Effects of the Eastern Mediterranean Sea circulation on the thermohaline properties as recorded by fixed deep-ocean observatories

Deep-Sea Research I 112 (2016) 1–13

Contents lists available at ScienceDirect

Deep-Sea Research I journal homepage: www.elsevier.com/locate/dsri

Effects of the Eastern Mediterranean Sea circulation on the thermohaline properties as recorded by fixed deep-ocean observatories Manuel Bensi a,n, Dimitris Velaoras b, Virna L. Meccia c, Vanessa Cardin a a

Istituto Nazionale di Oceanografia e di Geofisica Sperimentale (OGS), Borgo Grotta Gigante 42/C, 34010 Sgonico (TS), Italy Hellenic Centre for Marine Research (HCMR), Institute of Oceanography, Anavyssos, Greece c Institute of Marine Sciences-National Research Council, ISMAR-CNR, Trieste, Italy b

art ic l e i nf o

a b s t r a c t

Article history: Received 20 August 2015 Received in revised form 2 February 2016 Accepted 3 February 2016 Available online 3 March 2016

Temperature and salinity time-series from three fixed observatories in the Eastern Mediterranean Sea (EMed) are investigated using multi-annual (2006–2014), high-frequency (up to 3 h sampling rate) data. Two observatories are deployed in the two dense water formation (DWF) areas of the EMed (Southern Adriatic Sea, E2-M3A; Cretan Sea, E1-M3A) and the third one (Southeast Ionian Sea, PYLOS) lays directly on the intermediate water masses pathway that connects the DWF sources. The long-term variations of the hydrological characteristics at the observatories reflect the oscillating large-scale circulation modes of the basin (i.e. BiOS-Bimodal Oscillating System and internal thermohaline pump theories). In particular, between 2006 and 2014 an anti-correlated behaviour of the intermediate layer (200–600 m) salinity between the Adriatic and Cretan Sea observatories is verified. This behaviour is directly linked to reversals of the North Ionian Gyre, which appeared cyclonic during 2006–2011 and turned anticyclonic after 2011. Statistical analysis suggests that the travel time of the intermediate salinity maximum signal between the Cretan and Adriatic Sea is roughly 1.5 years, in good agreement with the analysis of additionally presented ARGO data as well as previous literature references. We argue that the understanding of such oscillations provides important foresight on future DWF events, as increased salinity may act as a crucial preconditioning factor for DWF processes. Additionally, energy spectrum analysis of the time-series revealed interesting short-term variability connected to mesoscale activity at the observatories. Hence, the sustain of permanent observatories able to monitor oceanic parameters at high sampling rates may play a key role in understanding both climatic and oceanic processes and trends. & 2016 Elsevier Ltd. All rights reserved.

Keywords: Quasi-decadal oscillations EMed circulation Dense water formation Thermohaline variability Adriatic Sea Ionian Sea Cretan Sea Short-term variability Deep-ocean observatories FixO3

1. Introduction The Mediterranean Sea is a semi-secluded oceanic basin of “concentration” type, since evaporation exceeds precipitation and river run off. The surface/subsurface (roughly 0–150 m) inflow of Atlantic Water (AW) that enters the Mediterranean through the Gibraltar Strait in order to compensate the water deficit of the basin, reaches the Eastern Mediterranean (EMed). There, in its easternmost part, AW is converted through winter open-sea convection into a dense and saline intermediate water mass known as Levantine Intermediate Water (LIW, Özturgut, 1976; POEM Group, 1992). Apart from the LIW, an additional intermediate water mass similar to the former in hydrological characteristics, namely the Cretan Intermediate Water (CIW), is formed inside the Cretan Sea through winter convection that locally reaches roughly 300 m n

Corresponding author.

http://dx.doi.org/10.1016/j.dsr.2016.02.015 0967-0637/& 2016 Elsevier Ltd. All rights reserved.

depth (Georgopoulos et al., 1989; Cardin et al., 2003; Velaoras et al., 2013). The CIW is considered slightly colder, saltier and denser than the LIW (Astraldi et al., 1999; Theocharis et al., 1999). It flows out from the Cretan Sea through a complex strait system of successive islands that separate Crete from the adjacent landmasses, such as the Antikythira Straits west of Crete (700 m deep) and the Kassos and Karpathos straits east of Crete (900 m and 850 m deep respectively, see Fig. 1). The LIW/CIW formation in the EMed drives a westward return flow at intermediate depths (roughly 150–500 m) carrying saline water masses into the Central and Western Mediterranean (WMed) seas. This surface eastward inflow-intermediate westward outflow system constitutes the upper thermohaline conveyor belt of the Mediterranean Sea. The Ionian Sea, due to its central position linking the WMed and the EMed, is a key region for the exchange of water masses (AW and LIW/CIW) between these two major Mediterranean sub-basins. The role of the saline intermediate masses (i.e. LIW, CIW) is of great importance for the dense water formation (DWF) processes

2

M. Bensi et al. / Deep-Sea Research I 112 (2016) 1–13

Fig. 1. General circulation of the EMed (LIW/CIW ¼ Levantine/Cretan Intermediate Water; AW ¼Atlantic Water) associated with the NIG (North Ionian Gyre) during the cyclonic (left panel) and anticyclonic phases (right panel). Red dots show the three fixed observatories part of the FixO3 network (E1-M3A, E2-M3A, and PYLOS). The color shading indicates the strength of the S signal associated with the water masses' flow. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article).

taking place in the Mediterranean Sea. They provide the DWF areas with the necessary salt amount that acts as a preconditioning factor favouring local winter convection. Two such DWF formation areas exist in the EMed, namely the Adriatic and the Aegean Seas (Zervakis et al., 2000; Gačić et al., 2002; Manca et al., 2002; Vilibić, 2003; Nittis et al., 2003a; Supić and Vilibić, 2006). It is in these DWF areas where the strong cold and dry winds, blowing during the winter period, provoke significant surface buoyancy loss. This last causes the sinking of surface water and the formation of dense waters that contribute to the Eastern Mediterranean Deep Water (EMDW, Hainbucher et al., 2006). Water masses dense enough to reach deep/bottom layers of the EMed are formed during winter mainly in the Adriatic Sea, but occasional very dense waters (able to reach the Ionian abyssal plain) have also been formed in the Aegean Sea. In the latter case, the most important episode was reported between 1987 and 1995, when a massive dense water outflow from the Cretan Sea altered the characteristics of the deep thermohaline cell of the EMed, making the EMDW saltier and warmer than in the past. This event was called Eastern Mediterranean Transient (EMT, Roether et al., 1996; Theocharis et al., 1999). Amongst the various synergetic causes that led to the EMT event and to the change of the main DWF sources in the EMed, there is the alternation of the AW and LIW water mass pathways in the Ionian interior as reported by Malanotte-Rizzoli et al. (1999). This alteration took place between 1987 and 1991 and resulted in the advection of AW masses towards the Northern Ionian while saline LIW was at the same time confined in the eastern part of the EMed. Consequently, there was a freshening of the Ionian basin accompanied by a respective salinity increase in the Aegean Sea that favoured DWF in that basin. By 1999, the AW pathway had re-established its eastward flow towards the Levantine (Theocharis et al., 2002), and the EMed has gradually been returning to its previous state with the Adriatic Sea as the main DWF source (Rubino and Hainbucher, 2007; Bensi et al., 2013; Cardin et al., 2015; Meccia et al., 2015). More recently, newly emerged theories have shown that the upper thermohaline cell of the EMed is subjected to periodical (quasi decadal) oscillations of circulation modes. In particular, Gačić et al. (2010, 2011, 2013) have proposed a feedback mechanism induced by the Adriatic dense water production, named

