Interactive comment on “ Extreme winter 2012 in the Adriatic : an example of climatic effect on the BiOS rhythm ”

This manuscript relates a reversal of the circulation in the northern Ionian Sea to an especially pronounced dense-water formation in the Adriatic in 2012, which the authors judge as showing a “remarkable sensitivity of Adriatic-Ionian BiOS” (which the authors have delt with in various previous papers) “on climate forcing”. Absolute Dynamic Topography (ADT) and the related surface geostrophic velocities, Argo profiler and altimetry data, and air-sea heat fluxes were taken into account; subsurface geostrophic currents were obtained from CTD data using the float velocities for referencing. The manuscript is well organized, the figures are helpful to assist the argumentation, the results are convincing, and the literature is adequately covered. However, I note some items that the authors must address.


Introduction
The Mediterranean Overturning Circulation (MedOC) is characterized by a number of processes and features (dense water formation, eddies, gyres, meanders and filaments) occurring in the world ocean as well (Bethoux et al., 1990) but having much smaller spatial and often temporal scales (see the map of the Mediterranean Sea in Fig. 1).Contrary to the thermohaline circulation (THC) in the world ocean, where differences in the heat content play a major role, the salinity contrast between the inflowing Atlantic water (AW) and the Levantine surface and intermediate waters (Robinson et al., 2001) drives and largely maintains the Mediterranean basinwide thermohaline circulation (MTHC).The high-salinity waters originating in the Levantine and, in general, in the eastern Mediterranean (EM) are related to the excess of evaporation over precipitation.The MTHC, due to the east-west climatic differences, is zonally oriented contrary to the THC which is essentially a meridional flow.In addition, in the world ocean the entire or a larger part of the water column is affected by the north/south flows.In contrast, the MTHC is mainly limited to the surface and intermediate layers due to two factors: the first one is that the Levantine intermediate water (LIW) in the EM is not dense enough to sink to larger depths and thus spreads in the intermediate layer (∼ 300 m); the second factor is the bathymetry in the Sicily and Gibraltar straits (with depths < 500 m).Deeper Mediterranean waters are also involved in the water exchange directly, or via upwelling and mixing into the intermediate layer.In addition to the thermohaline circulation, the wind-driven flow represents an important component of the MedOC.Winds are highly variable on seasonal and interannual scales as well as spatially and thus their influence on MedOC is not fully quantified.Generally, from available wind data, it was shown that in the northern portion of the Mediterranean Sea the wind curl is prevalently cyclonic while in its southern part is anticyclonic imparting positive and negative vorticity to the oceanic flow, respectively (Pinardi et al., 2013).
In contrast to the world ocean where vertical convection is an integral part of the Meridional Overturning Circulation (MOC), in the Mediterranean Sea (MS) the ventilation of the deep portion of the water column only partially interacts with the MedOC (Lascaratos et al., 1999).The winter convection and deep-water formation taking place in the northern parts of the MS, i.e. in the Gulf of Lion in the western Mediterranean (WM) and in the Adriatic/Aegean in the EM, make up part of the closed meridional thermally driven circulation cells.The coupling with the MedOC is achieved via LIW whose presence in the dense-water formation sites (Gulf of Lion and Adriatic/Aegean Sea) represents a key ingredient in facilitating the vertical convection and in determining the thermohaline properties of the deep water formed.Therefore, in the MS two types of circulation cells co-exist: the zonal MTHC, driven mainly by the east-west salinity gradient interacting with two secondary cells controlled by the northsouth temperature gradient, where the driving mechanisms are the winter air-sea heat losses and vertical convection.
Decadal variability of the MedOC is rather well assessed.The idea of the Mediterranean circulation being in a steady state was abandoned when in the early 1990s the eastern Mediterranean transient (EMT) was discovered.The phenomenon manifested essentially in the Aegean/Cretan Sea substituting the southern Adriatic as the dense water formation site (Roether et al., 1996).This abrupt change was attributed in a number of numerical studies to a meteorological forcing (Beuvier et al., 2010) and to the Levantine circulation changes bringing highly saline waters into the Cretan Sea.Salinity increase in the Levantine was also explained in some experimental studies in terms of a blocking of the LIW outflow through the Cretan Passage (Malanotte-Rizzoli et al., 1999).Subsequently, deep winter convection and bottom water formation in the Cretan Sea took place following severe winter climatic conditions.
Recently, cyclical occurrences of the high salinity in the Levantine and the EMT preconditioning have been explained in terms of the feedback mechanism called the Adriatic-Ionian bimodal oscillating system (BiOS), i.e. the decadal reversals of the Ionian upper-layer basin-wide circulation from cyclonic to anticyclonic and vice versa (Gačić et al., 2011).The Mid-Ionian Jet, which brings AW into the Levantine basin, is reinforced or weakened by the Ionian cyclonic or anticyclonic circulation, respectively.This then results in a varying intensity of the AW advection towards the Levantine and consequently in a varying dilution of the Levantine surface waters.Considering that the LIW is formed in the Levantine, in the area of Rhodes Gyre, obviously the LIW salinity will change as a function of the intensity of the Levantine surface water dilution by the AW.
During the Ionian anticyclonic mode the AW flow is mainly deflected northeastward affecting the northern Ionian and southern Adriatic.In that situation the flow of the AW towards the Levantine basin is reduced, the Levantine surface waters become saltier and the same applies to the LIW.The Ionian anticyclonic circulation mode is thus the preconditioning factor for EMT-like phenomena (Gačić et al., 2011).Possible occurrence of the EMT would then take place only if the air-sea heat losses in the Aegean Sea were strong enough to produce deep convection.
The Adriatic Sea as a dense water formation site is more prone to winter convection when the Ionian circulation is cyclonic, bringing there the salty waters of Levantine origin.In the opposite circulation pattern, under the influence of the AW, the vertical stability of the water column in the Adriatic hampers the vertical convection.In addition, the water formed is of a lower/higher density due to the freshening/salting of the upper part of the Adriatic water column.From these considerations, one can explain why the Levantine and Adriatic salinities are out of phase and why the Aegean and Adriatic were substituting each other as a source of the eastern Mediterranean deep water (EMDW) as it happened during the EMT.However, we have to take into consideration that the winter convection in the Adriatic depends on the interplay between air-sea heat fluxes and the buoyancy and thus winter meteorological conditions can modify both the efficiency of the Adriatic as a source of the EMDW and the Adriatic Deep Water (AdDW) density.
From the T -S data of the last 60 years (Gačić et al., 2013) it is clear that e.g. the salinity minimum in the North Ionian, presumably associated with the anticyclonic phase of North Ionian Gyre (NIG) occurred cyclically on timescales of between 4 and 11 years.From altimetric data, the occurrences of two anticyclonic phases in the Ionian (1987-1997 and 2006-2010) have been identified.In addition, two cyclonic phases have been documented, the first one in the period 1998-2005, and the second one which started in 2011.However, in 2012 an unexpected reversal from the cyclonic to the anticyclonic mode took place.The aim of this study is to analyse this "premature" inversion of the NIG, trying to explain it in terms of the extremely severe climatic conditions in the Adriatic during the winter of 2012, thus relating it to the spreading of the very cold and dense AdDW formed during that winter.

