seesaw variability of buoyancy fluxes which is clearly visible during that period.

4. Discussion

The intermediate water which is produced in the Cretan Sea and settles at $\sim 300\mathrm{m}$ in the Aegean Sea was named CIW by Schlitzer et al. based on observations from the F.S. Meteor during August–September 1987, just after the onset of the EMT [17]. At that time, the CIW was found outside of the Cretan Straits at the depth range between 500 $\mathrm{m}$ and 1200 $\mathrm{m}$, centered at 700 $\mathrm{m}$, while in 1991, in the same region, the CIW occupied the 150–700 $\mathrm{m}$ layer and the transitional layer occupied the 700–1200 $\mathrm{m}$ layer, as the CDW filled the horizons below 1200 $\mathrm{m}$ and uplifted the lighter water masses [6]. The CDW was produced in the North–Central Aegean Sea, filled the subbasins of the Cretan Sea and then the deep/bottom layers of the Eastern Mediterranean Sea. During the EMT, the CIW could be considered an intermediate layer in terms of settling depth both for the Ionian Sea and for the Aegean Sea [6]. In 1998, the CIW in the Ionian Sea became warmer, saltier and lighter compared to the early nineties [41]. CIW densities of $29.15 \mathrm{k}\mathrm{g} {\mathrm{m}}^{-3}<{\sigma }_{\theta }<29.25 \mathrm{k}\mathrm{g} {\mathrm{m}}^{-3}$ in the Aegean Sea were observed only during the EMT. From 2000 onwards, typical values are $\sim 29.05 \mathrm{k}\mathrm{g} {\mathrm{m}}^{-3}$ and the maximum near-surface densities rarely and marginally exceed the $29.1 \mathrm{k}\mathrm{g} {\mathrm{m}}^{-3}$ threshold in the Cretan Sea, at least in the in situ dataset we analysed. The present-day density in the deep Cretan Sea close to the straits’ sill depths (600–1000 $\mathrm{m}$) is a little higher ($\sim 0.05 \mathrm{k}\mathrm{g} {\mathrm{m}}^{-3}$) compared to the density of the CIW during the peak of the EMT.

Velaoras et al. has proposed to name the waters which form in the North–Central Aegean Sea during the post-EMT period, and which end up in the Cretan Sea with densities $29.1 \mathrm{k}\mathrm{g} {\mathrm{m}}^{-3}<{\sigma }_{\theta }<29.2 \mathrm{k}\mathrm{g} {\mathrm{m}}^{-3}$, deep Cretan Intermediate Water (dCIW), since their density is not high enough to reach the bottom of the Eastern Mediterranean Sea but settle at depths $<2000 \mathrm{m}$ [18]. This approach is convenient for studies not restricted inside the Aegean Sea, while the naming of the Aegean Sea water masses remains consistent with the EMT literature. Following these Eastern Mediterranean-centric conventions for the naming of water masses produced in the Aegean Sea, the CIW water is produced by open-sea convection in the Cretan Sea and settles around 300 $\mathrm{m}$, while the dCIW is produced in the North–Central Aegean Sea and ends up in the subbasins of the Cretan Sea. The CDW consists of waters produced also in the North–Central Aegean Sea but which end up in the Cretan Sea with densities higher than $29.2 \mathrm{k}\mathrm{g} {\mathrm{m}}^{-3}$. When waters with such densities are found at depths shallower than 1000 $\mathrm{m}$, or perhaps 800 $\mathrm{m}$, and in sufficient quantities, they can reach the abyssal layers of the Eastern Mediterranean Sea [3]. From the doming of the isopycnals and the deepest part of the TMW intrusions in the Cretan Sea, the upper limit of the CDW produced after the EMT seems to be around $29.25 \mathrm{k}\mathrm{g} {\mathrm{m}}^{-3}$. Intrusions of denser water that end up in the Cretan Sea could be discerned in our analysis but their volume was insufficient to uplift the $29.25 \mathrm{k}\mathrm{g} {\mathrm{m}}^{-3}$ isopycnal above the 1000 $\mathrm{m}$ sill depth of the East Cretan Straits, and affected only slightly—by decelerating the rate at which their density drops—the otherwise stagnating waters in the deepest parts of the Cretan Sea.

