Impacts of atmospheric modes of variability on Mediterranean Sea surface heat exchange

Transcription

1 JOURNAL OF GEOPHYSICAL RESEARCH, VOL. 116,, doi: /2010jc006685, 2011 Impacts of atmospheric modes of variability on Mediterranean Sea surface heat exchange Simon A. Josey, 1 Samuel Somot, 2 and Mikis Tsimplis 1 Received 28 September 2010; revised 24 November 2010; accepted 8 December 2010; published 22 February [1] The impacts of variations in the state of the first four modes of atmospheric variability in the North Atlantic/Europe region on air sea heat exchange in the Mediterranean Sea are considered. Observation based indices of these modes from the NOAA Climate Prediction Centre are used together with two reanalysis (NCEP/NCAR and ARPERA) surface flux data sets for the period to determine their relative influence on the mean heat budget of the full Mediterranean basin and the eastern and western subbasins. The modes considered are the North Atlantic Oscillation (NAO), East Atlantic pattern (EA), Scandinavian pattern (SCAN), and East Atlantic/ West Russian pattern (EA/WR). Similar results are obtained with both NCEP/NCAR and ARPERA. In each case, winter anomalies dominate the annual mean heat budget and the leading mode, the NAO, has a surprisingly small impact on the full basin winter mean heat budget, <5 Wm 2. In contrast, the EA mode has a major effect, of order 25 Wm 2, with similar impacts on both the eastern and western Mediterranean. The SCAN mode has the weakest influence of those considered. The EA/WR mode plays a significant role but, in contrast to the EA mode, it generates a dipole in the heat exchange with an approximately equal and opposite signal of about 15 Wm 2 on the eastern and western subbasins. A particularly strong impact in the Aegean Sea is observed for the EA/WR mode and this is discussed in the context of episodic deep water formation in this region. Citation: Josey, S. A., S. Somot, and M. Tsimplis (2011), Impacts of atmospheric modes of variability on Mediterranean Sea surface heat exchange, J. Geophys. Res., 116,, doi: /2010jc Introduction [2] The Mediterranean Sea is an active region of dense water formation which has exhibited significant variations in the location and intensity of the main convection sites in recent decades. The Gulf of Lions region in the northwestern Mediterranean has long been recognized as a site of deep convection and has experienced relatively recent major new deep water production in winters and [e.g., Schroeder et al., 2008, 2010]. In the eastern Mediterranean, the Adriatic Sea was long thought to be the sole region in which deep water formation occurred [Robinson et al., 1992]. However, that view point was overturned in the mid 1990s as a major convection event, widely termed the Eastern Mediterranean Transient (EMT), was observed to have taken place in the Aegean Sea between repeat hydrographic surveys in 1987 and 1995 [Roether et al., 1996]. The EMT has had significant long term consequences for the eastern Mediterranean and its evolution has now been traced over more than a decade [Roether et al., 2007]. 1 Ocean Observing and Climate Group, National Oceanography Centre, Southampton, UK. 2 Centre National de Recherches Météorologiques, Météo France, CNRS, CNRM GAME, Toulouse, France. Copyright 2011 by the American Geophysical Union /11/2010JC [3] Dense water formation is driven by a variety of factors. These include surface buoyancy loss through heat and freshwater exchanges, wind forced preconditioning of surface layer density through doming of isopycnals and advective changes in density via variations in the near surface temperature and salinity. Each of these processes potentially plays a significant role in driving events such as the EMT. However, disentangling their relative contributions has proved a complex task which has benefited from the use of ocean model simulations [e.g., Samuel et al., 1999; Beuvier et al., 2010; Herrmann et al., 2010]. Recently, Beuvier et al. [2010] find using an eddy permitting model driven by realistic interannual high resolution air sea fluxes that the major triggering elements for the EMT are the atmospheric fluxes and winds occurring in winters and In the western Mediterranean, Herrmann et al. [2010] find that both air sea fluxes and ocean initial conditions have an impact on the deep water formation event in [4] Extreme winter surface heat loss has been linked to both the EMT [Josey, 2003], deep Levantine convection in the Rhodes gyre [Sur et al., 1993] and the recent deep water production in the western Mediterranean from 2004 to 06 [Schroeder et al., 2010; Herrmann et al., 2010]. Furthermore, the enhanced heat loss may be favored by variations in the state of the major North Atlantic modes of atmospheric variability. In particular, the recent western Mediterranean 1of15

2 event has been linked by Schroeder et al. [2010] to the second mode of variability, termed the East Atlantic pattern. However, the relative contributions of the different atmospheric modes to the heat budgets of both the eastern and western Mediterranean have yet to be determined. As a consequence, the relationship between dense water formation in both halves of the Mediterranean basin and larger atmospheric scales remains unclear. Here, we resolve this uncertainty through a study of the impacts of the four major modes of atmospheric variability on Mediterranean Sea heat loss using fields from both unmodified and downscaled atmospheric reanalysis data sets. [5] For our analysis, we employ the NOAA Climate Prediction Centre (CPC) definitions of the leading atmospheric modes that have been widely employed in other studies. The CPC modes of relevance for interannual atmospheric variability over the Mediterranean region are the North Atlantic Oscillation (NAO), the East Atlantic pattern (EA), the Scandinavian pattern (SCAN) and the East Atlantic/West Russian pattern (EA/WR). The NAO has been the subject of much research [e.g., Hurrell et al., 2003] and its impacts on the Mediterranean can be seen in various fields (e.g., strength of the Corsica channel transport, [Vignudelli et al., 1999; Somot, 2005]; sea level [Tsimplis and Josey, 2001]). The EA pattern is less well known but its importance is becoming recognized, for example though its influence on