Levantine Intermediate and Levantine Deep Water Formation: An Argo Float Study from 2001 to 2017

: Levantine intermediate water (LIW) is formed in the Levantine Sea (Eastern Mediterranean) and spreads throughout the Mediterranean at intermediate depths, following the general circulation. The LIW, characterized by high salinity and relatively high temperatures, is one of the main contributors of the Mediterranean Overturning Circulation and inﬂuences the mechanisms of deep water formation in the Western and Eastern Mediterranean sub-basins. In this study, the LIW and Levantine deep water (LDW) formation processes are investigated using Argo ﬂoat data from 2001 to 2017 in the Northwestern Levantine Sea (NWLS), the larger area around Rhodes Gyre (RG). To ﬁnd pronounced events of LIW and LDW formation, more than 800 Argo proﬁles were analyzed visually. Events of LIW and LDW formation captured by the Argo ﬂoat data are compared to buoyancy, heat and freshwater ﬂuxes, sea surface height (SSH), and sea surface temperature (SST). All pronounced events (with a mixed layer depth (MLD) deeper than 250 m) of dense water formation were characterized by low surface temperatures and strongly negative SSH. The formation of intermediate water with typical LIW characteristics (potential temperature > 15 ◦ C, salinity > 39 psu) occurred mainly along the Northern coastline, while LDW formation (13.7 ◦ C < potential temperature < 14.5 ◦ C, 38.8 psu < salinity < 38.9 psu) occurred during strong convection events within temporary and strongly depressed mesoscale eddies in the center of RG. This study reveals and conﬁrms the important contribution of boundary currents in ventilating the interior ocean and therefore underlines the need to rethink the drivers and contributors of the thermohaline circulation of the Mediterranean Sea.


Introduction
The Mediterranean Sea (Figure 1) is composed of two basins of nearly equal size, the Western and the Eastern Mediterranean Sea, connected by the Sicily Channel. The general circulation of the Mediterranean Sea can be divided into three dominant scales of motion: the basin scale including the thermohaline circulation, the sub-basin scale including permanent and quasipermanent cyclonic and anticyclonic gyres, and the mesoscale with small but energetic temporary eddies [1,2]. All these scales are interacting.
Through the Strait of Gibraltar, the relatively fresh Atlantic water (AW) enters the Western Mediterranean Sea within the upper 100 to 200 m. It is modified flowing eastward, passes the Sicily Channel and the Ionian Sea and enters the easternmost part of the Mediterranean, the Levantine Sea. The salinity of the AW in the Levantine Sea depends on the circulation patterns during its path, mainly influenced by the variability of the circulation of the North Ionian Gyre (NIG) which varies significantly at seasonal and decadal scales ( [3][4][5]; Figure 1a).  [15] showed that little to no net sinking takes place at convection sites (blue arrows; from left to right: Gulf of Lion, South Adriatic, Aegean Sea, Rhodes Gyre (RG)) while boundary layer currents undergo net intense sinking (brown arrows). The yellow rectangle indicates the area of study. Adapted from [15].

Datasets and Methods
The datasets used for this study are Argo floats vertical profiles of temperature and salinity (T/S) collected in the NWLS during winter months (January, February, March-JFM) between 2001 and  [15] showed that little to no net sinking takes place at convection sites (blue arrows; from left to right: Gulf of Lion, South Adriatic, Aegean Sea, Rhodes Gyre (RG)) while boundary layer currents undergo net intense sinking (brown arrows). The yellow rectangle indicates the area of study. Adapted from [15].

