Mean circulation and EKE distribution in the Labrador Sea Water level of the subpolar North Atlantic.

: A long term mean flow field for the subpolar North Atlantic region with a horizontal resolution of 10 approximately 25 km is created by gridding Argo-derived velocity vectors using two different topography following interpolation schemes. The 10-d float displacements in the typical drift depths of 1000 m to 1500 m represent the flow in the Labrador Sea Water density range. Both mapping algorithms separate the flow field into potential vorticity (PV) conserving, i.e. topography following contribution and a deviating part, which we define as the eddy contribution. To verify the significance of the separation, we compare the mean flow and the eddy 15 kinetic energy (EKE), derived from both mapping algorithms, with those obtained from multiyear mooring observations. The PV-conserving mean flow is characterized by stable boundary currents along all major topographic features including shelf breaks and basin-interior topographic ridges such as the Reykjanes Ridge or the Rockall Plateau. Mid-basin northward advection pathways from the northeastern Labrador Sea into the Irminger Sea and from the 20 Mid Atlantic Ridge region into the Iceland basin are well-resolved. An eastward flow is present across the southern boundary of the subpolar gyre near 52°N, the latitude of the Charlie Gibbs Fracture Zone. The mid-depth EKE field resembles most of the satellite-derived surface EKE field. However, noticeable differences exist along the northward advection pathways in the Irminger Sea and the Iceland basin, where the deep EKE exceeds the surface EKE field. Further, the ratio between mean flow and the square root of the EKE, 25 the Peclet Number, reveals distinct advection-dominated regions as well as basin interior regimes in which mixing is prevailing. briefly describe the methods to separate < U > and an accompanying u' , obtained for each displacement vector from the difference between the observed displacement and the displacement projected to a PV contour. The fields obtained by two different gridding methods are 110 verified for internal consistency, and in comparison to independent measurements from mooring records. Next, a gridded velocity and an Eddy Kinetic Energy (EKE) field of relatively high spatial resolution (order 25km grid size) for the SPNA is created by both gridding procedures. We discuss the fields for internal consistency based on major flow features. Furthermore, the ratio of advective flow and diffusion (Peclet number) is estimated. The EKE field at depth is then compared with the EKE field at the surface, based on satellite data. The gridded data 115 sets are provided for download and further use e.g. for model/data comparison, so far we are not aware of an intermediate depth EKE map. three at the same Iceland water depth 1800m 2500m. all around the basin the close correspondence of the float trajectories and the is an indication of the PV following nature of the deep flow field. By subtracting the mean component, which we to be inherent in each of the measured displacement vectors, from the original data a current residual ( u’ ) could be calculated


Introduction
The subpolar North Atlantic has been in the focus of both observational and modelling efforts in hindsight of 45 circulation-and water mass changes as part of the climate relevant Atlantic Meridional Overturning Circulation (AMOC; reviewed e.g. by Daniault, et al., 2016). In this context the intermediate depth circulation, which also determines the spreading pathways of newly ventilated Labrador Sea Water (LSW) through the subpolar North Atlantic (SPNA), is of specific importance and has been investigated from observations and models for several decades. A better understanding of the mechanisms that control the transport properties at mid ocean depth 50 through the interplay of advection and diffusion is fundamental to our understanding of subpolar LSW circulation and export, and thus potentially subpolar AMOC contributions. Unlike the surface circulation, which can be analyzed for example from satellite and drifter data, the intermediate depth circulation and energetics is known to a much lesser extent. Studies that map energetics at the intermediate depth from observational data and at gyre scales are rare but identified for example as important evaluation metrics for basic verification of ocean 55 model simulations, including CMIPs models (Griffies et al. 2016).
In the late 1990ies, technology of profiling floats advanced such that investigations of the intermediate deep circulation could be undertaken. Two experiments were carried out in the western subpolar North Atlantic (mainly in the Labrador-and Irminger Seas) using Profiling ALACE (PALACE) floats and are of special interest to the investigation carried out herein. The first was by Lavender et al. (2000) with a large fleet of floats drifting 60 through the Labrador and Irminger Seas in 700m depth (the approximate depth level of upper Labrador Sea Water in the subpolar North Atlantic). A major result of the study was that the intermediate depth circulation could well be described as a cyclonic boundary current system along the topography and a series of anticyclonic recirculation cells adjacent to the Deep Western Boundary Current (DWBC). The second experiment was dedicated to the boundary current off Labrador, and conducted in summers 1997 and 1999 with 15 PALACE spatial resolution that resembles narrow circulation elements in higher resolution compared to what have been discussed in the past.

