The Denmark Strait Overflow (DSO) is an important contributor to the lower limb of the Atlantic meridional overturning circulation (AMOC). Determining DSO formation and its pathways is not only important for local oceanography but also critical to estimating the state and variability of the AMOC. Despite prior attempts to understand the DSO sources, its upstream pathways and circulation remain uncertain due to short-term (3–5 days) variability. This makes it challenging to study the DSO from observations. Given this complexity, this study maps the upstream pathways and along-pathway changes in its water properties, using Lagrangian backtracking of the DSO sources in a realistic numerical ocean simulation. The Lagrangian pathways confirm that several branches contribute to the DSO from the north such as the East Greenland Current (EGC), the separated EGC (sEGC), and the North Icelandic Jet (NIJ). Moreover, the model results reveal additional pathways from south of Iceland, which supplied over 16% of the DSO annually and over 25% of the DSO during winter of 2008, when the NAO index was positive. The southern contribution is about 34% by the end of March. The southern pathways mark a more direct route from the near-surface subpolar North Atlantic to the North Atlantic Deep Water (NADW), and needs to be explored further, with in situ observations.
The dense waters entering the Atlantic Ocean from the northern latitudes spill across the Greenland–Iceland–Scotland ridges through two main passages: the Denmark Strait and Faroe Bank Channel. The two overflows merge after entrainment to form the densest constituent of North Atlantic Deep Water (NADW), which is the main contribution to the deep branch of the North Atlantic meridional overturning circulation (AMOC) (Swift et al. 1980; Pratt and Whitehead 2008; Østerhus et al. 2019). The focus of this study is the origins of the dense water that cascades over Denmark Strait, known as Denmark Strait Overflow (DSO). The DSO is a major export route for dense waters in the Nordic Seas, feeding the deep western boundary current (Dickson and Brown 1994). The mean volume transport of the DSO is approximately 3.2 ± 0.5 Sv (1 Sv ≡ 106 m3 s−1) (Macrander et al. 2007; Jochumsen et al. 2017). Determining the overflow origins and pathways is important for estimating the state and variability of the AMOC and hence the climate system.
Furthermore, water flowing across a topographic barrier is a ubiquitous process in the ocean and the DSO is a prime example (Pratt and Whitehead 2008). The flows of water through narrow straits, canyons and over topographic features are similar in many ways; these dense overflows undergo significant mixing, entrain ambient fluid, which dilutes the temperature and salinity signal of the water, and increases the volume flow (Price and Baringer 1994; Legg 2012). Although the mixing in overflows is highly localized (North et al. 2018), it plays a significant role in influencing the large-scale ocean circulation (Koszalka et al. 2017). Studying the DSO helps us understand common features of climatologically important overflow processes in other parts of the global ocean.
The fate and downstream evolution of the overflow and its variability have been well studied through observations (Macrander et al. 2005; Tanhua et al. 2008) and realistic modeling both in Eulerian (Köhl et al. 2007; Haine et al. 2008; Almansi et al. 2017) and Lagrangian representations of the flow field (Koszalka et al. 2013; von Appen et al. 2014). The formation, sources, and pathways of the DSO have also been studied (Rudels et al. 2002; Köhl et al. 2007). However, despite various proposed circulation schemes based on hydrographic transects (Harden et al. 2016; Våge et al. 2013), the sources and details of the DSO upstream pathways are still uncertain due to mesoscale eddy features, short time scale (3–5 day) variability, mixing, and water mass transformation (Almansi et al. 2017; Spall et al. 2019; Moritz et al. 2019).
There are two main hypothesized sources of the DSO. The first source is the Atlantic-origin water (warmer than 0°C) resulting from transformation in the eastern part of the Nordic Seas (Mauritzen 1996) and the second source is the Arctic-origin water (colder than 0°C) resulting from transformation in the Greenland and Iceland Seas (Swift and Aagaard 1981). The first source is transported primarily by the East Greenland Current (EGC) that flows along the east Greenland shelfbreak. The EGC bifurcates upstream of the Denmark Strait, with one branch continuing along the east Greenland shelfbreak and a separated branch called the separated EGC (sEGC) (Våge et al. 2013) located farther offshore, near the base of the Iceland slope. The contribution of the EGC to the DSO was recognized through hydrography, age analysis, and isotope ratio measurements (Swift et al. 1980; Smethie and Swift 1989; Rudels et al. 2002; Tanhua et al. 2005b).
The second DSO source is carried by the North Icelandic Jet (NIJ), which is centered near the 650 m isobath approaching Denmark Strait from the Iceland slope (Jónsson and Valdimarsson 2004; Våge et al. 2011; Semper et al. 2019). Chemical oceanography studies suggest that there are other water masses contributing to the DSO (Jeansson et al. 2008; Tanhua et al. 2005a), but their percentage of contribution appears to be small (Mastropole et al. 2017). The NIJ and EGC currents are observed to be less distinguishable from each other closer to the Denmark Strait (Jónsson and Valdimarsson 2004). To distinguish the currents upstream of Denmark Strait, Harden et al. (2016) analyzed data from a densely instrumented mooring array deployed from September 2011 to July 2012 at the Kögur section, which is roughly 200 km north of Denmark Strait (see Fig. 1). The sEGC is typically found near the base of the Iceland slope and is often difficult to distinguish from the NIJ. In the yearly-mean hydrographic and orthogonal velocity at the Kögur section, the NIJ and sEGC appear as a single feature (Harden et al. 2016). The NIJ also interacts with the (inshore) northward-flowing relatively warm and saline North Icelandic Irminger Current (NIIC) when the bottom topography steers the two currents close together (Pickart et al. 2017).
