The data analyzed come from the South Adriatic regional facility, which is part of the EMSO-Eric observation network consisting of two sites, one of which is located in the center of the SAP (E2M3A-position: Lat 41.53°N, Lon 18.06°E, see Figure 1 ) and the other consisting of two moorings on the western side of the southern Adriatic continental edge (mooring BB-position Lat 41.34°N, Lon 17.19°E and FF-position Lat 41.81°N, Lon 17.04°E, see Figure 1 ). The combination of data collected at these three moorings allows us to observe the flow of gravity currents in the SAP after the dense water formation in the northernmost part of the Adriatic Sea. The monitoring strategy set up for the southern Adriatic observatory is based on continuous and high-frequency measurements of temperature (T), salinity (S), oxygen (O) and velocities, which are suitable for detecting both long-term changes, episodic processes over decades (Cardin et al., 2020b), and high-frequency processes (1h) such as waves and turbulence.

The E2M3A observatory, which has been in operation since 2006 (Bensi et al., 2014; Cardin et al., 2014a, 2015, 2018, 2020b, 2020a, 2024; 2025; Amorim et al., 2024), collects hourly data of T, S and O at seven different levels of the water column (150, 350, 550, 750, 900, 1,000 and 1,200 dbar). In this study, we focus on the period from December 2011 to December 2022 (a period in which the availability of data is greatest for the three moorings combined) for the deepest layers of the pit.

The mooring BB is placed at 600 m depth on the main branch of the Bari Canyon system and FF on the furrow area on the open slope at 733 m depth (Paladini de Mendoza et al., 2022b, 2024a, 2024b). The two mooring lines are 100 m long, have been in operation since 2012, and collect half-hourly T, S, and current data at different levels of the water column. The Teledyne RD Workhorse 307KHz ADCP in downward-looking mode measures the currents of the last 100 m near the seafloor every 30 min with an accuracy of ±0.5 cm s⁻¹ and a resolution of 0.1 cm s⁻¹. The ADCP is a four-beam, convex configuration with a beam angle of 20°, and the number of depth cells is set to 27 with a cell size of 4 m. A CTD probe, SBE 16plus V2 SeaCAT, is located approximately 10 m above the seafloor to record thermohaline parameters. The water conductivity data were measured with a sensor with an accuracy of 5 × 10⁻⁴ Sm⁻¹ and a resolution of 5 × 10⁻⁵ Sm⁻¹; the water temperature was measured using a thermometer, with an accuracy of 5 × 10⁻³ °C and a resolution of 1 × 10⁻⁴ °C. The moorings are also equipped with a temperature sensor SBE56 (accuracy ±2 × 10⁻³ °C and 1 × 10⁻⁴ °C resolution), which is installed in FF at a height of 37 and 20 m from the seabed and in BB at a height of 33 m from the seabed.

In this study, we consider the potential temperature (θ) and S; θ was derived using the Gibbs Seawater GSW Oceanographic Toolbox, which contains the TEOS-10 routines (McDougall and Barker, 2011). The Python language was used in this study through several libraries: the Numpy module (Harris et al., 2020), the Pandas module (McKinney, 2010), the Scipy module (Virtanen et al., 2020), and the StatsModels module (Seabold and Perktold, 2010). We used the Matplotlib module (Hunter, 2007) to display the results.

Copernicus reanalysis data for the Mediterranean Sea and ERA-5 datasets from the Copernicus-ECMWF portal were used to assess the surface heat fluxes and the influence of salinity on the formation of dense water and the thermohaline properties of the water masses in the three sub-basins of the Adriatic Sea (North, Middle, and South).

Net surface heat fluxes were derived using hourly ERA-5 datasets of Copernicus-EMCWF (European Centre for Medium-Range Weather Forecasts, https://doi.org/10.24381/cds.adbb2d47) portal [Q_tot ; (Herbasch et al., 2020; Hersbach et al., 2023)] on three different areas: North, Middle and South Adriatic. We used data from 1979 to 2022. The total surface heat flux:

is the sum of (from left to right) the short-wave radiation, the long-wave radiation, the latent heat flux, and the sensible heat flux. The data is sampled hourly, but in our study we use daily averages. The resolution of the grid is 0.25°. We consider a spatial average over each of the three basins from December to March (winter period) and derive the monthly integrated surface heat flux for each month of this period.

