In 2003 and in 2005, hydrographic data provided sufficient spatial coverage in the Labrador Sea to infer basin wide changes in the water mass characteristic of the Upper Labrador Sea Water (ULSW). The ULSW was considerably saltier and warmer in 2005 than in 2003. Although convection in the Labrador Sea leads to mixing with salinity-poor surface water and is opposed to the observed salinity trend, the increased vertical homogeneity of the CTD profiles, the increase in the ULSW thickness and the intensification of the potential vorticity minimum for the isopycnals σΘ = 27.700–27.734 kg/m3 in 2005 point to convection in winter 2005 which ventilated at least about 20% of the Labrador Sea region.
 As a response to global warming, many climate models predict a decline of the meridional overturning circulation (MOC), and in most of the models this decrease is caused by a reduction of Labrador Sea Water (LSW) formation [e.g., Stouffer et al., 2006]. The connections between LSW formation, strength of the MOC, and climate changes put close attention to the variability of LSW formation, and the hydrographic conditions prevailing in the Labrador Sea. The Labrador Sea is one of the few regions in the ocean, where deep convection occurs. Dependent on the heat loss in winter and on the vertical density stratification in the water column in late winter, the intensity of the convection process and the density layers affected change from year to year [e.g., Lazier et al., 2002]. The peak of intense deep convection was in the early 1990s, forming the coldest, freshest, and densest Labrador Sea Water in the last 50 years in the density range σΘ = 27.74–27.80 kg/m3. Water in this density range is sometimes called ‘classical’ LSW (here it is henceforth named LSW). Since that time, convection declined, caused by decreasing heat fluxes and subsequently increased vertical density stratification. From time to time, a lighter mode of LSW was formed (σΘ = 27.68–27.74 kg/m3), and this mode was named Upper LSW (ULSW) [Pickart et al., 1996]. In the last decade, the main formation area of ULSW was the northern central Labrador Sea [Stramma et al., 2004], but some was presumably also produced in the Irminger Sea [Kieke et al., 2006]. Between 1997 and 2005, convection did not reach the LSW [Lazier et al., 2002; Yashayaev and Clarke, 2006]. Weak heat flux anomalies prevailed in the Labrador Sea during winters of 2004 and 2005 [Hendry, 2006], so that presumably convection activity was weak. We use hydrographic data from the Labrador Sea from 2003 and 2005 to show that, in contrast to this expectation, convection did occur and did affect ULSW. As a consequence, this has led to a shift and to an intensification of the potential vorticity minimum for most of the ULSW, and to an increase in the volume of ULSW by an expansion of the layer thickness.
 Hydrographic data in the Labrador Sea from the following cruises are used (Figure 1): METEOR cruises ‘M59/2’ (July 2003) and ‘M59/3’ (September 2003), and N/O THALASSA cruises ‘SUBPOLAR’, (June/July 2005), and ‘WNA’ (July 2005). M59/3 and ‘WNA’ are focused on the western Labrador Sea. The four cruises are part of the German national program SFB 460, entitled ‘Dynamics of thermohaline circulation variability’. During all cruises, a Seabird SBE 911plus instrument was used for CTD/O measurements. Salinity was calibrated by analysing water samples taken from 10L Niskin bottles with an Autosal salinometer. Data accuracy for these cruises is 0.002–0.003°C for temperature and 0.002–0.003 for salinity. Data from the oxygen sensors were calibrated either by manual titration of water samples or by using an automated two-channel SIS Winkler titration system. The rms error between CTD-derived oxygen and bottle values varies from cruise to cruise between 0.04 and 0.06 ml/l. Due to various problems the absolute calibration of the oxygen measurements remains uncertain, and we therefore refrain from comparing oxygen data from different cruises. The data set is further complemented by CTD measurements carried out by the Bedford Institute of Oceanography (BIO, Dartmouth, Canada) along the WOCE line AR7W during July 2003. These data are available at the Marine Environmental Data Service (MEDS, Ottawa, Canada).
