The dense overflow plume was thicker and colder at the channel mooring section than at the downstream section (Figs.  and ). The bottom layer at the southernmost mooring in array C (C3) was the coldest, with a mean temperature at 26 m above bed (mab) of -0.3 ∘C over the 1-year deployment. However, although the very cold water occasionally reached up to 140 mab at mooring C3, it was mostly present only as a thin layer. Between 30 and 100 mab temperatures increased sharply where the bottom layer was capped by strong stratification. The overflow water was generally contained within the bottom 120 m.

In the centre of section C, at mooring C2, the plume was thicker. The presence of very cold water was not as frequent as at C3 – only about 60 % of the time even at the deepest level – but overflow water on average occupied the lower 200 m of the water column (Fig. ). Above about 50 mab, the mean temperature at mooring C2 was thus colder than at C3.

The northernmost mooring (C1) in array C was located outside the main path of the plume; very cold water occurred here only <5% of the time. When present, the overflow covered up to about 100 m from the seabed; however, 30 % of the time no overflow was recorded.

At mooring array S, the plume was thinner and warmer than at C, even in the bottom layer. Very cold water occurred only occasionally in the deepest layer at the two central moorings S2 and S3. At S2, the plume water (T<3 ∘C) reached up to about 60 mab, overlain by a 40 m thick interfacial layer. At S3, the bottom and the interfacial layer were equally thin. Above the interfacial layer there was a 100–200 m layer of intermediate water, and Atlantic Water was found above about 200–300 mab. The water column structure at mooring S4, 14 km further south-west, was markedly different as this southernmost mooring in the S array was located outside the plume. Overflow water was measured only about 20% of the time at the deepest level, coinciding with eddy activity, while intermediate water took up a large part of the water column at S4, occasionally reaching the bottom.

The mean current in the near-bottom layer was directed towards north-northwest at section C and turned more towards west-northwest at section S (Fig. ). The average current was in this along-stream direction throughout the part of the water column occupied by the cold, dense plume and the interfacial layer.

The maximum plume velocity was found in the bottom 50 m at mooring array S, where the time-average velocity perpendicular to the mooring section (u; see Sect. ) reached 90 cms-1 (mooring S3; see Fig. ). Similar but somewhat weaker velocities occurred at mooring C3. The thickness of the maximum velocity layer, in contrast, was significantly larger at the upstream array C. Defined as the height where the along-stream velocity u is reduced to half of the maximum velocity of that mooring velocity profile, the “kinematic interface height” , HIL, was 220 m at section C compared to 110 m at S (see Figs.  and ). The thinning of the plume with distance from the sill was reflected in both velocity and temperature profiles (Fig. ). The height above bed of the 3 ∘C isotherm was similar to HIL and correlated with it over timescales longer than 1 day, as previously observed by .

The high-velocity core of the plume was also faster but thinner at the southern side of the C array. Long-term mean velocities were thus higher at mooring C3 than C2, while the opposite was true for the thickness of the high-velocity layer. This is consistent with observations by showing a northward decrease in velocity and an increase in plume thickness across a section at the sill in the FBC (∼25 km upstream of our measurements). At the S array the velocities were similarly higher at S3 than at S2, but no southward thinning of the plume was observed; see Fig. .

Although high instantaneous current velocities occurred at intermediate and upper levels, the flow above the plume was characterized by oscillations and frequent reversals, resulting in weak mean flows (Figs.  and ). The varying upper layer currents contribute to modulating the period of mesoscale oscillations . The upper level mean current was generally directed along the plume, i.e. north-westerly, except for very weak flows in the opposite direction at the southernmost moorings in the S array. The mean counter-flow above 150 mab at mooring S4 was less than 2.5 cms-1.

In the centre of the C array (mooring C2) daily current vectors were always directed along-plume in the bottom 200 m. Above that, occasional reversals occurred, although rarely within 300 m above the seabed (Fig. ). C1 was located near the northern edge of the plume in the channel, characterized by weaker currents and frequent reversals.