85
Several attempts have been undertaken to estimate advective (long term mean) and diffusive contributions in the displacement vectors on the basis of statistical and physical constraints. While the displacements of the profiling floats may be well suited to determine the long term mean of the flow field, this is not straight forward for the eddy component of the flow field (Davis, 2005). The author suggested calculating the diffusivity from displacement anomalies u' calculated from the difference of the mean flow <U> and the measured displacement 90 vector U m . Here, we loosely follow the method proposed by Davis (1991), in which the mean flow is controlled by topography (f/H, with H is the water depth), an assumption that should hold true in the SPNA regime where weak stratification und small vertical current shear is encountered. Thus, we will estimate the advective part of the flow that is related to the concept of potential vorticity conservation (LaCasce, J.H., 2000), and the residual flow contribution that is attributed to the diffusive part of the flow. Validation of this principle has been 95 performed in the past (see Fischer and Schott, 2002;Fischer et al., 2004) through a comparison of deep displacements along curved topography in relation to moored (Eulerian) records.
We focus here on the SPNA north of 45°N and make use of the extended set of Eulerian (current meter moorings) and Lagrangian (floats) observations available in the region. Over the previous two decades 100 (regionally even longer) an impressive observing effort has been undertaken north of 45°N on the (intermediate) deep flow. Boundary currents are, thanks to their strength, the prominent circulation features in the SPNA and found all along the shelf edges in particular on the western side of the gyre. However, there are also interior circulation features of both advective-and eddy-dominated pattern, and the primary research objective of this effort is to discriminate the mean flow <U> from the turbulent (eddy) component u' of the flow field from which 105 the deep EKE field could be determined.
The paper is structured like follows: First, we briefly describe the methods to separate <U> and an accompanying u', obtained for each displacement vector from the difference between the observed displacement and the displacement projected to a PV contour. The fields obtained by two different gridding methods are 110 verified for internal consistency, and in comparison to independent measurements from mooring records. Next, a gridded velocity and an Eddy Kinetic Energy (EKE) field of relatively high spatial resolution (order 25km grid size) for the SPNA is created by both gridding procedures. We discuss the fields for internal consistency based on major flow features. Furthermore, the ratio of advective flow and diffusion (Peclet number) is estimated. The EKE field at depth is then compared with the EKE field at the surface, based on satellite data. The gridded data 115 sets are provided for download and further use e.g. for model/data comparison, so far we are not aware of an intermediate depth EKE map.

Material and Methods
Two quality controlled Argo displacement (deep and surface) sets exist, but cover somewhat different time 120 spans. Here, we use the Yomaha07 -Argo dataset (Lebedev et al., 2007)  of deep and surface currents using data of the trajectories from displacements between consecutive dives of Argo floats. This data set is updated frequently on a monthly basis. The area under investigation ranges from 45°N, the latitude just south of Flemish Cap, to 65°N, which is just south of Denmark Strait (Figure 1). The westernmost longitude is 62°W, i.e. the Labrador shelf break, and to the 135 east, the area is bounded at 7°W west of the British Isles. The evolution of the Argo data in this domain shows a rapid increase in data density in the first 5 years of the program (Figure 2), and from 2006 onwards the data density is adding around 2500 to 3000 displacement vectors per year; this is roughly equivalent to the number of T/S profiles gained through Argo per year. Maximum annual data increase is reached in 2011/12 with 4000 additional current vectors each year. Thereafter, the yearly data gain stabilizes at 2500 to 3000 current vectors.