These studies of the DSO sources and pathways have been mostly based on Eulerian measurements and moorings at historical hydrographic sections. Although they estimate transport and hydrographic properties, they do not conclusively show the connectivity between sections and the evolution of individual currents from one section to another (de Jong et al. 2018). The vertical sections are distant from each other, which makes it challenging to trace the origins of the DSO water through the currents that carry them from one vertical section to the other (Tanhua et al. 2008). There is only one Lagrangian study from observations that investigates pathways of dense water using acoustically tracked, high-resolution RAFOS float trajectories (de Jong et al. 2018). These float trajectories revealed new information on the connection between the EGC and the NIJ, and on the subsurface circulation in the Iceland Sea. But they are relatively few in number, and could not be tracked through the Strait itself, leaving some gaps in our knowledge of pathways leading up to the sill (de Jong et al. 2018). Given the complexity of observing pathways leading to the DSO, it is sensible to explore them in a realistic circulation model, in particular to perform backward Lagrangian particle tracking to investigate the origins of the dense waters flowing over the sill at Denmark Strait. Here we apply this technique using a well-studied regional model (Almansi et al. 2017) to trace the near-field origins of the overflow. The available model run, described in detail in the following section, is 15 months long, limiting our ability to study long term trends and seasonality with this particular simulation. Nevertheless, short-term evolution of the Lagrangian particle pathways and hydrographic properties leading up to the sill can be revealed in detail.
The words origin and the DSO need to be carefully defined. The question of the origin of a water mass is time and space dependent. The farther back in time, the more distributed the origin of a water mass in space becomes. It is important to determine how far back in time we are interested in the DSO sources. In this paper, we define the word origin as the farthest upstream location from the Denmark Strait in our computational domain within a three month time window, which is the shortest length within the time boundaries of our model run that reveals geographically distinct DSO sources. The DSO needs to be defined as well. In the literature various criteria based on temperature or density have been used (Cooper 1955; Tanhua et al. 2008; Behrens et al. 2017). In this study, we use the most common definition by Dickson and Brown (1994): The DSO is the water with σθ ≥ 27.8 kg m−3 at the Látrabjarg sill section (see Fig. 1) that flows southward after cascading over the Denmark Strait (by sill section we mean the saddle point in the bottom topography that the Látrabjarg section transects). In this paper, we address the following questions: What are the origins and pathways of the DSO? How do the model pathways compare with observed pathways? How do the temperature, salinity, depth and density properties evolve along the DSO pathways? How do the Lagrangian pathways improve our understanding of circulation in the area? How do the main currents interact? What is the best schematic representation of the DSO pathways?
The paper is organized as follows: section 2 presents the details of the numerical model, the particle tracking code, and the observational data used for this study. In section 3 model results are compared with observations, the Lagrangian particle tracking results are presented, and the DSO particle properties, trajectories, and preferred pathways are investigated. A summary and discussion are provided in section 4.
a. Numerical model configuration
The flow field is calculated using the hydrostatic version of the Massachusetts Institute of Technology General Circulation Model (MITgcm) developed by Marshall et al. (1997). It solves the Navier–Stokes equations under the Boussinesq approximation for an incompressible fluid in a rotating frame of reference. The model domain is configured for the Denmark Strait and it includes the sub-Arctic region from the Greenland Sea to Cape Farewell (Fig. 1). The model simulation period is from 1 September 2007 to 30 November 2008. The temporal resolution of the model is 30 s but the output is stored every 6 h. The horizontal grid resolution is 2 km in the center of the domain from 60° to 71°N (Fig. 1), and it decreases to 4 km moving toward the boundaries. The vertical grid resolution decreases from 2 to 15 m in the upper 120 m and is 15 m thereafter.
The model configuration is identical to that of Almansi et al. (2017), except that the atmospheric forcing at the surface (air temperature, specific humidity, wind, evaporation, precipitation, and radiation) is based on three hourly 15 km resolution fields from the regional Arctic System Reanalysis product (ASRv2; Bromwich et al. 2018). The model is initialized using the global 1/12° reanalysis of the Hybrid Coordinate Ocean Model (HYCOM) + Navy Coupled Ocean Data Assimilation (NCODA) (Cummings and Smedstad 2014), and a coupled ocean–sea ice data assimilation product for the North Atlantic Ocean and Arctic called the TOPAZv4 monthly reanalysis (Sakov et al. 2012). At the open boundaries, time series of velocity, temperature, and salinity from HYCOM + NCODA are used. The sea surface temperature is relaxed to the satellite data from the Operational Sea Surface Temperature and Sea Ice Analysis (OSTIA) global product with a time scale of 5 days (Donlon et al. 2012). The oceanic component is coupled with the MITgcm sea ice model (Losch et al. 2010). Freshwater forcing from runoff, and solid and liquid discharge is used along the Greenland coast from a combination of climate models, remote sensing, and terrestrial data (Noël et al. 2016; Bamber et al. 2012).
b. Passive Lagrangian particle tracking
Lagrangian trajectories of virtual particles are calculated using a three-dimensional particle tracking package developed by Koszalka et al. (2013) and Gelderloos et al. (2016) in MATLAB. For this study, the code is extended to compute the trajectories at a single depth level in isobaric mode as well. The algorithm employs a MATLAB built-in ordinary differential equation (ODE) solver to solve for particle position at each time step. The particles at each time step are advanced with the two- or three-dimensional velocity field linearly interpolated to the instantaneous position of the particle. The code does not have explicit diffusion, as we assume all the information about the flow is contained in the velocity field output of the MITgcm model. The velocity component normal to boundaries is zero and the particles slide along the solid boundaries of the domain. The evolution of tracer fields such as salinity and temperature along the Lagrangian paths are obtained by nearest-neighbor interpolation of the model output to the particle positions. The model equation of state (Jackett and McDougall 1995) is then used to compute the density of the particles.