Current velocities and hydrographic properties varied strongly on different timescales. Velocity power spectra (not shown) displayed a distinct peak at the semi-diurnal tidal frequency at all moorings. In addition, the C array of moorings showed a diurnal tidal peak. There was also elevated power spectral energy in the frequency band corresponding to mesoscale oscillations. The mesoscale spectral peak was found at frequencies corresponding to 4–8 days at C and 3–6 days at S.

Time series of monthly mean temperature at 25 mab from moorings along the main plume path are shown in Fig. . Although this frequency band contains only a small part of the variability (about 5 % for the series low-pass filtered with a 2-month cut-off compared to about 40 % in the mesoscale frequency band), there is a consistent pattern between the different moorings.

The temperature in the bottom layer (Fig. ) was lower than average during August through to October 2012 at both section C and further downstream. Warmer than average temperatures persisted from November 2012 to February 2013 at mooring C2 and through to April at M1 and S2. The maximum temperature in the bottom layer occurred in December at C2, and December through to February was also a warm period at M1 and S2. After the warm winter months, the temperature decreased in early spring to a minimum in May. At mooring S2 the monthly mean temperature during this coldest month was about 0.5 ∘C below the long-term mean.

The variation in monthly bottom layer temperatures might be interpreted as a seasonal cycle, with generally higher temperatures in the winter months and low temperatures the rest of the year. The temperature in July, however, deviated from this pattern by being clearly above the long-term mean at both sections C and S (with maximum monthly temperatures recorded at M1 and S2), in sharp contrast to the low mean temperature of August.

Volume transport varied strongly on daily to monthly timescales (Fig. ), but did not show a distinctive seasonal cycle (Fig. ). The daily mean transport of water colder than 3 ∘C ranged between 0.7 and 2.3 Sv at section C. The frequency spectra of volume transport through the arrays are shown in Fig. . Transport of cold water through section S varied most strongly at the timescale of mesoscale oscillations, while at section C low-frequency variation dominated the transport spectrum. The month-to-month variability in volume transport was largely in phase between sections C and S, with maxima at both sections in August and minima in November 2012 (Fig. ).

Figure  shows a subset of the data from the mooring array S, showing lateral temperature changes with time along the section. This mesoscale variability is typical of the entire record. The plume takes the form of domes of cold water moving past the mooring array, as previously noted by and . While the presence of cold water at moorings S2–S4 was associated with downslope velocities, it was associated with upslope movement at mooring S1. Velocity vectors rotated mainly clockwise at S1 and S2, and anti-clockwise at S3 and S4.

There were energetic fluctuations in volume transport in the mesoscale frequency band, particularly strong at section S (Fig. ). At section C, it was mainly the transport of overflow water (T<3 ∘C) that showed higher energy at mesoscale frequencies; at section S, there was high mesoscale variability also in the layer containing modified overflow water. The frequencies in the mesoscale band that are associated with elevated spectral energy levels correspond to periods of about 3–4 days and about 6 days, respectively. have recently shown that the oscillations in the channel alternate over time between these two periods, and that time periods dominated by the higher oscillation frequency (3–4 days) coincide with larger volume transport across the sill.

The mean eddy heat flux in the mesoscale band was calculated following . The mooring records were rotated to align with the mean current (100 mab) and band-pass filtered (2–10 days) to obtain fluctuations v′ and T′ in the dominant mesoscale frequency band. Finally the mean temperature flux v′T′‾ was calculated at levels where both velocity and temperature data were available (Fig. a). The temperature fluxes were multiplied by the density (ρ0) and heat capacity (cp) to obtain heat fluxes and then integrated up to the deployment mean position of the 6 ∘C isotherm (numbers adjacent to moorings in Fig. b). The position of the isotherm was above the mooring at C3 and roughly at the level of the instrument at S1 (80 mab); no value was therefore calculated for C3, while the value in parentheses at S1 is estimated assuming a constant value of the eddy heat flux from the bottom up to 80 mab. The temperature fluxes were mostly directed upslope, i.e. to the right of the mean current, and their magnitude was of the order of 0.1 Kms-1, reaching a maximum of 0.21 Kms-1 at S3, 70 mab.