The near-surface salinity variability is considered important for DWF in the Aegean Sea, and some studies suggest that the decadal DWF variability of the Eastern Mediterranean Sea is primarily salinity-driven, with the internal oceanic processes dominating over the atmospheric forcing [18,19,20,24]. Those ideas were put forward in the 2010s, shortly after Gacic et al. [24] observed a relation between the NIG reversals with the salinity of the upper-layers in the Ionian and Levantine Seas and the salinity of the deep/intermediate layers in the Cretan Sea. In the Aegean Sea, DWF strong enough to uplift the $29.15 \mathrm{k}\mathrm{g} {\mathrm{m}}^{-3}$ isopycnal in the Cretan Sea started after 2002 due to the increased surface buoyancy loss, and continued at least until 2009 despite the mild winters since 2006 [3]. After the peak of the DWF, which was observed around 2008, salinity increased substantially in the South Aegean Sea and the TMW was eroded almost completely by 2012 [3,15,19]. According to our results, a similar increase in salinity in the intermediate layers of the North–Central Aegean Sea occurred in 2008, in accordance with the sudden reduction in BSW inflow from 2007 to 2009. Although the NIG reversals are related to the variability of DWF in the Aegean Sea, their progressive weakening after the EMT cannot explain the increase in salinity and DWF in the Aegean Sea, in particular, after 2019. Similarly, the decadal variability of winter atmospheric forcing alone is not sufficient to increase substantially the formation of dense water in the subbasins of the Aegean Sea. A combination of forcing mechanisms in the Aegean Sea seems necessary to trigger DWF strong enough to affect unambiguously most of its parts, even though the years of peak surface buoyancy loss due to the decadal variability from 1992–1993 onwards were years of DWF. Moreover, identifying (and, especially, quantifying) the potential effects of climate change on the frequency and intensity of dense water formation episodes requires analysis extending well beyond the context and purpose of this paper, incorporating both comparison with results of long hindcast studies of the region, enabling buoyancy budget calculations and the determination of long-term trends of the particular buoyancy contributors and the application of recent potential data bias-removing algorithms [98].

The monitoring of the hydrography of the Aegean Sea can be divided into two eras and four periods according to the accuracy and the availability of observations. The first two periods, from 1945 to 1985, constitute the low-accuracy salinity era, and the last two periods, from 1985 until the present day, constitute the high-accuracy salinity era. From 1945 to 1960, only a handful of observations are available in the Aegean Sea. Still, increased DWF between the late 1940s and mid 1950s could be identified in our analysis. In principle, the observations from 1960 to 1985, although still scarce, should be sufficient to infer at least the quinquennial variability of DWF; however, salinity imprecision permits only major formation events to be identified with some certainty. Formation in the North–Central Aegean Sea appears much weaker during the 1960s and only the formation which started probably during the early 1970s and culminated after the mid-1970s was strong and widespread enough to be identified with certainty in most subbasins of the Aegean Sea. From 1985 to 2015, the data availability increased sharply, both in time and in the vertical direction, and salinity accuracy improved to present day standards. During the 1985–2015 period, observations were taken mostly from ship cruises, and in the 2000s, a network of fixed point observatories initiated the long-term and high-frequency monitoring of the Aegean Sea. This period, which has been studied in depth in the literature, revealed the onset and cessation of the EMT from 1987 to 1997, as well as the prolonged but weak DWF period from 2002 to 2009. Strong buoyancy loss in 2012 resulted in a strong DWF episode and intrusions of dense water below the TMW core were observed until 2014 in the Cretan Sea. The last period began in 2015, when the number of Argo float profiles increased significantly and became the main source of hydrographic information for the Aegean Sea [38]. More recently, gliders have also started operating routinely in the South and North Aegean Seas [79,99]. Since 2017, the Aegean Sea has entered again a state of increased DWF, with relatively strong formations taking place almost biannually (Figure 2) due to the variability in the surface buoyancy fluxes (Figure 7d) and the decreased BSW inflow (Figure 7b).