Several observational datasets are compared to the model output to build confidence in the realism of the model. Shipboard data, moorings, and hydrographic sections are used to perform the comparison in an Eulerian frame, while RAFOS float trajectories and properties (de Jong et al. 2018) are used to perform the comparison in a Lagrangian frame.
1) Kögur mooring array
To perform model–data comparison upstream of Denmark Strait, data from the Kögur mooring array is used. Harden et al. (2016) analyzed results from a year-long densely instrumented mooring array upstream of the Denmark Strait, across the Blosseville Basin and along the Kögur section (shown in Fig. 1). Each of the 12 moorings was equipped with instruments measuring temperature, salinity, pressure, and current velocity. The data coverage is from 29 August 2011 to 30 July 2012. The hydrographic and the velocity data are interpolated into a structured grid with a spatial resolution of 8 km in the horizontal and 50 m in the vertical direction. The temporal resolution of the gridded product is 8 h. Harden et al. (2016) explain the details of the data and interpolation scheme used for gridding the data in their appendixes A and B.
2) RAFOS float and shipboard data
To make a direct comparison between the model Lagrangian trajectories and the floats, we use isobaric RAFOS floats deployed near 70°N on a zonal section in July 2014 (de Jong et al. 2018). The complete RAFOS dataset contained 52 floats, but only the 11 floats that drifted near the Denmark Strait are considered here. The information about these floats studied here is listed in Table 1. They are isobaric subsurface drifters ballasted for the depth of interest and the hydrographic properties of the study region (Rossby et al. 1986). The floats were tracked by acoustic signals on a daily schedule from six sound sources moored in the Iceland Sea (de Jong et al. 2018). They provide pressure, temperature, and arrival times of acoustic signals along their trajectories. They were deployed in water with σθ > 28.0 kg m−3 in order to be embedded in the overflow water. At the RAFOS float deployment locations, shipboard conductivity–temperature–depth (CTD) measurements were taken. The CTD data are used to make a direct comparison of temperature and salinity at the initial positions of the RAFOS floats and the modeled Lagrangian particles.
3) Hydrographic data
Data provided by the Marine and Freshwater Research Institute (Malmberg and Valdimarsson 2003) are used to verify the realism of the model at the Látrabjarg section. The data are from a monitoring project in Icelandic waters that has been ongoing since 1990 (Malmberg and Valdimarsson 2003), and the research on the Icelandic waters hydrographic variability and environmental effects on Icelandic fisheries (Palsson et al. 2012). In addition, hydrographic profiles from the years 1980 to 2017 on the Iceland shelf are collected from various databases such as Unified Database for Arctic and Subarctic Hydrography (Behrendt et al. 2017), World Ocean Database (NOAA 2013), and International Council for the Exploration of the Seas (ICES 2006). These data are used to better understand the characteristics of water on the Iceland shelf.
a. Comparison with observations
1) Moorings at Kögur
The model results are compared with the year-long mean hydrographic properties and orthogonal velocity from the Harden et al. (2016) Kögur mooring deployment (Fig. 2). Although the year of observation (2011–12) does not coincide with the modeled year (2007–08), the hydrographic structures from the model (Figs. 2a,c) match the observations (Figs. 2b,d) well. The transition from warm and salty North Atlantic water with subtropical origin (referred to as Irminger Water) to the cold and fresh Polar Surface Water in the upper layer is evident in both the model and observations along the section from east to west. The cold and fresh Polar Surface Water extends across the section from the west in both cases. The Irminger Water region extends somewhat further in the observations than in the model.
Harden et al. (2016) observed two water masses below the 27.8 kg m−3 isopycnal: the Recirculated Atlantic Water and Arctic-origin Water. The Recirculated Atlantic Water, defined by potential temperature θ > 0°C and salinity S > 34.9, is observed between depths of 300 and 800 m. The Arctic-origin Water with θ < 0°C and σθ > 28 kg m−3 is found below 800 m closer to Greenland, but it can also be seen at shallower depths on the Iceland slope. These two water masses can be seen in the model, too; however, the model mean potential temperature is higher by about 0.2°C in the deeper part of the section (below 1200 m), and also the Arctic-origin Water is less extended to the Iceland side of the Strait.
The isopycnal structure in the model is similar to that of the observations in the middle and western parts of the section. However, the slope of isopycnals is steeper toward the Iceland shelf in the observations, which has dynamical implications for the strength of the NIJ. The model has a lower shear and slightly smaller orthogonal velocity in the year-long mean field as shown in Figs. 2e and 2f. The individual snapshots (not shown) have better consistency with observations in isopycnal structure. Moreover, the isopycnals are slightly deeper in the model, consistent with the deep warm bias found at Denmark Strait by Almansi et al. (2017).