Downstream of the Denmark Strait overflow, eddy heat fluxes were shown to be convergent in the central part of the plume , indicating that the horizontal stirring induced by the eddies there contributed significantly to the observed along-path warming of the plume. This was not in general true for the FBC region: Fig. b shows divergence between S4 and S3, i.e. in the central and lower parts of the plume, while the observations suggest convergence between C2 and C1 and between S3 and S1, i.e. for the shallow part of the plume.

The FBC overflow is a strongly sheared bottom-enhanced current characterized by energetic turbulence . Our unique data set is used here to present the kinematic stability in different layers of the water column throughout 1 year. The gradient Richardson number, Ri, representing the stability of stratified shear flow , is calculated over a 10 m vertical scale using the hourly and 1 m vertical gridded velocity and temperature profiles. Density is inferred from the temperature as described in Sect. . Calculations are made using data from two moorings, each sampling the core of the dense overflow plume: C2 at the channel, and S2 over the slope. For each hourly profile, the depth of the 1 to 6 ∘C isotherms is identified. Vertical 10 m segments centered at the corresponding depths are extracted. Vertical gradients of density and of horizontal components of the velocity are obtained as the slope of the linear fit against depth. Hourly Ri is then calculated as N2/S2 from the vertical velocity shear squared, S2=∂u∂z2+∂v∂z2, and the buoyancy frequency squared, N2=-gρ0∂σθ∂z. Monthly averages and standard deviations are then obtained after excluding outliers (the lowest and top 1 % quantiles) in each month. Standard error is assigned using the standard deviation and degrees of freedom assuming that every 13th sample is independent (to account for the semidiurnal variability).

The FBC overflow is characterized by small values of Ri. Very frequently, 60–80 % of the time, hourly Ri on the slope was below 1, and 10–30 % of the time Ri was below 1/4. In the channel, the core of the plume was associated with a 150 m thick high-velocity layer with weak shear, typically with temperatures less than 3 ∘C. Despite the weak shear, stratification was low, and monthly averaged Ri varied between 2 and 4 (Fig. d). In the sheared interfacial layer monthly Ri was approximately 2 or less, and in the sheared bottom boundary layer monthly Ri was approximately 1 or less (the bottom boundary layer is not captured by the 1 ∘C class, and hence is not shown in Fig. d).

Microstructure measurements reported in show that (i) the interior core of the plume in the channel is characterized by low dissipation rates, due to weak shear and lack of turbulent kinetic energy production, and that (ii) turbulence is enhanced in a 100 m thick high-shear bottom layer and 100 m thick interfacial layer. Ri values derived from C2 are consistent with these measurements and imply that, in our analysis, values of Ri approximately 2 and less can be associated with energetic, turbulent regions of the water column.

In the channel, the 3 ∘C isotherm capped the low-shear core of the plume and Ri was typically above 2 throughout the year. The interface above, however, was turbulent at all times, and particularly in May, with average Ri <1. In contrast to the channel profiles, the low-shear core did not exist over the slope (Fig. ). Ri was typically less than 2 below the 4 ∘C isotherm which separated the relatively quiescent stratified layer above (the 6 ∘C isotherm was characterized by average Ri >2 throughout the year). The vertical distribution of turbulence in the overflow thus differs between the channel section and the slope section.

From each mooring array (C and S), daily averages of volume transport, mean temperature, layer thickness, and cross-sectional area binned in temperature classes were used to construct a volume-averaged heat budget for the time period spanned by both arrays (30 May 2012–17 January 2013). The budget method assumes a steady state and follows and . Two layers were considered: one delineated by the seabed and the 3 ∘C isotherm and the other one above it, bounded by the 3 and 6 ∘C isotherms. Mooring array C defines the inflow section, while array S is the outflow section of the control volume (see Fig. ).