 In the following, CTD profiles which are weakly stratified in the density range occupied by ULSW are separated from the ones with a stronger vertical density gradient. We look for a criterion which is applicable to the data of 2003 as well as of 2005. All density profiles are at first inspected by eye. Further analysis reveal that a vertical density gradient of σΘ = 0.056 kg/m3 per 1000 m is the most suitable threshold for both years (Figures 2a and 2c). Profiles with a vertical density gradient weaker than this threshold are named ‘weakly stratified profiles’ (WSP). In 2003 and in 2005 the WSPs exhibit the lowest salinities (Figures 2b and 2d).
 Except for surface water influenced by the Arctic outflow and local glacier melt, the ULSW is the freshest water mass in the western subpolar North Atlantic. The West Greenland Current (WGC) carries saline Irminger Sea Water (ISW) into the Labrador Sea (Figures 1, 3a and 3b). Maximum salinities are found at about 150–350 m depth at the Greenland boundary and shallower than 50 m in the centre of the Labrador Sea The CTD data indicate that the region near the Greenland boundary is more saline than further offshore for all densities <σΘ = 27.81 kg/m3, i.e. the realm covered by ULSW and LSW. The features of the mean salinity distribution in the density range of ULSW (σΘ = 27.68–27.74 kg/m3) are typical for the water column above 2000 m (Figures 3a and 3b): the saline ISW in the WGC follows mainly the boundary current along the northwestern rim of the Labrador Sea and joins the Baffin Island Current to form the Labrador Current at the Canadian continental shelf (Figure 1). The water mass characteristics found in the WGC are also transported into the interior of the Labrador Sea by eddies [Lilly et al., 2003]. The freshest ULSW is found in the centre of the Labrador Sea, but reaches closer to the Canadian than the Greenland coast (Figures 3a and 3b). The main pathways for ULSW out of the Labrador Sea are the Labrador Current along the Canadian continental shelf [Stramma et al., 2004], offshore of the WGC into the Irminger Sea, and into the Northeast Atlantic [Kieke et al., 2006]. The latter can be seen in a band of fresher water, extending from the centre of the Labrador Sea farther east and into the Irminger Sea (Figure 3b).
 In 2003 and in 2005, WSPs with relatively small vertical and horizontal variations in salinity and temperature are found in the central Labrador Sea, enveloped on three sides by the 3200 m isobath (Figure 1). In 2005, the area covered by the WSPs is about 0.37 × 106 km2, i.e. about one third of the area of the Labrador Sea with depths >500 m (∼1 × 106 km2). The differences in the area between 2003 and 2005 are most likely an artefact of the different spatial coverage in the two years (e.g., Figures 3a and 3b). The WSPs co-locate with the low salinities in the ULSW (Figures 2b, 3a, and 3b) similar to 2003. Only two WSPs are observed in 2005 outside the central Labrador Sea in the low salinity exit path [Lavender et al., 2000] from the Labrador Sea to the Irminger Sea. (Figures 1 and 3b). Presumably, the pathways are topographically guided (Figure 1). Another unique feature of the WSPs is the oxygen maximum in the ULSW layer. This feature is present in 2003 and in 2005.
 The most obvious differences between 2003 and 2005 are that the WSPs in 2005 are saltier by about 0.02–0.025 (Figure 4a) and warmer by 0.15–0.2°C than observed in 2003. In 2005, only one profile at 57°12′N, 51°59′W shows salinities in 500 m and 800 m depth, which are comparable to the 2003 survey (Figure 4b) and are thus a remnant of the 2003 vintage. Yashayaev and Clarke  suggested that the warming and salinification of the Labrador Sea is caused by advection of warm and salty ISW from the WGC. This water is then transported into the central Labrador Sea by eddies, which pinch off the WGC, and mixes isopycnally with the fresher ULSW.
