Seasonal variability of the warm Atlantic water layer in the vicinity of the Greenland shelf break

The warmest water reaching the east and west coast of Greenland is found between 200 and 600m. While important for melting Greenland’s outlet glaciers, limited winter observations of this layer prohibit determination of its seasonality. To address this, temperature data from Argo profiling floats, a range of sources within the World Ocean Database, and unprecedented coverage from marine-mammal borne sensors have been analyzed for the period 2002–2011. A significant seasonal range in temperature (~1-2°C) is found in the warm layer, in contrast to most of the surrounding ocean. The phase of the seasonal cycle exhibits considerable spatial variability, with the warmest water found near the eastern and southwestern shelf break toward the end of the calendar year. High-resolution oceanmodel trajectory analysis suggests the timing of the arrival of the year’s warmest water is a function of advection time from the subduction site in


Introduction
The warmest water to reach the east and west coasts of Greenland is found in the 200-600 m depth range [e.g., Fratantoni and Pickart, 2007;Mauritzen, 1996]. However, harsh environmental conditions including seasonal sea-ice cover have restricted the number of year-round sub-surface observations near the Greenland shelf, making it challenging to (a) ascertain if the warm layer (hereafter WL) experiences a significant seasonal cycle, and (b) determine the characteristics of any such cycle. Nonetheless, quantifying and understanding the nature of the WL seasonal cycle are important for improving predictions of sea level rise because, temperature changes within the layer have been highlighted as a possible cause of accelerated melting of tidewater glaciers [Holland et al., 2008;Straneo et al., 2010;Mortensen et al., 2011;Sutherland et al., 2013]. In particular, the 1997 warming (relative to previous years) of the subsurface waters along the West Greenland coast has been attributed to an anomalously warm Irminger Current associated with a shift in the phase of the North Atlantic Oscillation [Holland et al., 2008;Zhu and Demirov, 2011]. Observations of the vertical temperature structure on the West Greenland coast have been used to estimate heat content and thus the ice-melting potential of both the WL and the overlying Polar Water [Myers and Ribergaard, 2013]. Such studies make implicit assumptions concerning the seasonal cycle of the water masses, for example that the temperatures observed in summer are representative of the annual mean, or that the Irminger Water summer temperatures are the annual maxima [Holland et al., 2008;Myers and Ribergaard, 2013].
In this study, the seasonal cycle of the subsurface WL is explored using in situ observations from Argo floats as well as the range of platforms collated by the World Ocean Data Base (WODB), complemented with a large number of temperature measurements taken from marine-mammal borne sensors between 2002 and 2011. A number of studies have used temperature measurements from animal-borne instruments to derive useful physical oceanographic insights. The enhanced spatiotemporal sampling enabled by the animals residing for periods in relatively restricted areas frequently yield data sets that differ radically from those obtained using conventional techniques [Boehme et al., 2008a;2008b Figure S1 • Movie S1 2011; Sutherland et al., 2013;Roquet et al., 2013]. Here, we combine observations from sensors deployed on different marine-mammal species deployed by US, Canadian, Greenlandic, and Norwegian research programs. The resulting improvements in spatial coverage near the Greenland shelf and in temporal coverage in the non-summer months provide a unique opportunity to explore the seasonal variability of temperature in the subsurface WL. In addition, a hindcast of an eddy-resolving ocean model is used to investigate the causes of the observed variability.

Data and Analysis Procedure
Observations from several different sources collected between January 2002 and December 2011 are used in our analyses ( Figure 1): (i) >110,000 WODB 2013 profiles [Boyer et al., 2013] from various platforms including Ocean Station data, XBTs, drifting buoys, and gliders; (ii) >46,000 Delayed-Mode Quality Controlled profiles from drifting Argo floats, not duplicated in WODB; and (iii) three temperature data sets from marine-mammal borne instruments. The first marine mammal data set consists of 48,712 profiles from hooded seals, harp seals, grey seals, and beluga whales that were tagged by scientists from Fisheries and Oceans Canada at shelf sites around the Labrador and Irminger Seas between July 2003 and May 2008 [Andersen et al., 2009[Andersen et al., , 2013Bailleul et al., 2012]. Further details, including quality control procedures, are found in Grist et al. [2011] and references therein. The second marine mammal data set consists of over 109,000 observations from sensors deployed on narwhals between December 2005 and March 2008 in the Baffin Bay region [Laidre et al., 2010]. The third source consists of 6048 profiles obtained from hooded seals tagged in the Nordic Seas by scientists from the Norwegian Polar Institute between July 2007 and April 2009 [Isachsen et al., 2014].