the western Mediterranean cited above and at larger scales in controlling the freshwater flux to the northeast Atlantic [Josey and Marsh, 2005]. The impacts of the SCAN pattern on Eurasian climate have been examined by Bueh and Nakamura [2007] and these include generation of northern European precipitation anomalies. The extent of its influence on the Mediterranean Sea remains to be determined. The EA/ WR pattern has been found to have an impact on rainfall in the Mediterranean [Krichak and Alpert, 2005]. As yet, no study has compared the relative impacts of these four modes on air sea heat exchange in the Mediterranean Sea and that is the central goal of this study. We aim to build up a consistent picture of how different modes of atmospheric variability influence the heat budget of the Mediterranean and in particular heat loss in the main dense water formation regions. Our focus is on the air sea heat flux rather than the freshwater flux (i.e., the net evaporation) as the latter has already been shown to be a weak influence on the density flux and hence dense water formation [Josey, 2003]. [6] The structure of the paper is as follows. In sections 2 and 3, the data sets and method used for the analysis are described. The results of the study are presented in section 4, with a focus on the heat flux fields associated with the largescale modes of variability, their variation with season and impacts on the basin and subbasin heat budgets. Finally, in section 5, the main findings are summarized and discussed in the context of Mediterranean Sea dense water formation. 2. Data Sets [7] The two main data sets used in the analysis are the well established NCEP/NCAR atmospheric model reanalysis [Kistler et al., 2001] and the recently developed higherresolution ARPERA fields [Herrmann and Somot, 2008]. The NCEP/NCAR fields consist of monthly means derived from 6 h forecast fields output by a numerical weather prediction model that assimilates a wide range of meteorological data. It is based on the NCEP global spectral model operational in 1995, which has a horizontal resolution of about 210 km (equivalent to 1.9 of latitude), full details may be found in the work of Kistler et al. [2001] and references therein. The ARPERA fields have been obtained by a dynamical downscaling of the ERA40 reanalysis [Uppala et al., 2005] produced at the European Centre for Medium Range Weather Forecasts (ECMWF) by the regional climate model ARPEGE Climate developed at CNRM. ARPEGE Climate is a stretched grid general circulation model with a zoom capability over an area of interest. In the current configuration the pole of maximum resolution is in the Tyrrhenian Sea and the resolution is 50 km over the whole Mediterranean Sea and large parts of Europe and North Africa. In a full climate mode (no nudging), the present climate of the stretched grid ARPEGE Climate model has been validated over the Mediterranean area in the work of Gibelin and Déqué [2003] and Somot et al. [2008]. The downscaling includes a spectral nudging which essentially allows the small scales (under 250 km) to freely evolve while the larger scales are driven by ERA40. Note that the last full year available with ERA40 is 2001, after this time fields from ECMWF analysis are used instead. These analysis fields are downgraded to the ERA40 resolution for consistency before being implemented in the spectral nudging. The ARPERA fields are increasingly being used in Mediterranean Sea studies, for example in analysis of airsea fluxes [Herrmann and Somot, 2008; Aznar et al., 2010] and for forcing high resolution ocean models [Herrmann and Somot, 2008; Tsimplis et al., 2008, 2009; Beuvier et al., 2010; Herrmann et al., 2010]. Note that the ability of regional climate models to reproduce the large scale atmospheric conditions of the driving general circulation model is not a source of concern as this has been tested previously by Sanchez Gomez et al. [2008] and found to be satisfactory. ERA40 and ARPERA have been compared by Herrmann and Somot [2008], which established the added value of the dynamical downscaling without losing the large scale chronology imposed by the reanalysis. The impacts of downscaling on the basin mean heat budget have been recently studied by Sanchez Gomez et al. [2011] in the context of the constraint provided by measurements of the heat transport through the Strait of Gibraltar. [8] We have chosen to focus here on NCEP/NCAR and ARPERA as these two data sets provide complementary representations of the air sea exchange in the Mediterranean Sea. The NCEP/NCAR fields are relatively coarse resolution and potentially do not include the full effects of finescale surface forcing in dense water formation regions. In contrast, the ARPERA fields are relatively high resolution and, as we will show, do contain fine scale regions of high heat loss. However, they contain some uncertainty arising from the downscaling process which is not present in NCEP. By considering both data sets we are able to test the robustness of the mode response to variations arising from data set resolution, different model physics and the inclusion of finer scale physical processes. We stress that our aim is not to investigate the effects of downscaling (for which the logical choice would have been ECWMF and ARPERA) 2of15

3 Figure 1. Composite fields of the NCEP anomalous winter (October March) sea level pressure (colored field) and 10 m wind speed (vectors) for the following modes of variability: (a) the NAO, (b) the EA, (c) the SCAN, and (d) the EA/WR. Note the horizontal vector in the bottom left corner of each image is for scale and indicates a wind speed of 5 ms 1. but rather to use different data sets to increase confidence in our results. For our study we consider the common period, , which is spanned by the two data sets and enables us to include within the time interval covered the strong winter heat loss events in the western Mediterranean in and [Herrmann et al., 2010; Schroeder et al., 2010]. [9] Northern Hemisphere teleconnection patterns have been determined in various studies [e.g., Barnston and Livezey, 1987; Rogers, 1990]. As noted in the Introduction, a widely employed analysis of the main modes of atmospheric variability is that carried out at the NOAA Climate Prediction Centre They characterize the main modes through a rotated principal component analysis [Barnston and Livezey, 1987] of the observed monthly mean 500 mb height anomaly fields in the region 20 N 90 N and provide monthly index values for each mode which are regularly updated. In this study, we use the CPC index values for the NAO, EA pattern, SCAN pattern and EA/WR pattern from telecontents.shtml. 