Data and methods
A fleet of 17 Argo profiling floats is used for this work.They operated between 2011 and 2012 in the Ionian Sea.They were deployed in the framework of the Argo programme in the Mediterranean Sea (MedArgo, see Poulain et al., 2007).The profilers were equipped with Sea-Bird CTD (conductivity-temperature-depth) sensors (models SBE 41 or 41CP) with accuracies of ±0.002 • C, ±0.005 and ±2.4 dbar for temperature (T ), salinity (S) and pressure (P ), respectively.The sampling depth, defined as pressure, was variable: in the upper layer [0, 100 dbar] it is set to 5 dbar; 10 dbar in the layer 100-700 and 50 dbar in the depth interval 700-2000 dbar including wherever possible the near bottom values.The Argo floats were programmed with a cycle length between 2 and 10 days, a drifting depth of 350 or 1000 m and a maximal profiling depth between 1000 and 2000 m.The data were processed and quality controlled using a basic set of tests at the Argo Data Assembly Centre (DAC).Prior to the data quality control, some notoriously spurious values were eliminated.A delayed-mode quality control of P , T and S data was then applied in accordance to the Argo Quality Control Manual (2013), Version 2.8, 3 January 2013, Argo Data Management (available at http://www.argodatamgt.org/content/download/15699/102401/file/argo-quality-control-manual-version2.8.pdf) and in particular the Owens-Wong method was adopted (Owens and Wong, 2009) to check the S data.The Argo float S profiles were also qualitatively compared to a reference data set (see details in Notarstefano and Poulain, 2008); this comparison detected no drift of the conductivity sensor for the majority of the floats.A significant negative offset (0.0151) in the S data was evidenced only for Argo float 6900952 (Notarstefano and Poulain, 2013), which was later corrected.After removing the spikes, the potential drifts or offsets of the T and conductivity sensors were found to be smaller than the natural variability of the water column and hence no delayed mode correction was deemed necessary.
The altimetry data used for this study are gridded (1/8 • Mercator projection grid) Ssalto/Duacs weekly, multimission, delayed time (quality controlled) products from AVISO (SSALTO/DUACS users handbook 2014, available at http://www.aviso.altimetry.fr/fileadmin/documents/data/tools/hdbk_duacs.pdf).Among the available versions of the delayed time products we have selected the reference data series, based on two satellites (Jason-2/Envisat or Jason-1/Envisat or Topex/Poseidon/ERS) with the same ground track.These data series were homogeneous all along the available time span, thanks to a stable sampling.Absolute dynamic topography (ADT) and corresponding absolute geostrophic velocity (AGV) data were downloaded.The ADT is the sum of sea level anomaly and synthetic mean dynamic topography (SMDT), estimated by Rio et al. (2007) over the 1993-1999 period.The error of the SMDT in the study area is of the order of 1 cm.
Annual and monthly averages of the ADT and AGV in the Ionian Sea were estimated over the 2008-2012 period.The AGV were sub-sampled in non-overlapping bins of 0.25 • × 0.25 • .Weekly zonal geostrophic components were spatially averaged in an area of the northern Ionian (see Fig. 2a and b).
Vertical profiles of horizontal geostrophic currents were derived at discrete depth levels by vertically integrating the thermal-wind balance equation.Interpolated (1 m vertical resolution) vertical profiles of potential density were derived from the available T -S profiles for the period June-November 2012.The required horizontal gradients of potential density were derived by locally fitting horizontal density planes to each point coinciding with the ADT grid using density measurements from nearby locations.The 350 m level, where float-derived velocity field was available, was chosen as reference level to obtain absolute velocity profiles.The absolute velocities were then compared with the surface geostrophic flow.
Heat fluxes were calculated at the air-sea interface every 6 h (00:00, 06:00, 12:00, 18:00 UTC) from the European Centre for Medium Range Weather Forecasts, Reading, UK (ECMWF) Operational Analysis Data Set using a 0.25 • interpolated latitude/longitude Gaussian grid (Cardin and Gačić, 2003).From the 6-hourly values, instantaneous flux components for long-wave (May, 1986), sensible and latent heat flux (Kondo, 1975) were obtained.These were subsequently averaged to obtain daily mean values, while the short-wave radiation was determined as an integral from sunrise to sunset, according to Schiano (1996).Adding its value to the mean of the other three components, the daily net heat flux was determined at each grid node.More information on formulas applied and coefficients can be found in Cardin and Gačić (2003).Mean area winter heat losses were then calculated for the period spanning from 1 December to 31 March using a grid of 157 nodes for the southern Adriatic basin for three consecutive winters i.e. 2009-2010, 2010-2011 and 2011-2012.Correlation between the ADT and the depth of the isopycnals 29.00, 29.10 and 29.18 kg m −3 obtained from in situ CTD casts were computed for the two periods June-November 2011 and June-November 2012.Correlation coefficients were calculated between the 6-month average ADT values and isopycnal depths at each CTD position.
Due to the paucity of float data for the period June-November 2011, the float data set were merged with CTD profiles from the POS414 cruise (R/V Poseidon), carried out in June 2011 in the eastern Mediterranean (Tanhua et al., 2013).All profiles from that cruise were sampled using a CTD equipped with a SBE19plus sensor, with overall accuracies of ±0.002 • C, ±0.003 for T and S, respectively.Potential density anomalies (reference pressure equaling zero) were computed with MATLAB using TEOS-10 thermodynamic equations of seawater (http:/www.teos-10.org).