Datasets and Methods
The datasets used for this study are Argo floats vertical profiles of temperature and salinity (T/S) collected in the NWLS during winter months (January, February, March-JFM) between 2001 and 2017. In total 879 T/S profiles from 20 floats were analyzed visually. Figure 2 shows the position of the Argo float profiles and the annual distribution of profiles for JFM between 2001 and 2017 for the area of study. The datasets used for this study are Argo floats vertical profiles of temperature and salinity (T/S) collected in the NWLS during winter months (January, February, March-JFM) between 2001 and 2017. In total 879 T/S profiles from 20 floats were analyzed visually. Figure 2 shows the position of the Argo float profiles and the annual distribution of profiles for JFM between 2001 and 2017 for the area of study. With the help of an external bladder, the Lagrangian floats descend to a programmed parking depth (350 or 1000 m) where they stay for a specified period (1-10 days, mainly 5 days [18]). Then, they descend to greater depths (up to 2000 m). During their ascent back to the surface, they measure temperature and salinity throughout the water column. At the surface they transmit the data to satellites and descend again. The transmitted data are stored at data assembly centers (DAC) which apply a quality control and provide open access to real time and delayed mode quality controlled data.
Quality controlled Argo float data were downloaded from the Ifremer Data Assembly Center (DAC; ftp://ftp.ifremer.fr/ifremer/argo/dac/coriolis/). For this study, only data with the best quality control (qc=1) were taken into account. Downloaded parameters included float number, position, time, pressure, temperature, and salinity.
Hydrographic properties were expressed as potential temperature, potential density, and salinity according to the practical salinity scale (PSU).
The visual inspection of the Argo float profiles is important due to the fact that the Argo floats may pass an area not exactly during the event of mixing or convection. They can instead sample days or weeks later when the recapping (i.e., a newly formed shallow MLD) already occurred. In such a case, MLD detection algorithms indicate a shallow MLD, but do not give any information about mixing or convection events before the recapping. While at the top, there can be already a newly formed MLD and the convection event can still be visible deeper in the water column. MLD detection statistics rarely give information about deep mixing events while the visual inspection of the form of the profile (potential temperature, salinity, and potential density) reveals clearly such events. Figure  3a shows the climatology of the winter maximum MLD derived from Argo float data and downloaded from [19]   With the help of an external bladder, the Lagrangian floats descend to a programmed parking depth (350 or 1000 m) where they stay for a specified period (1-10 days, mainly 5 days [18]). Then, they descend to greater depths (up to 2000 m). During their ascent back to the surface, they measure temperature and salinity throughout the water column. At the surface they transmit the data to satellites and descend again. The transmitted data are stored at data assembly centers (DAC) which apply a quality control and provide open access to real time and delayed mode quality controlled data.
Quality controlled Argo float data were downloaded from the Ifremer Data Assembly Center (DAC; ftp://ftp.ifremer.fr/ifremer/argo/dac/coriolis/). For this study, only data with the best quality control (qc=1) were taken into account. Downloaded parameters included float number, position, time, pressure, temperature, and salinity.
Hydrographic properties were expressed as potential temperature, potential density, and salinity according to the practical salinity scale (PSU).
The visual inspection of the Argo float profiles is important due to the fact that the Argo floats may pass an area not exactly during the event of mixing or convection. They can instead sample days or weeks later when the recapping (i.e., a newly formed shallow MLD) already occurred. In such a case, MLD detection algorithms indicate a shallow MLD, but do not give any information about mixing or convection events before the recapping. While at the top, there can be already a newly formed MLD and the convection event can still be visible deeper in the water column. MLD detection statistics rarely give information about deep mixing events while the visual inspection of the form of the profile (potential temperature, salinity, and potential density) reveals clearly such events. Figure 3a shows the climatology of the winter maximum MLD derived from Argo float data and downloaded from [19]   The SST and sea surface height (SSH) data were downloaded from Copernicus (marine.copernicus.eu).