3. Method [10] Basin mean values for the net heat flux have been determined by calculating area weighted averages for a full Mediterranean Sea region (defined here to exclude the Black Sea and to be bounded at the western margin by the Strait of Gibraltar) and the eastern and western subbasins separately (which are divided by the Sicily Strait). In addition, in order to isolate dense water formation region signals, averages have also been determined for boxes centered on the Aegean Sea, the southern Adriatic and a region including the Gulf of Lions in the western Mediterranean. The dense water formation boxes are shown subsequently in section 4 and Figure 6. [11] Anomalies of a given variable are determined in all cases with respect to the climatological monthly mean over the period The primary focus is on air sea exchanges during the winter centered half of the year as this spans the main period for deep water formation and the density flux anomalies in winter dominate the annual signal [Josey, 2003]. Winter centered means and anomalies have been obtained for the six month period from October March (referred to as winter hereafter for convenience but note that this period contains the outlying months of October and March which lie either side of a typical winter). In addition, we consider to a lesser extent the summer centered period from April September (subsequently referred to as summer for convenience). [12] For the analysis of the impacts of different modes of variability, composite fields of the heat flux signal associated with each mode have been determined as follows. First, 3of15

4 Figure 2. Time series of the winter centered atmospheric mode indices from 1958 to 2006 for (a) the NAO, (b) the EA, (c) the SCAN, and (d) the EA/WR. for each mode all individual months fields throughout the period with the CPC mode index value greater than a threshold value of 1.5 (or less than 1.5) are selected for both winter and summer; note that other threshold values could be adopted without impacting our conclusions. Second, the net heat flux field for each selected individual month is divided by the mode index value for that month and a weighted mean is then formed such that the resulting pattern represents the value of the net heat flux associated with a unit positive value of the index. This process is repeated for other variables under consideration, in particular sea level pressure, air temperature and wind speed. Note, the net heat flux sign convention adopted throughout the paper is for heat loss from the ocean to the atmosphere to be negative. 4. Results 4.1. Large Scale Modes of Atmospheric Variability [13] In this section, we set the large scale context for our subsequent analysis by using the NCEP/NCAR reanalysis to characterize the patterns of sea level pressure variability associated with each of the major atmospheric modes over a broad North Atlantic/European domain and considering their potential impacts on the Mediterranean Sea. We also consider the time dependence of each mode over the period [14] Composite fields of winter sea level pressure for a unit positive index value of each of the four major modes are shown in Figure 1, together with the associated wind speed pattern. The NAO is characterized by the familiar north south dipole structure with anomalously high/low pressure over the Azores/Iceland [Josey et al., 2001; Hurrell et al., 2003]. The high pressure region extends over the Mediterranean Sea and has been shown to have a significant impact on sea level through the inverse barometer effect [Tsimplis and Josey, 2001]. The composite associated with the EA pattern is dominated by a broad region of anomalously low (by up to 12 mb) pressure centered at about (55 N, 25 W) which is approximately midway between the two centers of the NAO dipole [Josey and Marsh, 2005]. It gives rise to strong cyclonic wind forcing of the North Atlantic centered on this location. In the western Mediterranean, it produces a relatively strong pressure gradient with the potential to generate a cold northerly airflow (in its negative state) and enhanced ocean heat loss in this region. The SCAN pattern has a southwest to northeast pressure dipole with the stronger pole centered to the east of Scandinavia. It is associated with anomalously low pressure over the Mediterranean but only a weak variation in this field. Finally, the positive state of the EA/WR pattern exhibits anomalously high pressure over the North Sea which is flanked by low pressure centers over West Russia and to a lesser extent the western North Atlantic at N. [15] Time series for each of the four major mode winter indices over the period are shown in Figure 2; note for clarity, on the temporal scale a value of, for example, 1960 indicates winter not An upward trend in the NAO index from the 1960s to the 1990s, followed by a leveling off over the subsequent decade, is evident and has been recognized elsewhere [e.g., Hurrell, 1995]. Less well known is the upward trend in the EA pattern index which has a more pronounced switch from primarily negative values in the period to positive values in the 1980s starting about 5 years earlier than the increase for the NAO. This EA pattern trend plays a significant role in freshwater flux variations to the eastern North Atlantic [Josey and Marsh, 2005]. In contrast, there are no clear trends in either the SCAN or EA/WR pattern indices; although there is some suggestion that the SCAN pattern has primarily negative values in the 1980s and 1990s Impacts on Mediterranean Sea: Winter [16] In this section, we focus in on mode impacts in the Mediterranean Sea using both NCEP/NCAR and the higherresolution ARPERA fields. 4of15