In steady state, the volume budget can be written as QBi=Q3+QBo+QEK,QUi+Q3=Q6+QUo, where indices B and U refer to the bottom and upper layers, respectively; i refers to inflow (through array C) and o refers to outflow (array S), and all quantities are time averaged over a suitable time window (defined below). The inflow and outflow are first integrated vertically over the layer and then horizontally across the section before time averaging. Spatial and temporal averaging operators are omitted from the equations for simplicity. QEK is the bottom Ekman transport, and Q3 and Q6 are the average vertical transports (positive upward) across the 3 and 6 ∘C isotherms, respectively.

The heat budget can then be expressed, using the time-averaged quantities, as QBiTBi=FH3+3Q3+QBoTBo andQUiTUi+FH3+3Q3=FH6+6Q6+QUoTUo. In this section we use kinematic heat flux, that is, the temperature flux vT. The diffusive heat flux, averaged over volume and time, is approximated as FH=-Kz(dT/dz)LW, where Kz is the average vertical diffusivity, L is the horizontal along-path separation between the arrays, W is the average plume width defined by the given isotherm, and dT/dz is the average vertical temperature gradient. This approximation is typically invoked in the budget calculations, and is valid if the local values of diffusivity and the vertical temperature gradient are not correlated. Our resulting estimates of average vertical diffusivity should thus be interpreted with this caveat in mind. L was chosen as 60 km. In the calculations we used transport weighted temperature for the layers see e.g.. For each layer, dT/dz is calculated as the difference between (transport weighted) average temperature divided by the difference in average thickness in each layer. We assumed that the upper layer underlies a 50 m thick top layer at a temperature of 7.5 ∘C.

The budget was calculated using averages over 30-day long segments of the daily time series which are moved by 1 day throughout the record (that is, 1-day moving 30-day width windows). The 30-day window sufficiently averages over the 3–5-day mesoscale variability. While the steady-state assumption is questionable, the variability in section volume transports is mainly contained in timescales less than 10 days (see Fig.  for the variance-preserving transport spectra). Budget calculations using time-average windows from 5 days to 3 months with 5-day increments (not shown) confirm that the results are not sensitive to the choice of the 30-day window after approximately 15–25 days. We thus expect the budget calculations to be fairly representative of the steady state, and to yield a more accurate estimate of entrainment rates compared to analysis of snapshots of hydrography and currents from single cruises. Because the mesoscale variability is averaged, the eddy fluxes discussed in Sect.  are not included in the formulation of the heat budget. The Ekman transport is QEK=(CDV2/f) L, where the drag coefficient CD=3.7×10-3 is from direct turbulence measurements of the FBC overflow , and V is the spatial mean flow of the plume at the given time. For V we use an average over both the bottom and upper layers and over both arrays (average transport divided by average cross-sectional area). The results are not sensitive to using only the bottom layer average. The long-term mean value of QEK estimated in this way was 0.3 Sv. QEK does not contribute to the heat budget because the near-bottom temperature is approximately 0 ∘C and thus equal to the reference temperature.

The magnitude of the different terms in the budget can be compared to justify the exclusion of lateral eddy fluxes and the heat flux associated with QEK. The average advective heat fluxes, QT, and diffusive heat fluxes, FH, are of the order 1 Sv ∘C with 1 order of magnitude smaller standard deviation. QT in the upper layer is 2 (section C) to 4 times (section S) larger than the values in the bottom layer (approximately 1.3 Sv ∘C at both sections), and the diffusive fluxes are approximately 0.7 Sv ∘C downward, across each layer. In comparison, the bottom Ekman heat flux averaged over both sections is -0.02 Sv ∘C (hence negligible). Using the magnitudes given in Fig. , the net average eddy heat flux integrated to the 6∘C isotherm, averaged over both sections, is of the order 100×105 W m-1, or integrated over L= 60 km (and converting to temperature flux) is 0.15 Sv ∘C. The uncertainty is not quantified and is possibly large. Nevertheless, the eddy heat flux term is 1 order of magnitude less than the advective heat fluxes and approximately 20 % of the diffusive flux. The contribution of the eddy heat flux to the heat budget is probably much less since the rotational component of the eddy flux is not removed (see Sect. ).

