Ocean temperature, salinity, and horizontal current data were measured at two mooring arrays across the continental slope north of Svalbard (Fig. ), for approximately one year from September 2018. The western array (18° E) was recovered in September 2019, and the eastern array (24° E) was recovered in November 2019. The data are available from and have been presented in . Each mooring array consisted of one upper slope mooring (W1 and E1) near the 300–400 m isobath, one middle slope mooring (W2 and E2) near the 700 m isobath, and one lower slope mooring (W3 and E3) near the 1200 m isobath, providing a good coverage of the boundary current (Fig. b). In addition, a mooring was deployed onshore of the western array (W0). In this study, we used data from the deeper mooring pairs W2–W3 and E2–E3 (Table ) as the distance within each pair – approximately 9 km – is small enough for calculating lateral gradients (see Sect. ).

Mooring instrumentation coverage and setup, and details of data processing can be found in . The vertical instrument coverage at the moorings for September–February and March–August is summarised in Fig. . Temperature, salinity, and horizontal current records were checked for inter-consistency and compared and corrected against ship-based profiles taken at mooring deployment/recovery. All time series were hourly averaged (the sampling interval of instruments varied between 5 min and 1 h) and linearly interpolated onto the same hourly time vector. At each hourly time step, instrument depths were obtained by linear interpolation between instruments equipped with a pressure sensor, to account for mooring knockdowns. The data were finally gridded onto a regular (t,z) grid with 1 h temporal and 10 m vertical resolution.

The analysis is based on data from the deeper mooring pairs which provided current profiles with a 5–10 m vertical resolution from the seafloor to approximately 300 m depth. Here, only three levels of density measurements at approximately 300, 500, and 700 m depth were available. Our analysis is therefore based on layer averages over 300 to 700 m (Sect. ), focusing on the levels close to the hydrographic sensors to limit the errors from interpolation. For context, we present seasonal mean sections of velocity and hydrography at both mooring arrays (Fig. ). For computing mean structure of the current velocity and hydrography across the slope, a spline-Laplacian routine was used for interpolation, described in detail in . While current measurements cover the water column at W1, there were no temperature and salinity measurements available, and the hydrographic structure over the upper slope was supplemented by historical hydrographic data. The details on the gridding and the associated uncertainties are described in the Appendix of .

In order to better evaluate and interpret the analysis based on the mooring observations, we used the output from a high-resolution ocean model applying ROMS (Regional Ocean Modeling System; ) in a domain covering the continental shelf and slope north of Svalbard . The model used the same configuration as the Norkyst model system documented in but was set up with a horizontal resolution of 500 m and 35 vertical stretched bathymetry-following levels. The model configuration also included a sea ice module . The operational ocean forecasting model Barents2.5 provided boundary conditions including tidal forcing, and the operational weather forecasting model AROME-Arctic provided atmospheric input. The model was validated by , and a brief summary of this validation is provided in Appendix .

The model simulation was initialised in April 2019, and we used two-year daily output from October 2019 to align seasonally with the mooring observations. Hourly saved fields were daily averaged, and the daily fields were then detided for the fortnightly and monthly tidal constituents using UTide . The across-slope structure of the modelled boundary current and hydrography is shown in Fig.  and compared with the mooring-based sections in Appendix .

Mean structure and seasonal variability of the boundary current observed in the mooring arrays have been reported in . Here, we summarise the mean hydrography and current structure relevant to our analysis (Fig. ) using three-monthly averages grouped into September–February and March–August periods.

The boundary current was on average stronger in September–February, with a velocity core at approximately 100 m depth over the 800 m isobath, carrying AW warmer than 3 °C at a speed exceeding 20 cms-1 (Fig. a and b). In March–August, the velocity core was weaker (10–20 cms-1), deeper, and relatively diffuse in both vertical and lateral directions (Fig. c and d). Water with the highest temperatures, exceeding 2.5 °C, was at 200 m depth seawards of the 500 m isobath, but did not coincide with the deeper velocity core.

