Title: Recurrent replenishment of Labrador Sea Water and associated decadal-scale variability
Abstract: Journal of Geophysical Research: OceansVolume 121, Issue 11 p. 8095-8114 Research ArticleOpen Access Recurrent replenishment of Labrador Sea Water and associated decadal-scale variability Igor Yashayaev, Corresponding Author Igor Yashayaev [email protected] Fisheries and Oceans Canada, Bedford Institute of Oceanography, Dartmouth, Nova Scotia, CanadaCorrespondence to: I. Yashayaev, [email protected] for more papers by this authorJohn W. Loder, John W. Loder Fisheries and Oceans Canada, Bedford Institute of Oceanography, Dartmouth, Nova Scotia, CanadaSearch for more papers by this author Igor Yashayaev, Corresponding Author Igor Yashayaev [email protected] Fisheries and Oceans Canada, Bedford Institute of Oceanography, Dartmouth, Nova Scotia, CanadaCorrespondence to: I. Yashayaev, [email protected] for more papers by this authorJohn W. Loder, John W. Loder Fisheries and Oceans Canada, Bedford Institute of Oceanography, Dartmouth, Nova Scotia, CanadaSearch for more papers by this author First published: 30 September 2016 https://doi.org/10.1002/2016JC012046Citations: 125AboutSectionsPDF ToolsRequest permissionExport citationAdd to favoritesTrack citation ShareShare Give accessShare full text accessShare full-text accessPlease review our Terms and Conditions of Use and check box below to share full-text version of article.I have read and accept the Wiley Online Library Terms and Conditions of UseShareable LinkUse the link below to share a full-text version of this article with your friends and colleagues. Learn more.Copy URL Share a linkShare onFacebookTwitterLinkedInRedditWechat Abstract Winter convective overturning in the Labrador Sea reached an “aggregate” maximum depth of 1700 m in 2015—the deepest since 1994—with the resulting Labrador Sea Water (LSW) “year class” being one of the deepest and thickest observed outside of the early 1990s. Argo float, annual survey, and moored measurements in recent decades provide an unprecedented view of important seasonal, interannual, and longer-term LSW variability in the Labrador Sea region. During the 2002–2015 “Argo” era, the average winter LSW pycnostad volume was about 70% larger in relatively strong convection years than in relatively weak ones. However, the winter-to-fall LSW disappearance volume was 180% larger, pointing to a factor of 2.8 difference in the potential LSW export rates from the region between relatively strong and weak convection years. Intermittently recurrent deep convection is contributing to predominant decadal-scale variations in intermediate-depth temperature, salinity, and density in the LS, with implications for decadal-scale variability across the subpolar North Atlantic and potentially in the Atlantic Meridional Overturning Circulation. Comparison of the LS ocean heat content changes and cumulative surface heat losses during the fall-winter cooling seasons indicates that anomalously strong winter atmospheric cooling, associated at least in part with the North Atlantic Oscillation, is continuing to be a major forcing of the recurrent convection. Key Points Deep convection in Labrador Sea (LS) in 2015 strongest since mid-1990s, resulting in one of largest Labrador Sea Water (LSW) year classes Long-term variability in middepth LS properties dominated by decadal-scale variation associated with intermittent strong convection Average winter-to-fall LSW disappearance volume in strong convection years about three times that in weak convection years 1 Introduction The northern North Atlantic (NA) is a key area to variability in the global climate system because of deep convection contributing to the Atlantic Meridional Overturning Circulation (AMOC) and the ocean storage of anthropogenic carbon and heat [e.g., Broecker, 1991; Dickson et al., 1996; Mauritzen et al., 2012; Khatiwala et al., 2013; Rhein et al., 2015], and strong decadal-scale variability contributing to hiatus periods in global warming [Chen and Tung, 2014; Drijfhout et al., 2014; Visbeck, 2014]. However, recent studies indicate that the AMOC and its variability are extremely complex. It has temporal variability on a wide range of time scales [Srokosz and Bryden, 2015], and there are uncertainties regarding the linkages between deep convection and the AMOC, and the origin and structure of AMOC variability more generally [Hurrell et al., 2006; Lozier, 2012; Buckley and Marshall, 2016]. On the other hand, there is increasing evidence that natural variability in atmospheric forcing of deep convection in areas like the Labrador Sea (LS; Figure 1a), in conjunction with the circulation of the subpolar NA gyre, can lead to broader-scale interannual-to-decadal variations in NA ocean heat content [e.g., Sarafanov, 2009; Msadek et al., 2014; Duchez et al., 2016; Williams et al., 2015]. Further, there are indications of implications for AMOC circulation and water property variability [e.g., Rhein et al., 2011; Yeager and Danabasoglu, 2014; Zhang and Zhang, 2015], and for climate variability on larger scales including other variables such as NA sea ice and sea level [Yeager et al., 2015; Delworth and Zeng, 2016]. With global climate models projecting surface ocean warming, freshening and increasing stratification at high latitudes, and weakening of the AMOC [e.g., Collins et al., 2013], it is important to have ongoing robust observations of various aspects of the AMOC, including water mass production, properties and transformation rates in major deep convection areas. Figure 1Open in figure viewerPowerPoint (left column) (a) Topography, major upper-ocean circulation features, and locations of the AR7W line, Argo and vessel-based profiles (yellow dots) used in Figures 2 and 4, and Hamilton Bank mooring (⋄) used in Figure 5; and climatological distributions of (b) salinity (S) at 500 m and (c) vertical difference in potential density referenced to 1000 dbar (Δσ1, kg m−3) between 1000 and 1500 m. (middle and right columns) Distributions of (from top to bottom) potential temperature (θ, °C), salinity, potential density referenced to the surface (σo, kg m−3), and dissolved oxygen (O2, mL/L) on the AR7W line during (d–g) 23–29 May 2008 and (h–k) 15–20 May 2015, with key water masses labeled (see text for definitions). The reference pressure of 1000 dbar is used in Figure 1c because it is more appropriate for examining intermediate-depth density changes than the usual surface reference depth used in section plots like Figures 1f and 1j. Here we report on recent hydrographic observations from the LS deep convection region, and place them in a historical context. Together with recent observations of enhanced convection in the Irminger Sea in 2015 [Fröb et al., 2016; de Jong and de Steur, 2016], these indicate that intermittent strong deep convection in part associated with variability in the North Atlantic Oscillation (NAO) is continuing to occur in the northern NA. In contrast to recent studies that report [Robson et al., 2014; Rahmstorf et al., 2015; Srokosz and Bryden, 2015; Somavilla et al., 2016; Yang et al., 2016] or project [e.g., Collins et al., 2013; Loder et al., 2015] recent or ongoing reductions in deep convection and/or AMOC strength, we show that the LS convection depths during the winters of 2014 and 2015 were the largest since the record values of the early 1990s and among the larger ones in the historical record. Using vertical profile data from ships and floats, we examine seasonal-to-interannual temperature, salinity, and associated density variations since 2002, and interannual-to-decadal variability in these properties since 1938. We also introduce a newly available moored near-bottom temperature time series from the 1000 m isobath on the Labrador Slope extending back to 1987, and point out similarities with (as well as differences from) the interannual and decadal-scale variability in the central LS. We suggest that intermittent deep convection on pentadal-to-decadal time scales is continuing to generate voluminous “year classes” of Labrador Sea Water (LSW, the NA's predominant intermediate-depth water mass) [e.g., Lazier, 1973; Talley and McCartney, 1982; Yashayaev, 2007, henceforth Y2007] with varying properties including its dynamically important density. Considering the recent studies mentioned above, there is high potential for broader-scale influences on the upper-intermediate subpolar NA such as reduced warming (or even cooling), increased ventilation and intensified circulation, and also on water properties and transports in the AMOC. We describe our data sources and methodology in section 2, and then the variability in hydrographic properties during the “Argo era” (since 2002) in relation to the longer historical record in section 3. In section 4, we present results on the variability of LSW density and pycnostad volumes and, in section 5, we provide a brief update on the relation of deep convection variability to atmospheric forcing in recent years. We make concluding remarks in section 6. 2 Data Sources and Methodology Our primary data sets are (i) full-depth temperature, salinity, and dissolved oxygen profiles on the AR7W (Atlantic Repeat Hydrography Line 7 West) line across the LS occupied during Fisheries and Oceans Canada's (DFO's) annual survey under its Atlantic Zone Off-shelf Monitoring Program (AZOMP) [Greenan et al., 2010], (ii) temperature and salinity profiles over the upper 2000 m in the LS region from the International Argo float program [Riser et al., 2016], (iii) available observed-level temperature and salinity data archived by other programs and national and international data centers [see Kieke and Yashayaev, 2015], and (iv) a near-bottom moored temperature time series from a long-term mooring maintained by the Bedford Institute of Oceanography (BIO) on the Labrador Slope (Figure 1a). The AZOMP survey is a continuation of annual DFO