Data analysis and discussion
In order to illustrate decadal inversions of the Ionian basinwide circulation for the last two decades, we present time series (Fig. 3) of the zonal component of the surface geostrophic currents computed for the northern portion of the Ionian (see Fig. 2a and b for the position of the area where the average geostrophic currents were calculated).We present both low-pass monthly time series filtered by the 13month moving average and the unfiltered data.From the latter time series, clear seasonal variability emerged.Positive zonal geostrophic component (eastward) represents the anticyclonic circulation while the negative one is associated with the cyclonic Ionian mode.Generally, the zonal component of the geostrophic current varies primarily at the decadal timescale.The seasonal basin-wide circulation reversals occurred only in the transitional phase between the two subsequent decadal circulation patterns like in the late 1990s, around 2006 and around 2009.From the presented time series there is evidence suggesting that the BiOS anticyclonic mode which started in late 1980s was of a longer duration than the anticyclonic phase starting in 2006.The same is very likely valid for the cyclonic modes.This could be explained in terms of the EMT, when much larger quantities of the Aegean dense waters (about 3 Sv (1 Sv = 10 6 m 3 s −1 ) average outflow between mid-1992 and late 1994 according to Roether et al., 2007) than the "normal" AdDW volume (0.3 Sv; Lascaratos, 1993) flooded the Ionian abyssal area.We will be analysing the mechanism potentially responsible for the 2012 "premature" reversal, documented from both yearly and monthly altimetric maps (Figs. 4 and 5).The low-pass zonal geostrophic flow in Fig. 3 shows only the weakening of the westward flow in 2012 and not the complete reversal.In fact some inconsistencies in timing of the flow reversal obtained from altimetric maps and from the zonal geostrophic flow are mainly due to the fact that the geostrophic currents are related to the averaging area often affected also by smaller-scale motions.Altimetric data and surface geostrophic currents (annual mean maps) from 2008 to 2012 show the second part of the anticyclonic phase in the late 2000s and the breaking up of the basin-wide anticyclonic meander into several mesoscale cyclonic and anticyclonic features.The complete reversal from the anticyclonic to the cyclonic mode took place in 2011 when the large anticyclonic meander retreated below the 36 • N latitude (Fig. 4).Subsequently, the "premature" reversal from the cyclonic to anticyclonic circulation took place in 2012 and manifested as a northward protrusion of the anticyclonic meander reaching 38 • N latitude.From the monthly mean altimetric maps for 2012 (Fig. 5), we can identify with more precision the reversal time.It is evident that the reversal of the NIG from cyclonic to anticyclonic initiated with the break-up of the sub-basin scale cyclone into two mesoscale cyclones in May 2012.Between these two cyclones, the establishing and strengthening of the anticyclonic meander became more and more evident protruding northward.This latter circulation pattern reached its maximum in the period August-November.
In December 2012, the anticyclonic meander started to withdraw leaving again room to the cyclonic circulation and very likely to an eventual re-establishment of the cyclonic BiOS phase which is evident in the altimetric maps for January, February and March 2013 (Fig. 5).To a certain extent, this change of pattern can be associated to intra-annual variability but as shown, the seasonal signal very rarely can invert the basin-wide circulation.
We will now try to explain the "premature" Ionian basinwide circulation reversal, which occurred in the period May-November in terms of the resulting effect of the cold and dense AdDW formed during the extremely cold 2012 winter.According to recent papers (Bensi et al., 2013;Mihanović et al., 2013;Vilibić et al., 2013) the winter of 2012 was very harsh in the northern Adriatic resulting in the formation of extremely dense bottom water in the shelf area.Density anomaly recorded in the northernmost end of the Adriatic attained 30.59 kg m −3 , the value reached only twice in the last 100 years.Mihanović et al. (2013) showed that the average annual transport of the northern Adriatic dense water reached a value that was approximately half the average dense water formation rate in the southern Adriatic.Therefore, a contribution of the northern Adriatic dense water to the total volume of the AdDW was probably much larger than about 10 % as estimated by Vilibić and Orlić (2002).Apart from the effect of the northern Adriatic dense water on the outflowing AdDW, the heat fluxes in the openocean convection area in the southern Adriatic were also very strong.More precisely, the comparison of the surface heat losses between three consecutive (2010, 2011 and 2012) winters (December-March period) for the southern Adriatic confirmed that they were strongest in winter 2012.Six episodes with the surface heat losses stronger than 400 W m −2 occurred in January-February of 2012, while in other two years such episodes were practically absent or less frequent (Fig. 6).Subsequently, cold