The mesoscale variability is demonstrated using records from W2, and shows elevated activity in autumn and winter (Fig. ), in the same period when the boundary current was strongest and carried the warmest water (Fig. a and b). The strongest velocity variability occurred on time-scales between 2 and 6 d and, like the elevated mesoscale activity, was observed in autumn and winter (Fig. a and c). From October 2018–January 2019, the offshore velocity component v and the temperature T co-varied on a 3–10 d time-scale (Fig. d). At that time, the current core was located near the 800 m isobath, close to W2, and carried AW (Fig. a).

The offshore eddy temperature flux, v′T′‾, at 250 m depth was positive in the autumn of 2018, peaking in mid-November, and became negative in December 2018 (Fig. b), consistent with mesoscale activity driving cross-isobath heat transport. Averaged over September–November 2018, v′T′‾ at W2 was positive at depths 150 to 350 m, spanning the temperature core. At W3, the eddy temperature flux in the upper water column could not be calculated in autumn due to mooring knockdown, but was positive at and below 400 m depth (not shown). The eddy temperature flux divergence between W3 and W2 was over four times higher in autumn and winter (September–February) than in spring and summer (March–August; not shown), suggesting stronger lateral heat loss from the boundary current in autumn and winter. While the relatively higher eddy temperature flux and its divergence in autumn and winter indicate increased mesoscale activity, we cannot directly quantify net lateral heat transport because the rotational and the divergent components of the flux could not be separated .

Despite their limited spatial coverage and sampling volume, the moorings enable year-round estimates of eddy kinetic energy and associated energy conversion rates. EKE and BT, together with its contributing terms, are shown in Fig. . BC and its contributing terms are shown in Fig. . To provide context for the magnitudes of conversion rates and their effect on changes in EKE, note that a conversion rate of 1×10-5 Wm-3 sustained over one day would produce 1×10-3 Jkg-1 corresponding to an EKE increase of 10 cm2s-2.

Several features are consistent at both mooring arrays. EKE was higher in autumn and winter than in spring and summer (Fig. a and b), with values at 300 m depth approximately four times larger than at 700 m (not shown). In contrast, BT remained weak year-round relative to the conversion rate expected from the observed EKE variability (Fig. g and h). It was negligible in summer and otherwise fluctuated within ±0.5×10-5 Wm-3. The persistently low BT reflects both cancellation of oppositely signed momentum fluxes (Fig. e and f) and periods of weak velocity shear (Fig. c and d). BC was generally largest in autumn and winter at both sites, ranging from -3×10-5 to 4×10-5 Wm-3, and decreasing to near-zero values in spring and summer (Fig. e and f). BC was generally strongest at the uppermost measurement level (310–320 m), where variability in density and cross-slope velocity was greatest, and decreased with depth (Fig. c and d). Despite these similarities, BC estimates showed substantial variability between depths and moorings, particularly in autumn and winter, primarily due to differences in eddy density fluxes despite being measured only 100 m apart vertically, e.g. in October and November 2018 (Fig. c and d).

At the western array, the autumn- and wintertime EKE reached nearly 40 cm2s-2, and decreased to below 20 cm2s-2 in summer (Fig. a). BT was generally weak, with two notable exceptions. In January 2019, BT was positive (Fig. g). Based on estimates at 100 m intervals (not shown), the values at the uppermost level (300 m) reached 1×10-5 Wm-3 driven by negative momentum fluxes combined with positive shear. This signal weakened with depth, reducing the layer-average BT. In late March 2019, a second event occurred, when strong positive velocity shear and positive momentum flux at W2 produced a similarly large but negative and shorter-lived BT (Fig. g, c, and e).

The eastern array had generally lower EKE than the western array, but with a comparable seasonal contrast. BT showed a similar overall behaviour to the western array. Typically low values increased in two events, but with opposite signs to the western array. BT was negative in January 2019 due to modest momentum fluxes and shear, while from late March to early April 2019, BT was positive due to a period of negative momentum fluxes and an offshore displacement of the boundary current (Fig. h, d, and f). During the February EKE maximum, BT remained negligible due to near-zero shear and momentum fluxes.