The orthogonal velocity in both Figs. 2e and 2f consists of two main equatorward currents, one on the Greenland side and the other on the Iceland side. The model agrees well with the observations in capturing the surface-intensified East Greenland Current on the Greenland shelf break. The EGC maximum velocity at its core is 0.3 m s−1 in both model and observations. As Harden et al. (2016) describe, on the Iceland side there are two distinct currents that appear as a single feature in the mean field: the NIJ, a middepth intensified flow positioned near the 650 m isobath, and the sEGC, a surface-intensified current located seaward of the NIJ. In the annual mean field, the NIJ and the sEGC have one core in both observations and model. However, the two currents appear as distinct features with separated cores in some individual snapshots (not shown). Using orthogonal velocity to distinguish the two currents can be difficult due to the transient nature of the NIJ and sEGC. Below, the Lagrangian particles help us understand the pathways and variability of the two currents.
Besides the equatorward currents, there are two regions with mean poleward velocity. One is the NIIC on the Iceland shelf that carries warm subtropical-origin water into the Nordic Seas. The other is a weak flow on the Iceland slope that Harden et al. (2016) attributed to a signature of recirculation of water that previously passed through the section. The model orthogonal velocities are consistent with these observations (Figs. 2e,f), and the model Lagrangian particle trajectories confirm the recirculation (see section 3b below).
2) Comparison with float and ship data
Figure 3 shows a comparison between data from the shipboard CTD survey on the release date of the RAFOS floats with the model results. The CTD survey took place in July 2014. As the observations comprise a single snapshot in time, in order for the comparison to be relevant, the mean potential temperature and salinity in the month of July from the model is used. The CTD data is also compared with daily snapshots in the month of July from the model (not shown), and no significant difference with the mean section was observed. The depth and overall structure of the isopycnals in the model are similar to the observations. The deepening and shoaling of 28.05 kg m−3 isopycnal varies in daily snapshots, but the average depth of the isopycnal is consistent with the observations. The surface temperature along the section is also consistent. East of 17°W the water column temperature and salinity structure are in good agreement with the data, but the model is slightly fresher. However, the subsurface temperature and salinity near the Greenland shelf from 50 to 380 m depth is higher by 1.15 ± 0.77°C and 0.4 ± 0.27 in the observations. Between 380 and 750 m, which includes the depth range where the floats are released, the model agrees well with the observations.
To determine how the hydrographic properties evolve in time, we also compare the model with observations in a Lagrangian framework. A set of 400 particles seeded at 70°N are tracked forward in time in isobaric mode at 600 m depth. This mimics the constraint placed on the RAFOS float trajectories listed in Table 1. The depth of 600 m is used because it is close to the average depth of the floats (560 m). Figure 4 shows the float trajectories and simulated Lagrangian particle trajectories and the evolution of temperature and pressure along these paths. The RAFOS trajectories are compared with the modeled particle trajectories for 139 days during the same time of the year from mid-July to the end of November, which was the last day at which the model outputs are available.
The observed float trajectories are within the range of the modeled particle trajectories. The EGC is visible in both the observed and modeled particle trajectories. The model particles also reveal a northward flow offshore of the EGC. This pathway is due to intermittent eddies visible in the Eulerian velocity field at 600 m depth, which was not likely to be captured by only the two floats released close to it. In Figs. 4a and 4b, the evolution of temperature and pressure for the 11 selected floats are shown. The spikes in one float’s (RFS1209) pressure record occurred when that float was traveling along the east slope of Greenland, embedded in the EGC (Fig. 4b). These pressure excursions are coincident with the float measuring warmer temperatures (~0.5°C warmer, Fig. 4a), and were likely due to the float being pushed upslope by strong upwelling. Evidence of upslope excursions, measured by other floats included in the complete RAFOS dataset, were found in other locations in the Iceland Sea, and most prevalent along the northwest slope of Iceland (de Jong et al. 2018). Some of the float temperature measurements fall within the envelope of model particle temperatures. However, some others are colder than model particles by 0.1 ± 0.045°C, which is well within the range of observed interannual variability (Lauvset et al. 2018). The RAFOS floats were not equipped with conductivity sensors; therefore, salinity evolution along the float trajectories is not possible.