The interpolated SST product (SST_MED_SST_L4_NRT_OBSERVATIONS_010_004_c_V2) has a daily temporal resolution and a spatial resolution of 0.04° × 0.04°. The interpolated SSH product (SST_MED_SST_L4_REP_OBSERVATIONS_010_021) has a daily temporal resolution and a spatial The SST and sea surface height (SSH) data were downloaded from Copernicus (marine.copernicus.eu). The interpolated SST product (SST_MED_SST_L4_NRT_OBSERVATIONS_010_004_c_V2) has a daily temporal resolution and a spatial resolution of 0.04 • × 0.04 • . The interpolated SSH product (SST_MED_SST_L4_REP_OBSERVATIONS_010_021) has a daily temporal resolution and a spatial resolution of 0.125 • × 0.125 • . Monthly means of SST superimposed with the geostrophic currents from SSH were used to describe the negative slope of eddies within the RG during intense convection events.
The freshwater fluxes were derived from ERA-INTERIM (daily) data. The downloaded parameters are evaporation (E) and total precipitation (P). The downloaded data have a time step of 12 hours, i.e., daily data at 00:00:00 and at 12:00:00 and a spatial resolution of 0.25 • × 0.25 • . The daily freshwater fluxes (FWF) were calculated as the subtraction of the daily means of E and P: FWF=E−P.
The air-sea heat fluxes were derived from ERA-INTERIM (daily) data. The downloaded parameters are: Surface net solar radiation (Q sw ), surface net thermal radiation (Q lw ), surface sensible heat flux (Q s ), and surface latent heat flux (Q l ). The heat budget can be expressed as the difference between the net shortwave solar radiation (incoming minus reflected) absorbed by the sea surface, the sum of the longwave back radiation, the sensible, the latent, and the advective heat flux. The advective heat flux (Q adv ) was not available at ERA-INTERIM and therefore not considered. The downloaded data have a time step of 12 hours, i.e., daily data at 00:00:00 and at 12:00:00 and a spatial resolution of 0.25 • × 0.25 • . The daily mean of each parameter as well as the daily net heat fluxes were calculated as the sum of the daily means of each parameter: Q net = Q sw + Q lw + Q l + Q s . The surface buoyancy flux B, composed by thermal (B T ) and haline (B S ) components, was calculated according to [20]: where α is the thermal expansion coefficient, g = 9.8 ms −2 is the gravity acceleration, Cp = 3.9715 × 10 −3 Jkg −1 K −1 is the specific heat capacity of sea water, ρ 0 = 1029 kgm −3 is a reference sea water density, β is the haline contraction coefficient and S 0 = 38.9 is a reference salinity. α and β were calculated at surface pressure, using monthly mean surface salinity and monthly mean surface temperature, downloaded from Copernicus (MEDSEA_REANALYSIS_PHYS_006_004). B is positive when surface water gets lighter and negative when surface water becomes denser (river inputs as well as horizontal and vertical advection also contribute to density changes, but were not considered due to lack of data).
The Turner angle (Tu) was computed to evaluate the relative roles of temperature and salinity gradients on the density gradients. Tu is defined as the four-quadrant arctangent [21], which units are degrees of rotation and was calculated with the Gibbs-SeaWater (GSW) Oceanographic Toolbox [22]. Argo float salinity and temperature were converted to absolute salinity and to conservative temperature, respectively. The conservative temperature represents more accurately the heat content [22]. Tu = 45 • indicates that temperature is the only contributor, while Tu = −45 • indicates that salinity is the only contributor to density changes; |Tu|< 45 • indicates stable stratification and in this condition both temperature and salinity contribute to the density change ; 45 • < Tu < 90 • shows that salinity is working against temperature and is also called the 'salt finger regime' with the strongest activity near 90 • ; −90 • < Tu < −45 • is called the 'diffusive regime' and shows that temperature is working against salinity, reaching the strongest activity near −90 • ; |Tu|> 90 • characterizes a statically unstable water column (where the Brunt-Vaisalaa frequency N 2 < 0).