Gridded temperature estimates and associated error bars were calculated at 1 × 1°resolution on standard World Ocean Atlas (WOA) levels using the objective analysis (OA) scheme and spatial correlation scales of Boehme and Send [2005]. For each month of the year, estimates were made at the middle of the month using a temporal decorrelation scale of 30 days and confining the source data to that from the corresponding month, the preceding month, and the following month (e.g., January, February, and March observations are the data source for the February analysis). As determining the seasonal variability was the goal of the study, the OA was conducted using data from each month irrespective of year. Interannual variability is thus neglected in the resulting estimates but reflected in the size of their errors. The source data available for August and February are shown in Figure 1. As Argo floats are confined to off-shelf areas and other WODB sources are greatly reduced in the winter, the importance of the enhanced coverage from marine mammals near the Greenland coast in non-summer months is evident. A recent gridded estimate of the mean annual temperature field used the on-shelf observations from marine-mammals borne sensors [Grist et al., 2011] and substantially improved the depiction of the temperature structure associated with regions boundary currents relative to the EN3 [Ingleby and Huddleson, 2007] and World Ocean Atlas [Levitus et al., 2009] gridded estimates. However, with regard to resolving the seasonal cycle, the data coverage is still incomplete in some areas. For example, in February, there are no data between 64°N and 66°N along the West Greenland coast (Figure 1b). Despite these constraints, increased winter coverage from marine mammals together with

Geophysical Research Letters
an OA scheme that accounts for the stronger spatial correlation along topographically steered currents [Boehme and Send, 2005] allows for an improved description of the seasonal cycle around Greenland.
Observational analysis is complemented by an examination of the 2001-2010 section of the 1978-2010 hindcast from the 1/12th degree NEMO ocean model [Madec, 2008]. The simulation, which is hindcast ORCA0083-N001 in the DRAKKAR data set of simulations [Barnier et al., 2006;DRAKKAR Group, 2007], is referred to as ORCA12 in this paper. The higher resolution "eddy-resolving" simulation is utilized for a more realistic depiction of boundary currents [e.g., Hecht and Smith, 2013]. Further details of ORCA12 are documented in Duchez et al. [2014]. ARIANE software [Blanke and Raynaud, 1997] is used to calculate water particle trajectories from ORCA12 in a manner similar to that reported by Jutzeler et al. [2014].

Results
The depth of the maximum annual mean temperature is shown in Figure 2a. Over most of the domain, the temperature maximum is at the surface or within the top 200 m (white in Figure 2a). However, a distinctive feature of the Greenland, Baffin Bay, and Labrador coast regions is the existence of a vertical temperature maximum between 200 and 600 m, below the cooler fresh surface boundary current [e.g., Tang et al., 2004;Fratantoni and Pickart, 2007 and supporting information Figure S1]. The same diagnostic calculated for the 10 year mean of ORCA12 shows similar characteristics but emphasizes the small horizontal scale associated with the sub-surface temperature maximum (Figure 2b). In Figures 2c and 2d seasonal ranges in temperature at the depth of the subsurface maxima in Figures 2a and 2b, respectively, are plotted. The magnitude is shown if, in the case of the observations, it exceeds the sum of the error associated with the coolest and warmest months and in the case of the model analysis, if it exceeds the sum of the interannual standard deviations of the coolest and warmest months. Along the Greenland shelf break, an annual cycle is detected in most of the areas where there is a subsurface maximum. The seasonal range in temperature is of the order 1-2°C. The locations at which a seasonal cycle is observed at the fixed level of 300 m are shown as magenta dots in Figure 2c to provide an indication of how common it is for there to be a detectable seasonal cycle at the depth of the WL. It is possible to detect a seasonal cycle near the shelf regions of Greenland, Baffin Bay, and Labrador. However, the seasonal variation in the subsurface temperature maxima near the Greenland shelf is not observed in much of the neighboring ocean at the same depth (Figure 2c).