The thermohaline properties of the water masses (S and θ) on the North Adriatic region are obtained from the Copernicus database (https://doi.org/10.25423/CMCC/MEDSEA/_MULTIYEAR_PHY_006_004_E3R1I) (Escudier et al., 2020; Nigam et al., 2021; Escudier et al., 2021). The data have a daily frequency and a horizontal grid resolution of 0.042°. Salinity is averaged over the 10 upper meters of the water column (and also over the entire water column in the Supplementary Figure S6 ), and we also consider monthly averages for salinity during the winter period. We use a dataset running from 2011 to 2022.

The PDA values are derived from daily S and θ datasets using the Gibbs Seawater (GSW) Oceanographic Toolbox, which includes the TEOS-10 routine (McDougall and Barker, 2011). We present PDA values corresponding to the mean of the PDA values in the upper 10 m of the water column.

Three-dimensional velocities were also obtained from the Copernicus database (https://doi.org/10.48670/moi-00021) (EU Copernicus Marine Service Information, 2024). The data considered are from 2012/01/01 to 2021/06/30. They have a daily frequency and a horizontal grid resolution of 0.083°. We consider an average of velocities at 1063 m corresponding to nodes of the model closer in latitude and longitude than 0.1° to the E2M3A position (four nodes).

The time series of S, θ, PDA, and O in the dSAP all show bursts of strong fluctuations that start in the late winter period in 2012, 2017, 2018, and 2022 and can be characterized by the fluctuation index defined in subsection 2.1.4 (see Figure 3 ). The strong fluctuations of oxygen during bottom ventilation years provide information about the ventilation that took place in the dSAP, as mentioned by (Gačić et al., 2002; Cardin et al., 2011; Bensi et al., 2014; Cardin et al., 2020b; Amorim et al., 2024). They cannot be the result of a deep convection event, as convection, during the studied period, never reached 900 m and could then not destroy the two-layer structure that was mentioned in (Amorim et al., 2024). The ventilation of the deep layer is therefore due to gravity currents. Here, we argue that the strong fluctuations are the result of gravity currents of North Adriatic origin entering the dSAP along the topography and accumulating there (Bensi et al., 2013; Mihanović et al., 2018), where it is subject to local mixing processes. The next gravity current arrives a year or a few years later.

Bursts of fluctuations of PDA at BB and FF are indicators of bottom ventilation of the SAP. If the water masses transported by these gravity currents are dense enough, they proceed to the dSAP, leading to turbulent mixing. In Figure 3 (and in Supplementary Figure S1 for SBE56 fluctuations at BB or FF in 2022), we observe that in 2012, 2017, 2018, and 2022, bursts of PDA fluctuations at BB were followed by bursts of fluctuations at the E2M3A mooring. The same applies to the FF mooring. The 15-day averages of PDA at the three moorings show a sawtooth behavior with a sudden increase in PDA, followed by a period of relaxation ( Figure 4 ). This sawtooth behavior is also discussed by (Querin et al., 2016; Cardin et al., 2020b) for the SAP and by (Mihanović et al., 2018) for the Jakuba pit. Querin et al. (2016) and Cardin et al. (2020b) explained this phenomenon as follows: the first phase consists of a linear density decrease in the SAP, which is due to local mixing driven by flow instabilities and mesoscale turbulence generated by the large-scale cyclonic circulation of the South Adriatic sub-basin. In contrast, the second phase consists of the abrupt increase of density, which is the result of gravity currents transporting NAdDW formed during the strongest winters. The second phase effectively ventilates the dSAP, which remains unventilated during the first phase. The saw-tooth behavior for oxygen observed at E2M3A was explained by Marini et al. (2006) for the Jakuba pit, where bottom water masses lose oxygen between two ventilation events due to mineralization, while nutrients increase. Martellucci et al. (2024) also observed the same saw-tooth behavior of oxygen at the BB mooring during the 2017, 2018, and 2019 ventilation events where dissolved oxygen increased to more than 235 µmolkg⁻¹.