Heat and Freshwater Fluxes within the Northwestern Levantine Sea
The intensity of the mixing and convection events depends mainly on the surface buoyancy fluxes B, which in turn depend on the heat fluxes through the air-sea interface, scaled by the thermal expansion coefficient α, and the freshwater fluxes, scaled by the haline contraction coefficient β. Monthly surface buoyancy fluxes and their thermal and haline (freshwater) components, integrated over the center of RG (longitude: 28-31 • E, latitude: 34-36 • N), are shown in Figure 4.  The climatology of monthly heat fluxes Qnet for the center of RG from 2001 to 2017 shows that the largest heat losses which induce the preconditioning phase occurred mainly in November and in December ( Figure 5). The subsequent heat losses in JFM induce the formation of dense water and therefore lead to convection and mixing. The haline components (B S ) dominate the surface buoyancy fluxes (Figure4, Lower Panel), i.e., that intense evaporation, especially during the preconditioning phase (e.g., Figure 8 for winter 2006), controls the surface buoyancy loss. Detected events of pronounced (i.e., with a MLD deeper than 250 m) dense water formation by the Argo float data, are indicated with blue (RG) and yellow (coastline) circles ( Table 1).
The climatology of monthly heat fluxes Qnet for the center of RG from 2001 to 2017 shows that the largest heat losses which induce the preconditioning phase occurred mainly in November and  The climatology of monthly heat fluxes Qnet for the center of RG from 2001 to 2017 shows that the largest heat losses which induce the preconditioning phase occurred mainly in November and in December ( Figure 5). The subsequent heat losses in JFM induce the formation of dense water and therefore lead to convection and mixing.    (Table 1). More than 800 profiles of 20 floats were analyzed, but only four floats (Tables 1 and 2) captured pronounced dense water formation, being at the right place at the right time. To document the dense water formation events the float had to be either inside or pass later through the area of dense water formation. Float WMO 6900098 had an exceptionally long lifetime of nearly 6 years and therefore it was able to capture one event of pronounced LIW formation along the Northern coastline and four events of pronounced DWF. Unfortunately, it stopped measuring at 600 m depth. The WMO numbers of the Argo floats that found pronounced dense water formation events within the center of RG and along the coastline are listed in Table 2.  (Figure 6a). Hoevmueller plots of salinity, potential temperature, and potential density describe two pronounced events of mixing and convection during this winter (Figure 6b-d). The first event occurred by the end of January until mid-February and led to LDW formation (13.7 • C < potential temperature < 14.5 • C, 38.8 psu < salinity < 38.9 psu) while the second event around mid-March led to LDW and 'lower range' LIW (temperature about 15 • C and salinity about 39 psu) formation.  Figure 6a). Hoevmueller plots of salinity, potential temperature, and potential density describe two pronounced events of mixing and convection during this winter (Figure 6b,c,d). The first event occurred by the end of January until mid-February and led to LDW formation (13.7°C < potential temperature < 14.5°C, 38.8 psu < salinity < 38.9 psu) while the second event around mid-March led to LDW and 'lower range' LIW (temperature about 15°C and salinity about 39 psu) formation. The MLD deepens from 50 m within December to about 100 m in the beginning of January and the high surface salinity is mixed to intermediate layers. The heat and freshwater fluxes integrated over the center of RG show an intense preconditioning phase during December 2005, due to strong dry and cold winter winds which led to heat losses ( Figure 8a) and evaporation (Figure 8b) and consequently to high surface salinity values. Additional heat losses by the end of January and the beginning of February coincide with the LDW formation event within the RG described above. The heat losses in mid-March coincide with the second dense water formation event within RG which led to a mixture of LDW and 'lower range' LIW formation. The heat and freshwater fluxes integrated over the center of RG show an intense preconditioning phase during December 2005, due to strong dry and cold winter winds which led to heat losses ( Figure 8a) and evaporation (Figure 8b) and consequently to high surface salinity values. Additional heat losses by the end of January and the beginning of February coincide with the LDW formation event within the RG described above. The heat losses in mid-March coincide with the second dense water formation event within RG which led to a mixture of LDW and 'lower range' LIW formation.