occupations of the AR7W line by BIO since the start of the World Ocean Circulation Experiment (WOCE) in 1990 [Lazier et al., 2002; Yashayaev et al., 2014, 2015]. Since 2004, the survey has been carried out in May with approximately 30 CTD (conductivity-temperature-depth) and water sampling (e.g., for dissolved oxygen, nutrients, transient tracer) stations occupied between Labrador and Greenland. The pressure, temperature, conductivity, salinity, and dissolved oxygen data sets have been quality controlled and calibrated to meet WOCE standards, using water sample (e.g., Autosal salinity and Winkler titration), SBE35 temperature recorder and laboratory calibration data. Argo float temperature and salinity profiles, available since 2002, have been quality controlled through comparisons with vessel CTD and water sample data and comparisons between floats and by performing critical analyses of spatial and temporal deviations. The historical and other recent data have also been quality controlled and processed through similar critical analyses [see Yashayaev and Seidov, 2015, for a summary]. Following Yashayaev and Loder [2009; henceforth YL2009], but now including all available Argo and other data from July 2002 to December 2015 and June 2015, respectively, time-depth series of spatially averaged potential temperature, salinity, and potential density with weekly-to-monthly (dependent on Argo and ship survey data coverage) resolution have been computed for an area of ∼60,000 km2 in the central LS (Figure 1a). Peak winter mixed layer depths have historically occurred in this area [e.g., de Boyer Montégut, 2004; Våge et al., 2009; Loder et al., 2015], which is characterized by weak horizontal gradients in the climatological distributions of water properties such as salinity and vertical density difference (Figures 1b and 1c). Note the similarities between the patterns of the salinity at 500 m depth and the vertical density difference between 500 and 1000 m. There is a relatively homogeneous area in the central LS that tails eastward to the south of Greenland (also see section 4), and a relatively sharp gradient with the surrounding boundary current zone to the northeast, north, and west [e.g., Straneo, 2006]. The climatological fields in Figures 1b and 1c have been derived through an iterative spatial interpolation and temporal filtering process which reduced interannual and lower-frequency influences on the resulting spatially gridded climatology, thereby providing a clearer indication of the intermediate layer's (or LSW pycnostad's) typical horizontal structure. In each iteration, the low-frequency variability was approximated by polynomial functions fitted to the deviations of the observed values from the spatially gridded fields. These gridded fields were in turn constructed from “detrended” data (i.e., with the low-frequency signal removed). The iterations were repeated until the changes between successive ones were insignificant. To increase the robustness of the method (by reducing the sensitivity to random outliers), each observation used in the polynomial fitting and spatial gridding was assigned a weight based on its deviation from a longer-term (typically 3 year) median. The vertical gradient of potential density (Δσ1/Δz) has been computed for each hydrographic profile, using vertical differences over 50 m intervals at 10 m increments, with particular interest in identifying the extent of the pycnostads [Talley and McCartney, 1982] associated with various LSW year classes [Y2007]. With our focus on LSW, potential density referenced to 1000 dbar (σ1) is used in this study, except for the use of potential density referenced to the surface (σo, which is also commonly referred to as σθ) in the section plots in Figures 1f and 1j. Four novel aspects of our analysis have been the developments of (i) a new set of estimates of the “aggregate” maximum depth of convection in the central LS in each year; (ii) time-σ1 displays of the occurrence of particular σ1 values in the upper 2000 m of the central LS during the Argo era, from which density layer thicknesses can be computed; (iii) spatial maps showing the seasonal and interannual evolution of the thickness (and hence three-dimensional extent) of the LSW pycnostads; and (iv) estimation of the volume of LSW that disappears from the LS region between late winter and midfall (i.e., during the “restratification” season) in strong and weak convection years. The “aggregate” maximum convection depth was defined as the 75th percentile of the deepest extent of the pycnostad in the available profiles from each winter. The σ1 layer thicknesses were computed from the occurrence of 1 m vertically averaged (or vertically interpolated, depending on resolution of the observed profiles) σ1 values in overlapping 0.005 kg m−3 σ1 intervals spaced at 0.0025 kg m−3, with the overlap included so that occurrence peaks spread across two adjacent nonoverlapping σ1 intervals were not missed. The