Table 1 shows that the years in which all criteria are confirmed are 2012, 2017, 2018, and 2022 (fluctuations at BB and FF for T can be seen in Supplementary Figure S1 ). The occurrence of gravity currents in two consecutive years (2017/2018), as compared to a five-year period (Querin et al., 2016; Cardin et al., 2020b), possibly indicates an exceptional period of water mass changes for the south Adriatic, as mentioned by (Vilibić et al., 2023; Amorim et al., 2024).

Exceptional PDA fluctuations are observed in 2015, 2016, 2019, and 2021 in BB and FF, but no remarkable fluctuations were observed in the dSAP ( Figure 3 ), which is due to the fact that the water in BB and FF was not dense enough to reach the bottom ( Figure 4 ). We observed nevertheless fluctuations in the upper layer (900 and 1,000 dbar) (not discussed in the article). These waters intrude the pit at a higher level (measured at about 900 dbar, not discussed here). For instance, Martellucci et al. (2024) observe a lighter NAdDW than usual in 2015 and 2016 that filled the intermediate layer of the SAP. In 2013, strong fluctuations were also reported at BB and FF, but in the absence of data at E2M3A this year, no conclusion can be drawn about the presence of a gravity current reaching the dSAP or intruding at higher levels. As for the other years, none of the criteria were met in 2014 and 2020 ( Table 1 ) and no significant fluctuations are observed in neither in BB nor in FF ( Figure 3 ).

Pulses of dense water in the dSAP due to gravity currents are associated with the generation of NAdDW in the North Adriatic during winter periods, as shown by Di Biagio et al. (2023), where a moderate correlation (0.68) between the high concentration of dissolved oxygen in the deeper layers of the SAP with the intermittent inflow of NAdDW was found for the period 2012-2014. Evidence for the arrival of these dense water pulses is associated with the high concentration of dissolved oxygen, as mentioned by Manca et al. (2006). To determine this arrival, we compare the annual variability of PDA maxima in the North and Middle Adriatic with the gravity current events observed in the dSAP. To do so, we used PDA derived from conservative temperature (CT) and absolute salinity (SA) (which are calculated using Copernicus T and S data, see Figure 5 ) for the entire Adriatic Sea and determined the area where the NAdDW formed and the date of the PDA maximum. We considered a two-week centered mean around this date for the North-Middle Adriatic surface waters with a threshold of minimum water PDA to be able to reach the dSAP, which is the minimum value registered at the E2M3A bottom ventilation years, as will be discussed in the next paragraphs.

Results show that a gravity current event is detected in the pit when this criterion (Equation 2) is verified. Several years meet this criterion: 2012, 2017, 2018, and 2022 (bottom ventilation years) and 2013. These results are consistent with (Neri et al., 2022) who found NAdDW signatures with PDA values above 29.2 at bottom depth (between 40 and 60 m) at the sampling station SG05 (about 30 km from the coast near Senigalia) during the winters of 2012, 2017, and 2018. During these years, most of the water column was characterized by pools of waters with S above 38.5. In addition, (Mihanović et al., 2018) reported a maximum of PDA around mid-February in 2012 along the western Adriatic shelf, a result consistent with Table 2 .