3) Comparison at the Látrabjarg section and evidence of dense water on the Iceland shelf
The fidelity of the model has already been studied at the Látrabjarg section by Almansi et al. (2017), who showed that the model hydrography resembles shipboard CTD observations between 1990 and 2012 at Látrabjarg (Fig. 4 in Almansi et al. 2017). That study concluded that the model captures all of the major currents (the NIIC, the NIJ, and the EGC) along the section. The hydrographic structures are shown to be consistent with the detailed observational studies performed by Mastropole et al. (2017) at Denmark Strait. The comparison by Almansi et al. (2017) was focused on the western side of the Denmark Strait where the bottom intensified DSO is located. The DSO is the dense (σθ ≥ 27.8 kg m−3) southward flow banked against the Greenland side of the trough in Denmark Strait. However, observations show evidence for the existence of dense water (satisfying the DSO density criterion) on the Iceland shelf occasionally in winter (Våge et al. 2015). In Fig. 5 the model-mean section in February is compared with observations from February 1997. They both show steep isopycnals from 27.6 to 27.8 kg m−3 on the Iceland shelf. The 27.8 kg m−3 isopycnal on the Iceland shelf lies farther westward in the model than the observations by approximately 70 km, and it continues through the trough and changes slope on the Greenland side. The same isopycnal structure is evident from the observations but for the 27.6 and 27.7 kg m−3 isopycnals. The continuity of isopycnals from the Iceland shelf to the trough is seen from January to March in the model with the isopycnals being steep and outcropping to the surface by the end of February and early March. The model shows biases in the surface salinity (about +0.1 or less), which could be the potential reason why the dense water at the shelf appears to occupy a larger area than the observations. The dense water on the Iceland shelf has been observed in other years as well; however, it is sparse in winter. Our exploration of the Unified Database for Arctic and Subarctic Hydrography (Behrendt et al. 2017), World Ocean Database (WOD 2013), and International Council for the Exploration of the Seas (ICES 2006) database revealed that out of 3700 individual profiles near the Iceland shelf (within about 100 km periphery of Iceland’s west coast) spanning from 1980 to 2018, about 800 of them were collected in the winter (JFM). Despite the limited winter observations, 20 out of 200 profiles show the dense water (σθ ≥ 27.8 kg m−3) on the shelf along Látrabjarg section (in depths shallower than 230 m). These 20 profiles are spread over the years 1981–84, 1990, 1993, and 1995, which have relatively high North Atlantic Oscillation (NAO) index.
b. Backward tracking of particles released at Látrabjarg
To determine the near-field origins and pathways of the DSO, passive particles are released at the Látrabjarg section at the end of each month from 30 November 2007 to 31 October 2008. There is an ensemble of 12 particle release experiments (see Table 2). Each ensemble member is tracked backward in time for 3 months in three-dimensional space. This length of time was chosen because it is the shortest time period at which the geographical distribution of pathways, the water masses and properties of the DSO sources can be identified. We explored backtracking of longer time period (4, 5, and 6 months) to be sure 3 months was an appropriate time scale. The particle seeding resolution is 0.5 km in the horizontal and 25 m in the vertical direction. All particles are seeded below the 27.8 isopycnal satisfying the conventional overflow threshold in the literature, which we refer to as dense particles (see Table 2). The total number of backward tracked dense particles over all ensemble members is 18 399.
The backward tracking alone does not reveal what fraction of the dense water eventually cascades over Denmark Strait. Therefore, each ensemble member is also advanced forwards for 30 days. Particles that appear south of Denmark Strait after 30 days are considered to have participated in the DSO, regardless of their final density. The total number of these so-called DSO particles is 13 708, which is 74.5% of all the dense particles. It is found that the ratio of the DSO to the dense particles varies between ensemble members. The information about each ensemble member is listed in Table 2. Notice that although the number of dense particles is highest in February and March, the number of DSO particles does not increase proportionally. In fact, the ratio of the DSO to the dense particles decreases in March. This is because the Irminger Current is colder in winter and spring (by 1.8°C; Mastropole et al. 2017), and becomes denser as it moves northward until some of it satisfies the σθ ≥ 27.8 kg m−3 criterion. This dense water continues northward and mixes with surrounding water, namely, the southward flowing EGC and NIJ, and some fraction of it cascades over Denmark Strait within 30 days. This splitting and retroflection of the Irminger Current in the vicinity of the Denmark Strait was demonstrated by Valdimarsson and Malmberg (1999) and discussed in Rudels et al. (2002) as well. However, in that study, the location of the splitting was not fully identified. The Lagrangian trajectories in the following section reveal the time-varying location of this bifurcation. For the rest of the paper, the DSO particle trajectories and hydrographic properties are discussed; studying the properties and pathways of dense water that does not cascade over the Denmark Strait (the difference between columns 5 and 6 in Table 2) is beyond the scope of this research.
The backward trajectories of the DSO particles reveal different sources for the DSO. Figure 6a shows the geographical distribution of the particles three months before reaching the Látrabjarg section, at their origin, color coded by ensemble member (see also Table 2). The DSO particles come from both north and south of the Denmark Strait, with 83.6% originating from the north. The geographical distribution of the DSO particles at their origins are nearly identical throughout the year between all ensemble members, but the fractions from north and south of the sill vary. To study this variability for each ensemble member, the DSO particles are categorized into different subsets, based on geographical origin. They are naturally broken down into the northern- and southern-origin subsets. The southern origin subset is further separated into two groups, depending on whether they arrive at the Icelandic side of the Látrabjarg section east or west of 26.5°W (star marker on Fig. 6a). Figure 6b shows the variation in the fraction of the DSO particles flowing from the north (abbreviated as N), the south and west of 26.5°W (abbreviated as S), and the south and east of 26.5°W, on the Iceland shelf (abbreviated as ISh) for each ensemble member. In summary, the DSO particles are categorized into N, S, and ISh subsets based on where they originate. The trajectories and properties of the particles in each individual subset are now discussed in detail.
The fraction of the DSO particles from the south (S + ISh), which is persistent throughout the year, is highest at the end of March (ensemble member V), with a mean contribution of 16.3%. The contribution from the ISh subset is present only between 31 December to 30 April (ensemble members III, IV, V, and VI) with a peak of 12.8% of DSO particles at the end of February. The depth–longitude positions of the DSO particles at the Látrabjarg section can be seen in Fig. 7, color coded by the origin subsets. Note that the S and N subsets both occupy the western side of the section and the ISh subset, separated from the others, occupies the eastern side of the section. The presence of DSO particles in depth–longitude space is consistent with the observations of dense water at the Látrabjarg section, discussed earlier (Fig. 5).