At the western array, the most notable BC event occurred at 320 m at W2 from mid-October through November 2018. BC increased from negligible values to 4×10-5 Wm-3 and remained elevated for nearly a month. This event coincided with a strong mean current, high EKE (Fig. a), a modest and persistent negative isopycnal slope, and an elevated eddy density flux from an increased variability and covariance of v and ρ, with dominant periods of 2–10 d (similar to Fig. d). During and following this event, EKE at W2 remained high, peaking at 88 cm2s-2 at 300 m depth. However, such concurrent increases in EKE and BC were not commonly observed. A similar but shorter-lasting event occurred at 310 m depth at E2 from mid-November to early December 2018 (Fig. f).

In summary, current strength, temperature variability, EKE, eddy momentum and density fluxes, BC, and to a lesser extent BT, were all generally stronger in autumn and winter than in spring and summer. Although the measurements do not resolve the upper ocean and surface layer, the available vertical coverage indicates a tendency for enhanced EKE, BT, and BC at the uppermost measurement levels. While EKE was often elevated during periods of higher fluxes and BC, the relationship between them was neither consistent nor robust.

Advection of eddies past the moorings complicates the interpretation of our estimates in several ways. A fraction of the calculated EKE may be attributed to eddies advected by the mean flow, adding to the local conversion from mean to eddy energy, potentially leading to an overestimation of locally generated EKE. In addition, the use of a band-pass filter to estimate fluctuations may attenuate contributions from mesoscale features when the filter cutoff exceeds the eddy advection time scale. The advection time scale is 24–40 h, based on along-slope velocities of 10–15 cms-1 and an eddy lateral length scale of twice the baroclinic Rossby radius of deformation (∼7 km). This range overlaps with, and partly exceeds, the 35 h cutoff used in the EKE calculations. Alternative calculations using a shorter cutoff of ≈24 h, excluding the diurnal tidal band while retaining most mesoscale variability, show no significant differences (Appendix ; Fig. a and b).

We analyse output from the high-resolution model to assess whether the conversion estimates using the simplified expressions for BT and BC, based on limited depth levels and mooring locations, are representative of a broader region. To do so, we calculate EKE, BT, and BC using the full conversion terms (Eqs.  and ) and vertically average over 100–2000 m to obtain their spatial structure (Fig.  and Sect. ). We then show the temporal evolution of the energetics along the boundary current pathway by averaging over a 20-km diameter region centred on the 1400 m isobath (Fig.  and Sect. ). Additionally, we apply the simplified formulations (Eqs.  and ) to velocity and density time series extracted at virtual moorings, using the same method as for the moorings (Fig.  and Sect. ). These results are compared with both the observational estimates and the spatially averaged values along the boundary current pathway.

We observed higher EKE during autumn and winter from the mooring observations (Fig. a and b). This agrees with who found higher EKE between October and March in a high-resolution model and with who found increased geostrophic surface EKE in the autumn and winter months. Upstream at the 78°50′ N mooring array, the WSC is most energetic from December to May, with a peak in February . In AW inflow regions, such as the Barents Sea Opening, Fram Strait, and the western Nansen Basin, wintertime EKE can be relatively higher. This is attributed to the typically lower sea ice cover in these areas, which permits stronger wind forcing and supports weaker stratification and higher instability .

During the same periods of elevated EKE in autumn and winter, we also observed larger energy conversion, with BC as the dominant contribution (Figs. g and h and e and f). The important role of BC in autumn and winter seen in the mooring observations is supported by the model, lending confidence to the simplified calculations from the moorings. Previous studies from upstream regions in Fram Strait and west of Spitsbergen have similarly shown that the boundary current is baroclinically unstable . The high wintertime EKE in the WSC has been linked to baroclinic instability due to weak stratification and a stronger vertical shear . Similarly, in the Eurasian Basin, reported a spatial agreement between EKE and BC, suggesting that baroclinic instability supplies eddy energy locally. Our finding that both EKE and BC are greater in autumn and winter indicates that baroclinic instability may similarly drive higher EKE north of Svalbard as well.