The heat budget was used to estimate rates of entrainment and detrainment (Fig. a). Throughout the whole period there was always detrainment from the bottom layer and entrainment into the upper layer: Q3 was positive throughout, with a long-term mean of 0.2 Sv, and Q6 negative, with a mean of -0.2 Sv. Entrainment in the upper layer peaked in June and August, and was at minimum levels in October and December. December was also a time of minimum detrainment in the bottom layer.

The mean vertical diffusivities inferred from the heat budget were (120± 43)×10-4 for the bottom layer and (30± 15)×10-4 m2s-1 for the upper layer. The sensitivity of these results to the choices of various parameters in the heat budget was tested. Increasing the value of L chosen to represent the distance between the mooring arrays by 10% reduced Kz in each layer by 8–9%. An increase in assumed thickness of the top layer overlying layer U from 50 to 100 m led to an increase in mean KzU by 14%, while keeping the thickness at 50 m, and instead increasing the temperature by 1 ∘C reduced the mean KzU by 20%. Determining dT/dz from temperature and thickness values at only one mooring array, instead of averaging over both, led to a 13% increase (using values from C) or decrease (values from S) in KzU. The sensitivities are thus less than the error bounds (1 standard deviation) given for Kz.

The topography, with the narrow channel opening onto the Iceland–Faroe slope, affects the anatomy and mixing of the FBC overflow. The overflow plume became markedly thinner between mooring sections C and S. Not only the average temperature but also the shape of both the velocity and temperature profiles changed, so that the thickness of the well-mixed bottom layer decreased (see Fig. ). The reduction in height reflects the spreading and thinning of the plume on the open slope after escaping the topographic constraint of the channel. The thinning of the plume with distance downstream was accompanied by increased velocities (see Figs.  and ), as previously observed by e.g. and .

Inside the thick core of the plume in the channel, there was a 150 m thick weak shear layer. As shown by , the flow in this layer was relatively stable (higher Ri) compared to the more turbulent layers above and below it (in the interface and towards the bottom, respectively). The thinner and faster plume at section S, however, had no such quiescent layer. The low values of Ri (average <1 throughout the year) below about 150 mab at S and for a layer centred at about 300 mab at C demonstrate the high potential for turbulence generation due to the strongly sheared flow, despite the strong stratification. Similar results from earlier process studies are thus corroborated by our longer-term data set.

At the mooring section in the channel, the plume was faster, thinner, and colder toward the southern (Faroe Bank) side of the section. This is in agreement with the observation by that the core velocity decreases and the thickness of the plume increases toward the right (looking downstream), and is a typical structure for a gravity current confined to a channel in a rotating system cf.. The stronger thermal stratification found on the southern side is an example of the “isotherm pinching” observed on the southern side of the channel e.g..

In classical descriptions of dense overflows, the volume transport of an overflow is assumed to increase gradually – or abruptly at mixing hotspots – due to entrainment of the overlying waters . For the FBC overflow, intensive mixing has been observed in locations just west of the channel exit . The mixing results in large changes in water mass properties; south of Iceland the boundary current contains only about half overflow water and half admixed upper layer water . If there was no detrainment as suggested by, a doubling of the overflow volume transport might thus be expected . On the basin scale, the total transport of Iceland Scotland Overflow Water increases with distance along its path through the North Atlantic by on average 1.8 Sv over 1000 km which would correspond to 0.1 Sv between sections C and S. observed an increase in dense water transport (although the densest classes vanished) between CTD sections progressing downstream in the FBC outflow region. A recent model study of the FBC overflow by showed a downstream increase in overflow volume transport (for temperatures ≤6 ∘C) of about 0.6 Sv over 90 km.