The DSO particles are tracked backward from their release point at the Látrabjarg section. However, for the sake of visualization, analysis, and discussion their trajectories are plotted forward in time to the Látrabjarg section. Figure 8 shows the trajectories of the DSO particles approaching Denmark Strait. To avoid clutter, instead of plotting the full trajectories of individual particles, 10-day-long pathlines of every 5th DSO particle in each subset are plotted. The visualized trajectories are representative of trajectories of all the DSO particles. Note that, not all the DSO particle trajectories at all times look like the trajectories in Fig. 8 (see the supplemental animations). The northern pathway (in blue) is present in all months. The DSO particle trajectories in subset N depict the conventional route of the DSO along the east Greenland shelf break; that is, the EGC. The bifurcation of the EGC is also evident from the trajectories (and better visualized in the animations): sometimes the separation takes place north of the Kögur section (≈68°N in ensemble members I, IV, VI, and IX), and other times the separation takes place south of it within the southern part of the Blosseville Basin (≈67°N in ensemble members V and VIII). The particle trajectories show that the separation from the EGC takes place in multiple locations along the east Greenland shelf (mostly south of ≈70°N). The trajectories of DSO particles in the N subset show that sometimes the bifurcated branches reunite with the EGC (e.g., in ensemble members III and IX), and other times they shift to the middle of the basin and continue south. Concurrently, the NIJ is evident from the Lagrangian particle trajectories on the northwest Iceland slope: sometimes very close to the Kolbeinsey Ridge (e.g., in ensemble members I, IV, X, XI, and XI) and sometimes about 130 km southwest of the ridge, closer to the Kögur section (e.g., in ensemble members V–VII). The DSO particle trajectories also reveal that sometimes the sEGC swings toward the Iceland shelf, interacts with the NIJ and the two currents merge and continue south along the Iceland shelf. This interaction is clearly visible from the animations of particle trajectories, provided as supplemental material. Notice that, although the particles follow the currents, they do not necessarily stay in one current at all times. In other words, the Lagrangian trajectories are in principle distinct from the currents (velocity maxima). The separation from the EGC and its interaction with the NIJ is also observed by Våge et al. (2013). The interaction happens at different times and places along the path for each ensemble member. As the simulated trajectories cover only one year, detecting the sEGC and the separation frequency is hard; nevertheless, there is no strong evidence of seasonality. Besides, as discussed above, the sEGC appears to be intermittent.
The DSO particles in the S and ISh subsets have a similar spatial distribution at their origins for ensemble members III–VI (Fig. 8). The majority of particles in the ISh subsets, however, move along shallower isobaths (depth ≤ 300 m, Fig. 12c) than the particles in the S subsets (depth ≈ 400–800 m, Fig. 11c). Particles in both subsets flow northward toward the Denmark Strait. The particles in the S subset then encounter the southward flowing particles in the N subset, retroflect, and flow south again, to reach the western side of the Látrabjarg section. In contrast, the northward flowing ISh particles arrive at the eastern side of the Látrabjarg section and predominantly continue northward with the NIIC. Most of these particles return south within 20 days of crossing the Látrabjarg section in the forward run. Figure 9 shows the particle trajectories at the end of the forward run for each ensemble member. The particles in the S and N subsets cascade over Denmark Strait and trace multiple recirculations in the Kangerdlugssuaq Trough. These cyclonic recirculations are also detected by Koszalka et al. (2013). Finally, the recirculated particles continue to flow south along the east Greenland shelf following the traditional DSO path (Dickson and Brown 1994). Similarly, the particles in the ISh subsets cascade over Denmark Strait but with a time lag of about 20 days compared to the other subsets. They then follow a similar path to the particles from the N and S subsets. Notice that very close to the Látrabjarg section immediately before and after cascading the Denmark Strait, the particles from all subsets follow nearly identical pathways. If the particles of each subset were not color coded differently, they would be indistinguishable. Here, the backward Lagrangian particle tracking elucidates the origins and trajectories of the DSO even in the regions of high mixing.
The evolution of hydrographic properties of DSO particles in the N, S, and ISh subsets are shown in Figs. 10–12, respectively. The mean temperature and salinity of the DSO particles in the N subset over all ensemble members (dash–dotted black lines) are nearly constant, with a slight increase (, ) approaching the Denmark Strait sill (from −90 to 0 days). After cascading there is an increase in their temperature and salinity due to entrainment of the surrounding warmer and saltier water mass. The range of temperature before cascading is narrower (from −0.11° to 2°C) and it gets wider after cascading the Denmark Strait (from 1° to 4.5°C). The range of salinity however, is wider at the origin (34.3–34.9) and narrows approaching the Denmark Strait and after the cascade (34.7–35). The abrupt change is also evident in their potential density and depth. The DSO particles’ mean potential density is nearly constant (less than 0.06 kg m−3 increase along the path) with the value of from the origin until they reach the Denmark Strait. After cascading, their mean potential density decreases by about 0.15 kg m−3. The depth of the DSO particles is also nearly constant approaching the Denmark Strait and they sink (900 m on average) as they cascade over the cataract. Notice that the DSO particles in ensemble members VI–IX are shallower than particles in ensemble members I–V at their origin and along their path to the Denmark Strait, but after cascading the DSO particles in all ensemble members experience similar initial deepening at least for the first 15 days (from approximately 360 to 1000 m).