LIW and LDW Formation within the Northwestern Levantine Sea
The heat and freshwater fluxes integrated over the center of RG show an intense preconditioning phase during December 2005, due to strong dry and cold winter winds which led to heat losses ( Figure 8a) and evaporation (Figure 8b) and consequently to high surface salinity values. Additional heat losses by the end of January and the beginning of February coincide with the LDW formation event within the RG described above. The heat losses in mid-March coincide with the second dense water formation event within RG which led to a mixture of LDW and 'lower range' LIW formation.   (Figure 6b-d).
The mesoscale eddy during that time shows a diameter of about 60 km which is within the typical mesoscale eddy diameter within the Levantine Sea (40-80 km).     Figure 11 shows the T/S plots for the two events of dense water formation during JFM 2006. Water masses above 100 m were not taken into account for the T/S plot to exclude shallow MLDs and recapping and to capture the events of pronounced intermediate and deep water formation. Water masses from 100 to 500 m are plotted with a green dot while water masses under 500 m are plotted with blue dots. Figure 11a shows the T/S plot for the first dense water formation event: The potential temperature exhibits values smaller than 14.5°C, the salinity shows values smaller than 39 psu, and the potential density shows a constant value of about 29.17 kg/m 3 . The typical ranges for LDW for potential temperature are between 13.7°C and 14.5°C and for salinity between 38.8 to 38.9 psu [6,9,10]. The potential density line of 29.17 kgm −3 represents the upper deep-water boundary density for the NWLS, corresponding to approximately 1000 m depth [23]. All potential temperatures and salinity   Figure 11a shows the T/S plot for the first dense water formation event: The potential temperature exhibits values smaller than 14.5 • C, the salinity shows values smaller than 39 psu, and the potential density shows a constant value of about 29.17 kg/m 3 . The typical ranges for LDW for potential temperature are between 13.7 • C and 14.5 • C and for salinity between 38.8 to 38.9 psu [6,9,10]. The potential density line of 29.17 kgm −3 represents the upper deep-water boundary density for the NWLS, corresponding to approximately 1000 m depth [23]. All potential temperatures and salinity values lay on the line of constant potential density of 29.17 kgm −3 , i.e., that the formed water masses sank to at least 1000 m, until the same potential density was reached. Therefore, the T/S plot confirms that LDW took place during the first event by late January until mid-February (Figure 11a).
For the dense water formation event in March 2006, the T/S plot shows potential temperatures smaller than 15 • C, salinities smaller than 39.1 psu, and potential densities between 29.125 kg/m 3 and 29.17 kg/m 3 . By nearly reaching 15 • C and with some salinity values above 39 psu, a part of the water mass reaches the lower range of LIW water mass characteristics (Figure 11b). This indicates a mixture of LDW and 'lower range' LIW formation during the second dense water formation event within RG.
(2) LDW formation took also place from the end of February until mid-March 2004 within another cyclonic mesoscale eddy located in the western part of the RG. Hoevmueller plots for DJFMA for salinity, potential temperature, and potential density are shown in Figure 12. During January and February, the MLD deepens constantly and the event of LDW formation occurs by the end of February until mid of March when the minimum surface temperature was reached. The examination of single profiles shows convection down to at least 600 m. (2) LDW formation took also place from the end of February until mid-March 2004 within another cyclonic mesoscale eddy located in the western part of the RG. Hoevmueller plots for DJFMA for salinity, potential temperature, and potential density are shown in Figure 12. During January and February, the MLD deepens constantly and the event of LDW formation occurs by the end of February until mid of March when the minimum surface temperature was reached. The examination of single profiles shows convection down to at least 600 m. The T/S plot of JFM2004 shows mainly LDW formation ( Figure 14). Green dots represent water masses between 100 and 500 m while blue dots represent water masses between 500 and 600 m. The water mass characteristics show LDW and 'lower range' LIW (potential temperature around 15°C and 39 psu < salinity < 39.1 psu) formation. The T/S plot of JFM2004 shows mainly LDW formation ( Figure 14). Green dots represent water masses between 100 and 500 m while blue dots represent water masses between 500 and 600 m. The water mass characteristics show LDW and 'lower range' LIW (potential temperature around 15 • C and 39 psu < salinity < 39.1 psu) formation.  The T/S plot of JFM2004 shows mainly LDW formation (Figure 14). Green dots represent water masses between 100 and 500 m while blue dots represent water masses between 500 and 600 m. The water mass characteristics show LDW and 'lower range' LIW (potential temperature around 15°C and 39 psu < salinity < 39.1 psu) formation.   Monthly means of satellite SST are shown in Figure 16. The satellite SST reached a minimum between mid-February and mid-March, i.e., during the event of deep water formation, and coincides with the surface potential temperature measured by the Argo float ( Figure 12). Monthly means of satellite SST are shown in Figure 16. The satellite SST reached a minimum between mid-February and mid-March, i.e., during the event of deep water formation, and coincides with the surface potential temperature measured by the Argo float ( Figure 12).