the Adriatic–Ionian Bimodal Oscillating System (BiOS), which can influence the salt distribution in the whole EMed. Additionally, Theocharis et al. (2014) and Velaoras et al. (2014) have suggested the existence of an internal thermohaline pumping mechanism that takes into account the whole upper thermohaline cell of the EMed, which regulates the salinity distribution and the DWF processes in the respective formation areas of the EMed. Both aforementioned theories attribute the circulation oscillations to internal thermohaline mechanisms. Other studies (e.g. Cessi et al., 2014; Pinardi et al., 2015) point out the effect of the wind stress as the main driving factor for the circulation changes. They argue that the wind works together with the buoyancy fluxes, both contributing in comparable portion, to the support of the mechanical energy of the circulation in the whole Mediterranean Sea. The upper thermohaline cell oscillations are manifested by reversals of the Ionian upper circulation. The latter are expressed in different phases of the North Ionian Gyre (NIG), which presents alternations between cyclonic and anticyclonic rotation schemes (Fig. 1). These reversals affect the salt distribution within the EMed by modifying the AW pathways towards the Levantine or Northern Ionian Sea thus producing out-of-phase variations between the salt content in the Adriatic or in the Aegean (Cretan) Seas. In turn, these salinity oscillations favour DWF in each of these marginal seas in an anti-correlated, competitive way. Hereafter, we will refer generally to the NIG in cyclonic or anticyclonic phase, to understand the effects induced by different circulation modes on the thermohaline properties of the Adriatic and Aegean (Cretan) Seas. Observations during the last two decades have indeed shown that the NIG experiences periodical reversals that consequently govern the salt content changes in the Adriatic and Cretan Seas. Borzelli et al. (2009), Bessières et al. (2013), Gačić et al. (2011, 2014), Poulain et al. (2012) have confirmed that in 1997, 2006, and 2011 the Ionian Sea upper circulation demonstrated reversals (1997–2006 cyclonic NIG, 2006–2011 anticyclonic NIG, and again cyclonic after 2011). Additionally, both Gačić et al. (2014) and Mihanović et al. (2015) showed how these reversals could be rapid or slow, depending on the governing basin-scale processes that drive them. Results from time-series, often retrieved from repeated cruise stations, have been very helpful to understand long-term ocean processes at different temporal scales (Schroeder et al., 2013). For

M. Bensi et al. / Deep-Sea Research I 112 (2016) 1–13

that reason, there is a growing need for developing and maintaining fixed observational sites, able to provide long-term timeseries of in situ datasets in key regions of the ocean. In the recent past, a European Ocean Observatory Network for open-sea oceanographic sites has been established and consolidated within the European community, firstly through the EU-FP7 Collaborative Project EuroSITES project (http://www.eurosites.info/), and then through the FixO3 project (The Fixed Open Ocean Observatory Network, http://www.fixo3.eu/). Both projects created the necessary structure to integrate technical and scientific aspects needed to promote and develop the activities of fixed ocean platforms. The purpose of this study is twofold: on the one side, we aim at contributing to the understanding of the long-term thermohaline variability of the EMed, while on the other side we attempt to highlight the importance, and the potential, of data retrieved through open-sea deep observatories in the EMed. Maintaining long-term observatories is indeed a rather costly and risky task (Schroeder et al., 2013), which priority is often considered lower with respect to new emerging biogeochemical technologies. Hence, taking advantage of the data collected at three fixed-point observatories (part of the FixO3 network) located in key areas of the EMed (Adriatic, Cretan, Ionian seas, Fig. 1), we aim at interpreting their thermohaline variability in the context of the general circulation of the Central Mediterranean. Moreover, the continuous high frequency sampling strategy allows us to explore the mesoscale variability and relate it with larger scale processes. The paper is organized as follows: firstly, we will describe the datasets and methods applied; then, we will present and discuss the results regarding the relation between the observatories as well as the long-term and high frequency thermohaline variability; and finally, we will summarize the main conclusions.

3

2. Datasets and methods 2.1. Data sources The datasets used in this study comprise time-series of temperature (T) and salinity (S) collected at three open-sea observatories, namely E2-M3A, PYLOS, and E1-M3A (Fig. 2), all located in the EMed. E2-M3A and E1-M3A are characterized by a common configuration, since they were designed in the framework of the Mediterranean Moored Multi-sensor Array (M3A) concept during the first half of 2000 (Nittis et al., 2003b). Indeed, their architecture was based on the use of a principal mooring holding a surface buoy (Bozzano et al., 2013) and a subsurface mooring equipped with instruments for ocean interior monitoring. Moreover, E1-M3A and PYLOS work in the framework of the POSEIDON-II and KRIPIS projects, forming the national network of observatories planned to monitor the Greek seas. E2-M3A observatory is moored in the center of the Southern Adriatic Sea (41.5°N; 18.0°E) at about 60 nm from the shore and at about 1180 m depth (Fig. 1). Its system is constituted by two components. The principal mooring hosts the surface buoy that provides meteorological and radiometric measurements at the air-sea interface, telemetry and services (2.5 m above the waterline). It also supports physical-biochemical measurements in the surface layer (T, S, dissolved oxygen, pCO2 and pH at 2 m depth) with sampling rates of 1 h (T, S, dissolved oxygen, pH), and 4 h (pCO2). The secondary deep mooring hosts sensors (CTD, current meters) at 300 m, 350 m, 550 m, 750 m, 900 m, 1000 m, 1200 m depth, to measure physical and biochemical parameters (T, S, dissolved oxygen, light transmission, currents) with sampling rates of 1 h for all instruments with the exception of CTDs at 550 m and 750 m (3 h). Data used in this work comprise those

Fig. 2. Time-depth diagrams of θ (potential temperature) and S at E2-M3A (panel a, Adriatic Sea), PYLOS (panel b, Ionian Sea), and E1-M3A (panel c, Cretan Sea). (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article).