5 Figure 3. Climatological winter mean net heat flux (Wm 2 ) for the Mediterranean Sea from (a) NCEP/ NCAR and (b) ARPERA for the period Climatological Mean Winter Air Sea Heat Exchange [17] The climatological winter mean air sea heat flux fields are discussed first prior to considering the impacts of the different modes of variability. Climatological winter mean fields of the net heat flux from NCEP/NCAR and ARPERA for the common period are shown in Figure 3. The NCEP/NCAR fields are relatively coarse and show net heat loss over the whole basin, with relatively little spatial structure although there is some tendency for stronger net heat loss from west to east and, to a lesser extent, from south to north. In contrast, the ARPERA fields show clearly defined localized regions of intense heat loss over both the Gulf of Lions in the western Mediterranean and the Aegean Sea with extreme values up to 200 Wm 2. These reflect the effects of the higher resolution in ARPERA which enables smaller scale flows associated with orographic effects [Herrmann and Somot, 2008] and are consistent with other downscaled fields (e.g., Ruiz et al., 2008). Thus, we have two data sets with quite different representations of the mean winter forcing of the Mediterranean Sea with which to examine the impacts of large scale modes of variability. [18] It is interesting to consider the climatological winter mean net heat flux values for each data set for the full basin and subbasins (see Table 1, top row). For the basin as a whole, the enhanced heat loss in the dense water regions found for ARPERA results in a mean net heat flux of 112 Wm 2 which is 20 Wm 2 stronger than the value obtained for NCEP ( 92 Wm 2 ). This difference arises primarily in the eastern subbasin for which the NCEP value ( 102 Wm 2 ) showing slightly greater losses than ARPERA ( 99 Wm 2 ). In contrast, the western subbasin values are quite similar with the ARPERA value ( 102 Wm 2 ) showing slightly greater losses than NCEP ( 99 Wm 2 ). The difference in behavior between the two subbasins reflects the greater spatial extent of the ARPERA high heat loss region in the eastern basin Impact of Modes on the Net Heat Flux: Winter [19] Spatial fields of the anomalous winter net heat flux for a unit positive index value of each of the four main modes are shown in Figure 4 for NCEP/NCAR. Note that, given the sign convention adopted, negative heat flux anomalies represent stronger ocean heat loss than normal. Values for the corresponding basin and subbasin averaged heat flux anomaly together with the climatological winter means are listed in Table 1. The NAO shows an east west gradient and reversal of the sign of the heat flux anomaly between subbasins (Figure 4a). This results in slightly reduced heat loss in the western Mediterranean (average 8Wm 2, consistent with earlier studies reporting a link to the NAO in this subbasin [e.g., Somot, 2005] and slightly enhanced heat loss in the eastern Mediterranean (average 5 Wm 2 ); values for individual grid cells range up to +/ 20 Wm 2. These cancel out such that the full basin average is 0 Wm 2 ; that is, the NAO has no net impact on the full Mediterranean Sea winter mean heat budget according to NCEP. [20] In contrast, the positive state of the EA pattern gives rise to reduced heat loss across the whole basin (Figure 4b) with strongest individual values (up to 40 Wm 2 ) in the Gulf of Lions. Thus, the negative state of the EA pattern results in significantly stronger than normal winter heat loss across the basin and, in particular, has the potential to enhance dense water formation in the western Mediterranean. The average heat flux anomaly for the western subbasin associated with the EA pattern is 28 Wm 2 which when compared with the climatological winter mean heat loss of 102 Wm 2 can be seen to be a major signal (i.e., more than a quarter of the total value). For the full basin the corresponding values are 24 Wm 2 and 92 Wm 2. [21] Turning to the remaining two modes, the SCAN pattern (Figure 4c) results in slightly stronger than normal heat loss across the Mediterranean but does not have a major impact, with the basin averaged anomaly being just 3 Wm 2. In contrast, the EA/WR pattern net heat flux anomaly field (Figure 4d) has an east west dipole structure which is markedly stronger than that associated with the NAO (Figure 4a). Consequently, the EA/WR mode has a major impact on the winter heat budgets of the western (17 Wm 2 ) and eastern ( 18 Wm 2 ) subbasins although only a minor Table 1. Basin and Subbasin Averaged Values of the Climatological Winter Mean Net Heat Flux for the Period a Full Basin Western Basin Eastern Basin NCEP ARPERA NCEP ARPERA NCEP ARPERA CWM NAO EA SCAN EA/WR a Climatological winter is October March. Also shown are the basin averaged values of the anomalous winter net heat flux for a unit positive value of each of the four atmospheric modes, units Wm 2. Anomalies greater than 10 Wm 2 are bold. 5of15

6 Figure 4. Anomalous winter (October March) NCEP/NCAR net heat flux (Wm 2 ) for a unit positive value of the index for each of the four main modes: (a) NAO, (b) EA, (c) SCAN, and (d) EA/WR. impact on the full basin. Thus, the winter heat loss associated with the EA/WR pattern has the potential to influence dense water formation in each subbasin but in the opposite sense in each region (i.e., enhanced heat loss in the eastern subbasin is accompanied by reduced heat loss in the western subbasin in the positive state of the EA/WR pattern and vice versa for the negative state). This raises the interesting possibility that variations in the sign of this mode of atmospheric variability could lead to see saw behavior in the location of dense water formation in the Mediterranean. The various basin and subbasin means are shown in a bar chart representation on Figure 5a, and this highlights the major impacts of the EA and EA/WR patterns relative to the NAO and SCAN pattern. [22] The above analysis has been repeated for the higherresolution ARPERA data set (see Figures 5, 6, and Table 1). The anomaly fields associated with each of the modes have broadly similar structure in ARPERA as noted for NCEP above albeit with greater resolution. The east west gradient is again present for the NAO but some localized forcing of the Aegean Sea as large as 40 Wm 2 is now evident (Figure 6a) and this may reflect modulation of the signal associated with this mode by the better representation of orography possible with ARPERA (which allows cold and dry air to be advected from the northeast, in contrast this is not possible in NCEP as it contains no channel between Greece and Turkey). As a consequence, the impact of the NAO on the eastern subbasin with ARPERA ( 10 Wm 2 )is twice that found for NCEP ( 5 Wm 2 ). For the basin as a whole, ARPERA has a mean NAO signal of 5Wm 2, i.e., a small additional heat loss, compared with the no net impact value of 0 Wm 2 obtained using NCEP. Note, as part of a heat budget closure study using ECMWF fields, Pettenuzzo et al. [2010] find that the NAO has some influence on the overall Figure 5. Regionally averaged values of the NCEP/NCAR anomalous winter net heat flux (Wm 2 ) for a unit positive index value of each of the four major atmospheric modes determined from (a) NCEP/NCAR and (b) ARPERA. Black bars indicate the whole Mediterranean Basin, dark gray bars indicate the western Mediterranean, and light gray bars indicate the eastern Mediterranean. 6of15

