In this study, we use a combination of historical hydrographic data, recent hydrographic transects collected during late autumn, and year-long current and hydrography time series from a mooring. The mooring was deployed at 70 m depth on Hopenbanken, a shallow bank, between the Storfjordrenna trough and the interior of the Barents Sea (Fig. ).

The historical data include hydrographic profiles extracted from the University Centre in Svalbard (UNIS) Hydrographic Database (UNIS HD; ), covering the vicinity of the mooring area from 1930–2019.

Recent hydrographic transects were collected on 14 November 2019 aboard Kronprins Haakon , on 16 October 2020 aboard G. O. Sars , and on 13 November 2021 aboard Kronprins Haakon .

From 29 September 2018 to 14 November 2019, temperature, salinity, and ocean currents were measured using instruments on the mooring located at 77°16.116′ N, 24°24.402′ E (Fig. ). The mooring was a trawl-proof bottom frame equipped with one conductivity–temperature–depth (CTD) recorder (Sea-Bird MicroCAT SBE 37-SM (unpumped)) and one acoustic Doppler current profiler (ADCP; Nortek Signature250), both housed within the frame with transducers pointing upwards. The MicroCAT instrument recorded temperature, salinity, and pressure near the bottom (at about 68 m) for the entire period at a sampling interval of 15 min. The ADCP profiled current speed and direction through the water column using 25 cells with a thickness of 3m at an averaging interval of 20min. All data from the mooring were interpolated onto a common hourly time vector prior to analysis.

We corrected the velocity direction for a magnetic declination of 15°. The ADCP compass was further validated by comparing the tidal ellipses obtained from the observed current velocity with those from the Arctic Ocean Tidal Inverse Model on a 5 km grid, developed in 2018 (Arc5km2018) . For the M2 tidal constituent, the ellipse inclinations from Arc5km2018 and the depth-averaged measured current were approximately the same. For the other significant semidiurnal constituents (S2, N2, and K2), the ellipse properties from the measured currents were in fairly good agreement with Arc5km2018 – i.e. ellipse inclinations agreed within 14°, and semi-major axes agreed within 1cms-1. Since M2 is the dominant constituent in the study area, no further compass corrections were deemed necessary. We then referenced the current velocity profiles to the surface to account for sea-level variations detected by the ADCP's altimeter.

We extracted data on wind speed and direction for the region from the ERA5 reanalyses – this included hourly data on single levels with a spatial resolution of 0.25° (). Additionally, we extracted data on sea surface temperature (SST) and sea ice concentration (SIC) from the Global Ocean OSTIA Sea Surface Temperature and Sea Ice Reprocessed product, which comprises daily data with a spatial resolution of 0.05° ().

All data used in this study are openly available, as detailed in the “Data availability” section.

The historical hydrographic data have been optimally interpolated onto horizontal and vertical climatological sections to compare data from the past 2 decades (2000–2019) to data from the previous century (1930–2000). Due to the temporally and spatially sparse data coverage, only summer (July–October mean) sections were created. A description of gridding and interpolation, along with details on data coverage and standard deviations for the climatological maps and vertical sections, is given in Appendix .

Temperature and practical salinity from the recent hydrographic transects, the mooring, and the optimally interpolated climatological sections were converted to Conservative Temperature and Absolute Salinity, respectively, following the Thermodynamic Equation of Seawater – 2010 (TEOS-10) . In the following, we use the subscript “b” (for bottom) to refer to temperature Tb and salinity Sb near the seafloor. AW is defined as water with Conservative Temperature Θ≥2 °C and Absolute Salinity SA≥35.06gkg-1 . We refer to Atlantic-origin waters as AW that is modified or transformed en route from the WSC through the Storfjordrenna trough into relatively warm and saline water compared to the surrounding water masses; however, it is colder and less saline than pure AW. This inflow of warm water into the Barents Sea affects deep-basin stratification and sea ice growth and melt.