4

M. Bensi et al. / Deep-Sea Research I 112 (2016) 1–13

gathered from the secondary mooring and cover the period from late 2006 to late 2014. PYLOS observatory (36.83°N; 21.61°E) is a multidisciplinary mooring located in the crossroad of Adriatic and Eastern Mediterranean basins, in the easternmost Ionian Sea (5 nm from the shore at about 1670 m depth, Fig. 1). It is placed in a very geologically active area with lots of earthquakes and landslides as well as a potential source of tsunamis that might affect the EMed. Apart from meteorological and wave field sensors, it is equipped with T and S sensors at surface, 20 m, 50 m, 75 m, 100 m, 250 m, 400 m, and 500 m (600 m from late 2007 to late 2008) below the surface with a sampling rate of 3 h. Data cover the period from late 2007 to early 2013. E1-M3A observatory (35.78°N; 24.92°E) is located at approximately 22 nm from the shore at about 1440 m depth (Fig. 1). It is a multidisciplinary mooring located to the north of Crete Island, an area of open-sea conditions, characterized as extremely oligotrophic, where dense waters with intermediate and occasionally deep characteristics are formed. Apart from meteorological and wave field sensors, it is equipped with T and S sensors at surface, 20 m, 50 m, 75 m, 100 m, 250 m, 400 m, 600 m and 1000 m below the surface with a sampling rate of 3 hours. Data cover the period from mid-2007 to late 2014. In order to provide a temporal and spatial coverage of data in the easternmost region of the Ionian Sea, between the exit of the west Cretan Straits in the Southeast Ionian Sea and the Adriatic Sea, S profile data from ARGO floats were used. They span from 2004 to 2014 and are positioned inside a  1° wide strip between 36 °N and 39.7 °N. Data are freely provided by the Copernicus (Marine environment monitoring service) Mediterranean In-Situ TAC data base available through http://marine.copernicus.eu (ex-MyOcean service). To help in the interpretation of the results regarding the high frequency variability, we used model outputs from the physical reanalysis component of the Mediterranean Forecasting System. The reanalysis are also available through Copernicus. The Ocean General Circulation Model consists in the Nucleus for European Modelling of the Ocean-Ocean Parallelise (NEMO-OPA) version 3.2 (Madec et al., 1998). A detailed description of the high-resolution reanalysis is given in Adani et al. (2011). In this work, modeled daily fields of potential temperature and salinity were used. Finally, we used maps of Absolute Dynamic Topography (ADT) generated by the SSALTO/DUACS delayed time altimeter data provided by AVISO (http://www.aviso.altimetry.fr). ADT is the sum of sea level anomaly and mean dynamic topography, both being referenced over a twenty-year period 1993–2012. For the Mediterranean Sea, provided data are gridded on a 1/8° regular grid. 2.2. Methods Data obtained from the three observatories were quality checked by means of direct comparisons with CTD casts periodically carried out at the sites. Accuracies for T and S data are 70.002 °C and 7 0.005, respectively. Each time-series was corrected, if needed, after post-calibration phase. Wherever necessary, in order to obtain a homogeneous dataset, time-series were subsampled at an interval of 3 hours. Data have been processed using MATLAB to be cleaned and despiked according to MyOcean in-situ quality control standards and methodology described in http://catalogue.myocean.eu.org/static/resources/user_manual/ myocean/QUID_INSITU_TS_OBSERVATIONS-v1.0.pdf. Finally, to obtain sub-inertial non-tidal flow a 33 h low pass Hamming filter was applied to the time-series (Flagg et al., 1976). Potential temperature (θ) and potential density (sθ) were computed with MATLAB using TEOS-10 thermodynamic equations of seawater (http:/www.teos-10.org). ARGO float profiles that provided data coverage at least in the

layer between 50 m and 600 m were selected. Vertically integrated S values were calculated for the 150–400 m layer that characterizes the saline intermediate LIW/CIW horizon in the Ionian Sea. Taking the integrated values, we calculated the mean annual integrated value of S and the standard deviation for each year. The same procedure was applied separating the strip into two areas: the extreme North area (north of 38.5°N) and the extreme South area (south of 37.5°N) along the eastern Ionian region covered by the data. To identify events of relatively high frequency variability, a wavelet analysis was performed to the filtered time-series of θ and S. This methodology is useful to study non-stationary signals. In contrast to the Fourier analysis that returns an average amplitude and phase for each harmonic, the wavelet analysis produces instantaneous coefficients yielding information on the evolution of non-stationary processes (Meyers et al., 1993). The continuous wavelet transform Wn(s) of a discrete sequence xn, n ¼ 0;…;N-1 with uniform time steps δt, is defined as the convolution of xn with the scaled and normalized wavelet function ψ(η), which depends on a non-dimensional time parameter η: N −1

Wn (s )=

∑ n′= 0

⎡ (n′ − n) δt ⎤ x n′ ψ * ⎢ ⎥⎦ ⎣ s

where the * indicates the complex conjugate. By varying the wavelet scale s and translating along the localized time index n, it is possible to obtain the amplitude of any features versus the scale and how this amplitude varies with time (Torrence and Compo, 1998). The idea behind the continuous wavelet transform is to apply the wavelet as a bandpass filter to the time-series. A function must have zero mean and be localized in both time and frequency space to be admissible as a wavelet (Farge, 1992). One particular wavelet function is the Morlet function consisting of a plane wave modulated by a Gaussian:

WMorlet = π −1/4eiω 0 ηe−η

2 /2

where ω0 is a dimensionless frequency and η is a dimensionless time. To satisfy the admissibility condition, ω0 should be equal to 6 (Farge, 1992). When this methodology is applied to extract features, the Morlet function is recommended, since it provides a good balance between time and frequency localization (Grinsted et al., 2004). In this work, the MATLAB wavelet toolbox from the University of Colorado (http://paos.colorado.edu/research/wavelets/software.html) was used and the Morlet wavelet function (with ω0 ¼6) was applied. To perform the analysis, data evenly spaced in time are required. Hence, it was chosen to split the whole period of data collection and perform the analysis to sub-periods with a small amount of missing data (less than 5%). Data gaps were filled by means of a cubic interpolation. For each series, the mean value and the linear trend were subtracted before applying the methodology. Finally, the wavelet analysis was performed individually to each time-series, vertical level, and variable (θ, S). To help us in the interpretation of the wavelet analysis we used ADT maps of the Cretan Sea constructed averaging the daily ADT gridded data over a 5-days interval (July–August 2008), and synoptic fields of θ and S (daily mean reanalysis) of the Adriatic Sea from the numerical model described in Section 2.1. Some of the figures shown in this work were elaborated through Ocean Data View (Schlitzer, 2002).

3. Results and discussion 3.1. Relations among the three observatories: intercomparison of and S

θ

Fig. 2 shows the time-depth diagrams of θ and S at E2-M3A (i.e. Southern Adriatic), PYLOS (i.e. Southeast Ionian), and E1-M3A

M. Bensi et al. / Deep-Sea Research I 112 (2016) 1–13

(i.e. Cretan) sites. Firstly, an anti-correlated behaviour between the Adriatic and Cretan Seas emerges from this figure, especially in the intermediate layer spanning from 200 m to 800 m depth. Indeed, when both θ and S are relatively low in the Adriatic, they show relative maxima in the Cretan. In particular, the period that spans from 2009 to 2012 was characterized by a low S phase (and a slightly low θ phase) in the Adriatic. During that period, the NIG was anticyclonic (at least until 2011, Bessières et al. 2013; Gačić et al., 2014) hence, the AW flow stream entering the Ionian Sea from the Sicily Strait was able to flow directly towards the Adriatic through the Northern Ionian (Fig. 1). Data from the Ionian site instead, at around 300 m depth, show a relative warm and salty phase at the same period (2009–2012), concurrently with the Cretan data. A large similarity between thermohaline variability at E1-M3A and PYLOS (especially below 200 m depth) could be ascribed, in first approximation, to the vicinity of the two sites. However, the best explanation is provided by the fact that PYLOS lays on the LIW/CIW outflow from the Aegean Sea. Hence, major long-term variations captured by E1-M3A are expected also at PYLOS, with a certain delay that we will discuss in Section 3.2. Secondly, the low salt content in the Adriatic with respect to the Cretan and Ionian sites emerges from Fig. 2. This feature is due to a twofold cause: on the one hand, the LIW/CIW flowing towards the Adriatic Sea is progressively diluted along its way; on the other hand, the Northern Adriatic during favorable winter conditions produces relatively fresh (due to river runoff) and dense (due to strong cooling) waters that replenish the deep, and occasionally the intermediate layers of the Southern Adriatic (Vilibić and Orlić, 2002). Thirdly, the presence of a relatively fresh (So38.9) surface/ subsurface layer (o200 m) as well as a strong seasonal surface cycle at PYLOS emerge from Fig. 2. In particular, the presence of a low S surface/subsurface water was more evident until 2011, hence during the anticyclonic phase of the NIG. This fact could suggest that the AW during the NIG anticyclonic phase was able to reach the PYLOS site more easily than during the cyclonic phase. It is possible that during the former the AW moves closer to the Greek coast than during the latter, when the AW pathway is mainly zonal, away from the eastern Ionian coast (Bessières et al., 2013). However, the reduced AW presence after 2011 could be also