and very dense water formed locally in the southern Adriatic with an important contribution of the northern Adriatic dense water spread southward along the western flank of the Ionian.It generated the pressure gradient force oriented from the coast toward the open sea, and consequently the deeper-layer southward geostrophic flow.Due to the presence of the dense water along the western Ionian flank, the sea level along borders became lower than in the centre.Therefore, the southward motion of the colder and denser Adriatic water along the western Ionian flanks generated the northward surface geostrophic flow.This is In order to compare the sea level structure and the isopycnal depth, we calculated the linear correlation between the altimetric data and the isopycnal depths for 2011 and 2012 (Fig. 7).If such a relation exists, it would mean that the Ionian behaved as a two-layer basin, i.e. that the upper-layer cyclonic or anticyclonic flow was in the opposite direction with respect to a deep layer circulation as hypothesized earlier in this paper.In other words, this would suggest that the surface geostrophic circulation was related to the deeper layer flow and that the sea level pattern was representative of the horizontal distribution of isopycnal depths.We have chosen 29.00, 29.10 and 29.18 kg m −3 isopycnal depths for the comparison with the sea surface height.The average depth of the 29.00 kg m −3 isopycnal was about 200 m; the 29.10 kg m −3 isopycnal corresponds to the LIW and occupied depths of around 300 m.The 29.18 kg m −3 isopycnal can be considered as the lower limit of the AdDW density in the Ionian located around 1100 m depth.We have compared semi-annual mean sea level data (June-November) with the isopycnal depth at a given point obtained from the floats.We have chosen the 6-month period when rather homogeneous float data coverage was present in both 2011 and 2012.In addition, we expect that in the chosen period June-November the newly formed AdDW very likely reached the Ionian.Considering all the float data within the 6-month period, we hypothesized small intra-annual variations in the density field.This was a reasonable assumption for the portion of the water column below the seasonal pycnocline.However, some noise over the entire depth range may derive from the mesoscale variability affecting float trajectories.In 2012, the regression was close to linear for the first two isopycnals with the decreasing values of the correlation coefficient with depth.On the other hand, the data points in the sea level −29.18 kg m −3 isopycnal diagram were highly dispersed.During 2011 the  float data coverage in the Ionian was poorer than in 2012 (Fig. 2) and thus we combined data from the oceanographic campaign carried out in June by R/V Poseidon and the float data.In the upper part of the water column (isopycnals 29.00 and 29.10 kg m −3 ) there were no important differences between 2011 and 2012 in the relationship between the isopycnal depth and the free surface; both years showed a statistically significant linear correlation.The 29.18 kg m −3 isopycnal depth showed higher correlation with the free surface for 2011 than for 2012.This difference can be explained by the fact that the spreading of the very dense AdDW in 2012 introduced spatial heterogeneity in the density field in the deeper part of the water column, which then resulted in the lower correlation between the free surface and the 29.18 kg m − 3 isopycnal depth.In fact, as we will show later, a current reversal took place somewhere below the 750 m depth probably close to the depth of 29.18 kg m −3 and this could explain the weak correlation between the free surface and the 29.18 isopycnal.The statistically significant correlation confirmed the interdependence of the surface and deeper layer circulation showing that from the surface geostrophic current pattern it is possible to reconstruct the deeper layer flow.In the specific case of 2012, the massive inflow of the dense AdDW generated the reversal of the Ionian surface circulation which, as shown from the linear regression between the sea surface height and isopycnal depth, was concomitant with the deeper layer circulation inversion.It is also important to stress that data points, independently of different floats and their positions were scattered along the same straight line.
Next, we analyse in more detail the temporal variations of the water density from the float measurements.We looked for evidence on the evolution of the horizontal density gradient in the deeper portion of the water column between the centre of the Ionian and its flanks, which resulted in a reversal of the surface pressure gradients and of the geostrophic flow.We will estimate the timing of the change of the horizontal density gradient due to the spreading of the highly dense AdDW and compare it with the sea level gradient reversal.The series of profiles (Fig. 8a) revealed the density increase in the northwestern Ionian starting in June and related to the arrival of the newly formed AdDW.Thus, we should expect that the reversal of the density and surface pressure gradients first took place in that area.The temporal evolution of the average density in the deep portion of the