 The 2003 and 2005 data reveal several other features, which have changed during this time period. The vertical salinity variations in the WSPs in 2005 are on average more homogeneous than the ones measured in 2003 in the same months (Figure 4b). The vertical ULSW thickness expanded on average by 120 m between 2003 and 2005 and reached a mean value of 1070 ± 20 m in 2005 (Figures 3c and 3d). For these calculations only the WSPs from 2003 and 2005 have been used. In 2005, the upper bound of the ULSW is found in the WSP area at a mean water depth of 180 m. In the central Labrador Sea this isopycnal occupies even shallower depths between 120 m and 150 m (green dots in Figure 1), co-located with the lowest salinities (Figure 3b). In 2003, the mean depth of the upper bound is 200 m.
 The maximum ULSW thickness in both years is found in the central Labrador Sea. The horizontal gradient of the thickness within the WSPs is more pronounced in 2005 than in 2003 (Figures 3c and 3d) and the patch with the maximum thickness is more locally restricted. The thickness increase between 2003 and 2005 is mainly owed to the 110 m increase of the layer bounded by the isopycnals σΘ = 27.725 kg/m3 and σΘ = 27.735 kg/m3 The maximum thickness between these isopycnals is 660 m in 2005, i.e. it covered about 60% of the entire ULSW layer.
 From 2003 to 2005, an intensification of the minimum of the potential vorticity (PV = f/ρdρ/dz, [Talley and McCartney, 1982]) is observed for most of the ULSW density range, i.e. from σΘ = 27.700 to 27.734 kg/m3. Figures 3e and 3f present the change of the PV at the isopycnal σΘ = 27.730 kg/m3. The 2005 minimum is located further north than in 2003 and the region with an intensified minimum is larger (Figure 3f). The absolute PV minimum in 2005 is found at σΘ = 27.734 kg/m3, whereas in 2003, the minimum is located at a higher density (σΘ = 27.740 kg/m3). If we take the PV minimum as the indicator of the penetration depth of convection [McCartney and Talley, 1982], the maximum convection depth in 2005 is 1200 m. The vertical range extends from 960–1200 m, with a mean of 1100 m for the WSPs. If we assume that all density layers with a thickness increase have been affected by convection, convection reached at least down to the isopycnal σΘ = 27.735 kg/m3 located at a maximum depth of 1300 m in 2005.
 The influence of the summer mixed layer is restricted to the water column above 200 m [Lazier, 1988]. In 2003 and in 2005 the densities at this depth increase from the boundary towards the centre of the gyre (Figure 3g). In 2005 the doming is intensified (Figure 3h): the densities in the centre have increased well into the ULSW density domain, and the region with densities >σΘ = 27.68 kg/m3 is larger than in 2003. The dense water is also found in the Labrador Current in both years (Figures 3g and 3h). In 2005 denser water is also observed on a band leading into the Irminger Sea (Figure 3h). The stronger doming in 2005 compared to 2003 as shown here for the 200 m isobath is representative for all depths between 100 m and 500 m.
 Cooling and freshening of ULSW due to convection in the Labrador Sea is opposed by horizontal fluxes of heat and salt from the boundary current, carrying warm and saline ISW into the Labrador Sea. Lazier et al.  found that the salinity in the 150–1000 m depth range did not change significantly from 1990 to 2000. In contrast to that, the salinity in this depth interval increased by 0.02–0–025 from 2003 to 2005. The increase of salinity in the ULSW in the central Labrador Sea, which was also reported by Yashayaev and Clarke , is usually seen as an indication that convection was absent or weak, since convection in the Labrador Sea would lead to a freshening, and higher salinities are attributed to advection of ISW.