5

attributed either to local circulation changes (i.e. interannual/ seasonal variations of the Pelops anti-cyclone in position and/or strength) or to the advection of more salty water masses (especially below 100 m) observed at the site during the same period. As far as the seasonal variability at PYLOS is concerned, during the studied period (2007–2014), minima and maxima of θ and S in the surface layer (down to 100 m depth) can be clearly detected during the first and second half of each year, respectively (Fig. 3). In fact, maxima in θ at the surface correspond to minima in S, and viceversa. The only exceptions seem to occur in summer 2009 and 2012, when subsurface (20 m depth) θ maxima correspond also to S maxima. We hypothesize that the vicinity of PYLOS to the Pelops anticyclonic gyre could influence these local phenomena. In particular, Pelops gyre could cause downwelling of warm and less saline surface water (AW) especially during the late summer season when the anticyclone is intensified (Mkhinini et al., 2014). Moreover, the largest S maximum at  20 m depth occurred at the end of 2012 (Figs. 2 and 3), immediately after the beginning of the freshening phase at the E1-M3A. It is possible that the peak of S at PYLOS during 2012 was enhanced by the outflow of very salty (S 439.2) CIW/LIW waters from the Cretan Sea at the end of the anticyclonic phase of the NIG. This hypothesis is supported by the later abrupt S decrease down to 50 m depth at PYLOS that started in 2013 (Fig. 3). Table 1 shows long-term trends extracted from the time-series of all the observatories presented in this study. Positive trends in θ and S are observed at all the three sites in the intermediate/deep layers, with the exception of θ at 600 m and S at 250 m depth in the Cretan Sea and of θ and S at 600 m depth in the Ionian, which show a slightly negative trend. Especially for the Cretan Sea however, it is within the sensors’ accuracy range. Mentioned changes in θ and S have affected sθ, which experienced a slight decrease at all sites in the upper part of the water column. The analyzed period is still too short to hypothesize a trend towards a general warming and salinification of these sub-regions of the EMed, since the length of the presented time-series (less than 8 yrs) is comparable to the quasi-decadal period of the NIG reversals. However, there are studies that indicate the progressive warming and salinification of the Mediterranean Sea (e.g. Potter

Fig. 3. θ and S time-series at PYLOS site in the layer 0–100 m.

6

M. Bensi et al. / Deep-Sea Research I 112 (2016) 1–13

Table 1 Linear trends. θ and S negative values are highlighted in bold-yellow.

and Lozier, 2004; Borghini et al., 2014). At the studied sites, changes at depths between 200 m and 800 m can be attributed either to vertical movements of density surfaces, or to changes in the water masses properties occupying that layer due to lateral advection of water masses with different thermohaline characteristics. Long-term observations need to be maintained for the next years, in order to draw robust conclusions about possible trends linked to climatological changes. Only when the time-series are long enough we may be able to robustly answer the question whether there is a general trend or not, or if rather we are dealing with periodical oscillations. 3.2. Relations among the three observatories: correlation of S Here, the correlations among the three observatories are studied using only S variations. This choice is motivated because S signal appears stronger and, in general, is less affected by the seasonal cycle since it is more conservative than θ. The question to be addressed here is to what extent the variability observed at the three fixed observatories is correlated to each other. Hence, the correlation coefficients between couples of S time-series were computed in order to investigate possible relationships among

time-series at different depths in the three sites. Table 2 reports the results of these calculations. Additionally, using cross-correlation calculations we estimated the time delay of the S signal among the three observatories to assess the propagation time for the LIW/CIW from the Cretan Sea to the Adriatic Sea. Some limitations in the results have to be taken into account due to the gaps in the datasets. We will discuss them further on. S varies coherently at Ionian and Cretan sites (Fig. 2 and Fig. 4c). The highest correlations are 0.73 and 0.74 comparing time-series at 400–400 m and 250–400 m, respectively (Table 2). As far as the statistical correlation of data collected at the Ionian and Cretan sites in the subsurface layer (at 100 m depth) is concerned, the value is low (0.33) if compared with those obtained from the correlations among deeper time-series, probably due to the seasonal effects that are stronger at the Ionian than at the Cretan site (see Fig. 2). The time lag of the S signal between them, obtained from the cross-correlation, is around 4 months. S variations in the Southern Adriatic Sea, as discussed above, are opposite to those observed both at the Ionian and at the Cretan sites (see Fig. 2, Fig. 4c, and Table 2). The absolute highest statistical correlations between different couples of Adriatic and Cretan time-series span between  0.63 (Adr550m-Crt400m) and  0.73

Table 2 Statistical correlation values calculated between couples of salinity time-series at different depths. Best correlations for each comparison are highlighted in bold-red.

M. Bensi et al. / Deep-Sea Research I 112 (2016) 1–13

7

Fig. 4. S variability in the easternmost Ionian region, obtained from all available float profiles (http://marine.copernicus.eu). Panel a shows the geographical distribution of the data (black squares indicate moorings position). Panel b shows the 4th order polynomial fit curve over the annual mean integrated data in the layer 150–400 m for area north of 38.5 °N (blue curve) and south of 37.5 °N (red curve). Data from PYLOS have been superimposed. Panel c shows data from the three moorings.

(Adr750m-Crt600m), while between Adriatic and Ionian timeseries span between  0.45 (Adr750m-Pyl400m) and  0.75 (Adr350m-Pyl500m) (see Table 2). Comparing Adriatic data at 550 m with Cretan data at 250m-400m-600m we obtain negative correlations always larger than 0.63, and time lags of 19, 18 and 14.5 months, respectively, while the comparison among the Adriatic data at 750 m with Cretan data at 250m-400m-600m gives correlations 4 0.61 and time lags of 19, 11, and 10 months. It must be said that data at 550 m and 350 m depth are the most representative of the LIW signal that enters the Adriatic following the NIG reversals (i.e. cyclonic NIG means high salinity and viceversa). However, since the Adriatic data at 350 m has a limited time coincidence with the Cretan ones, we based our time lags calculation on the Adriatic data at 550 m. Hence, on average, the LIW/CIW signal takes approximately 16-18 months to reach the Adriatic from the Cretan Sea. In order to explain the S variability in the eastern Ionian region (between PYLOS and the Strait of Otranto) along the LIW/CIW northwestward path (Millot and Taupier-Letage, 2005), we used available profiles carried out by floats between 2004 and 2014. This period was chosen because it covers at least a decade and includes a NIG reversal, occurred approximately in 2006 (Bessières et al., 2013). Their data distribution and annual mean S values, vertically integrated in the layer 150-400 m, are shown in Fig. 4. The region covered by the data was divided into two non-adjacent sub-areas to emphasize the differences between them. In both sub-areas, north of 38.5°N and south of 37.5°N, S varies coherently with the data collected at PYLOS and E1-M3A sites (Fig. 4b, c).