140
The regional data density ranges from approximately 10 to more than 100 per bin (2°longitude x 1° latitude; Figure 1); the bin size corresponds with the typical area in which data will be used for the interpolation to a certain grid point. Given the barotropic nature of the flow field in the SPNA we merged the displacement vectors from the two drift depths 1000m and 1500m depth. Considering the temperature and salinity data recorded by the floats the mean potential density field at 1500 m varies between   = 27.72 kg m -3 and 27.92 kg m -3 with an 145 average density of   = 27.77 kg m -3 . This density is slightly lower than the commonly used lower boundary of classical LSW at   =27.80 kg m -3 and thus, the resulting circulation pattern represents the core depth of the LSW.

Auxiliary data
the analysis here (1000-1500m) were used to locally evaluate the results of the gridded data product (see table 1 for an overview). Given the floats inherent sampling at 10 days, the moored records were smoothed accordingly.

Separating mean flow and its fluctuation and interpolation of the results
Two interpolation methods were used to map the displacement vector data: the first is a weighted Gaussian interpolation (GI) and the second is an optimum interpolation (OI) procedure. Both methods use the same 165 physical constraints, and both operate on an identical grid of 0.5° longitudinal range and 0.25° latitudinal range.

Gaussian interpolation method
The strategy of the GI-method was to include two constraints in the interpolation procedure, namely a weighted distance between target (grid) point and data point, and the second is to reduce the influence of data points 170 located in regions with very different water depth. The latter is a manifestation of our assumption that flow in the region follows PV-contours. Thus, data points across the boundary current at steep topography would only weakly be influenced from nearby but much deeper/shallower locations outside the boundary current (a topography following mapping).
The weights used have a Gaussian shape described by two parameters for each dimension: for the distance 175 weighting we chose 40 km for the half width of the Gaussian and 80km for the cut-offsuch that points outside a radius of ~80km around a selected grid point will not be used. For the other dimension (water depth difference between data location and target location as a measure of PV difference) we chose 200 m half width and 600m cut-off range. The choice of these values was guided by the dimensions of the boundary current along steep topography (e.g., the Labrador shelf break), with the width of the DWBC (Zantopp et al., 2017) between 100 km 180 and 150 km, and a change of water depth across the DWBC from about 1000m to 3000 m. Through this procedure Boundary Currents would be conserved and not smeared out, while in the basin interior with flat bottom the weight is more toward distancewith only little influence of the depth-difference.
We analyzed the impact of different weights over a wide range of scales, but the selection applied here appears to generate the most robust result with a clear definition of the circulation elements described hereafter. Using a 185 higher resolved grid (smaller scales) result in noisier flow field with larger overall variance, while a coarser grid (larger interpolation scales) result in a smoother field and certain details of the flow field are suppressed. The procedure could be applied to both, irregular target locations, and regular grid locations.
In a first processing step we separate the measurements into a mean flow contribution <U> and a fluctuating part u' that will be used later to determine the EKE field. Around each measurement location we selected all data 190 within the cut-off radius and by using the selected weights (see above) we estimate a mean flow vector at the measurement location by applying the above described algorithm. Thus, we generate a velocity field that has the dimension of the original data set, and it contains only the weighted, PV-related ensemble-mean contribution ( Figure 3a). As an illustration how well the PV-constraint works we show three floats that were deployed at roughly the same location in the northern Iceland Basin at water depth around 1800m to 2500m. This depth Subsequently we applied the mapping procedure to the measured velocity field (U m ) to obtain a field on a 200 regular 0.5° longitude x 0.25° latitudinal grid and the result is <U> on a regular grid, which is considered as one of our final data products.
The Eddy kinetic Energy (EKE) is estimated independently for each of the two interpolation methods (GI and OI). Assuming that the separation of the measured displacement vectors into <U> (advective) and a fluctuating (eddy) component is successfully performed by the above methods, it allows to calculate u' and v', the

Optimum Interpolation method
The second procedure uses the method of optimum interpolation (OI-method), similar to the one described in detail in Schmidtko et al. (2013). Data was only mapped if the grid points have a water depth deeper than 1200m according to the topographic data set. All data within a radius of 110km and at locations with similar water 215 depthsless than 1000m differencewere used in the OI. Linear gradients in latitudinal direction , logitudinal direction and water depth were fitted to the data. For the covariance matrix a diagonal value of 1.5 was used as an estimate for the signal to noise ratio (see Schmidtko et al. 2013 for details). The background field used in the optimal interpolation was taken from a least squares linear and quadratic fit of the data using depth, longitude and latitude.