profile (1000-1200 m layer) obtained from floats in the centre of the basin was compared with the density variations along the Ionian western flanks.For that purpose, we used data from two floats, which were mainly moving near the flanks -one in the northern part just downstream of the Strait of Otranto and the other one further southward along the Calabrian coast (Fig. 9a).Then the density data were compared with those in the centre of the basin obtained from one float whose motion was mainly limited to the central area of the Ionian Sea,  representative of the density variations in the basin interior.The prominent density increase in the northwestern Ionian occurred from May through September reaching its maximum in August, a period when high-density AdDW reached the area as revealed from the density profiles (Fig. 8a).The density in the centre of the basin remained unaltered, hence during the AdDW spreading period a building up of a density gradient between the basin flank and its centre was taking place.Subsequently, the signal of the relatively high density reached the Calabrian coast around November (Fig. 9b), with an average speed of about 5 cm s −1 .It was rather weak probably because we compared the average density in the layer 1000-1200 m for all three sets of data.At the beginning of its spreading in the Ionian, the AdDW occupied mainly the layer around the 1000 m depth and thus the signal was very strong.This was particularly evident from the density profiles at the float positions indicated by the triangle and the cross in Fig. 8, which were located on the seaward side of the 1000 m isobath.The dense water vein deepened by about 300 m in the 10-day interval of 29 July to 9 August.Due to its further deepening downstream, the AdDW was not so evident in the layer between 1000 and 1200 m.
We compare the evolution of the sea level differences (between the northwestern area centred at the location where the AdDW signal in the density field was strongly present, as evidenced from the float profiles in Fig. 8a, and the centre of the basin) and the average density variations (Fig. 9).This comparison was done in order to reveal whether the sea level gradient and the surface geostrophic circulation reversals were concomitant with the density increase due to the arrival of the dense Adriatic waters.We considered the sea level differences between the northern part of the Ionian and its cen- variability due to heating and cooling.On the other hand, the noise related to the presence of mesoscale eddies was very pronounced.Thus, we apply a low-pass filter to the time series of the weekly sea level differences.The low-pass curve of the sea level differences showed clearly the lowering of the sea level in the northern area with respect to the central zone of the Ionian basin, related to the increase of the 1000-1200 m average density starting in May 2012.The sea level differences remained negative i.e. the sea level along the basin boundaries remained lower than in the centre until the beginning of 2013, although the vertically averaged density after September reached the same values as those at the open sea.This can be explained by the fact that the float in September left the area affected by the AdDW (see Fig. 9a) and thus reached the zone having densities rather close to those of the centre of the Ionian.Further, with the view to investigate the vertical flow structure and obtain the boundary between the upper and lower layer, we estimated the absolute current field.We computed the geostrophic shear from the float density field and added it to the measured Lagrangian velocity at the 350 m parking depth of the ARGO floats.Although floats were programmed to reach the maximum depth of 1000 or 2000 m we reconstructed the absolute velocity profile from the surface to the 750 m depth (Fig. 10) in order to maximize the spatial coverage of the Ionian.Large discrepancies at some locations between the surface geostrophic flow calculated from the altimetric data and absolute currents at the sea surface obtained from floats can be explained in terms of the large interval we chose for the data analysis (June-November) when seasonal variability in the surface layer is rather important.From Fig. 10 it is evident that the entire water column from the surface to the 750 m behaved mainly as a single layer, meaning that the inversion took place below that depth.Only at one location, which was positioned in the northwestern part of the basin and probably invaded by the AdDW, was the twolayer structure evident, with the velocity at 750 m being in the opposite direction with respect to the surface one.Due to the scarcity of data, their uneven spatial distribution and the method for the horizontal density gradient calculations, we were not able to cover in a more detailed manner the area in the close vicinity of the coast where we could expect increased horizontal density gradients and large values of the vertical shear caused by the presence of the AdDW.Another feature arising from the geostrophic shear calculations is that mesoscale structures (like e.g.large anticyclonic gyre in the centre of the Ionian) at the open sea were rather deep, extending from the surface down to at least 750 m depth.