Our data set, however, shows no corresponding increase from mooring array C to array S. To explore this surprising result, we must first ask ourselves whether our measurements covered the entire plume at both sections.

At array C, overflow water on average occupied the lower 120 m at the southernmost mooring (see Sect. ). The southern boundary of the plume must thus be south of the mooring section, but because the flow here is constrained by topography, the plume cannot extend far to the south. Our horizontal extrapolation adequately compensates for the edges of the plume; our mean volume transport estimate for water colder than 3 ∘C of 1.3± 0.3 Sv is within the range of earlier estimates for the FBC overflow varying between 1.1 and 2.1 Sv . The estimated volume transport at array C is about two-thirds of the kinematic overflow estimate at the sill and Appendix .

The S section was not located in the channel, but on the open slope. The question of how that mooring array was positioned with respect to the plume path is therefore important. First, we consider the statistics of the observed temperature and velocity at the moorings at section S. There was a clear presence of overflow water throughout the year at the two moorings S2 and S3 in the centre of the array, while at the two moorings at the edges, cold plume waters were only occasionally present. At the northernmost mooring, S1, overflow waters (T≤3 ∘C) were found at 80 mab (the only measurement level here) less than 10 % of the time and modified overflow waters (T≤6 ∘C) less than 35 % of the time. At the southernmost mooring, S4, the corresponding percentages at the same height above seabed were less than 2% for T≤3 ∘C and about 36 % for ≤6 ∘C. At both moorings, the lowest temperatures were correlated with stronger along-stream velocity, a clear indication of plume presence on such occasions. Cold water was present at these moorings only during periods with energetic oscillations , suggesting that it is brought here by the eddy motion. Most of the time, however, the overflow was found in the centre of the array while absent at the outer moorings (i.e. the boundaries of the plume were found within the mooring array).

There are no channel walls on section S to contain the overflow; however, the shallow topography north of array S likely prevents large excursions of the plume to the north (upslope). There is no evidence in the literature of a plume path north of mooring S1. The mean current of the overflow mainly follows isobaths , until it begins to descend under the influence of friction . Are there then downslope excursions that make the overflow pass south of section S? The occurrence of cold water at mooring S4 was rare, and the long-term average velocity there was approximately 0 (see Figs.  and ). Our observations thus do not show any evidence of significant overflow at (or beyond) the southern end of array S. However, we cannot rule out the possibility that a fraction of the overflow plume was not captured by our array, especially during periods of strong mesoscale activity.

A second important issue to consider is the bifurcation of the flow. The overflow is known to split into two branches. The shallower branch flows westward, approximately following the isobaths along the Atlantic flank of the Iceland–Faroe Ridge, and the deeper branch turns southwestward and descends more directly into the southern Iceland Basin . The splitting appears to occur between the secondary sill at about 9∘ W and a topographic “bump” or plateau further downstream . In designing our mooring array S, we chose a location to the east (i.e. upstream relative to the overflow) of the suspected bifurcation location to ensure as complete a lateral coverage of the plume as possible. Does the weak flow at mooring S4 indicate that the bifurcation occurred upstream of array S and that S4 thus was located between the branches? If, for example, there is more substantial detrainment due to bottom Ekman transport than we have assumed (estimated long-term mean QEK= 0.3 Sv), might this detrained water form a separate branch that turns southward before reaching array S? These questions can only be answered speculatively based on our present measurements. We turn to the literature to find indications of where the bifurcation is most likely to occur.