220
The field, resulting from the OI was used as mean flow field <U> which was then used to compute the residual flow (u' and v') from each displacement vector (U m ) by linear four point interpolation. An individual EKE value was computed for each displacement. To exclude extreme outliers an inter quartile range filter was applied, rejecting data points 2.2 times the inter quartile range above the third quartile or that range below the first quartile. This is similar to a 99.98% standard deviation filter in case of normal distributed data. The EKE data 225 was then mapped in an identical procedure as the mean field.

Labrador Sea
The intermediate circulation in the Labrador See shows narrow cyclonic boundary circulation where the topography is steep, i.e., along the East Greenlandand Labrador shelf breaks (Figure 3a), while in regions with a gentler slope (e.g. northern part of Labrador Sea) the boundary current widens considerably. From the 255 boundary current to the interior Labrador Sea the flow reveals stable but weak recirculation cells with cyclonic rotation, and the interior of these elongated cells is almost stagnant, as is also seen in time series measurements (Fischer et al., 2010;) at location K10, where the mean 1500 m-flow is 0.8 cm s -1 northwestward, and at K9 where the mean flow is 12.5 cm s -1 but southeastward (Figure 5b; Table 1).
The nearly stagnant, weakly anticyclonic rotation is observed for the area were deep convection take place. Here

Irminger Sea
The Irminger Sea has several characteristic flow patterns in intermediate depth (Figure 3a) . The most pronounced feature is the deep East Greenland Current (dEGC) that exists over the whole western part of the basin. On the opposite side the Irminger Sea is bounded by the Reykjanes Ridge that is a barrier for most of the flow beneath 1000m depth. Further south, several gaps in the ridge allow the water from the eastern basin to spill 275 over the ridge and a northward deep boundary current forms along the western flank of the ridge. This is one source of the dEGC. A second source of the intermediate dEGC is the mid-basin current band that is fed from the Labrador Sea and extends up to 64°N where it enters the dEGC; the cyclonic circulation that this mid-basin vein forms is sometimes called the Irminger Gyre. Within the Irminger gyre a number of long term moorings have been maintained for more than a decade to record the thermohaline evolution of the gyre center and possibly 280 deep convection underneath the Greenland Tip Jet (e.g. Pickart et al., 2003); the moorings are nowadays incorporated in the international OSNAP program and the OOI initiative. The mid-basin current band appears to have a number of meanders which are also visible in the 1500m geopotential derived from the Argo profile data.

Iceland Basin
The Iceland Basin, which is less well investigated, has two major topographic features that influence the 285 circulation strongly. The western limit of the Iceland Basin is the RR and that show the already discussed boundary current. At the location of the Charlie Gibbs Fracture Zone (CGFZ), which is at about 52°N, forms dynamically the southern boundary of the basin and were the circulation at the LSW depth is eastward in connection to the North Atlantic Current (NAC) supplying water towards the eastern SPNA.
Two branches of the NAC are evident (Figure 3a): the majority of the floats (not shown) drift far eastward in a 290 strongly meandering current band until arriving at the topography (still at the latitude of the CGFZ, i.e., 52°N).
Thereafter the flow follows the topography northward into the Rockall Trough west of Ireland. On the western flank of the Rockall Plateau a narrow eastern boundary current forms and flows northward until it reaches the Iceland-Scotland-Ridge, where it feeds the southwestward boundary current (discussed above) that eventually becomes a 'western' Boundary Current along the Reykjanes ridge. However, the broadest inflow comes from the 295 mid-basin flow regime extends from the NAC northward from about 27°W, and follows the deep trench northward to 62°N. This mid-basin flow is characterized by stable advection and several large wavenumber meanders.

The North Atlantic Current regime
The southern exit of the Labrador Sea is the region where the NAC meets the DWBC (Figure 5a) , and while the

Gridded Mean Flow
By application of the GI-method, the velocity field was interpolated to a regular grid of 0.25° latitude and 0.5° longitude (Figure 5a). As for the raw data maps (Figure 3a) the gridded data reflects all the major circulation There is also a connection between the NWC and the convection area by a low density anomaly that is not associated with the DWBC, but with the reverse circulation into the Labrador Sea.
Although, we only show the mean gridded flow field from the GI-method we do obtain the same results from the 320 OI method. The differences of the two estimations mainly contain small scale elements that reflect the scales of the influence radii by either method.