Nevertheless, the reduction in the density of the NAdDW due to mixing on the way south also plays an important role in the cascading of gravity currents in the dSAP. Therefore, an estimate of this reduction was made for bottom ventilation years ( Table 2 ; Figure 4 ), which in 2012 amounted to 1.18 kg m⁻³, 0.83 kg m⁻³ in 2017, 0.56 kg m⁻³ (minimum value) in 2018 and 0.76 kg m⁻³ in 2022. Looking at the moving centered average of the speed with a window of 7 days (not shown here) at the current meter at E2M3A ( Figure 2 ), it can be seen that the highest values correspond to years with bottom ventilation, with the minimum at 13.89 cm s⁻¹, which corresponds to the value for the minimum density reduction found in 2018. Other bottom ventilated years depict higher velocities, such as 17.98 cm s⁻¹ in 2017 and 15.65 cm s⁻¹ in 2022. This indicates that the higher the turbulence level and mixing, the higher the speed in the dSAP.

The dense water formed during the bottom ventilation years also varies in its physical properties ( Table 2 ). While in 2012 a rather low θ value (5.80°C) and a high S value (38.68) defined the water mass at the time and place of maximum PDA formation, in 2017 and 2022 salinity played a more important role (38.65 and 38.83 respectively and θ around 9°C). In 2018, the situation between the two parameters as driving factors is more balanced. Potential temperature of 7.11°C is between the values of 2012 and 2017/2022 with the lowest S value among the four bottom ventilation years (38.12). This observation agrees with (Supić and Vilibić, 2006) who stated that some years are dominated by either low T or high S values, but there are also years when the situation is more complicated and both parameters contribute.

Dense water formation is driven by heat loss and the preconditioning phase of the water column. We investigate the influence of these factors on the formation of NAdDW, firstly calculating the monthly average values of the integrated surface heat fluxes for the North, Middle, and South Adriatic Figures 6 , 7 (delimitation of these three regions is shown in the Supplementary Figures S2 – S4 ). In the following, we determine the annual variability of spatially and monthly integrated surface heat fluxes by normalizing it, taking into account all available data sets (1979–2022) from ERA-5 (Hersbach et al., 2023; Herbasch et al., 2020). We define 〈Q_m 〉 the average over all the years of the surface heat flux for the month m:

Secondly, monthly average values of salinity values integrated over the upper 10 m ( Figure 7 ) and over the whole water column (S6) are also derived for the North Adriatic.

The analysis of the time series shows that in 2012 strong heat losses occur (−5 × 10⁸ Wm⁻²) ( Figure 6 ) in February as compared to other years of our dataset ( Figure 7 ). At the same time, high saline waters fill the surface layer in the North Adriatic, especially in January and February (37.9 and 37.8, respectively) ( Figure 7 ), leading to the formation of very dense waters with a maximum observed on 13/02/2012 ( Table 2 ). (Mihanović et al., 2013) and (Janeković et al., 2014) showed that the increase of 0.3 in S during the preconditioning of the water column in the North Adriatic (summer–autumn 2011) was one of the main factors that triggered the 2012 event.

In 2017 and after, the oceanographic condition of the Adriatic changed due to an overall increase of S, as mentioned in (Amorim et al., 2024). Regarding the North Adriatic, that year’s strongest integrated heat loss with a value of −6 × 10⁸ Wm⁻² took place in January ( Figure 6 ) and exhibits the maximum value among all Januaries of the period considered ( Figure 7 ). Salinity values are also among the highest [with a mean value greater than 38.3 on the whole water column in January (S6) and nearly 37.9 for the upper 10 m ( Figure 7 )].

These results explain the high PDA values we observe in January 2017 in Table 2 with a maximum on the 26/01/2017 in the Northern Adriatic of 30.11 kg m⁻³.

In 2018, as already mentioned, the situation is more balanced. We observe relatively low S values in the North Adriatic compared to the other years ( Figure 7 ). As for the winter period, the beginning was characterized by strong heat losses (−4 × 10⁸ Wm⁻², Figure 6 ) in December, no significant losses in January, and a subsequent strong loss (−3.7 × 10⁸ Wm⁻², Figure 6 ) in February. The latter destabilized the water column, which was already less stable due to the intrusion of high-saline water in early 2017, as shown by (Amorim et al., 2024). The combination of these conditions could explain that the high-density water ( Table 2 ) triggered the gravity current observed this year in the deepest part of the pit, where the highest density value was observed at the end of April ( Figure 4 ).