We now examine if the phase of the seasonal cycle maximum is constant around the western and eastern shelf. Figures 3a and 3c show the seasonal temperature cycle of the WL around both the East and West Greenland shelf. Specifically, it shows the difference in observed monthly temperature from the annual mean temperature for the depths of 300 and 400 m, respectively. This is shown for each latitude band around the east and west coast of Greenland (where the grid cell chosen from each latitude is the one closest to the coast at that depth). It is clear that substantial spatial variability exists in the timing of the seasonal maximum. On the east coast, and south of 65°N on the west coast, the phase of the annual cycle appears to be relatively constant, with the temperature maximum occurring between September and December and the minimum occurring between February and June. Conversely, the northern region of the west coast exhibits a different phase. Typically, north of Davis Strait (65°N) the maximum occurs later (that is February or March), and the coolest months are between August and January. Although there appears to be little difference between the nature of the seasonal cycle at 300 and 400 m (Figures 3a and 3c), the signal is slightly less robust at 400 m. This is to be expected as the amount of available data diminishes with depth. The caveat to our analysis is that, while our data coverage is greatly improved, some areas are particularly sparsely covered and only have data from a single year. In such regions the error (which we associate with interannual variability) will be an underestimate; hence, it would be premature to assign temperature changes to purely seasonal as opposed to interannual variability, echoing the findings of previous studies [e.g., Straneo et al., 2012]. Nonetheless, there remains strong evidence for considerable spatial variability in the phase of the WL annual cycle, and this interpretation is supported by analysis of numerical model output (see below). In particular, the annual temperature maximum appears to occur later in the northern section of the west coast than at locations along the south-west and eastern coasts. An implication of this finding is that at many locations along the shelf, the warm layer often has its highest temperature during winter months when conventional (i.e., not from marine mammals) observations are few in number.
The observational analysis documented in Figures 3a and 3c has been repeated for ORCA12 (Figures 3b and  3d). The model analysis shows a similar magnitude in the range of the annual cycle. Like the observations, the peak month occurs between August and November on the east coast and south of 65°N on the west coast, with a later peak to the north on the west coast. The model's higher resolution and complete fields are better able to depict a gradual northward propagation of the peak month along the west coast from October at 61°N to May at 75°N. The annual progression of the warmest waters around at the Greenland shelf break is further illustrated dynamically in the supporting information.
An indication of the significance of the seasonal cycle is provided in Figure 3e, which shows the ratio of the seasonal range in temperature compared to the sum of the error of the warmest and coolest months for the observational analysis. For the model analysis the ratio of the seasonal range to the sum of the interannual standard deviations of the coolest and warmest month is used. The strongest seasonal signal (relative to the error range or interannual variability) is found in south-east Greenland. This signal reduces with increasing latitude along the west coast. Along the east coast north of the Denmark Strait the strength of the seasonal cycle diminishes markedly, to the point where it is about half of the error range or interannual variability in the observations and model, respectively.
The seasonal cycle and its northward propagation on the west coast are more evident in the model north of the Davis Strait at 300 m than at 400 m (Figures 3b and 3d). The suggestion is that there is a surface influence on the annual cycle in this region that is less evident on other parts of the coast. This is consistent with a large seasonal range in both surface salinity and temperature in the region associated with the seasonal advance and retreat of sea ice. An additional surface influence on the seasonal cycle in this region may also be important in producing a larger annual cycle in the vicinity of the Davis Strait, which is evident in both observations (Figure 2c, Figures 3a and 3c) and the model (Figures 2d and 3b).