Repeat hydrographic sections from the multi-vessel Overflow survey in June 1960 showed the cold core of the overflow centred at about 800 m depth, 10∘ W station F1 at 61∘54′ N, 10∘06′ W – very close to the “bump” centred at 61∘54′ N, 10∘07′ W; cf.. However, maps of overflow thickness and percentage overflow water based on the same survey data suggested a bifurcation of the overflow beginning already upstream of this see Figs. 3:108 and 3:109 in. Synthesizing the observations of the 1960 survey, the Overflow 1973 experiment , and some evidence from geological studies, suggested an “alternative path” (see their Figs. 43 and 49) for the overflow, turning southwestward and descending more directly and deeper in the (south-eastern) Iceland Basin. In a more recent study, the bottom temperature distribution measured by gliders indicated a branch of the overflow descending deeper than 1000 m, somewhere in the region close to our mooring section S purple line in Fig. ; cf.. This branch could not be resolved because of the 1000 m depth rating of the gliders.

During a survey conducted in June 2000, found dense waters and high velocities as deep as between the 1000 and 1200 m isobaths at their section G, only about 5 km downstream of our array S (Fig. ; for comparison, mooring S4 was located at a depth of 1082 m). The authors remark, however, that the occupation of section G shown in the paper (their Figs. 23 and 24) took place after a storm when the overflow transport doubled. A better temporal averaging is provided from the survey in June 2012 reported in . In a section repeated five times over the course of 70 h, located about 10 km upstream of our mooring array S, the high-velocity core of the overflow was mostly centered close to, or shallower than, the 1000 m isobath, with reduced flow at the outer two stations .

mention that a division into two branches at about 900 m depth is indicated at their westernmost section, H, located about 50 km downstream of our array S. (Section H is marked by green dots but not labelled in Fig. ; it is the section the lower part of which is parallel and close to Geyer's section A.) They suggest that the bifurcation occurs somewhere between their sections G and H, where the flow decelerates and is “separated by a plateau in the topography” p. 910. Most likely this flow separation is thus caused by the same topographic feature that refer to as the “bump” (see Fig. ).

Mean current velocity vectors from July to November 1999 presented by show a split into two branches at about 11∘ W (their Fig. 3), also west of our mooring section S (see Fig. ). Note that this is a split of a deeper flow around a different topographic feature, southwest of the bump of . The location of the knoll dividing the flow in the paper by fits with the observation by that data sets from the Overflow 1960 and 1973 expeditions both show a “similar three-way splitting of the overflow tongue at about 61∘45′ N and 11∘ W”.

The complex topography that varies strongly on small scales is a major factor controlling overflow pathways in simulations . Tracer-weighted average plume paths from model experiments by and generally turn more southward and descend below the 1000 m isobath close to (but typically west of) the location of array S.

Further evidence of the overflow splitting into a deeper branch and a shallower branch in the region just west of array S is found in velocities measured at 64 mab by a deep towed vehicle Fig. 7 in. The current below about 900 m veered downslope at a section located about 20 km downstream of array S (Fig. ). Two out of the three bottom-following float trajectories shown by followed roughly along the 1000 m isobath until at least 11∘ W, while one track descended earlier (closer to mooring array S). It should be noted that for the latter, as well as for one of the other floats, the positions along this part of the track were only estimates not directly determined by sound ranging.

In summary, mooring array S is located in the general region where the FBC overflow divides into branches, and we cannot rule out the possibility that a branch has been diverted before reaching S. A detailed comparison with the literature, however, suggests that the majority of earlier works place the split a short distance downstream of array S. The bifurcation (or one bifurcation) appears most likely to be caused by the topographic bump centred at about 61∘54′ N, 10∘7′ W, approximately 30 km downstream of array S.