LIW Formation along the Coastline
One example of LIW formation along the coastline is given in this subsection. While LDW was formed inside RG, typical LIW was instead formed along the Northern Turkish coastline, i.e., along the AMC. The salinity, potential temperature, and potential density values characterize a typical LIW formation event by the end of March ( Figure 17).
In December and January, very high surface salinity values (S > 39.3 psu) can be seen in the upper 50 to 100 m (Figure 17b). However, deep mixing and convection is still prevented by relatively high surface temperatures (T > 17°C) during this period. Surface temperature has a decreasing trend from about 20°C by the very beginning of December to about 17°C in March when deep convection occurs. By the end of April, the surface temperature gradually increases reaching 17.5°C. The observed surface water temperature along the coastline is about 1°C to 3°C warmer than in the open sea (Figures 6 and 12).
The MLD deepens gradually from about 50 m within December to about 250 m during February; therefore, the high surface salinity is mixed throughout intermediate layers (Figure 17b).
By the end of March, when lowest surface temperatures are reached (Figure 17c), dense water formation starts to occur. Surface potential density reaches values between 29 and 29.1 kg/m 3 during this period. The examination of single profiles shows that the mixing event takes place down to about 550 m.

LIW Formation along the Coastline
One example of LIW formation along the coastline is given in this subsection. While LDW was formed inside RG, typical LIW was instead formed along the Northern Turkish coastline, i.e., along the AMC. The salinity, potential temperature, and potential density values characterize a typical LIW formation event by the end of March (Figure 17).
In December and January, very high surface salinity values (S > 39.3 psu) can be seen in the upper 50 to 100 m (Figure 17b). However, deep mixing and convection is still prevented by relatively high surface temperatures (T > 17 • C) during this period. Surface temperature has a decreasing trend from about 20 • C by the very beginning of December to about 17 • C in March when deep convection occurs. By the end of April, the surface temperature gradually increases reaching 17.5 • C. The observed surface water temperature along the coastline is about 1 • C to 3 • C warmer than in the open sea ( Figures 6  and 12).

Climatology of Winter Mean MLD from 2000 to 2018
The climatology of the winter (JFM) mean MLD from 2000 to 2018 for the Levantine Sea is shown in Figure 21. Within the cyclonic RG, the mean winter MLD is quite shallow (around 70 m). Deeper mean winter MLDs are found within anticyclonic eddies (IE, MME, CE, ShE; see Figure 1b for position of eddies) and along the coastlines, indicating dense water formation along boundary currents.

Climatology of Winter Mean MLD from 2000 to 2018
The climatology of the winter (JFM) mean MLD from 2000 to 2018 for the Levantine Sea is shown in Figure 21. Within the cyclonic RG, the mean winter MLD is quite shallow (around 70 m). Deeper mean winter MLDs are found within anticyclonic eddies (IE, MME, CE, ShE; see Figure 1b for position of eddies) and along the coastlines, indicating dense water formation along boundary currents.

Discussion and Conclusions
The present study is focused on the LIW and LDW formation events in the NWLS (Figure 22a) as detected by Argo float data from 2001 to 2017. The new and most interesting result is that the typical LIW (potential temperature > 15°C and salinity > 39 psu) formation mainly occurred along the Northern coastline (Figure 18), while 'lower range' LIW (potential temperature about 15 °C and