Gridded Eddy Kinetic Energy
From the individual u' and v' fields we generated a smoothed and gridded version of the EKE (Figure 5b) using 325 the same interpolation parameters as for the mean fieldi.e., both fields have the same length scales in consideration, and the grid is identical. 'Smoothed' also means that some de-spiking and noise reduction during the gridding operation was applied, as there were a few individual spikes along the edges of the mapping environment, i.e. in regions where the mapping area intersects the 1500m topography and where floats might have become bottom-stuck. These spikes could be easily detected and accounted to less than 2% of the data 330 contributing to an individual grid point. As a result the cleaned EKE distribution is smoother and more reliable.
We note several intense EKE hot spots in the Labrador Sea, in the 'Northwest Corner' of the North Atlantic Current, and in the eastern SPNA located east and west of the Rockall-Plateau. While it is not surprising that the retroflection of the NAC (i.e. the 'Northwest Corner') shows large EKE values exceeding 250 cm 2 s -2 , it is surprising that the zonal basin-crossing of the NAC has relatively weak EKE at LSW levels. The second 335 strongest EKE is located in the northeastern Labrador Sea and is generated by instabilities and eddy shedding of forcing (winter wind-stress maximum). It appears that there is a significant difference of the EKE intensity on both sides of the Labrador Sea; while the northward flowing WGC is subject to intense eddy formation and hence high EKE, the southward flowing Deep Labrador Current is much more stable with remarkably low EKE levels. A possible reason is the PV-conservation that stabilizes the flow when progressing southward (towards lower f) and consequently the flow is driven toward the stabilizing topography, while for northward flow there is 350 a tendency to move into deeper water with smaller topographic Beta. This is further supported by the weak EKE in the southward flowing East Greenland Current.

Advection versus Diffusion --Peclet Number
The western subpolar basin has very different regimes regarding mean flow and EKE pattern. Even at larger depth, there are narrow boundary currents along the topography, there are interior persistent current bands, and

Verifications of the results
We verified our results in three different ways: First the results from GI and OI were compared in order to identify a superior interpolation method. Then we compared the mean flow and mean EKE fields with similar quantities derived from Eulerian time series data from moored stations (EKE moor ) in the region; and the third way of verification was a comparison between the deep EKE and the EKE surf from satellite SLA data.

Consistency of interpolation techniques
The mean flow fields from the two gridding methods are surprisingly similar and there are no significant differences between the velocity and speed fields.. The overall speed-difference is -0.16 cm s -1 which illustrates that there are no systematic differences (biases) between the two speed estimates as a result of the gridding 390 technique. The difference field are patchy in structure with patch-scales of the order of the interpolation radii.
Thus, by choosing the GI method the current map (Figure 5a) is considered representative and independent of the two mapping procedures applied.
Likewise the difference in GI and OI interpolated EKE fields ( Figure 6) agreed well. Most of the EKE differences occurred in the range ± 5 cm 2 s -2 with the strongest deviations around the NAC path across the SPNA

395
here, the GI method produces somewhat larger values. In contrast, the Northwest Corner reveals larger EKE values for the OI method. A patchy structure is observed with scales associated to the influence radii of the gridding methods (roughly 100km). The difference has an overall Gaussian distribution but with a slight bias of 1 cm 2 s -2 toward larger EKE in the OI method map. The magnitude of this bias depends on de-spiking method used in any of the two processing methods. The strongest impact is due to the removal of individual large spikes 400 in the EKE in the GI method, which leads to a regional reduction of the corresponding EKE field. The removal of only 1% of the largest velocities results in an increased in bias to 4 cm 2 s -2 . No explicit de-spiking has to be used in the OI method, as it is inherent in the method itself (see Schmidtko et al., 2013), while in the GI-method we explicitly had to remove some large outliers.