We identify the timescales of variability using spectral analysis. Cartesian and rotary spectral components were estimated following the multitaper method from , with Slepian data tapers applied e.g.. To analyse the time variability in the spectral components, we used wavelet transforms following , employing generalised Morse wavelets . The values for the parameters defining the Morse wavelets were set as γ=3 for symmetric wavelets and β=5 to resolve variability in both frequency and time on timescales ranging from semidiurnal to a few weeks.

The spectral analysis revealed three frequency bands with dominant variability: semidiurnal tides (periods of 10 to 14 h), “weather-band” processes (28 h to 10 d), and lower-frequency activity (7 d to 6 weeks). To study these fluctuations, we filtered the time series with a Butterworth band-pass filter. We used the highest possible order while ensuring stability for each frequency band – i.e. we employed an order of 5 for semidiurnal tides and weather-band processes and used an order of 3 for lower-frequency activity.

To describe and visualise the background conditions in hydrography and currents, we used a low-pass Butterworth filter with an order of 6 and a cutoff frequency corresponding to 10 d.

We analysed the current velocity data from the mooring to study the mean flow properties and flow variability over timescales associated with tidal variability, weather-band processes, and lower-frequency events. For this, we used two different coordinate system rotations. The coordinate system was rotated by -42° to align with the long-term mean flow along the isobaths. The rotated velocity components are denoted as uR and vR for the along-isobath (roughly southeast) and across-isobath (roughly northeast) directions, respectively. To analyse variability at selected timescales, we applied another rotation to the velocity components, denoted as ur and vr. These components were oriented along and across the direction of the highest variance within the frequency bands associated with weather-band processes and lower-frequency activity. This coordinate system was rotated by -28°, orienting ur roughly towards the east-southeast and vr towards the north-northeast.

Eddy temperature fluxes were calculated as ur′T′‾ and vr′T′‾, where T′ denotes fluctuations in near-bottom temperature and ur′ and vr′ are the velocity component fluctuations along (-28° relative to east) and across (62° relative to east) the direction of highest variance in the frequency bands associated with weather-band processes and lower-frequency activity. The fluctuations (denoted by primes) were obtained by applying a band-pass filter to the time series. The velocity data were layer-averaged over the bottom half of the water column. Although the water column was at times stratified in temperature, available CTD profiles suggest that the lower half of the water column was typically well mixed. Therefore, temperature measurements recorded close to the bottom may be considered representative of the lower half of the water column. The overline denotes a 30 d moving average.

The wind stress vector was calculated as τ=(τx,τy)=ρairCD|u10m|(u10m,v10m), where ρair=1.25kgm-3, representing the density of air; u10m=(u10m,v10m), denoting the wind vector with eastward and northward components of wind velocity at 10 m, extracted from ERA5; |u10m| is the magnitude of the wind velocity; and CD is the drag coefficient. We used CD values that depend on sea ice concentration, following . Ekman transport was calculated from the wind stress using (UE,VE)=1ρ0f(τy,-τx), where ρ0=1027kgm-3, indicating the reference density of seawater, and f is the Coriolis parameter.

Time series of SST, SIC, wind velocity, and wind stress at the mooring location were obtained by spatially interpolating onto the mooring location from the original grids, i.e. from 0.05° for SST and SIC and from 0.25° for wind velocity and wind stress.

The hydrographic environment in the Storfjordrenna trough and the shallow banks Hopenbanken and Storfjordbanken (Fig. ) has experienced warming over the past 2 decades. Summer climatology maps of depth-averaged temperature from 1930–2000 (Fig. a) and 2000–2019 (Fig. b) indicate warming across most of the area. The observed increase in average temperature exceeds 1 °C over large parts of Hopenbanken and Storfjordbanken (Fig. c). This warming is observed both at the surface and at depth (not shown). The 34.7gkg-1 isohaline extended further northeastwards in 2000–2019 compared to 1930–2000 (Fig. a and b).