Discussion and Conclusions
The present study is focused on the LIW and LDW formation events in the NWLS (Figure 22a) as detected by Argo float data from 2001 to 2017. The new and most interesting result is that the typical LIW (potential temperature > 15 • C and salinity > 39 psu) formation mainly occurred along the Northern coastline (Figure 18), while 'lower range' LIW (potential temperature about 15 • C and salinity about 39 psu) and LDW (13.7 • C < potential temperature < 14.5 • C, 38.8 psu < salinity < 38.9 psu) formation took place within mesoscale eddies located within the center of RG (Figure 22a; Figure 11a, Table 1).  The schematic summary of the results for the winter seasons from 2001 to 2017 is evident in Figure 22b. Blue and brown arrows describe the convection and net sinking areas of the Mediterranean Sea as derived from theoretical models by [15,16]. Red (Figures 6, 12 and 17), in agreement with the results of [24,25]. In January-February 2006, the Argo float data detected the LDW formation in the core of a cyclonic mesoscale structure located in the center of RG. This structure (diameter of about 60 km) shows the typical horizontal scale of convective chimneys. The T/S plot in Figure 11a reveals the LDW characteristics of the convection event. All potential temperatures and salinity values lay on the line of constant potential density of 29.17 kgm −3 , revealing the sinking of the formed water masses to at least 1000 m as previously observed [6,9,10].
The intensity of the mixing and convection events depends mainly on the surface buoyancy flux B, which in turn depends on its thermal (B T ) and haline (B S ) components. The calculation and plot of the surface buoyancy flux and its components revealed that the haline component dominates over the thermal component ( Figure 4, lower panel), i.e., intense evaporation (B S < 0) controls the surface buoyancy loss, especially during the preconditioning phase (e.g., Figure 8 for winter 2006). Therefore, the area of the RG is an area of net buoyancy loss, driven by the haline component, as shown by [26].
The influence of salinity and temperature gradients to the density gradients are described in the Hoevmueller diagrams of the Turner angle (Figures 7, 13 and 18). During the preconditioning phase (November, December) and the constant deepening of the MLD in the beginning of January, the influence of the salinity gradient was dominant, while during strong unstable conditions and consequently dense water formation, also the temperature gradient was influential. The Turner angle also approximately indicated the depth of the convection events.
The deep dense water formation events within the area of the RG can be described by the following phases: The whole process is influenced by the cyclonic rotation of the RG leading to the upwelling of cooler waters to the surface. In November and December, the preconditioning phase starts (Figures 4 and 5): the heat losses due to cold and dry winds lead to increased surface salinity (Figures 6, 12 and 17) through the evaporation and to a steady deepening of the MLD. Additional, temporarily outbreaks of strong winds during January, February, or March cause strong heat losses ( Figures 5 and 8), which cause further cooling of surface waters. When lowest surface temperatures are reached (Figures 10, 16  and 19), dense water formation starts to occur. Within hours, the newly formed dense water sinks down rapidly to a depth of equal density where it spreads horizontally, forming an anticyclonic circulation due to the influence of the Coriolis force.The convection event also implicates a stretching of the water column leading to a change in vorticity, an increased geostrophic velocity, and a depression of the SSH. In fact, all pronounced dense water formation events documented by the Argo floats were indicated by a strong depression of satellite SSH (Figures 9, 15 and 19) and by lowest SSTs (Figures 10, 16 and 19).
The Argo float data revealed that LDW formation took place within the RG during winter months and showed the key role of the boundary currents for the LIW formation. The climatology of the mean MLDs of the Levantine Sea ( Figure 21) reveals that, despite the deep convection events, little to net mean sinking takes place within the center of RG, while the deeper MLDs along the coastline indicate dense water formation occurring along boundary currents. Therefore, the drivers, sources, and main contributors of the Mediterranean thermohaline circulation have to be rethought not only within the Levantine, but also within the Mediterranean Sea. Deployments of additional Argo floats to survey boundary currents and deployment of deep Argo floats within the main Mediterranean convection sites, i.e., the Gulf of Lion, the South Adriatic, and the RG area, will contribute to further understanding of dense water formation processes.
A better understanding of the Mediterranean thermohaline circulation is needed not only for a wider knowledge of the effects of climate change, but also for the impact on the ecology. Newly formed intermediate or deep waters can be polluted (with oil, microplastics, nutrients from extensive agriculture, heat due to global warming etc.), e.g., the Northern Levantine coastline, where pronounced dense water formation occurs, has the highest coastline plastic pollution within the Mediterranean Sea [27]. The newly formed water masses with the above-mentioned water properties and pollutants are transported throughout the Mediterranean to finally reach the Atlantic Ocean. The full impact (in terms of pollution and different water mass characteristics) will only be seen by future generations when these water masses emerge after decades or even centuries at different places within the Mediterranean Sea.
Therefore, it is obvious and evident that scientists and policy makers are obliged to join forces now to support and make commitments towards a real sustainable world that is not threatening, but protecting our ecosystems and lives.

Funding:
The float data were collected and made freely available by the International Argo Program and the national programs that contribute to it (http://argo.jcommops.org). The Argo Program is part of the Global Ocean Observing System.