Comparison with local Eulerian measurements
The second method for verification was a comparison between the derived mean fields (<U> and EKE) and selected locations were time series data from moored instrumentation was available (Figure 5b; table 1).  (Figure 5a). Short time scales dominate the variability of the flow (Figure 7), and the spectra indicate In the boundary current system a number of records could be analyzed. In general the flow is rather stable and strong (compare also Table 1). Representative for the DWBC at 53°N (K9, Zantopp et al., 2017) the long term 430 mean flow along the topography is 12.5 cm s -1 and the EKE (again for periods less than 180d) is 62 cm 2 s -2 .

Labrador-and Irminger Seas
Farther toward the topography the mean speed is even larger and the EKE smaller, as the DWBC appears to be more focused by the steep topography at 53°N. In any of the boundary current records a large energy contribution is on timescales less than 10 to 20 days (Fischer et al., 2015), which are not captured by the Argo displacement vectors and different from the basin interior were the flow variability is on timescales longer than a 435 month and thus better resolved by 10 day displacement vectors from Argo ( Figure 8).
The records in the center of the DWBC at Hamilton Bank the total EKE of the moored record is larger than that from the float displacement but similar to K9 (Table 1). For a better comparison we calculated the EKE fraction that Argo would represent in their 10 day displacement vectors by low-pass filtering the mooring data (10d cut-440 off period of the filter). Then, the EKE values coincide much better as is demonstrated by the colored mooring numbers in Figure 5b.
Near the offshore edge of the DWBC, at mooring K10, the flow speed is rather low, as the mooring lies in the transition regime between the DWBC and the recirculation pathway in the upper 2000 m, while at deeper depths 445 it is still part of the DWBC (Zantopp et al., 2017). At 1500 m the flow is mainly reverse to the DWBC direction and the EKE is rather small, but in good agreement with the EKE from Argo.
Associated with weak mean speeds (only 10% of the DWBC speed is found at locations offshore of K10) and moderate EKE moor values coincide when the resulting Peclet Numbers (Figure 5c; Table 1) are low and indicate 450 sufficient diffusion in the presence of weak advection. This structure is reflected in the Argo flow pattern, which shows an increasing advective contribution further toward the basin interior, and from the mean current map and the density field it is tempting to assume this route as one of the supply routes for the deep central Labrador Sea.

Subpolar Locations
Moored observations in the Iceland-Scotland-Overflow Water were available at 4 positions (Named I, S, O, W; see Kanzow and Zenk, 2014). Only three (S, O, W) moorings delivered data in the appropriate depth range for this study. While S was located in the area of low deep EKE, the fluctuations increase toward east with mooring 460 W located in the EKE max along the northward flow (Figure 5a).
North of the I S O W array the Iceland Array is located at the shelf break south of Iceland, and the northern mooring direct at the topography reflects the low EKE moor typical for topographically guided currents, while the one further offshore is located in the northern extension of the EKE maximum of the Iceland Basin.
During the Jasin program in the late 1970s a number of moorings were deployed in the northern Rockall Trough 465 (Gould et al., (1982)) and these moorings reflect the intermediate intensity of the deep EKE moor that is also present in the Argo derived values ( Table 1).
The EKE from the moorings represent mean regional variations. In order to compare the high resolution time series with the Argo data a 10d low-pass filter is applied. There is a remaining discrepancy between EKE from

470
Argo and from moorings with a tendency that in region with low EKE (taking now the Argo derived map as a reference) the Argo estimates are larger than the 10d-lowpass filtered mooring estimates, while in regions of high EKE the situation is reversed. We interpret this discrepancy by the inherent (nonlinear) temporal filtering in the EKE derived from Argo that tend to low-pass filter the field with an unpredictable filter characteristics (depending at which times the floats enter the corresponding interpolation radius).

Surface EKE versus intermediate depth EKE
In addition to the deep EKE we estimated the surface EKE (EKE surf ) field calculated from remote sensing-based ADT observations. The geostrophic surface flow from SLA contains variability over a wide range of 480 frequencies, and some of the long term components are not generally thought to be part of the turbulent eddy field. Thus, we extracted the intra-seasonal variability by applying a high-pass filter (Hanning window) with a cut-off period at 180d. The result is a field of geostrophic fluctuations from which we calculate EKE surf (Figure   8a). This field is independently derived, and thus allows an independent comparison of the Argo-derived fields (here, the deep circulation and EKE).