While BC was stronger in the mooring observations, the volume-averaged conversion rates in the model show similar contributions of BT and BC (Fig. c and d). Averaged along the boundary current path, BT was comparable to and larger than BC for most of the time (Fig. b and c). Considering the uncertainties and limitations of point-based estimates of conversion rates, highlighted by the limited ability of the virtual moorings to reproduce the volume-averaged conversion rates along the boundary current pathway (Sect.  and Fig. ), we may assume there can be a substantial contribution to energy conversion from barotropic instability. Upstream of our mooring site, the boundary current west of Spitsbergen has been shown to be barotropically unstable , and the results of also indicate that barotropic conversion is non-negligible.

Conversion rates have been estimated in several other studies using mooring observations and models. Across the pathway of the WSC in Fram Strait, conversion rates inferred from moored observations were highest near the shelf break during winter, with a pronounced seasonal cycle, peaking in February and reaching its minimum in August . Their largest monthly mean BT, 2.5×10-5 Wm-3 at 75 m depth in February, is about ten times higher than our layer-averaged maximum during January 2019 at the western array. At the same location, their BC value reaches 15×10-5 Wm-3, which exceeds our largest monthly mean, 2.9×10-5 Wm-3 at 320 m in November. In our data, both BT and BC tended to be larger in the shallower part of the 300–700 m depth range, so it is plausible that values at 75 m could be closer to those of . It should be noted that we used slightly different passbands than to obtain fluctuations. A comparison between choices of passbands and filters is shown in Appendix . If we use their passband of 48 h to 30 d instead of our 35 h to 14 d, our November-mean BC increases by 25 %. Conversion rates in Fram Strait estimated using two high-resolution models , reach magnitudes of 5×10-5 m3s-3 along the slope. Depth-integrated values in m3s-3 are numerically equivalent to depth-averaged values with units Wm-3 for a ∼1000 m-thick layer of seawater with density ∼1000 kgm-3.

On the southern flank of the Yermak Plateau at 80° N, further downstream of the mooring observations of , estimated conversion rates from mooring records. Similar to our results, they observed that both BT and BC were highly variable in depth and time, with no consistent or strong correlation with EKE. Their observed EKE levels exceeded what would be expected from the local conversion rates and could be partly accounted for advection from an energetic region upstream of the moorings reported in .

Negative conversion rates, observed in both our mooring records (lasting weeks) and the model (lasting months in some regions), have also been reported in other regions e.g.. Theoretically, negative BT implies a transfer of energy from eddies to the mean flow, while negative BC implies energy transfer that steepens isopycnals by lifting denser water toward the shelf. Our observations provide some support for these interpretations. For example, a period of negative BC in January–February 2019 was followed by an increase in isopycnal slope (Fig. c and a). In the model, a wide zone between 80 and 110 km along the isobath with predominantly negative BT (Fig. b), shows a significant negative correlation (r-0.47 to -0.72) with the along-slope current velocity (not shown). Here, decreasing BT is associated with increasing current strength, consistent with a transfer of energy from eddies and to the mean flow. Correspondingly, the mean velocity transitions from below- to above-average values downstream (Fig. a, right axis of upper panel), suggesting a local strengthening of the boundary current. However, this relation between negative BT and an increase in u¯ is not found in all regions. Due to uncertainties in the estimates from the moorings, we cannot determine whether the observed change in the isopycnal slope is a direct consequence of negative BC or a coincidental alignment. However, the occurrence of negative conversion rates in both observations and high-resolution models supports the interpretation that energy can be transferred to the mean flow or into available potential energy.