Winter 2022 can be considered a very mild winter for the North Adriatic, as there are no strong losses compared to the other years (positive normalized values, Figure 7 ) ( Figure 6 ). However, in contrast to the other years analyzed above, very strong heat loss events occurred in December and January in the Middle Adriatic (−5.9 × 10⁸ Wm⁻² and −4.8 × 10⁸ Wm⁻², respectively, Figure 6 ). In terms of salinity, the winter months in the north have the highest value for the entire water column compared to all years, with a mean value around 38.45 (S6). This high amount of salt in the water column leads to a density of PDA = 30.03, which is among the highest observed during the bottom ventilation years ( Table 2 ). The strong heat losses in the Middle rather than in the Northern Adriatic could explain the observation of gravity currents in the pit earlier than in the other years (19/02/2022).

In this section, we first consider the typical time scales of the duration of the gravity current event at BB and at E2M3A. Secondly, we estimate the travel time and speed of the gravity current from the generation site in the North Adriatic to BB/FF and from BB/FF to E2M3A. Results are presented in Table 3 .

To determine the duration of the gravity current event, we use the ADCP data at BB. A gravity current is present when the observed speed V is greater than 0.17 ms⁻¹, which corresponds to the average speed plus one standard deviation. We further consider speeds greater than 0.17 ms⁻¹ along the down-slope axis (V_c ). This threshold can be justified considering the study of Paladini de Mendoza et al. (2022a) where it was observed that the passage of gravity currents in FF or BB are marked by a decrease of water temperature and/or high salinity values which is more important for velocities greater than 0.17 ms⁻¹. The beginning of bursts of fluctuations in the dSAP was defined as the time when η ρ exceeds two times σ s , the standard deviation of fluctuations during non-bottom ventilation years.

The speeds recorded at BB along the down-slope axis ( Figures 8B–E ) exhibit regimes formed by a series of pulses or consisting of isolated points, which are in the following neglected, (section 2.1.3) to determine the beginning and the end of the pulse regime in the speed time series ( Figure 8 ; Table 3 ). The results show that pulses end earlier in 2018 and 2022 (mid-May) than in 2012 and 2017 (beginning of June) ( Table 3 ) which lead to an average duration of the pulses of 106 days during the bottom ventilation years ( Table 3 ). Moreover, we determined the average travel time between BB (FF) and E2M3A, using bursts of fluctuations of PDA, of 15.5 days (14 days) ( Table 3 ). These results show, firstly, that the mean velocity of the gravity current from BB (FF) to the dSAP is on average of 0.057 ms⁻¹ (0.078 ms⁻¹) ( Table 3 ). Secondly, they show that the end of the cascading of gravity currents in the dSAP (and so the beginning of the relaxation period) ends no later than 29/06 in 2012, 20/06 in 2017, 01/06 in 2018, and 02/06 in 2022. Considering the average time travel observed between BB and E2M3A, these values are consistent with those of Martellucci et al. (2024), who mentions an outflow of NAdDW from the Bari canyon toward the SAP usually in spring from March to June. These values indicate then that the duration time of bursts of fluctuations in the dSAP varies from 58 days in 2018 to 111 days in 2012 ( Table 3 ) which is also consistent with the fact that among bottom ventilation years, 2018 is the year when the less dense water was formed in the North ( Table 2 ). We observe that in 2017 and 2018 the duration of pulses at BB is greater than the duration of bursts of fluctuations in the dSAP ( Table 3 ). The opposite was true in 2022, which is explained by water passing though FF played a greater role in the renewal of the dSAP water mass ( Supplementary Figure S1 ) and pulses began earlier there (10/02, not shown here).