The properties of the DSO particles in the S subset evolve differently (Fig. 11). Initially, the particle properties are warmer and saltier than the DSO particles in the N subset at their origin location, but the S subset particles cool and freshen as they move toward the Denmark Strait. The rate of change of temperature is steeper for the I–V than for the VI–XII ensemble members . The rate of change of salinity over all ensemble members is similar with an average of 0.15 decrease approaching the Denmark Strait from the origin. The DSO particles in the S subset are shallower and lighter at their origin in comparison to the DSO particles in the N subset but they rapidly densify while meeting the southward flowing DSO particles in the N subset (cf. Figs. 10a,c with Figs. 11a,c). After cascading, the DSO particles in the S subset have slightly higher average potential density (0.08 kg m−3) than those in the N subset within the first 15 days after cascading. Ultimately, their properties become indistinguishable from the N subset. This is expected considering the intense mixing downstream of Denmark Strait shown earlier in Fig. 9.
The DSO particles in the ISh subset exist only in winter, for ensemble members III–VI (Fig. 12). They exhibit a similar cooling and freshening of their properties as the S subset, but the changes in their temperature happen at much faster rates. The cooling rate is more rapid for ensemble members III, IV, and V , and slower for ensemble member VI . The changes in the salinity are relatively small; the DSO particles in ensemble members III and IV slightly freshen but those in ensemble members V and VI slightly salinify . The DSO particles in the ISh subset are generally shallower than the particles in the S subset (mean depth of 250 m) and they rapidly densify approaching the Denmark Strait. After crossing the sill (0 day) the DSO particles in the ISh subset still experience cooling, freshening and densification for approximately 15–20 days as opposed to the particles in the S subset. That is because the ISh subset particles still move northward along the shelf after crossing the Látrabjarg section and meet the southward flowing particles from the N subset, mix, and return south by cascading over the Denmark Strait within about 20 days of their forward run trajectories. Therefore, there is a lag in their property evolution with respect to the S and N subsets downstream of Denmark Strait.
To identify the water masses that the DSO particles carry, their distribution in potential temperature–salinity (θ–S) space are plotted in Fig. 13 for all ensemble members, color coded based on their origins. The θ–S diagrams represent the water masses of the DSO particles at three different times: 1) at the origin, 2) at the release time at Látrabjarg, and 3) at the end point, 30 days after crossing the Denmark Strait. The DSO particles in the N subset at the origin (Fig. 13a) mark two hydrographic branches that indicate the conventional water mass end members observed north of Denmark Strait: the recirculated Atlantic Water, Polar Water, and Arctic Water (Harden et al. 2016). The particles in the ISh and S subsets contain the Irminger Water end member. These water masses are not distinguishable anymore at the Denmark Strait and farther downstream (Figs. 13b,c).
To identify when and where along the trajectories the water mass transformation takes place, the mean geographic location of the DSO particles in the latitude–longitude space as well as their mean trajectories in θ–S space are plotted (Fig. 14). The DSO particles in the N subset trace a short trajectory in θ–S space nearly along the 27.9 kg m−3 isopycnal from their origin to the Denmark Strait (Fig. 14b, blue). This indicates that the changes in their properties are small (Fig. 10a). However, their trajectory (in θ–S space) after cascading the Denmark Strait is longer, which means they have a larger and faster change of properties due to the mixing downstream of the Denmark Strait. The DSO particles’ mean trajectory in the θ–S space crosses the 27.8 kg m−3 isopycnal and the DSO particles become lighter within 17 days after their cascade and continue to become more buoyant until the end of the simulation. As the N subset DSO particle trajectories are present both along the east Greenland shelf and west Iceland shelf, their mean geographic trajectories pass the middle of the Blosseville Basin, and after they cascade they continue along the east Greenland shelf.
The DSO particles in the S and ISh subsets behave differently (Fig. 14 red and yellow trajectories, respectively). The particles in the S subset cool and freshen along their trajectories from the origin to the Denmark Strait. They also densify until their potential density exceeds 27.8 kg m−3 approximately 9 days prior to reaching the sill. Their trajectories in θ–S space from the origin to the Denmark Strait are quite long, which implies that they experience a faster rate of change in their properties than the N subset particles. After the cascade, they experience a shorter mean trajectory in θ–S space and become slightly lighter until their properties as well as their spatial location converges to that of the N subset DSO particles.
While moving along the Iceland shelf, the DSO particles in the ISh subset keep their salinity constant and cool until they are densified enough to cross the 27.8 kg m−3 isopycnal. This transformation takes place within 66 days of traveling from their origin to within about 50 km from the Denmark Strait. They still densify until they reach the Strait (green circle in Fig. 14). Then they continue along the 27.85 kg m−3 isopycnal for 24 days until they cross the Látrabjarg section while flowing northward along the Iceland shelf. They still continue moving north (in geographical space) and along the 27.85 kg m−3 isopycnal (in θ–S space). The mean trajectory continues along that isopycnal even after cascading for about 10 days. Finally, the mean trajectories of the ISh subset of DSO particles converge to that of the N and S subsets both in geographical and in θ–S space. This reflects the mixing downstream (cf. Figs. 9, 13b with Fig. 14). Note that the particles in the ISh subset, although starting as the shallowest, warmest and saltiest, become, after cascading, nearly as cold and fresh as the DSO particles in the N and S subsets.