7 Figure 6. Anomalous winter (October March) ARPERA net heat flux (Wm 2 ) for a unit positive value of the index for each of the four main modes: (a) NAO, (b) EA, (c) SCAN, and (d) EA/WR. The black outlined regions in Figure 6a shows the areas chosen for the deep water formation region analysis in section 4.4. basin heat budget, unfortunately they did not consider the other modes so it is not possible to place their results in the context of the full range of atmospheric variability. [23] The EA pattern again results in reduced/enhanced heat loss for the positive/negative state across the full basin in ARPERA. However, the EA signal in the western Mediterranean is noticeably greater than before (again probably because of the inclusion of orographic enhancement of the strength of the wind field and thus the heat loss) such that the subbasin averaged anomaly associated with this mode is now 38 Wm 2 for ARPERA compared with 28 Wm 2 for NCEP. This is the largest signal found for any of the modes and more than 5 times greater than that found for the NAO (7 Wm 2 ). Thus, the EA pattern is the dominant mode of atmospheric variability influencing heat loss, and by extension, dense water formation in the western Mediterranean. This is consistent with the recent suggestion by Schroeder et al. [2010] that the deep convection observed in the Gulf of Lions in the winters of and is linked to a negative state of the EA pattern. [24] The results for the SCAN pattern with ARPERA are very similar to those found with NCEP with no clear impact on the heat exchange. Close agreement is also obtained between the ARPERA and NCEP budgets for the EA/WR pattern which is again seen to have a striking east west reversal in the sign of the heat flux anomaly. Furthermore, ARPERA reveals more intense features associated with this mode in the potential dense water formation sites of the Aegean and Gulf of Lions. Note that the Adriatic Sea, which until the mid 1990s was thought to have been the sole region of dense water production in the eastern Mediterranean lies in a transition region as regards the impact of the EA/WR pattern. [25] The impact of the different modes obtained using ARPERA are summarized on Figure 5b which again reveals the major impacts of the EA and EA/WR patterns when compared to the NAO and SCAN pattern (see the corresponding results for NCEP in Figure 5a). There is clearly close agreement between the results obtained with the coarse resolution NCEP and fine resolution ARPERA data sets (the latter being downscaled from the ECMWF coarse resolution model that has different physics to that used in the NCEP reanalysis). This agreement indicates that our conclusions regarding the impacts of the different modes are robust to changes both in the resolution of the data set employed and in the physics of the atmospheric model Driving Mechanisms for Ocean Heat Loss [26] In this section, the driving mechanisms for the air sea heat flux anomalies associated with each of the different atmospheric modes are considered. The net heat exchange is the sum of four components: two turbulent heat loss terms (the sensible and latent heat flux) and two radiative flux terms (the longwave and shortwave flux). Of these components, the turbulent heat loss terms, in particular the latent heat flux, and to a lesser extent the sensible heat flux, dominate anomalies in the winter net heat exchange [Josey, 2003]. The latent and sensible heat flux are essentially driven by the product of the wind speed and the sea air humidity and sea air temperature difference, respectively, according to well known bulk formulae [e.g., Josey et al., 1999; Josey, 2011; Romanou et al., 2010]. In order to determine the driving mechanisms for heat loss associated with the different atmospheric modes, it is thus necessary to consider the anomalous wind speed and air temperature fields (note the atmospheric humidity field tends to closely follow that for air temperature and is not considered separately here). We adopt this approach, with a focus on insights gained from the ARPERA fields; broadly similar results are obtained from NCEP but at coarser resolution. [27] Spatial fields of the anomalous winter 2m air temperature, 10 m wind speed and sea level pressure for each of the four modes are shown in Figure 7. For the NAO, SCAN and EA/WR patterns we present fields for unit positive 7of15

8 Figure 7. Anomalous winter (October March) ARPERA 2 m air temperature (colored field, C) and 10 m wind speed (vectors) for a unit value of the index (NAO, SCAN and EA/WR positive; EA negative) for each of the four main modes: (a) NAO, (b) EA, (c) SCAN, and (d) EA/WR. Corresponding sea level pressure anomalies are contoured in intervals of 1 mb. Note the horizontal vector above Figure 7a is for scale and indicates a wind speed of 2 ms 1. values of the mode index. However, for the EA pattern, the field shown in Figure 7 is for a unit negative value of the mode index. We have chosen this approach for the EA pattern as it is the negative state of this mode which gives rise to the strongest heat loss signals over the Mediterranean (as discussed in 4.2) and we want to examine the associated cold airflow signal. The fields are shown at a larger scale than in the preceding section in order to examine the forcing mechanisms over a broader domain. [28] The air temperature anomaly field for the NAO (Figure 7a) reveals warm air over the northwest Mediterranean and colder air over the southern part of the basin, the transition between the two occurring at approximately 40 N. The associated sea level pressure and wind fields show the western half of the Azores High centered at about (40 N, 20 W) and a westerly (i.e., from the west) airflow over Europe and the northwest Mediterranean, hence, the relatively warm air in this region. This is counterbalanced by a 8of15