Indeed, between 2008 and 2012, S increased in the intermediate layer (150-400 m depth). However, from south to north S experienced a decrease of about 0.07 (Fig. 4b) in response to the dilution that affect the LIW/CIW flowing towards the Adriatic. Furthermore, a certain delay between the south and the north time-series is observable. We may observe from the time-series that the S minimum in the southern sub-area lays between 2006 and 2008 while the S maximum between 2012 and 2013. Interestingly, in this region the LIW signal is stronger (larger S) during the anticyclonic phase of the NIG. It confirms that when a portion of AW is deviated by the NIG towards the Adriatic Sea, the overall salt content in the upper intermediate layer of the Levantine basin increases. Additionally, Fig. 4b shows that in the northern sub area of the eastern Ionian, both minimum and maximum S signals appear with a delayZ6 months. However, we are aware that there is a considerable error margin in this approach, as the available data do not cover every year and they do not have the same yearly profiles density. Regarding the obtained estimations of the travel time of the LIW/CIW signal from the Cretan Sea to the Adriatic Sea, some additional considerations have to be taken into account, especially if we want to compare these values with those reported in literature so far. Gačić et al. (2013) estimated a travel time of the LIW signal from its formation area (Rhodes Gyre) and the Sicily Channel of about 11 yrs. Roether et al. (1998), instead, used transient tracers to estimate a travel time of 8 yrs (upper limit) for the LIW along the same pathway. In this work, we consider a significantly shorter pathway (in terms of distance) for the LIW/CIW

8

M. Bensi et al. / Deep-Sea Research I 112 (2016) 1–13

towards the Adriatic Sea. Furthermore, the estimated time lag is obtained, not from CTD casts as in the aforementioned studies, but from high frequency almost continuous time-series, which obviously contain intrinsically some local and seasonal variability that can affect this calculation. However, being aware of the limitations that each method can exhibit, and considering a roughly approximated geographical distance at least 3 (lower limit) to 4 (upper limit) times (considering that water masses do not move directly to one point to another) shorter than the one considered by Roether et al. (1998) and Gačić et al. (2013), our results seem to indicate a slightly faster time travel of the LIW/CIW from the Cretan to the Adriatic Sea. Moreover, although Roether et al. (1998) did not present any estimation for the intermediate water travel time between the Cretan and the Adriatic Seas, Fig. 4 as appearing in their article shows that this travel time has an upper limit of no more than 2 yrs, which agrees well with our estimations. 3.3. High-frequency variability: identifying mesoscale events In this section, we take benefit from the high frequency data sampling to study the high frequency variability. We aim at attempting to identify energetic events of mesoscale variability and to find the possible forcing mechanisms related to it. The analysis was applied to time-series of θ and S for the Adriatic (E2-M3A) and Cretan (E1-M3A) sites, two regions where it is supposed to find DWF events and eddies passages. It is worthwhile to emphasize the importance of a continuous sampling design to address this study. The global wavelet spectra, which is the integration of the

wavelet power over time (not shown), revealed that in general the spectral energy is concentrated in the frequency bands around 0.067; 0.033 and 0.017 cycles per day (cpd). These scales correspond to periods of around 15, 30 and 60 days, respectively. In what follows, we analyze more in detail the variability around 15 days, since we are interested in the mesoscale variability associated with the passage of eddies. Figs. 5 and 7 display the results for the Adriatic (from November 2006 to February 2009) and Cretan (from November 2007 to April 2009), respectively. In these figures, time-series of θ and S vertical profiles are plotted in panels a) and c), respectively. The averaged wavelet power in the frequency bands corresponding to periods between 13 and 17 days is plotted as a function of time and depth in panel b) for θ and d) for S. To construct these figures, the wavelet power was normalized by the value corresponding to the 95% confidence level. Hence, values less than 1 indicate energy lower than the 95% confidence level and they are not contoured in the figures. On the other hand, values larger than 1 correspond to energy higher than the 95% confidence level and are the results plotted in the figures. Regarding the results for the Adriatic (Fig. 5), it seems that events of variability around 15 days are isolated episodes and occur for both θ and S. Some episodes seem to propagate throughout the water column, while others seem to be restricted to a specific depth range. Two episodes resulted particularly energetic during the analyzed period. The first one occurred during winter 2008 (Jan.–Feb.), clearly visible in the θ time-series from 350 m down to 750 m. The second energetic episode, instead, took

Fig. 5. Time-depth diagram of a) θ and c) S at E2-M3A (Southern Adriatic) from November 2006 to February 2009. Averaged wavelet spectral energy in the frequency bands corresponding to periods around 15 days as a function of time and depth for b) θ and d) S. Values of wavelet power are normalized to the 95% confidence level value. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article).

M. Bensi et al. / Deep-Sea Research I 112 (2016) 1–13

9

Fig. 6. Synoptic fields of θ (at 450m) and S (at 750m) resulted from the daily mean reanalysis of the Mediterranean Forecasting System close to E3-M3A during January 2008 (a) and June 2008 (b). Red points represent the mooring position.

place during spring/summer 2008 and it is evident from the S time-series at intermediate depths, in particular around 750 m. This event resulted more energetic for S but it is also present in the θ time-series. First, we keep our attention on the episode of θ variability occurred in January 2008. The wavelet methodology helped us to identify such mesoscale event with periodicity around 15-days. To illustrate the time evolution of the synoptic fields during that episode, θ at approximately 450 m was extracted from the ocean daily mean reanalysis of the Mediterranean Sea. The resulted fields are plotted in Fig. 6a for the period between 27 December 2007 and 21 January 2008, with snapshots each 5 days. Red dot represents the positions of the E2-M3A. The figure shows the evolution of a relatively cold gyre, which is propagating in the southeast direction. A cold lens that arises between two relatively warm lenses characterizes the modeled field of December 27 (Fig. 6a). By January 6, the cold lens is intensified and starts to move southward. Finally, by the end of January the structure is modified since the cold lens is passing through the Strait of Otranto in the southeast direction. Ursella et al. (2011) analyzed Acoustic Doppler Current Profile data collected during 5 months in the Strait of Otranto. They reported rotational events that resulted

vertically uniform on a time scale of about 10 days. The authors associated them with passages of eddies in the southward direction and suggested that such events are related to a maximum outflow from the Adriatic Sea. Regarding our results, our explanation is that the event of θ variability found by the wavelet methodology is explained by the passage of eddies that likely were generated by the DWF that took place during winter 2008. To explain the energetic episode associated with S variability at around 750 m during spring/summer 2008, we look again at the model results, in particular at the S fields. Fig. 6b shows the evolution of the synoptic fields near the E2-M3A in June 2008, when the event was identified by the wavelet methodology. In this case, the horizontal fields at approximately 750 m depth seem to be characterized by a zonal gradient where saltier waters arise along the eastern boundary and fresher waters from the western one. The evolution of those fields plotted every 7 days, shows mesoscale features causing S fluctuations at the E2M3A (Fig. 5c). This way, we suggest that the reported S variability with periodicity of 15 days was caused by lateral advection, likely associated with mesoscale features. Regarding the results for the Cretan (Fig. 7), energetic events with periodicity of about 15 days are in general, simultaneously

10

M. Bensi et al. / Deep-Sea Research I 112 (2016) 1–13

Fig. 7. Time-depth diagram of a) θ and c) S at E1-M3A (Cretan Sea) from November 2007 to April 2009. Averaged wavelet spectral energy in the frequency bands corresponding to periods around 15 days as a function of time and depth for b) θ and d) S. Values of wavelet power are normalized to the 95% confidence level value. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article).

present in both θ and S. Similarly to what observed at the E2-M3A, some events at the E1-M3A occurred throughout the water column, while others are limited to some vertical levels. In particular, a significant episode occurred in July 2008. High energetic episodes are also found at 600 m and near surface in May 2008 and September 2008. The event of July 2008 cannot be linked with DWF events since it is not a DWF season. As the E1-M3A site lays in an area surrounded by various eddies (Cardin et al., 2003) we may suggest that these signals are connected to eddy activity but further investigation is needed in order to propose their generating mechanisms. Here, we used ADT maps to evaluate the presence of dynamical features in the Cretan Sea, since we were more interested in the variations of the upper water column. In particular, the temporal evolution of ADT (5-days averaged snapshots) at the position of the E1-M3A (Fig. 8) shows that from mid-July to mid-August a strong cyclonic circulation develops in the central Cretan Sea. A minimum of ADT values is placed north of E1-M3A (Fig. 8). The upwelling caused by this cyclone is observed in Fig. 7 as a doming of the isotherms and isohalines during the same period as well as by the intensification of wavelet energy.