To elucidate the cause of the spatial variability in the annual cycle, in particular for the south-east and west coast of Greenland where the warmest temperatures occur in winter, we conducted a particle trajectory analysis using ARIANE tracking software [Blanke and Raynaud, 1997]. For this, we seeded the model with particles in the subsurface temperature maxima (200-600 m) in the Irminger Basin (Figure 4a), a location close to that at which Atlantic Water at the surface descends to become separated from the direct influence of solar  heating. We then released these particles on 2 September 2000 and tracked them for 15 months. Particles swiftly entered, circumnavigated, and left the Labrador Sea. Consistent with observations and previous model studies [Lilly et al., 2003;Chanut et al., 2008], Cape Desolation (48°W, 60°N) was a favored location for eddying departures from this direct route. About 3% of the particles continued through the Davis Strait into Baffin Bay along the West Greenland coast. We repeated the analysis for years [2001][2002][2003][2004][2005][2006][2007][2008][2009], and these trajectories were related to the spatial variability of the phase in the annual cycle of WL temperature. For each time step we relate the travel time of the particle from the Irminger Basin to the seasonal cycle at that grid point and, in particular, to the month in which the maximum temperature occurs. Figures 4c and 4d show how the month in which the maximum temperature occurs is related to the particle travel time from the Irminger Sea. (We note that this is only relevant to regions downstream of the Irminger Sea, so that other processes must determine seasonal variability in and to the north of the Denmark Strait.) There is a clear indication that the longer the transit time from the Irminger Sea, the later the seasonal temperature maximum is reached. That is, the phase of the WL temperature maximum reflects the time that has elapsed since the parcel of water last reached its seasonal maximum at the surface. Although we emphasize that more winter sub-surface observations are needed to fully constrain the annual cycle of the WL and to examine other processes that might be involved, our current explanation is consistent with both our model analysis and the available observations.

Summary and Conclusions
A unique set of observations has been used to improve our description of the sub-surface seasonal cycle of water masses around Greenland between 2002 and 2011. The analysis shows a sub-surface temperature maximum associated with the coastal boundary current that has a significant seasonal cycle, with a magnitude of the order 1-2°C. In context, this is similar in magnitude to the 1.6°C increase in spring and summer temperatures between 1995 and 1997 for this layer near Disko Bay, in West Greenland [Holland et al., 2008;Hansen et al., 2012]. This interannual change is thought to have played a significant role in accelerating the melt of Jakobshavn Isbrae outlet glacier. Attempts to quantify the relationship between ocean temperature and glacier melt have yielded a range of results. However, a conservative example suggests that a 1°C temperature increase would be associated with at least a doubling of the ice-melt rate [see Holland et al., 2009]. We emphasize, that our analysis does not directly address the extent that the seasonally varying warm layer accesses outlet glaciers to reach the Greenland Ice Sheet, as this question will depend on fjord-specific processes [e.g., Inall et al., 2014;Sutherland and Straneo, 2012;Johnson et al., 2011]. However, as a point of reference, a rare "winter" profile in the vicinity of Helheim Glacier (south-east Greenland) recorded WL temperatures that were 1°warmer in March 2010 than the previous summer (August-September 2009) . It is unclear if these summer-winter differences were due to interannual, or seasonal variability . Nonetheless, the magnitude of the seasonal temperature range found here implies that it would be important for any assessment of interannual WL changes to take into account the time of year at which observations were obtained.
There is considerable spatial variability in the phase of the WL seasonal cycle around Greenland. In particular, the annual peak in the WL temperature occurs significantly later to the north along the west coast. Particle tracking analysis in a high-resolution ocean model suggests that this spatial variability is determined by the travel time of water from the subduction sites in the Irminger Basin.
Finally, in addition to a significant seasonal temperature range, our results suggest that in some regions the WL may often be warmest in winter, i.e., the most poorly observed period of the year. A recent study of two Greenland fjords suggested that this is also the time of year when fjord-shelf exchanges allow warm subsurface ocean waters to most strongly influence glacier melt rate [Jackson et al. 2014]. This highlights the need for year-round subsurface temperature measurements, for a complete description of the WL seasonality, and in particular to ensure that the ice-melting potential of the WL is not underestimated.