9 northeasterly flow over the Aegean which brings the relatively cold air to the southern half of the basin. Comparison of Figures 7a and 6a reveals a close correspondence between the net heat flux and air temperature anomalies which indicates that air temperature (and related humidity anomalies, not shown) play a major role in driving the NAO related ocean heat loss anomalies. For the basin as whole (and individual subbasins) the warm and cool air temperature anomalies tend to cancel. Hence, the overall impact of the NAO on the heat budgets is small as discussed above. [29] In contrast to the NAO, we found earlier that the EA mode has a major impact (20 40 Wm 2 ) on the heat budgets and the reason for this is clear in Figure 7b. The negative state of the EA mode has intense high pressure over the midlatitude West Atlantic with the Mediterranean lying on its southeast flank and experiencing a northeasterly airflow which brings cold, dry air from continental Europe over the full basin. Close to the northwestern Mediterranean coast, and in the northern Adriatic, the mean air temperature anomaly is around 1 C colder on average than normal which enables the strong heat losses seen previously. Moreover, the cold temperature anomalies are maintained over the full basin, although becoming progressively weaker toward the south and east and it is this basinwide cold (and dry) signal which results in the major heat budget anomalies noted in section 4.2 primarily through a combination of sensible and latent heat flux contributions. The latent heat flux anomalies are typically a factor of between 2 and 3 times greater in magnitude than the sensible heat loss terms as is to be expected given that the latent heat flux is typically stronger than the sensible heat flux. For example, additional calculations reveal that the western basin ARPERA winter net heat flux anomaly of 38 Wm 2 associated with the EA pattern (Table 1) contains contributions of 24 Wm 2 from the latent heat flux component and 10 Wm 2 from the sensible heat flux. [30] Turning to the remaining two modes, the SCAN pattern has a relatively weak signal, of varying sign, in air temperature over the Mediterranean (Figure 7c). Hence, the small impact on the heat budgets seen previously. Of much greater interest is the EA/WR pattern which was earlier found to generate equal and opposite signals in the heat flux over the eastern and western halves of the basin (Figure 5b). These signals are seen to be due to anomalously high pressure centered over the North Sea which results in a cold northerly airflow over the eastern Mediterranean (and the neighboring Black Sea) and a warmer southeasterly airflow in the western Mediterranean (Figure 7d) Atmospheric Mode Impacts on Surface Heat Flux at Dense Water Formation Sites [31] We now consider the impacts of the different modes of variability on the air sea heat exchange at the three main locations where dense water formation has been known to occur in the Mediterranean: the northwest of the basin centered on the Gulf of Lions, the southern Adriatic and the Aegean Sea [e.g., Roether et al., 2007; Schroeder et al., 2008]. Time series of the winter heat flux anomaly for the regions outlined on Figure 6 which have been chosen to cover each of these sites are shown in Figure 8. Given the high resolution of ARPERA and variations in the location where the dense water is formed, different boundaries for each region could be adopted. The sensitivity of the results to the precise boundary definition has been explored and similar results obtained for different boundary choices. Correlation coefficients between the heat flux anomaly time series and the different atmospheric mode indices have also been determined and are given in Table 2. [32] The northwestern Mediterranean time series (Figure 8a) shows a recent transition between positive winter heat flux anomalies (i.e., reduced ocean heat loss) in the range Wm 2, throughout most of the 1990s and early 2000s, and negative values (i.e., stronger winter heat loss) in the mid 2000s, centered on (which had an anomalous loss of 50 Wm 2 ). It is worth remembering that this anomaly figure is an average value for the entire winter centered half of the year; that is, the northwestern Mediterranean region experienced on average 50 Wm 2 greater heat loss than usual over the 6 month period from October 2004 to March This is consistent with observations and modeling of new dense water formation in this region at this time [e.g., Schroeder et al., 2008; Herrmann et al., 2010]. Note also that the 1960s and early 1970s also saw predominantly strong winter heat loss in the northwestern Mediterranean and the first measurements of dense water formation there were made during this period [Lacombe et al., 1970]. The correlation statistics for the NW Mediterranean in Table 2, reveal that the EA pattern has a strong influence, r = 0.52, on the heat loss in this dense water formation area, as expected from the analysis presented earlier, and this further supports the suggestion that this mode plays a dominant role in the event by Schroeder et al. [2010]. The NAO is also likely to play some role although the correlation value is relatively small, r = Note that in an earlier study, with a narrower definition (January March) of winter, Rixen et al. [2005] find a stronger correlation of heat flux anomalies with the NAO, they did not consider other modes of variability so it is difficult to put their results in context. [33] The southern Adriatic time series (Figure 8b) bears some resemblance to that of the northwestern Mediterranean with positive heat flux anomalies in the 1990s transitioning to negative values in the 2000s. This fits with understanding of the EMT, discussed in the Introduction, which saw a switch in the main location of dense water formation