Understanding heat loss from AW along the boundary current is key to regional climate and oceanography north of Svalbard, with implications for the broader Arctic Ocean. estimated the seasonal evolution of along-stream heat loss of AW between the western and the eastern mooring arrays and showed that air–sea heat fluxes alone cannot account for the observed cooling rates in winter, spring, and summer. This implies that additional mechanisms contribute to the along-stream heat loss of AW, such as lateral exchange proposed by and .

Several observations from the slope north of Svalbard indicate the presence of such lateral processes. Temperature variability exhibited an onshore–offshore mode with a 6–7 d periodicity that was correlated with across-slope velocity variability, and surface geostrophic EKE was elevated near the current core . Anticyclonic eddies are a likely agent for this exchange. Limited observations north of Svalbard showed anticyclones carrying warm anomalies offshore from the boundary current and other eddy features linked to baroclinic instability . Anticyclonic eddies carrying AW or warm anomalies have also been identified in high-resolution modelling studies both north of Svalbard and upstream . The model results presented here also indicate substantial conversion rates both near the mooring sites and further offshore (Fig. ).

Seasonal contrasts further point to a role for mesoscale processes in the heat budget. In autumn, along-stream heat loss can largely be explained by direct atmospheric cooling, when the AW core is warm, shallow, and the region is ice-free . In winter, however, along-stream heat loss increases substantially while the AW core subducts and becomes less exposed to the atmosphere. During this period, EKE remains elevated, suggesting that lateral exchange associated with mesoscale activity becomes more important. In spring, surface heat loss is strongly suppressed by sea ice, while mesoscale activity persists at reduced levels, again indicating a contribution from eddy-driven lateral exchange. Consistent with this interpretation, eddy temperature flux divergence estimated from the moorings (Sect. ) indicates stronger lateral heat loss from the boundary current in autumn and winter. Although these local estimates are uncertain, their seasonal structure is consistent with the periods when atmospheric heat fluxes cannot account for the observed cooling. Model-based estimates of integrated eddy temperature flux divergence along the boundary current show a similar seasonal signal, with consistently positive divergence during winter and the largest monthly values occurring in November–February (Sect. , Fig. e).

In contrast, summer is characterized by weak mesoscale activity, small EKE (Figs. c and a and b, and a), negligible observed (Figs. g and h and e and f) and below-average modelled (Fig. b and c) energy conversion rates, and small eddy temperature fluxes and their divergence. Mesoscale eddies are therefore unlikely to be the dominant agents for heat loss in summer. However, estimated average along-stream cooling of AW in summer comparable to winter, suggesting that all the heat loss in summer must be lateral or downward by turbulence at the base of the AW layer. A likely candidate for summer lateral heat loss is topographic Rossby waves, which have previously been linked to increased heat loss from the WSC . An ongoing study (F. Nilsen, personal communication, 2025) indicates that the shape of the continental slope near the western mooring array, together with the reduced stratification and slower current observed in summer, could support topographic Rossby waves with diurnal periods, energized by the diurnal tide.

Our observations and model results, together with previously published evidence, indicate mesoscale activity and eddy-driven exchange play an important role in the heat loss of AW from the boundary current.

The pattern of alternating signs in the energy conversion rates along the continental slope (Fig. c) is consistent with findings from other regions such as Fram Strait and the Lofoten escarpment . North of Svalbard, these alternating patterns were connected to veering isobaths (Sect. ), indicating a close connection to bathymetry, as also noted by . This connection is likely a manifestation of the necessary condition for barotropic instability, which depends on spatial variations in bathymetry through the topographic β (defined in Sect. ).

Our metric for the lateral sign change of the PV gradient (Fig. d) identifies several areas along the 1400 m isobath where the necessary condition for barotropic instability could be satisfied. However, the spatial and temporal patterns of BT are not directly aligned with these regions (Fig. b). For example, BT was positive in several regions such as upstream of the eastern array, where the condition for instability was not met, while in some areas where the condition was fulfilled, BT was weak or negative. The latter is expected since the necessary condition does not ensure instability will occur. Overall, the connection between the potential for barotropic instability and the observed barotropic conversion is inconclusive. This can partly be explained by advection: instabilities or eddies generated upstream, in regions where the instability condition is met, may be transported into downstream areas with different local conditions. Similar observations were made by e.g. .