4. Conclusions In this work, we used θ and S data collected at three EMed open-ocean observatories, the E1-M3A (Cretan Sea), the PYLOS (southeastern Ionian), and the E2-M3A (Southern Adriatic) to

investigate long-term variability and possible relationships existing among these three regions characterised by complex internal dynamics. Time-series collected between 2006 and 2014 revealed that by 2011 S increased in the intermediate layer (200–800 m) both in the Cretan and eastern Ionian Seas. This increment was stronger in the southeastern Ionian, although the maximum S value (  39.15 at 250 m) resulted the same in both areas. From 2011 onwards, S decreased again. Statistical analysis showed that the high S signal appeared first in the Cretan Sea followed by similar signal at Ionian site with  4 months delay. This increase was also pointed out by Krokos et al. (2014) in the upper 300 m layer of the eastern Levantine using ARGO data between 2005 and 2010. This fact provides evidence that the S increase in the intermediate layer observed in the second half of the 2000s affected the whole eastern part of the EMed. The S increment seems then to have propagated from the Levantine to the Aegean and further on towards the Ionian Sea and then to the Adriatic. Time-series collected in the Adriatic Sea revealed that, during the period of relatively high S values at Cretan and southeastern Ionian sites, the intermediate layer reached a minimum value of  38.65 at 350 m in 2011, while further on it increased again. Thus, during the period of observations there was a clear anti-correlated behaviour between the S variations observed in the Adriatic and the Cretan/ Southeast Ionian Seas. This anti-correlation is attributable to the reversals of the upper Ionian Sea circulation (NIG reversals). Indeed, the NIG changed from cyclonic to anticyclonic in 2006 causing the deflection of fresher AW masses entering the Ionian

M. Bensi et al. / Deep-Sea Research I 112 (2016) 1–13

11

Fig. 8. ADT (cm) maps of the Cretan Sea averaged over a 5-day period for July–August 2008. The red dot indicates the E1-M3A position.

Sea through the Sicily Strait directly towards the Northern Ionian/ Adriatic basins (Fig. 1). The NIG reversed again to cyclonic after 2010/2011, reflecting the redirection of AW masses towards the eastern part of the EMed. The water mass connecting the Cretan and Adriatic Seas at intermediate depths, thus acting as a salt carrier between these basins, is a mixture of LIW and CIW. The statistical analyses showed that the time delay of the intermediate layer S signal observed in the Cretan Sea to reach the same layer in the Southern Adriatic is roughly 1.5 years, while the time delay for the same signal between the southeastern Ionian and the Southern Adriatic is less than 1 year. This finding is further validated by the data analysis of the ARGO floats profiles in the eastern part of the Ionian Sea, along the LIW/CIW path towards the Adriatic Sea (Fig. 4). Taking into consideration the results presented herein as well as the theories concerning the quasi-decadal oscillations of the upper thermohaline circulation of the EMed (e.g., Gačić et al., 2011; Theocharis et al., 2014), we may argue that the observed anti-correlated behaviour is directly linked to these circulation oscillations, which characterize the different functioning modes of the upper thermohaline conveyor belt of the EMed. The observed S anti-correlation highlights the existence of a competitive dipole in the EMed between the Adriatic and Aegean Seas. Increased S in one of these basins acts as a DWF preconditioning factor locally, while at the same time a lower salt content in the competitive basins hinders DWF processes therein. Indeed, outflow of denser than usual CIW from the Cretan Sea was observed during the high salinity

period (2008–2010) in that basin (Krokos et al., 2014; Velaoras et al., 2014), while dense water was formed in the Adriatic Sea during the high salinity period (winter 2012) in the Adriatic Sea (Gačić et al., 2014). It is therefore clear that the understanding of such oscillations and their implications on the preconditioning factors of the two DWF areas of the EMed provides important foresight on future DWF events. The wavelet approach applied to the Adriatic and Cretan timeseries to explain the short-term variability (  15 days) suggested that this could be mostly associated with eddy activity in both areas. In some cases, their generation can be enhanced by DWF events during the winter season. This was the case for the Adriatic site in winter 2008, when large averaged wavelet spectral energy from θ time-series in the frequency bands corresponding to a period of  15 days were distributed throughout the water column (from 350 m to 750 m depth). Differently, in summer 2008, the strong energetic phase caused by S fluctuations at 750 m, appeared associated with lateral intrusions of (alternately) fresher and saltier waters, likely transported by mesoscale eddies. This work highlights the importance of continuous measurements by fixed observatories. Indeed, such operational oceanography platforms do not only contribute to the evaluation of longterm trends of oceanic parameters but are the only means available that can resolve high frequency processes. Hence, the deployment and maintenance of such observatories is a key factor in understanding climatic trends and/or oceanic processes in both long and short time scales.

12

M. Bensi et al. / Deep-Sea Research I 112 (2016) 1–13

Acknowledgments This work was supported by the EU-FP7 FIXO3 project (grant agreement n °312463) and the Italian Ministry of Education, University and Research (MIUR) under the RITMARE (Ricerca ITaliana per il MARE) project, the POSEIDON-II project co-funded by the European Economic Area (EEA) Financial Mechanism and the Hellenic Ministry of National Economy, and the KRIPIS – Integrated Observatories in the Greek Seas project (MIS 451724) of the National Strategic Reference Framework (NSRF) co-funded by EU and Hellenic funds. Thanks are due to Leonidas Perivoliotis and George Petihakis for PYLOS and E1-M3A data, to the OGS technical staff (F. Brunetti, P. Mansutti, A. Bubbi, S. Küchler, G. Siena, F. Arena, R. Nair, and N. Medeot) for the maintenance of the E2-M3A observatory, and to the HCMR technical staff (D. Ballas, A. Chondronasios, M. Doumas, D. Kassis, P. Pagonis, M. Pettas, and M. Potiris) for the maintenance of the E1-M3A and PYLOS observatories. We would also like to acknowledge the contribution of the late K. Nittis for the precious work he did on the Mediterranean operational observatories network in the past years.