from the southern Adriatic to the Aegean in the 1990s [e.g., Roether et al., 2007]. Prior to the 1990s, the southern Adriatic experienced both anomalously strong and weak heat loss winters, and observations in the mid 1980s revealed it be the main site of dense water formation in the eastern Mediterranean at that time [Roether et al., 1996]. The r values in Table 2 reveal that the EA pattern is the only atmospheric mode which is significantly correlated (r = 0.41) with heat loss in the southern Adriatic dense water formation region; note the strength of this correlation is weaker than in the northwestern Mediterranean. [34] The main features of note in the Aegean time series (Figure 8c) are the strong heat loss winters of and occurring in the middle of the EMT period discussed above which otherwise experienced close to normal heat loss. The causes of the heat loss in these winters have been discussed in detail in an earlier study [Josey, 2003]. The main factors were enhanced latent and sensible heat loss over the region associated with an anomalously northerly airflow. The study of Josey [2003] employed the NCEP/ 9of15

10 Figure 8. Time series of ARPERA winter heat loss anomaly (units Wm 2 ) in the three main dense water formation regions: (a) northwestern Mediterranean, (b) southern Adriatic, and (c) Aegean. Dashed vertical lines indicate the EMT period spanned by the two hydrographic surveys reported by Roether et al. [1996]. NCAR reanalysis (as the higher resolution ARPERA product was not available at that time) and only considered one atmospheric mode, the NAO, which was found not to be a significant factor in Aegean Sea heat loss. Here we consider all four modes, for which the correlation statistics are listed in the last row of Table 2. The NAO is again found to have an insignificant influence and the same is true for the EA and SCAN patterns. However, the EA/WR pattern is seen to be strongly negatively correlated (r = 0.59) with the air sea heat flux in the Aegean Sea dense water formation region. Thus, the positive state of the EA/WR pattern results in strongly enhanced heat loss over the Aegean Sea (see Figure 6d) and explains 35% of the interannual variance in winter heat loss in this region. For the other dense water formation regions, it is the EA mode which dominates explaining 27% of the winter heat loss variance in the NW Mediterranean region and 17% of the variance in the southern Adriatic. The maximum variance explained by the NAO is just 9% for the NW Mediterranean region and the SCAN mode values are smaller still; this indicates that while the NAO and SCAN modes may well be of importance for other processes in the Mediterranean Sea they do not take a dominant role in establishing heat loss anomalies in the key deep convection regions. As an illustration of the mode heat flux relationship, we show scatterplots of the Aegean Sea net heat flux anomaly against the value of each of the four major mode indices in Figure 9. The strong relationship with the EA/WR mode in contrast to the other 3 modes considered is clearly evident. Returning to the time series (Figure 8c), outside of the EMT period, strong winter heat loss was seen in and but this was preceded by very weak heat loss in and this together with circulation changes accompanying the EMT may have prevented the resumption of dense water formation in the Aegean at this time. Note also that winter had strong winter heat loss and this may have played a Table 2. Values of the Correlation Coefficient, r, Between the Dense Water Formation Region Winter Net Heat Flux Anomaly Time Series in Figure 8 and the Corresponding Winter Index Values for Each of the Four Main Atmospheric Modes Over the Period a NAO EA SCAN EA/WR NW Mediterranean S Adriatic Aegean a Boldface values are significant at the 95% level. 10 of 15

11 Figure 9. Scatterplots showing the variation of anomalous winter net heat flux in the Aegean Sea from ARPERA with each of the four main modes: (a) NAO, (b) EA, (c) SCAN, and (d) EA/WR. significant role in preconditioning surface waters for the subsequent EMT event [Beuvier et al., 2010] Impacts on Mediterranean Sea: Summer Climatological Mean Summer Air Sea Heat Exchange [35] For completeness, we now consider the climatological mean air sea heat exchange in the summer centered half of the year (April September) and its response to the different atmospheric modes. We note in advance, that our results indicate a much weaker response of the ocean to atmospheric variability in summer so we restrict the discussion here to the key points. We also note that the patterns associated with each mode are somewhat different in summer from their winter counterparts but do not discuss these in detail here given the weak impact on the heat flux in the summer season. [36] The climatological summer mean air sea heat flux fields for NCEP and ARPERA are shown in Figure 10 with corresponding basin mean values in Table 3 (top row). The two fields are broadly similar in the magnitude of the net heat flux which in the summer half of the year is dominated by solar radiation (i.e., the shortwave flux) and exhibits a net ocean heat gain in the range Wm 2 depending on region. The NCEP/NCAR fields tend to have greater heat gain than ARPERA and this is reflected in a basin mean value of 94 Wm 2 compared to 80 Wm 2 for ARPERA. As was found to be the case for the winter fields (Figure 3), the ARPERA fluxes indicate the existence of fine scale regions of modified heat exchange over the Aegean Sea and northwestern Mediterranean where the net heat flux is at the low end of the range (around 60 Wm 2 ) because increased latent heat loss (not shown) in these areas offsets to some extent the strong shortwave gain Impact of Modes on the Net Heat Flux: Summer [37] By analogy with Figures 4 and 6, spatial fields of the anomalous summer net heat flux for a unit positive Figure 10. Climatological summer mean net heat flux (Wm 2 ) for the Mediterranean Sea from (a) NCEP/NCAR and (b) ARPERA for the period of 15