Baroclinic instability requires that the cross-stream gradient of the effective potential vorticity changes sign with depth . With only two moorings at each array, we cannot reliably assess this condition. However, small values of the geostrophic Richardson number, Ri=N2/S2, may indicate conditions favourable for baroclinic instability , as strong stratification tends to suppress, while strong shear promotes instability. We use the background fields of stratification, N2=-g/ρ0∂ρ/∂z, and shear, S2=∂u¯/∂z2, instead of the thermal-wind shear, which are representative of the geostrophic scales. Overall, Ri was significantly smaller during winter than in summer, consistent with the enhanced EKE and stronger BC in winter observed in the mooring records. Median values of Ri in summer were 2–20 times larger than in winter, depending on the mooring and the measurement level analysed. During winter, Ri was less than 50 21% of the time at W2 (30% at W3), compared to only 5% during summer. Strong vertical shear, connected to the horizontal density gradient through the thermal wind relation, was the main factor reducing Ri.

The mean flow advecting eddies can also explain differences between measured energy conversion and observed EKE. Eddies generated in regions of high instability can be transported downstream by the mean circulation – for instance, eddies observed in the eastern Eurasian Basin have been traced as far back upstream as the Yermak Plateau, the area north of Svalbard, and the slope between Svalbard and Franz Josef Land – resulting in enhanced EKE away from the generation sites . Furthermore, some EKE can originate from meandering of the current, hence elevated EKE levels do not necessarily indicate local eddy occurrence .

In summary, several factors complicate the interpretation of instability diagnostics, conversion rate estimates, and local EKE: the necessary condition for barotropic instability identifies regions susceptible to instability, but may still not coincide with regions of high positive barotropic conversion; local conversion rates may not align with local EKE due to advection of eddies from upstream; and EKE levels themselves may not reflect the eddy formation or occurrence.

As noted by earlier studies e.g., conversion rates based on mooring observations are subject to significant uncertainty and therefore should be interpreted with caution. A primary limitation is that moorings provide fixed-point measurements, which may not capture the full spatial and temporal variability of energy conversion. Model results show that conversion rates are patchy and intermittent, as illustrated by differences between volume-averaged rates and estimates from virtual moorings or cross-slope segments. Sufficient averaging in time and space is therefore essential for reliable conversion rate estimates.

Another limitation is that mooring arrays do not allow estimation of all terms in the conversion rate equations. This limitation is illustrated using the estimates from the simplified formulas at virtual mooring and segments vs. the full equations. In addition, a comparison of the four terms in the BT formula (Eq. ) showed that the two divergent terms were more often positive, while the rotational terms were occasionally negative. In a similar comparison for the two terms in the BC formula (Eq. ), the signs often opposed. Hence, a negative value in one term estimated from moorings does not preclude a positive total conversion rate, or vice versa. Even under the assumption that point-based estimates from one term adequately represent regional conditions, uncertainties would still arise from methodological and instrumental limitations associated with the mooring data: for example a larger mooring spacing relative to the Rossby radius will underestimate the isopycnal slope and thus BC ; and similarly, inadequate resolution of the lateral velocity shear will underestimate BT.

Our model results suggest that much of the conversion occurs deeper and further offshore than our moorings (Fig. ). Nonetheless, a considerable fraction of the conversion occurs over the slope and at the depth sampled by the moorings, as supported by both analyses of virtual moorings (Fig. ) and by comparing full-depth cross-slope segment estimates with those at limited-depth segments with shorter lateral span. The virtual mooring results further demonstrate that the substantial spread observed with the BC rates at various depths (Fig. ) is to be expected. As such, mooring-based estimates are better regarded as order-of-magnitude approximations indicating the possible conversion for that time in that region.