During bottom ventilation years, the minimum travel time from the formation site in the North Adriatic to the dSAP (interval between the date of the PDA maximum in the North Adriatic and the date of the start of the bursts of fluctuations at E2M3A) ranges from 25 days to 73 days, see Tables 2 , 3 . However, in 2022, we obtain a negative value indicating that the start of the bursts of fluctuations in the dSAP is due to either dense water that formed well before 08/03/2022 (date of the PDA maximum, Table 2 ) or dense water formed in the Middle Adriatic (see section 3.2 above). Assuming that the maximum of the PDA at the formation site in the North Adriatic generates the maximum of the PDA in the dSAP, a more accurate travel time ranges from 52 days in 2018–99 days in 2017 ( Table 3 ), which is consistent with the order of magnitude of 2–4 months for the travel time from the North to the Southern Adriatic mentioned in the literature (Vilibić and Orlić, 2001, 2002; Rubino et al., 2012; Benetazzo et al., 2014). These travel times lead to the velocity of the dense water masses ( Table 3 ) ranging from 0.083 ms⁻¹ in 2012 to 0.16 ms⁻¹ in 2018.

In the literature, (Vilibić and Orlić, 2001) mentioned a time of two months for the arrival of the NAdDW near Bari, which means a velocity of 7−8 cm s⁻¹ (a value also mentioned by (Hendershott and Rizzoli, 1976). However, as shown by (Vilibić, 2003) in his study around the Palagruza Sill, the NAdDW current can be very variable and reach velocities of up to 20 cm s⁻¹ in the southeast. This variability in NAdDW velocity was also mentioned by (Benetazzo et al., 2014), who found an NAdDW velocity in the order of 30 cm s⁻¹ when leaving the northern basin in the early stage of formation and 9 cm s⁻¹ on average after this early stage. Another hypothesis for the short time travel in 2018 is a higher ambient velocity of the water masses, to which the gravity currents are submitted on their way southward. To investigate this feature, maps of the difference in 7-day mean projected meridional velocity along the Adriatic axis in the near bottom (15−25 m above the bottom) (V′) between 2018 and the other bottom ventilation years (not shown here, see Supplementary Figure S8 for the difference between 2018 and 2017). We observed higher V′ values southward in 2018, explaining the higher dense water velocity observed. This result is confirmed by Table 4 ; Supplementary Figure S7 , which take into account the three areas on the western part of the Adriatic Sea sub-basins for 2012, 2017, and 2018 in the near-bottom region and which depict higher ambient velocities in each basin in 2018 as compared to the other years.

The water masses in the dSAP after a bottom ventilation event are a mixture of intruding waters, mainly through BB and FF, and the water present before the event (see section 2.1.2). If the mixing is due to turbulent motion of water parcels, the difference in the molecular diffusion coefficients does not matter and the mixing ratios of Equation 1 apply. The different water masses are presented in θ/S diagrams ( Figure 2 ), and can be distinguished based on their thermohaline properties (see section 2.1.2). A mixing of water masses shown in this diagram leads to a new water mass consisting of the weighed combination of all water masses involved. Therefore, in our case, the water mass after the bottom ventilation event must lie within the triangle represented by the water mass at E2M3A before the fluctuation and the water masses at BB and FF (see section 2.1.2). In terms of Equation 1, this means that all ratios (R) lie in the interval [0,1] and their sum equals unity. This condition is satisfied by the thermohaline properties for all the bottom ventilation years considered, that is, 2012, 2017, and 2018 (no thermohaline data is present for 2022 at BB/FF) ( Figure 9 ). This finding presents criteria 6, further indicating that a gravity current event is responsible for bottom ventilation.

In addition, it is interesting to note that for all three years, the θ/S properties differ. Indeed, the percentage of water mass originating from BB or FF involved in changing the properties of the deepest water mass at E2M3A differs between the different bottom ventilation events ( Figure 9 ) with values ranging from 10% to 23% for BB in 2012 and 2017, respectively, and from 12% to 48% for FF in 2018 and 2012, respectively. This means that in 2012 and 2018, the water mass involved in the change of the deepest waters of the dSAP passed mainly through FF, while in 2017 it passed mainly through BB. In 2022, in Supplementary Figure S1 , fluctuations seem to occur mainly at FF mooring, indicating that the water flowing into the pit came mainly from FF in that year.