 ULSW salinity and temperature profiles below about 700 m depth are more homogeneous in June/July 2005 than in summer 2003 (Figure 4b) and exhibit in both years an oxygen maximum. Theoretically, smoothed T and S profiles can be accomplished by isopycnal mixing with the warm and saline surrounding, provided the mixing continues over a sufficient time to equalize the salinity difference between the centre of the Labrador Sea and the boundary. The horizontal salinity gradient between the centre and the boundary in the ULSW density range decrease from 2003 to 2005. While the centre gets saltier (Figures 3a and 3b), it still remains fresher than the boundary. If isopycnal mixing would be the dominating process, the oxygen maximum in the ULSW layer would be eroded, because oxygen is much lower in the boundary region and in the stratified profiles compared to the WSPs. To satisfy both, the increased vertical homogeneity of the T/S distributions and the conservation of the oxygen maximum as well as the warming and salinification, both processes, i.e. mixing and convection are necessary: the vertical variability of salinity and temperature in the ULSW layer is increased by isopycnal mixing but is then smoothed out in February–March through the convection process. The vertical mixing during convection also adds fresher and oxygen-rich surface water. Both processes together leave the WSPs saltier than in former years, but they remain the freshest profiles and the ones with the strongest oxygen maximum observed during an individual year.
 The comparison between the two data sets indicates changes between two consecutive winters instead of one. The winter heat flux anomalies in the central Labrador Sea in 2004 and 2005 were lower than normal by about 26–27 W/m2 [Hendry, 2006]. The observed smoother profiles, the ULSW thickness increase, the intensification of the PV minimum for most of the ULSW density range, the shifting of the PV minimum, and the strengthening of the doming observed in summer 2005 compared to June/July 2003 are most likely caused by convection in late winter 2005, otherwise, these features would have been partly eroded. Judging from Figures 3c and 3d, the ULSW thickness increased over an area of 0.22 × 106 km2 from summer 2003 to 2005. Depending on the chosen criteria convection reached maximum depths of 1200–1300 m. Based on Argo float data and moored instruments Avsic et al.  reported a convection depth of 700 m in March 2004 compared to 1400 m in March 2003 and 1300 m in March 2005. The latter fits well to our 2005 estimates. Avsic et al. also reported a diminished sea-air heat loss during April 2003 and March 2004, including the convection period in 2004. These results support our conclusion that the winter 2005 was a year with more intense convection than winter 2004.
 The ventilated volume, i.e. the volume of water with contact to the surface ocean is calculated by multiplying the area of 0.22 × 106 km2 with the range of observed convection depths (960–1200 m) inferred from the location of the PV minimum and is (0.21–0.26) ·106 km3. This is equivalent to a ventilation rate of 6.7–8.4 Sv. Most of this ventilated volume consists of water which was already in the density domain of ULSW before convection started in 2005. It was thus ventilated but not newly formed, i.e. it was not transferred from lighter densities into the density realm of ULSW. These numbers are therefore not comparable to the ULSW formation rates reported, for instance, by Kieke et al. . The input of anthropogenic CO2 into the deep ocean, however, is more related to the ventilated volume, while the formation rate might be more important for the strength of the meridional overturning.
 Despite the weak heat flux anomalies, the salinification and the warming of the Labrador Sea since 2003, convection occurred in late winter 2005 reaching at least the isopycnal σΘ = 27.735 kg/m3. It affected about 20% of the Labrador Sea region. The mean density profiles of the central Labrador Sea from 1996 to 2005 show that the stratification of ULSW weakened considerably since 2001, with a minimum in 2005 (Figure 4c), and increased for water denser than ULSW. The most significant change is the deepening of the isopycnal σΘ = 27.74 kg/m3, the lower bound of the ULSW, from a depth of about 400 m in 1996 to 1300 m in 2005. Between 2003 and 2005, however, the upward shift of the lower ULSW bound was larger than the downward movement of the lower bound. These changes in the stratification will favour convection and ventilation of the ULSW in the following winters, whereas ventilation of water denser than the ULSW will become more difficult to achieve.
 We thank captains and crews of the R/V METEOR and N/O THALASSA cruises for valuable assistance. Data from THALASSA cruise ‘WNA’ are courtesy of Jürgen Fischer, IFM-GEOMAR Kiel, and Allyn Clarke (BIO) of the Canadian cruise in 2003. This work was supported by the Deutsche Forschungsgemeinschaft as part of the SFB 460 ‘Dynamics of Thermohaline Circulation Variability.’