References Adani, M., Dobricic, S., Pinardi, N., 2011. Quality assessment of a 1985–2007 Mediterranean Sea reanalysis. J. Atmos. Ocean Technol. 28 (4). http://dx.doi.org/ 10.1175/2010JTECHO798.1. Astraldi, M., Balopoulos, S., Candela, J., Font, J., Gačić, M., Gasparini, G.P., Manca, B., Theocharis, A., Tintoré, J., 1999. The role of straits and channels in understanding the characteristics of Mediterranean circulation. Prog. Oceanogr. 44 (1–3), 65–108. http://dx.doi.org/10.1016/S0079-6611(99)00021-X. Bensi, M., Rubino, A., Cardin, V., Hainbucher, D., Mancero-Mosquera, I., 2013. Structure and variability of the abyssal water masses in the Ionian Sea in the period 2003–2010. J. Geophys. Res. Oceans 118, 931–943. http://dx.doi.org/ 10.1029/2012JC008178. Borzelli, G.L.E., Gačić, M., Cardin, V., Civitarese, G., 2009. Eastern Mediterranean transient and reversal of the Ionian Sea circulation. Geophys. Res. Lett. 36, L15108. http://dx.doi.org/10.1029/2009GL039261. Borghini, M., Bryden, H., Schroeder, K., Sparnocchia, S., Vetrano, A., 2014. The Mediterranean is becoming saltier. Ocean Sci. 10, 693–700. http://dx.doi.org/ 10.5194/os-10-693-2014. Bozzano, R., Pensieri, S., Pensieri, L., Cardin, V., Brunetti, F., Bensi, M., Petihakis, G., Tsagaraki, T.M., Ntoumas, M., Podaras, D., Perivoliotis, L., 2013. The M3A Network of Open Ocean Observatories in the Mediterranean Sea, in 2013 MTS/IEEE OCEANS-Bergen, 2013, pp. 1–10. Bessières, L., Rio, M.H., Dufau, C., Boone, C., Pujol, M.I., 2013. Ocean state indicators from MyOcean altimeter products. Ocean Sci. 9, 545–560. http://dx.doi.org/ 10.5194/os-9-545-2013. Cardin, V., Gačić, M., Nittis, K., Kovačević, V., Perini, L., 2003. Sub-inertial variability in the Cretan Sea from the M3A buoy. Ann. Geophys. 21, 89–102. http://dx.doi. org/10.5194/angeo-21-89-2003. Cardin, V., Civitarese, G., Hainbucher, D., Bensi, M., Rubino, A., 2015. Thermohaline properties in the Eastern Mediterranean in the last three decades: is the basin returning to the pre‐EMT situation? Ocean Sci. 11 (53–66), 2015. http://dx.doi. org/10.5194/os-11-53-2015. Cessi, P., Pinardi, N., Lyubartsev, V., 2014. Energetics of semienclosed basins with two-layer flows at the strait. J. Phys. Ocean 44, 967–979. http://dx.doi.org/ 10.1175/JPO-D-13-0129.1. Farge, M., 1992. Wavelet transforms and their applications to turbulence. Annu. Rev. Fluid Mech. 24, 395–457. http://dx.doi.org/10.1146/annurev. fl.24.010192.002143. Flagg, C.N., Vennersch, J.A., Beardsley, R.C., 1976. 1974 MIT New England shelf dynamics experiment (March 1974), data report. Part II: The moored array, MIT Rep. 76-1. Mass. Inst. of Technol., Cambridge, p. 22. Gačić, M., Civitarese, G., Miserocchi, S., Cardin, V., Crise, A., Mauri, E., 2002. The open ocean convection in the southern Adriatic: a controlling mechanism of the phytoplankton bloom. Cont. Shelf Res. 22, 1897–1908. http://dx.doi.org/ 10.1016/S0278-4343(02)00050-X. Gačić, M., Eusebi Borzelli, G.L., Civitarese, G., Cardin, V., Yari, S., 2010. Can internal processes sustain reversals of the ocean upper circulation? The Ionian Sea example. Geophys. Res. Lett. 37, L09608. http://dx.doi.org/10.1029/2010GL043216. Gačić, M., Civitarese, G., Eusebi Borzelli, G.L., Kovačević, V., Poulain, P.-M., Theocharis, A., Menna, M., Catucci, A., Zarokanellos, N., 2011. On the relationship between the decadal oscillations of the northern Ionian Sea and the salinity distributions in the eastern Mediterranean. J. Geophys. Res. 116, C12002. http: //dx.doi.org/10.1029/2011JC007280. Gačić, M., Schroeder, K., Civitarese, G., Cosoli, S., Vetrano, A., Eusebi Borzelli, G.L., 2013. Salinity in the Sicily Channel Corroborates the Role of the Adriatic–Ionian