Assuming on this basis that mooring array S did not miss a substantial branch of the overflow, our observations indicate that the simplified view of along-path volume increase may not be accurate for the FBC overflow. As the plume proceeds into the stratified ambient water, there is substantial detrainment from the deeper layer, of comparable magnitude to the entrainment into the interfacial layer. Although the maximum plume velocity was measured at the downstream array, the volume transport of plume water with T≤3 ∘C decreased to 0.8± 0.4 Sv at section S. As expected, the volume transport of the coldest waters disappeared (T≤ 0 ∘C) or was reduced (0∘C <T ≤ 1 ∘C) between section C and section S. A reduction in the low-temperature classes of FBC overflow transport within the first few hundred kilometres downstream of the sill was shown by , while found a downstream decrease in transport regardless of water class definitions used. Similarly, found no increase in volume transport in the bottom cold layer despite dilution typically indicating entrainment. They suggested that the dilution might instead be caused by warmer interface layer water being supplied to the bottom layer by the transverse circulation, replacing bottom water lost through bottom Ekman drainage. Throughout our year-long measurements, there was persistent detrainment from the bottom layer and entrainment into the interfacial layer; the decrease in cold water transport between sections C and S was compensated for by a roughly equal increase in transport in the intermediate temperature range 3∘C<T≤6 ∘C. The total mean transport of overflow and modified overflow together thus did not change between the channel and the slope sections. Simple estimates of the contribution of the FBC overflow to the global thermohaline circulation assume doubling of the volume transport due to entrainment. Our observations suggest that, despite substantial entrainment, volume transport does not increase significantly because of detrainment. In that case, the FBC overflow could be less important for the production of North Atlantic Deep Water and for assessing changes in the global thermohaline circulation than is usually assumed .

No clear seasonal cycle was detected in the volume transport of the FBC overflow, but there was variability at monthly timescales. Transport peaked at both sections in June and August 2012, coinciding with maximum entrainment into the upper layer. Minimum entrainment in the upper layer occurred in October 2012, a month that was marked by minimum bottom layer temperatures both at section S and the mid-mooring M1. found no seasonality of plume thickness, temperature, or velocity in year-long measurements, but showed a seasonal variation of kinematic overflow at the sill of about 10 % of the mean over a 10-year period. The relatively low amplitude of the seasonal cycle compared to shorter-term variations means it only becomes apparent in data sets covering several years. The seasonal variability in the outflow has been linked to seasonality in the barotropic northward flow of Atlantic Water in the Faroe–Shetland Channel . showed that variability in volume flux on shorter timescales is linked to the local barotropic forcing.

Mean diffusivities inferred from the two-layer heat budget constructed for the overflow were (30± 15)×10-4 m2s-1 for the upper layer and (120± 43)×10-4 m2s-1 for the bottom layer, within the range of values observed during research cruises in 2010 and 2012 . Earlier estimates have ranged between 50 and 500 ×10-4 m2s-1 cf.; however, they suffer from not resolving the mesoscale variability. The vertical diffusivity in both layers shows a factor of 5 increase in late August and late December. These changes coincide with periods of high eddy kinetic energy reported in , suggesting that the energetic mixing is associated with mesoscale variability.

propose that the mesoscale variability in the FBC is caused by baroclinic instability of the plume. In their model experiment, the baroclinic instability and resulting variability are energized downstream of the sill, and high baroclinic conversion rates occur 40–60 km downstream of the sill. This is consistent with our Fig. , which shows more energetic mesoscale transport fluctuations at section S than at C.

Eddy heat fluxes were divergent in the central and deep part of the plume and convergent between the northernmost (upper) moorings on each section, indicating that the effect of eddy heat transports is a net cooling of the deep part of the plume and a net warming of the shallower part. The observed temperature fluxes of the order of 0.1 Kms-1 were about 10 times larger than the values reported by from the region downstream of the Denmark Strait overflow and in agreement with values reported by and the modelled values by . Since density is largely determined by temperature in the FBC, the upslope temperature flux is equivalent to an upslope buoyancy flux, i.e. a release of potential energy. This is consistent with eddy generation due to baroclinic instability, shown by to be the mechanism at play in the FBC. The observed heat or buoyancy fluxes are the sum of a (dynamically unimportant) rotational component and a residual, divergent component which is the one responsible for the energy conversion . While our data set does not allow us to separate the two components, we note that, in this region, the total flux field and the divergent flux field are relatively similar in the model study by .