In all three bottom ventilation years, we observe an increase in density due to gravity currents ( Figure 9 ). This increase is 1 × 10⁻² kg m⁻³ in 2012, 5 × 10⁻³ kg m⁻³ in 2017, and 1.2 × 10⁻² kg m⁻³ in 2018. Considering the period studied, for which the density has a variability of −0.1 kg m⁻³, this means that the density changes are 10%, 5%, and 12% of the total variability in 2012, 2017, and 2018, respectively, and they are opposing the general trend (increasing in density, Figure 3 ). These changes are non-negligible because, in order to reach the bottom, the density of the water mass transported by the gravity current has only to be slightly superior to the water mass in place. The information about potential temperature and salinity shows which variable was the driver for the gravity current. In 2012, we observed a decrease in S and θ by almost 2.5 × 10⁻² and 0.13°C, respectively, suggesting that temperature was the driver of the gravity current. In contrast, 2017 shows an increase in both S and θ by almost 4 × 10⁻² and 0.14 °C, respectively, suggesting that salinity is the driver. In 2018, we observed both a decrease in θ by about 1 × 10⁻² °C and an increase in S by about 1.5 × 10⁻² °C ( Figure 9 ). This year, the change in water mass properties is almost perpendicular to the isopycnals, which leads to a significant increase in density. In 2012 and 2017, the change was mostly along the isopycnals, the former exhibiting a decrease and the latter an increase in spiciness.

The intrusion of the water transported by the gravity current into the dSAP is mainly turbulent, since no replacement of the previous water mass was found. This is evidenced by the substantial value of R_b for all three bottom ventilation events. Furthermore, for the different bottom ventilation years, the change vector at 1,200 dbar at E2M3A before and after the fluctuations is close to the vector at 1,000 dbar at E2M3A and almost perpendicular to the vector defined by the difference of these two water masses before the fluctuations ( Figure 9 ). This means that the mixing between the water masses at 1,000 dbar and at 1,200 dbar, which would lead to a variation along the initial water mass differences, is small.

A θ/S diagram also provides information that helps to exclude convection being responsible for the change of water mass properties in the dSAP as seen on Figure 9 , where the point corresponding to the water mass after the bottom ventilation at 1,000 dbar has a lower density than the point corresponding to the same situation at 1,200 dbar for all three years. This means that convection, which homogenizes the water masses in the vertical, is not involved here and that the dSAP is ventilated by gravity currents.

The observation of gravity currents through the Copernicus reanalysis is subject to the performance of the model to detect mesoscale processes. To compare E2M3A Lagrangian observations and Copernicus Eulerian results, we derive a 15-point average of Copernicus PDA around the E2M3A position ( Figure 10 ). Results show that bursts of fluctuations are not well represented in the Copernicus reanalysis, as no significant event is reported during bottom ventilation years by the fluctuation index ( Figure 10 ). Furthermore, the correlation coefficient between the two time series is equal to 0.24 (and 0.13 for a weekly average), which indicates that this model does not have a vertical resolution high enough to detect gravity currents cascading to the dSAP, see (Laanaia et al., 2010). During bottom ventilation years, we can observe fluctuations in the Copernicus reanalysis, but, considering the whole time series, they are not significant in terms of the fluctuation index (see Figure 10 ). Nevertheless, as presented in Figure 11 , potential temperature and salinity of the Copernicus reanalysis follow the general trend of the E2M3A time series. For the horizontal velocities, the amplitudes in the Copernicus reanalysis are typically five times smaller, and gravity currents are not discernible. The reanalysis poor results depicted in Figure 10 are due to a next-to-compensation of the influence of potential temperature and salinity when density is considered and a deficiency of the Copernicus data to capture high-frequency (weekly) changes. Today’s fine resolution numerical models of the Adriatic do not allow sufficient resolution to represent the topographic features important for the dynamics of a gravity current and its nonlinear dynamics.