Bimodal Oscillating System (BiOS) in Shaping the Decadal Variability of the Mediterranean Overturning Circulation. Ocean Sci. 9; , pp. 83–90. http://dx.doi. org/10.5194/os-9-83-2013. Gačić, M., Civitarese, G., Kovačević, V., Ursella, L., Bensi, M., Menna, M., Cardin, V., Poulain, P.-M., Cosoli, S., Notarstefano, G., Pizzi, C., 2014. Extreme winter 2012 in the Adriatic: an example of climatic effect on the BiOS rhythm. Ocean Sci., 513–522. http://dx.doi.org/10.5194/os-10-513-2014. Georgopoulos, D., Theocharis, A., Zodiatis, G., 1989. Intermediate water formation in the Cretan Sea (South Aegean Sea). Oceanol. Acta 12 (4), 353–359. Grinsted, A., Moore, J.C., Jevrejeva, S., 2004. Application of the cross wavelet transform and wavelet coherence to geophysical time series. Nonlinear Proc. Geophy. 11, 561–566. http://dx.doi.org/10.5194/npg-11-561-2004. Hainbucher, D., Rubino, A., Klein, B., 2006. Water mass characteristics in the deep layers of the western Ionian Basin observed during May 2003. Geophys. Res. Lett. 33, L05608. http://dx.doi.org/10.1029/2005GL025318. Krokos, G., Velaoras, D., Korres, G., Perivoliotis, L., Theocharis, A., 2014. On the continuous functioning of an internal mechanism that drives the Eastern Mediterranean thermohaline circulation: The recent activation of the Aegean Sea as a dense water source area. J. Mar. Syst. 129, 484489. http://dx.doi.org/ 10.1016/j.jmarsys.2013.10.002. Madec, G., Delecluse, P., Imbard, M., Levy, C., 1998. OPA 8.1 Ocean General Circulation Model Reference Manual No. 11. Institut Pierre-Simon Laplace, p. 91. Manca, B.B., Kovačević, V., Gačić, M., Viezzoli, D., 2002. Dense water formation in the Southern Adriatic Sea and spreading into the Ionian Sea in the period 1997– 1999. J. Mar. Syst. 33-34, 133154. http://dx.doi.org/10.1016/S0924-7963(02) 00056-8. Malanotte-Rizzoli, P., Manca, B.B., Ribera, M., D’ Alcala, A., Theocharis, S., Brenner, G., Budillon, Ozsoy, E., 1999. The Eastern Mediterranean in the 80s and in the 90s: the big transition in the intermediate and deep circulations. Dynam. Atmos. Oceans 29, 365–395. http://dx.doi.org/10.1016/S0377-0265(99)00011-1. Meccia, V.L., Borghini, M., Sparnocchia, S., 2015. Abyssal circulation and hydrographic conditions in the Western Ionian Sea during Spring–Summer 2007 and Autumn–Winter 2007–2008. Deep Sea Res. Pt I 104. http://dx.doi.org/10.1016/j. dsr.2015.06.007. Mihanović, H., Vilibić, I., Dunić, N., Šepić, J., 2015. Mapping of decadal middle Adriatic oceanographic variability and its relation to the BiOS regime. J. Geophys. Res. Oceans 120. http://dx.doi.org/10.1002/2015JC010725. Meyers, S.D., Kelly, B.G., O’Brien, J.J., 1993. An introduction to wavelet analysis in oceanography and meteorology: with application to the dispersion of Yanai Waves. Mon. Weather Rev. 121, 2858–2866. http://dx.doi.org/10.1175/ 1520-0493(1993)121 o 2858:AITWAI42.0.CO;2. Millot C., Taupier-Letage, I., 2005. Circulation in the Mediterranean Sea. In: The Mediterranean Sea, Hdb. Env. Chem. Part K, vol. 5, Springer-Verlag, Berlin, Heidelberg, pp. 29–66. http://dx.doi.org/10.1007/b107143. Mkhinini, N., Coimbra, A.L.S., Stegner, A., Arsouze, T., Taupier-Letage, I, I., Béranger, K., 2014. Long‐lived mesoscale eddies in the eastern Mediterranean Sea: Analysis of 20 years of AVISO geostrophic velocities. J. Geophys. Res. 119 (12), 8603–8626. http://dx.doi.org/10.1002/2014JC010176. Nittis, K., Lascaratos, A., Theocharis, A., 2003a. Dense water formation in the Aegean Sea: Numerical simulations during the Eastern Mediterranean Transient. J. Geophys. Res. 108, 8120. http://dx.doi.org/10.1029/2002JC001352. Nittis, K., Tziavos, C., Thanos, L., Drakopoulos, P., Cardin, V., Gačić, M., Petihakis, G., Basana, R., 2003b. The Mediterranean moored multi-sensor array (M3A): system development and initial results. Ann. Geophys. 21, 75–87. Özturgut, E., The Source and Spreading of the Levantine Interme-diate Water in the Eastern Mediteranean, Saclant ASW Research Center Memorandum SM-92, 1976, La Spezia, Italy, 45 pp. Pinardi, N., Zavatarelli, M., Adani, M., Coppini, G., Fratianni, C., Oddo, P., Simoncelli, S., Tonani, M., Lyubartsev, V., Dobricic, S., Bonaduce, A., 2015. Mediterranean Sea large-scale low-frequency ocean variability and water mass formation rates from 1987 to 2007: A retrospective analysis. Prog. Oceanogr. 132, 318–332. http://dx.doi.org/10.1016/j.pocean.2013.11.003. POEM Group, 1992. General circulation of the Eastern Mediterranean. Earth-Sci. Rev. 32 (4), 285–309. http://dx.doi.org/10.1016/0012-8252(92)90002-B. Potter, R.A., Lozier, M.S., 2004. On the warming and salinification of the Mediterranean outflow waters in the North Atlantic. Geophys. Res. Lett. 31, L01202. http://dx.doi.org/10.1029/2003GL018161. Poulain, P.-M., Menna, M., Mauri, E., 2012. Surface geostrophic circulation of the Mediterranean Sea derived from drifter and satellite altimeter data. J. Phys. Oceanogr. 42 (6). http://dx.doi.org/10.1175/JPO-D-11-0159.1. Roether, W., Manca, B.B., Klein, B., Bregant, D., Georgopoulos, D., Beitzel, V., Kovaĉević, V., Luchetta, A., 1996. Recent changes in Eastern Mediterranean deep waters. Science 271, 333–335. http://dx.doi.org/10.1126/science.271.5247.333. Roether, W., Klein, B., Beitzel, V., Manca, B.B., 1998. Property distributions and transient-tracer ages in Levantine Intermediate Water in the Eastern Mediterranean. J. Mar. Syst. 18, 71–87. http://dx.doi.org/10.1016/S0924-7963(98) 00006-2. Rubino, A., Hainbucher, D., 2007. A large abrupt change in the abyssal water masses of the eastern Mediterranean. Geophys. Res. Lett. 34, L23607. http://dx.doi.org/ 10.1029/2007GL031737. Schlitzer R., 2002. Ocean Data View, 〈http://www.awi-bremerhaven.de/GEO/ODV〉. Schroeder, K., Millot, C., Bengara, L., Ben Ismail, S., Bensi, M., Borghini, M., et al., 2013. Long-term monitoring programme of the hydrological variability in the Mediterranean Sea: a first overview of the HYDROCHANGES network. Ocean Sci. 9, 301–324. http://dx.doi.org/10.5194/os-9-301-2013. Supić, N., Vilibić, I., 2006. Dense water characteristics in the northern Adriatic in

M. Bensi et al. / Deep-Sea Research I 112 (2016) 1–13 the 1967–2000 interval with respect to surface fluxes and Po river discharge rates. Estuar. Coast. Shelf S. 66 (3–4), 580–593. http://dx.doi.org/10.1016/j. ecss.2005.11.003. Theocharis, A., Nittis, K., Kontoyiannis, H., Papageorgiou, E., Balopoulos, E., 1999. Climatic changes in the Aegean Sea influence the Eastern Mediterranean thermohaline circulation (1986–1997). Geophys. Res. Lett. 26 (11), 1617–1620. http://dx.doi.org/10.1029/1999GL900320. Theocharis, A., Klein, B., Nittis, K., Roether, W., 2002. Evolution and status of the Eastern Mediterranean transient (1997–1999). J. Mar. Syst. 33–34, 91–116. http: //dx.doi.org/10.1016/S0924-7963(02)00054-4. Theocharis, A., Krokos, G., Velaoras, D., Korres, G., 2014. An Internal Mechanism Driving the Alternation of the Eastern Mediterranean Dense/Deep Water Sources, In “The Mediterranean Sea: Temporal Variability and Spatial Patterns”. Borzelli, G.L.E., et al. (Eds.), AGU Geophys. Monogr. Ser 202. John Wiley, Oxford, U. K, pp. 113–137. http://dx.doi.org/10.1002/9781118847572.ch8. Torrence, C., Compo, G.P., 1998. A practical guide to wavelet analysis. Bull. Am. Meteor. Soc. 79, 61–78. Ursella, L., Kovačević, V., Gačić, M., 2011. Footprints of mesoscale eddy passages in the strait of Otranto (Adriatic Sea). J. Geophys. Res. 116, C04005. http://dx.doi.

13

org/10.1029/2010JC006633. Velaoras, D., Kassis, D., Perivoliotis, L., Pagonis, P., Hondronasios, A., Nittis, K., 2013. Temperature and salinity variability in the Greek Seas based on POSEIDON stations time-series: preliminary results. Medit. Mar. Sci. 14 (3), 5–18. http://dx. doi.org/10.12681/mms.446. Velaoras, D., Krokos, G., Nittis, K., Theocharis, A., 2014. Dense intermediate water outflow from the Cretan Sea: A salinity driven, recurrent phenomenon, connected to thermohaline circulation. J. Geophys. Res. Oceans, 119. http://dx.doi. org/10.1002/2014JC009937. Vilibić, I., Orlić, M., 2002. Adriatic water masses, their Rates of formation and transport through the Otranto Strait. Deep Sea Res., Pt I 49 (8), 1321–1340. http: //dx.doi.org/10.1016/S0967-0637(02)00028-6. Vilibić, I., 2003. An analysis of dense water production on the North Adriatic shelf. Estuar. Coast. Shelf Sci. 56 (3–4), 697–707. http://dx.doi.org/10.1016/ S0272-7714(02)00277-9, ISSN: 0272-7714. Zervakis, V., Georgopoulos, D., Drakopoulos, P.G., 2000. The role of the North Aegean in triggering the recent Eastern Mediterranean climatic changes. J. Geophys. Res. 105 (C11), 26103–26116. http://dx.doi.org/10.1029/2000JC900131.