4. Summary and conclusions
In this paper we presented a detailed analysis of upstream sources and pathways of the DSO. To do so, we performed a Lagrangian analysis on the output of a high-resolution ocean model to investigate the origins and pathways of the DSO. The model time period is from 1 September 2007 to 30 November 2008 and its outputs are evaluated by comparison with observations. This kinematic study gives a comprehensive understanding of the time-varying contribution to the overflow and evolution of previously identified northern pathways, i.e., the NIJ, the EGC, and the sEGC (blue curve in Fig. 6b). It also reveals an additional pathway from south of the Denmark Strait, which contributed to the DSO. This southern pathway itself is divided into two subsets, depending on whether the Lagrangian particles arrive at the sill east or west of 26.5°W. Those that arrive to the east of 26.5°W are shallower than 300 m and are present only during the winter. The average contribution from the south is 25.8% during the winter (JFM) of 2008, which has a relatively high NAO index. The entrainment south of Denmark Strait suggested by the literature, (e.g., cold entrainment of the northward Labrador Seawater (McCartney 1992), and warm entrainment of Irminger Current (von Appen et al. 2014) into the overflow at middepth and farther south of the sill) are different from the southern pathways we discovered. This subsurface southern pathway could potentially mark a shortcut for northward-flowing waters that densify and turn around to cascade over the Denmark Strait and eventually feeding the NADW. Although observations show evidence for the existence of the dense water on the Iceland shelf (Fig. 5; Våge et al. 2015) as well as presence of Irminger Water in the deep overflow (Mastropole et al. 2017), these southern pathways need more confirmation from observations. Future studies will benefit from models that incorporate coastal runoff on the Iceland shelf, and are several years long to account for interannual variability and to elucidate the relevance of these southern pathways in a changing climate.
The evolution of hydrographic properties of particles from each subset were also presented. The northern origin DSO particles (N subset) transform to become warmer, saltier and slightly less dense, but the southern origin DSO particles (S and ISh subsets) transform to become colder, fresher and denser along their path to the Denmark Strait. The DSO particle properties in all subsets converge after cascading over the Denmark Strait (Figs. 13b,c), and they have overlapping pathways (Fig. 9). Therefore, water mass analysis would not be able to discern the DSO sources at the Látrabjarg sill section or south of it. This study shows that the combination of backtracking and along-track water mass analysis can locate the water mass transformation sites continuously along the pathways (Fig. 14). In this way both time evolution and spatial distribution of pathways are required to determine the origins of the DSO water masses (see the introduction).
Finally, the schematic of the overflow water is updated by incorporating direct pathways of the DSO (Fig. 15). The circulation schematic shows that the separation from the EGC happens at multiple locations and it is intermittent. The DSO particles represent a clear distribution of pathways but they do not necessarily stay in one current (i.e., EGC, sEGC, and NIJ) at all times. Therefore, they do not indicate the currents, but rather trajectories of the particles. The particle pathways did not reveal an enduring source for the NIJ to the east of Kolbeinsey Ridge, but some particles show an intermittent eastern pathway (marked by dashed blue line in Fig. 15). The particle trajectories clearly marked the NIJ pathway from the north and west of Kolbeinsey Ridge toward Denmark Strait (e.g., Fig. 8, ensemble members II, IV, XI, and XII). This is similar to what was observed with the RAFOS floats (de Jong et al. 2018). We also see the core of the NIJ to the east of Kolbeinsey Ridge in the model mean Eulerian velocity field, consistent with Semper et al. (2019). Therefore, it is possible that an eastern source would be found if the Lagrangian simulation was multiyears long.
The southern pathways (revealed by the S and ISh subset particles) are the main addition to the previous schematics by Våge et al. (2013), Harden et al. (2016), and de Jong et al. (2018) although this contribution is likely to vary interannually depending on the local surface forcing. These newly identified pathways (S + ISh) are shown to supply up to 34.3% of the DSO water, and is an important contribution to understanding the overflow water sources and composition. The highest contribution from the south is seen when there is more dense water on the Iceland shelf. The sparse in situ observational evidence suggests that there is a correlation between the existence of dense water (σθ ≥ 27.8 kg m−3) on the shelf and a high winter NAO index, consistent with strong atmospheric cooling. However, the dynamical relationship between the atmospheric state and variability of dense water on the shelf needs more investigation and is left for future work. These dense water observations are mostly from the 1980s and 1990s when a warming trend of 0.2°–0.4°C decade−1 close to the bottom in the depth range of 100–250 m on the Iceland shelf was observed (Jochumsen et al. 2016). The warming can be another explanation for the missing dense profiles in recent years, in addition to the phase of the NAO. Determining the formation mechanism of the southern DSO sources is beyond the scope of this study, and could be a fruitful topic of future research.
This work was financially supported by the U.S. National Science Foundation under Grant Numbers OAC-1835640, OCE-1633124, OCE-1433448, and OCE-1259210. The authors thank Henrik Søiland, Institute of Marine Research, Bergen, Norway, for providing the CTD data at release time of RAFOS floats, and Benjamin E. Harden for providing the Kögur mooring array data. We thank Aleksi Nummelin, Mattia Almansi, and Soroosh Mani for providing constructive comments.
Denotes content that is immediately available upon publication as open access.