resulting σ1 histogram was then converted to layer thicknesses for the observed σ1 values by summing the occurrences and accounting for the overlapping σ1 intervals. Time series of parameter values at selected depths have been estimated from all individual profiles in the central LS and then low-pass-filtered to provide annual indices of variability since 1987. A complementary weekly temperature time series is available from the moored near-bottom Aanderaa current meter redeployed by BIO for successive 1 year periods at 20 m above bottom on the 1000 m isobath on the Labrador (Hamilton Bank) Slope (Figure 1a) for most years since 1978. Using concatenation, interpolation across the short mooring turn-around periods and time series filtering, a continuous uniform record has been created for the period from June 1995 to May 2015 at a common site (55°7.2′N, 54°5.5′W), and data for most of the time between September 1978 and June 1992 available for nearby sites to north have been used to construct a moored near-bottom temperature record for that period. With the assumption of negligible along-isobath variability these two long-term moored records can be considered as a single time series spanning nearly four decades. Further, to place the recent variability in a historical context, we use annual time series of temperature, salinity, and density averaged over the 200–2000 m vertical interval in the central LS back to 1938 as long-term indices of these variables over its intermediate-depth waters. These were derived from time series for selected depths like those discussed above and previously reported data sets from earlier years [Y2007; Yashayaev et al., 2008], with some additional early years (prior to 1950) included here as well. Finally, in examining the influence of atmospheric forcing on temperature variability, two annual indices have been used: the winter (January to March) NAO based on station sea level pressures available from the U.S. Global Climate Observing System (GCOS) website; and the cumulative air-sea heat flux integrated over individual-year cooling seasons, computed following the procedure described in YL2009 using 6 hourly heat flux and daily-mean radiation data obtained from the U.S. National Centers for Environmental Prediction (NCEP) Reanalysis (R2). 3 Variability During the Argo Era in a Longer-Term Context The full-depth distributions of (potential) temperature, salinity, (potential) density, and dissolved oxygen on the AR7W line in May 2015 are shown in Figures 1h–1k, together with those from May 2008 in Figures 1d–1g [YL2009]. Prior to 2015, the latter year had the deepest LS convection (to ∼1600 m) since the record 2400 m depth in 1994 [Y2007] (note the 2008 aggregate maximum convection depth, discussed below, was 1550 m). Description of the distributions in the upper 2000 m and their origin is best done in consideration of the time-depth evolution of temperature, salinity, and density (Figure 2) in the central LS during the Argo era and of the associated evolution of vertical density gradient (Figure 3a). Included in Figure 2 is the new set of estimates of the “aggregate” maximum depth of convection in the central LS in each year. Note that these aggregate maxima are generally less than the “extreme” maxima found in the individual profile(s) with the deepest pycnocline (due to eddies affecting the latter; see later), but are more robust and representative estimates of typical convection depths across the central LS. Figure 2Open in figure viewerPowerPoint Time-depth evolution of weekly-to-monthly (a) potential temperature (θ, °C), (b) salinity (S), and (c) potential density (σ1, referenced to 1000 dbar; kg m−3) in the upper 2000 m of the central LS (Figure 1a) during the Argo era (2002–2015). The “aggregate” maximum convection depth in each winter, computed as the 75th percentile of the pycnostad's deepest extent in the available profiles, is indicated by the horizontal bars in each plot. Figure 3Open in figure viewerPowerPoint (a) Time-depth evolution of weekly-to-monthly vertical gradient of potential density (Δσ1/Δz, kg m−3 per 100 m) in the central LS, corresponding to that for θ, S, and σ1 in Figure 2. (b) Time-density display of “(potential) density (σ1) layer thickness” (for σ1 intervals of 0.005 kg m−3, offset by 0.0025 kg m−3 for display purposes) showing the evolution of particular LSW year classes. The time-depth displays for 2002–2015 show the seasonal development and deepening of a warm fresh surface layer starting each spring, followed in fall by surface mixing extending to depths of 600–1700 m by late winter (Figure 2). Overall cooling and densification of the upper-intermediate layer in each winter are apparent, with the change in vertically averaged salinity less clear. A longer-term historical context is provided by the time-depth (200–2800 m) displays of annual temperature, salinity, and density since 1938 (Figure 4), which indicate that deep convection