485
The EKE surf also resembles major (deep) circulation elements, such that the zonal flow in the CGFZ region located underneath the zone of maximum EKE surf gradient at the surface. (Note in Figure 8  i.e. ln (EKE surf / EKE) , such that the ratio becomes negative when the deep EKE is larger than that at the surface. Generally, in a baroclinic ocean one would expect positive ratios, with the EKE surf sufficiently larger than the EKE at depth, as is the case for the region south of the North Atlantic Drift, i.e., south of 52°N. A global 500 much coarser map of such a ratio reveals that this is the case for almost the whole Atlantic Ocean (Ollitraut and de Verdiere;. In their paper, the subpolar North Atlantic appears as broad negative area in which the deep EKE exceeds the upper layer or is of similar magnitude. The much higher resolution of the field generated herein (Figure 8b), allows a more detailed view, which reveals two centers of deep EKE dominance. The first is associated with the DWBC all along the Labrador shelf break and the strongest signal around Hamilton Bank.

505
The second center is associated with the deep action center south of Cape Farewell that shows both, stable advection and EKE at depth, while at the surface these components are rather weak. This zone extends far north into the Irminger Sea where it appears to be related to the deep EGC and its variability. This behavior indicates that for the interbasin spreading and mixing of newly formed water masses the deep EKE field contains important information , which is not easily available elsewhere; at least not from the surface variability alone.

510
Besides the boundary current related anomalies there is one additional zone in which the deep EKE is close to the surface EKE, and that is along the CGFZ at the northern flank of the NAC. In this area the flow is guided by the deep topography and advection appears to be dominating the zonal flow (relatively large Pe).

515
The results of the investigation can be summarized as follows: 1) Based on nearly 17 years of quality controlled Argo displacement vectors a high resolution (~25km grid) map of mean flow in the depth layer of the LSW was constructed for the Subpolar North Atlantic.
Robust circulation elements were identified consisting of boundary currents along topographic slopes, mid-basin advective pathways, and stagnation regimes with very low mean speeds.

520
2) The mapping procedures were twofold: Gaussian Interpolation (GI) and Optimum Interpolation (OI), both methods were applied using potential vorticity constraints, and the resulting mean flow fields were very similaralmost identical.
3) The second product was the fluctuating (eddy --u', v') velocity component, which was determined as the residual after subtracting the average and potential vorticity conserving contribution from the 525 individual measurements (displacement vectors). The u', v'-fields were used to map the mean EKEdistribution; to our knowledge for the first time.
4) The ratio of mapped mean flow to the square root of the EKE, the Peclet-Number (Pe), was estimated and showed regions that are advection dominated (boundary currents and internal LSW routes), and regions with low PE, in which eddy diffusion prevails.

530
5) The mapped fields were analyzed for consistency between the OI and GI method. In addition velocity time series from moored sensors were used to estimate mean flow and EKE in an attempt to verify the mapped fields locally with independent data. While the general pattern of high and low EKE regimes are consistent, but the mooring EKE appears to be larger than EKE from Argo but the differences By focusing on the Labrador Sea, the "surprisingly rapid spreading" of LSW throughout the subpolar North Atlantic (Sy et al., 1997) is well supported by our gridded mean flow field: newly formed LSW is exported by the mid-basin advective pathway into the Irminger Sea (Figures 3a and 5a) and eastward through the pathway that connects the western SPNA with the northern Iceland Basin through the NAC and its northern pathway.

550
Individual floats released in the DWBC off Labrador used that path to drift within a few (3-4) years far north into the Iceland Basin.
More regional aspects were discussed in the float release experiments performed in the late 1990, i.e. before Argo started officially. On the basis of these investigations, export pathways for LSW out of the Labrador Sea were discussed (e.g. Straneo et al., 2003) in which the export of LSW into the Irminger Sea, and the Boundary 555 Current Export around Flemish Cap were identified as major export routes. While the Irminger Sea route appears strong and robust, the flow along the topography (Flemish Cap and Grand Banks) is relatively weak. Instead, the second major export route is into the eastern SPNA via the NAC route.
Traditionally the upper ocean eddy variability represented by the EKE distribution has been investigated from SLA data (Brandt et al., 2004, Funk et al. 2009