This work has merged data from a range of original sources (Nansen/Niskin bottle profiles, MBT/XBTs, CTDs, etc) to identify major DWF formation episodes in the instrumental period prior to 1985. Recently, Zhang et al. [98] introduced a method to identify and correct potential biases in the estimation of historical temperature profiles and the related trends, arising from the use of a wide range of different instruments, and produced a global database of bias-corrected temperature profiles. While this product is extremely useful in identifying climate trends, we believe that its use in the present work would not significantly alter our results, regarding the two major Aegean DWF formation episodes prior to 1985, for the following reasons: (a) the formation of the late 1940s–early 1950s and the signals of the dense-water propagation to the South Aegean was recorded by two different cruises in the region, while the corresponding process of the 1970s was recorded by four different cruises, and (b) the signals recorded vary with depth and are very strong in the North Aegean, and gradually weaken as they approach the basins to the south, a behaviour characterising the natural process of water-mass mixing and transformation and not a constant bias of the measurements. Thus, we believe that any potential bias due to the use of different instruments is minimal compared to the strength of the signals.

Among the findings of the present work is the eroding role of intrusions of dense water exported from the North–Central Aegean Sea on the TMW layer in the South Aegean Sea. This may appear somewhat contradictory with the compensatory nature of TMW inflow into the South Aegean during the export of CDW to the Levantine and Ionian Seas [15,41,78,100,101]. Indeed, the signature of the TMW in the South Aegean is concurrent and proportional to the intensity of events of CDW export from the basin, thus following very intense DWF episodes in the Central and North Aegean Sea, like during the EMT period or the post-2012 period of the re-appearance of the TMW layer in the South Aegean Sea [5,15]. However, the present work showed that weaker annual formation and export of dCIW to the South Aegean in the form of intrusions, contributes to the gradual erosion of the TMW signal in the former basin.

The BSW inflow is an important forcing mechanism for DWF in the Aegean Sea; however, it is the least known among the four mechanisms we mentioned. Consider the upper layer entering the Aegean Sea from the Dardanelles Strait, with T, S and Q its daily temperature, salinity and volume flow, respectively. Using the daily data from Maderich et al. [89], and denoting the time-mean with a bar, the anomaly from the monthly mean climatology with a prime, and the variance with Var, then $\mathrm{Var}\left(\overline{\mathrm{Q}}{\mathrm{T}}^{\prime }\right)/\mathrm{Var}\left({\mathrm{Q}}^{\prime }\overline{\mathrm{T}}\right)$∼$0.09$ and $\mathrm{Var}\left(\overline{\mathrm{Q}}{\mathrm{S}}^{\prime }\right)/\mathrm{Var}\left({\mathrm{Q}}^{\prime }\overline{\mathrm{S}}\right)$∼$0.11$. Thus, combining the existing temperature and salinity climatologies with the observed inflow would dramatically improve our knowledge of the lateral heat and salt fluxes, leaving unknown only ∼$10\%$ of their variability. Although the volume flow estimates of the Dardanelles Strait using the Dardanos HF radar system were low compared to that found in the literature [102,103,104], we consider its operation to be among the Aegean Sea’s oceanographic research priorities, as the long-term monitoring of surface currents to the east of Lemnos Island and their successful assimilation in hydrodynamic models could have far-reaching implications, not only for the improvement of the short-term circulation predictions but also for the buoyancy budget of long-term simulations of the Aegean Sea.

5. Conclusions

The main results of this study can be summarised as follows:

• The Aegean Sea has been in a state of increased dense water formation from 2017 to 2022. Deep Cretan Intermediate water was formed almost biannually due to seesaw atmospheric variability.
• Record-high salinity has been observed in the upper and intermediate layers of the Aegean Sea from 2019 to 2023. The BSW inflow appears to be very low during that period, suggesting that the dilution of the Aegean Sea by BSW plays a significant role for the near-surface salt budget of the basin, but also for the salt budget of the intermediate layers through enhanced DWF in the North–Central Aegean.
• Intrusions of dense water from the North–Central Aegean Sea contribute to the erosion of transitional Mediterranean waters in the South Aegean Sea. The increased DWF around the early 1950s, mid-1970s, and after 2017 in the North–Central Aegean Sea coincided with the erosion of the TMW in the South Aegean Sea.
• The density variability of the CIW depends mostly on temperature; thus, a salinity-driven DWF variability should be relevant mostly in the North–Central Aegean Sea. The intrusions of dense waters from the North–Central Aegean Sea probably play a role by uplifting the intermediate waters in the Cretan Sea.