extended to 1500 m or more in the late-1930s, early-1950s, mid-1970s, and early-1990s (as well as in 2008, 2014, and 2015; Figure 2). Figure 4Open in figure viewerPowerPoint Time-depth evolution of annual (a) potential temperature (θ, °C), (b) salinity, (S) and (c) potential density (σ1, referenced to 1000 dbar, kg m−3) between 200 and 2800 m (below the surface) in the central LS (same area as in Figure 2), using all available data between 1938 and 2015. The box in the upper right of each plot indicates the time-depth domain of the Argo coverage in Figure 2. The AR7W section plots (Figures 1d–1k) show large pycnostads with relatively uniform properties extending across most of the LS in the upper ∼1500 m in both 2008 and 2015, below a relatively shallow, warm, fresh, light, and oxygenated surface layer (which developed between the end of winter convection and each May survey). However, it is apparent that the 2015 pycnostad was wider, deeper, cooler, fresher, denser, and more oxygenated than its 2008 counterpart, thereby representing a more voluminous year class of LSW. Following Y2007's year class terminology in which major LSW pycnostads are named for the strong atmospheric-forcing winters during which they developed, we tentatively refer to the 2015 pycnostad as LSW2014–2015 to reflect the preconditioning that occurred in winter 2014 [Kieke and Yashayaev, 2015; Josey et al., 2015]. Alternatively, with the 2012 and 2013 convection viewed as preconditioning, and recent Argo profiles indicating convection to comparable depths in 2016, LSW2012–2016 could be regarded as more appropriate (to be discussed in a follow-up paper). In the longer-term context, LSW2014–2015 (or LSW2012–2016 in a broader sense) is the most voluminous year class since the record-thick LSW1987–1994 class, exceeding LSW2000–2003 [Yashayaev et al., 2008] as well as LSW2008, and possibly the third largest in the historical record (also after LSW1977–1978). The section plots (Figures 1d–1k) also show the two deep water masses, the more saline, and less-oxygenated Northeast Atlantic Deep Water (NEADW) and the less saline and more-recently oxygenated Denmark Strait Overflow Water (DSOW), that form the lower limb of the AMOC and pass through the LS [e.g., Yashayaev et al., 2015]. Also, the upper portion of NEADW is included in the historical time-depth display (Figure 4). However, the variability of these important water masses [e.g., Dickson et al., 2002; Yashayaev and Dickson, 2008] will be further discussed elsewhere. The interannual and seasonal variability of temperature, salinity, and density at intermediate depths in the central LS during the Argo era is reflected in the time-depth plots in Figure 2 and in the time series at 1000 and 1500 m in Figure 5a. LSW2000–2003 resulted in relatively cool, fresh, and dense conditions (compared to later Argo years) at intermediate depths in the early 2000s. This was followed by two multiyear periods (2004–2007 and 2009–2011) of overall warming (by ∼0.3°C and ∼0.1°C at 1000 and 1500 m, respectively) associated with reduced winter convection, interrupted by the cooling (by ∼0.2°C at 1000 m) and freshening associated with LSW2008, and then culminating in the warmest and saltiest intermediate-depth conditions during the Argo era in fall 2011 [also, see Kieke and Yashayaev, 2015]. Above-average convection started again in 2012, with aggregate maximum depths and temperature, salinity, and density returning in 2014 [e.g., Kieke and Yashayaev, 2015] to values comparable to those in 2008 (∼1500 m), and then reaching their recent extreme values (∼1700 m) in winter 2015 (Figure 2). Figure 5Open in figure viewerPowerPoint (a) Time series of LS and related variables since 1987. The uppermost curves are individual-year values of the winter NAO index (note the inverted scale), the change in ocean heat content in the central LS during the oceanic cooling season, and the cumulative surface heat flux (from the atmosphere to the ocean) during the atmospheric cooling season. Below these are individual-profile estimates (dots) of potential temperature (°C) at 1000 m (red) and 1500 m (blue), and salinity (magenta) and potential density (green; σ1, kg m−3, inverted) at 1000 m in the central LS, together with annual-mean estimates (faint solid curves). In the middle is weekly temperature at 20 m above the seafloor (solid brown curve) from the mooring on the 1000 m isobath on the Hamilton Bank slope, together with low-pass (∼annual-mean) estimates (faint background curve). (b) Time series of annual-mean potential temperature (red) and salinity (magenta) for the 200–2000 m vertical interval in the central LS (open circles connected by lines), and of the corresponding annual potential density anomaly (green; σ1, kg m−3; inverted scale). Thinner lines are used over multiyear periods with inadequate data. From the perspectives of anthropogenic climate change and subannual natural variability, it is critically important