12 Table 3. Basin and Subbasin Averaged Values of the Climatological Summer Mean Net Heat Flux for the Period a Full Basin Western Basin Eastern Basin NCEP ARPERA NCEP ARPERA NCEP ARPERA CSM NAO EA SCAN EA/WR a Climatological summer is April September. Also shown are the basin averaged values of the anomalous summer net heat flux for a unit positive value of each of the four atmospheric modes, units Wm 2. index value of each of the four main modes are shown in Figures 11 and 12 for NCEP/NCAR and ARPERA. The corresponding basin and subbasin averaged heat flux anomalies are listed in Table 3. [38] In most cases, the spatial pattern of anomalous heat exchange associated with each mode in summer is similar to that found in winter for both NCEP and ARPERA but is weaker in amplitude. The basin and subbasin mean anomalies do not exceed 10 Wm 2 for any of the modes considered. As before, use of the ARPERA output results in more localized heat loss features than NCEP but the overall patterns remain similar. [39] The main point of interest is a change in the air sea heat exchange pattern associated with the SCAN mode for which Figures 10c and 11c now show a clear east west asymmetry in the sign of the summer heat flux anomaly which was not present in winter. However, this signal is fairly weak, the western/eastern subbasin anomalies for a unit positive SCAN index are 5/3 Wm 2 for NCEP and 2/6 Wm 2 for ARPERA. These values are small compared to the subbasin summer means which range from 75 to 103 Wm 2 depending on the data set considered. [40] In summary, the atmospheric modes have a small impact on air sea exchange in the Mediterranean in summer with the largest signal being just 9 Wm 2. This stands in sharp contrast to winter, when mode related anomalies in the budgets associated with the EA/WR pattern ranged in magnitude up to 19 Wm 2 and with the EA pattern up to 38 Wm 2. These results are indicative of a fundamental difference in the nature of the coupled ocean atmosphere system in the Mediterranean region between the two halves of the year. In winter, large scale patterns of atmospheric variability have the potential to generate major changes in the strength of the coupling between the two components while this is not the case in summer. 5. Conclusions [41] Surface flux fields from two major atmospheric model data sets have been analyzed in order to establish the impacts of the first four modes of atmospheric variability in the North Atlantic/Europe region on air sea heat exchange in the Mediterranean Sea. The first of these data sets is the relatively coarse resolution NCEP/NCAR atmospheric reanalysis while the second is the higher resolution AR- PERA data set produced by dynamical downscaling of fields from the ECMWF reanalysis. By considering these two data sets, which span a range of model physics and spatial resolution, we have been able to increase the confidence in our results. [42] The analysis is carried out for the period and similar results are obtained from both NCEP and ARPERA. In each case the largest mode related heat flux anomalies are found in winter, with relatively minor contributions in summer. The NAO, has a surprisingly small impact on both the full basin and subbasin winter mean heat budgets. The magnitude of the full basin anomaly for a unit NAO mode index value is less than 5 Wm 2. In contrast, the EA pattern has a major effect, of order 25 Wm 2, with strong impacts on both the eastern and western Mediterranean, the largest signal being in the western part of the basin. The SCAN mode has the weakest influence of those considered and can probably be ignored in future analyses of heat budget variability in the basin. However, the EA/WR Figure 11. Anomalous summer (April September) net heat flux (Wm 2 ) for NCEP/NCAR composited on the four main modes: (a) NAO, (b) EA, (c) SCAN, and (d) EA/WR. 12 of 15

13 Figure 12. Anomalous summer (April September) net heat flux (Wm 2 ) for ARPERA composited on the four main modes (a) NAO, (b) EA, (c) SCAN, and (d) EA/WR. mode plays a significant role but, unlike the EA, it generates a dipole in the heat exchange with an approximately equal and opposite signal of Wm 2 on the eastern and western halves of the basin. [43] Thus, ocean atmosphere heat exchange in the Mediterranean Sea is dominated at basin and subbasin scales in the key winter half of the year by two modes, the EA and EA/WR patterns, not previously recognized for their importance in this context while the NAO plays a secondary role (see Figure 5 for quantitative details). The impacts of the different modes on the Mediterranean Sea in winter are shown schematically in Figure 13. The EA pattern (in its negative state) has a pervasive influence over the whole basin due to its associated northeasterly (i.e., from the northeast) flow of cold dry air which steepens the sea air temperature and humidity gradients facilitating stronger than normal heat loss. The EA/WR pattern produces a northerly flow of cold dry air over the eastern basin and a southerly flow of relatively warm moist air over the western basin leading to significant heat flux anomalies which are opposite in sign as shown. Thus, variations in the sign of the EA mode can be expected to impact the heat budget of the whole basin, while variations in the EA/WR mode have the potential to lead to a see saw variation in the budgets of the eastern and western subbasins. The NAO does not strongly modify the overall heat budget with either data set considered, although we recognize that it has significant impacts on other physical fields. In particular, the relatively uniform sign of the NAO related pressure signal over the Mediterranean has been shown to significantly modify sea level through the inverse barometer effect [Tsimplis and Josey, 2001] even though it does not produce the anomalous airflows seen for the EA and EA/WR which are conducive to modified air sea heat exchange. At a local scale, the impacts of the different modes have also been analyzed in the context of the surface forcing for the three recognized sites of dense water formation: the northwestern Mediterranean, the southern Adriatic and the Aegean Sea. Here, the EA pattern has the dominant effect on the northwestern Mediterranean Figure 13. Schematic representation of the surface high pressure anomaly, airflow, and air sea heat flux anomalies associated with the two main modes which influence the Mediterranean Sea heat budget (a) EA and (b) EA/WR. Separate arrows indicating the sense of the heat flux anomaly are shown for the eastern and western basins. 13 of 15