Conclusions
Here we have documented from altimeter data that the NIG circulation, as a part of the BiOS decadal variability, passed from the anticyclonic circulation to the basin-wide cyclonic mode in 2011.In 2012, however, unexpectedly the NIG became anticyclonic interrupting the most recent cyclonic BiOS phase.We have shown that the winter of 2012 in the southern Adriatic was extremely severe, as already evidenced in the literature for its northern part, and we looked for a possible link between strong air-sea heat losses and the Ionian circulation inversion.We set up the hypothesis that extreme winter air-sea heat losses in both the northern and southern Adriatic resulted in the formation of the cold and highly dense Adriatic bottom waters, spreading into the Ionian, which inverted the bottom pressure gradient and generated the NIG reversal.We showed first that the sea level anomalies were linearly related to the depth of the three chosen isopycnals (29.0, 29.10, 29.18 kg m −3 ) which confirmed that the Ionian behaved as a two-layer fluid.Yet, from the vertical geostrophic shear and the absolute geostrophic velocity we were not able to determine the basin-wide distribution of the no-motion level because the float profiles often do not reach the maximum nominal sampling depth (2000 m).There were, however, indications that near the northwestern flank the no-motion level or the boundary between the upper and lower layers was situated at depth shallower than 750 m.Analysis of the profiling float data also revealed evidence of the temporal evolution of the spreading of the extremely dense Adriatic waters along the northwestern Ionian flanks, showing that it started in May 2012 and affected the density field until September.This was also the period of the major horizontal density gradient between the basin interior where the density did not change, and its western flanks.Computations of the sea surface differences between the north Ionian flank and the basin interior, and implicitly the geostrophic velocity, confirmed that in concomitance with the arrival of the Adriatic dense waters, the seaward sea surface pressure gradient was set up.With the decrease of the horizontal deep density gradients towards the end of 2012 the surface geostrophic flow, as part of the anticyclonic NIG, weakened.Finally, we documented from altimetric data that the NIG circulation returned to a cyclonic one at the beginning of the 2013.
In summary, here we evidenced that the BiOS mechanism is highly sensitive to the climatic conditions, more specifically to the air-sea heat fluxes in the Adriatic dense water formation areas, which can affect its cycle.In fact, we have presented here supporting elements to the thesis raised by Mihanović et al. (2013), that future climate variability will have an important effect in changing the BiOS temporal scale if an increase of cold air outbreaks in the 21st century takes place as predicted.This study also implicitly confirmed that the NIG reversal can be explained solely in terms of the variations of thermohaline properties of the AdDW.