that the LSW variability during the Argo era be considered in relation to the historical record. The temperature, salinity, and density time series at 1000–1500 m in Figure 5a, together with the time-depth displays back to 1938 in Figure 4 and the time series for the 200–2000 m interval in Figure 5b, provide this context. The coolest, freshest, and densest intermediate-depth (∼500–2000 m) conditions in the central LS occurred in the early 1990s associated with LSW1987–1994. Then, there was relatively strong warming, salinification, and lightening during the late 1990s [e.g., Lazier et al., 2002], such that the conditions in the early 2000s associated with LSW2000–2003 (which we described above as cool, fresh, and dense relative to the Argo era) were actually near-average compared to conditions over the past ∼75 years. The annual 200–2000 m indices indicate that the recent Argo-era peak temperature and salinity values in 2011 were not as high as those in the early 1970s, pointing to the need for caution in inferring long-term trends in this region (e.g., anthropogenic warming) from records over the past three decades or less [e.g., Terray, 2012; Loder and Wang, 2015; Parker and Ollier, 2016]. These time series also indicate that longer-term variability of these intermediate-depth LS water properties in the modern observational record has been dominated by decadal-scale variability rather than a long-term trend, especially for temperature. However, this does not rule out a long-term trend in the surface conditions such as the warming indicated for the western LS in gridded historical data sets [e.g., Hartmann et al., 2013; Loder and Wang, 2015] or the freshening expected around Greenland with anthropogenic climate change and ice-sheet melting [e.g., Collins et al., 2013; Yang et al., 2016]. Important questions regarding the variability in LS deep convection and its controlling factors are the relative importance of temperature and salinity variability to LSW density variability, and whether the subannual temperature and salinity changes are density-compensated (i.e., whether the density changes associated with one property are offset by those associated with the other) [e.g., Mauritzen et al., 2012]. There is an indication in Figures 2 and 3 of significant density changes with the temperature contribution predominating in the central LS on seasonal, interannual, and decadal time scales. It can be seen that the increases in density during winter in general, and those associated with the major LSW year classes, in particular, are associated with cooling and freshening, with the temperature contribution to density generally dominating over the salinity contribution. Their relative importance on interannual and decadal scales can be seen more explicitly in Figure 6, in which the lower plot shows the anomalies in annual potential density at various depths, relative to the long-term means at these depths. The upper two plots show the annual potential density anomalies associated with each of temperature and salinity separately. It can be seen clearly that the intermediate-depth positive density anomalies (Figure 6c) associated with the major convection periods of 1976–1977, 1987–1994, 2000–2003, and 2014–2015 were predominantly associated with the contribution (positive density anomalies in Figure 6a) from relatively low temperatures, with the contribution from salinity being of opposite sign (corresponding to relatively low salinity water descending as part of convective overturning; Figure 6b). However, during some convection periods (e.g., 1938–1939, mid-1950s, 2008), the relative importance of these contributions is less clear, with some indication that both temperature (low) and salinity (high) anomalies contributed to relatively high densities, consistent with previous suggestions for some periods [e.g., Yashayaev et al., 2015; Yang et al., 2016]. Figure 6Open in figure viewerPowerPoint Time-depth displays of the annual potential density anomalies (relative to the long-term means at each depth) associated with (a) temperature and (b) salinity variations, and of (c) the total density anomaly, between 200 and 2800 m in the central LS (corresponding to the observations in Figure 4). The box in the upper right of each plot indicates the time-depth domain of the Argo coverage in Figure 2. The seasonal resolution available with the Argo data (Figure 2) provides an indication of why it is difficult to resolve this issue for some convection periods, especially for the earlier ones without year-round data coverage. It can be seen that, during 2008, there were salinity anomalies of differing signs in different seasons and at different depths in the upper-intermediate layers. As a result, the annual anomalies in Figure 5 may not be representative of those affecting winter convection [also see YL2009; Våge et al., 2009; Kieke and Yashayaev, 2015; Yeager et al., 2016, for more detail on this convection event]. Further