Figure 2 .
Figure 2. Spatial distribution of CTD profiles for 2011 and for 2012.Different colours represent different floats.Red dots in the left panel indicate the CTD stations referred to the POS414 cruise.Black rectangles denote areas where the average sea level heights were calculated.The zonal component of the surface geostrophic flow (shown in Fig. 3) was computed for the northern rectangle.

Figure 3 .
Figure 3. Zonal component of the surface geostrophic velocity computed in the area denoted by the northern rectangle (see Fig. 2).Continuous line represents the time series smoothed by the 13month moving average while the dashed line represents unfiltered data.

Figure 4 .
Figure 4. Mean annual maps of the sea surface height for the period 2008-2012.

Figure 5 .
Figure 5. Mean monthly altimetric maps for the period January 2012-June 2013.

Figure 6 .
Figure 6.Daily values of the surface air-sea heat losses in South Adriatic for three consecutive winters (2010 -black line, 2011blue line and 2012 -red line).

Figure 7 .
Figure 7. Scatter plots of isopycnal depths versus sea surface height for the years 2011 (upper panels) and 2012 (lower panels), for density anomaly values 29.0 (a and d), 29.1 (b and e) and 29.18 kg m −3 (c and f).Colours of data points refer to different floats and cruise, as in Fig. 2.

Figure 8 .
Figure 8. Vertical density profiles from floats (a) in the northwestern Ionian area (the time of sampling is denoted next to the curve).Sampling locations are indicated on the map (b).
Figure 9. (a) Float sampling locations within the three selected zones and (b) the time series of the mean density within the depth interval 1000-1200 m, averaged over all the available profiles within each zone.The colour of curves corresponds to the selected zones.The sea level differences (weekly data, thin grey line, low-passed data thick grey line) in (b) are computed between the two rectangles (northern rectangle minus the southern one, see Fig. 2 for their positions).The low-pass procedure consisted of a zero-phase forward and backward digital infinite impulse response filtering.A nine-point symmetric Hanning window has been chosen as a weighting function.

Figure 10 .
Figure 10.Absolute velocities calculated from the geostrophic shear using the 350 m Lagrangian measurements as a reference velocity.