Vertical climatological sections extending from the Storfjordrenna trough, across the saddle on Hopenbanken, and into the Olga Basin (Fig. ) show that Atlantic-origin waters have shoaled in the water column and now extend further east towards the saddle. This warming has occurred throughout the water column, and the saddle area has warmed by 1 to 2 °C. Additionally, salinity has increased by 0.05 to 0.35gkg-1 over intermediate depths in the Storfjordrenna trough (not shown). We also note that the past 2 decades has seen a deepening of the 34.4gkg-1 isobath and a corresponding decrease in temperature towards the Olga Basin (Fig. b and c).

Recent hydrographic transects from the saddle region, collected in November 2019, October 2020, and November 2021 (Fig. ), show that the mooring is situated in an area with notable thermal and haline gradients. Local maxima in temperature and salinity were observed near the mooring position in late autumn.

In November 2019, the transect along the saddle was cold and relatively fresh but retained some heat (Fig. a). The warmest water (-1.2 °C) was found by the mooring position in the lower half of the water column. On the northern side of the transect, temperatures were near the freezing point, accompanied by relatively high salinity (Fig. a). At this time, AW was present in the Storfjordrenna trough and close to the mooring (not shown). Specifically, 70km southwest of the mooring, water temperatures ranging from 2.9 to 3.7 °C and salinity between 34.84 and 35.02gkg-1 were found at depth. These temperatures are within the range for pure AW, but the salinities are slightly lower than the threshold of 35.06gkg-1 . Additionally, relatively warm water (around 0.5 °C) was detected at a depth of 10km to the west of the mooring position.

In October 2020, warm and saline water of Atlantic origin occupied the lower half of the water column (Fig. b). The warmest water, with a temperature of 3.3 °C and a salinity of 34.6gkg-1, was located 10km south of the mooring position. Water warmer than 2°C reached the surface at the mooring position and also between 15 and 37km south of the mooring position. The coolest and least saline water was found on the northern side of the transect. At a station located 60km to the northeast, we observed water with a temperature of 1 °C and a salinity of 35.0gkg-1 below 150m depth (not shown).

In November 2021, a maximum temperature of 1 °C was found at (and south of) the mooring position (Fig. c), which was warmer than the maximum temperature in November 2019 (Fig. a). The isohalines were also aligned with the isotherms, with salinities exceeding 34.2gkg-1 for water at depth with temperatures ≥0 °C on the southern side. The northern side of the transect remained below -1.75 °C, consistent with observations from November 2019.

Early in the record, both near-bottom temperature, Tb, and salinity, Sb, increased, reaching values approaching those of pure AW properties by the middle of October (Fig. c). Near-bottom temperature was higher than SST, and the increase in Tb and Sb coincided with a relatively strong and long-lasting surface-intensified current directed southeastwards along the isobath (uR; Fig. d). Following the maxima in near-bottom temperature and salinity, the water column gradually cooled until mid-January 2019, when both Tb and SST nearly reached the freezing point (Fig. c). During these months, uR oscillated between -5 and 10cms-1 on a 2-week timescale, with negligible depth variability over the measured range (Fig. d). Temperature and salinity co-varied with uR, with Tb and Sb lagging behind uR by 3.3 and 4.8 d, respectively. Sea ice was first observed across the saddle in late December (Fig. b).

From mid-January to mid-June 2019, there was partial sea ice cover across the saddle (Fig. b). Additionally, Tb and SST were generally at or close to the freezing point (Fig. c). Sb increased, probably as a result of brine release from sea ice formation, reaching its highest value (35.12gkg-1) by the end of March, and then gradually decreased (Fig. c). During this time, the current was more unidirectional than during autumn – i.e. uR was generally positive (Fig. d) and typically varied between 5 and 10cms-1, occasionally exceeding 20cms-1.

Although the saddle was covered in ice and the water was generally near the freezing point during winter and spring, there were several episodes when Tb and SST increased. These increases coincided with a reduction in SIC and an enhancement of uR (Fig. b–d). The first and largest increase in Tb and SST occurred in early February 2019, following a sharp reduction in SIC at the end of January. During this period, sea ice cover that had built up over 2 weeks due to northerly winds disappeared within 2 d. The wind shifted to a southerly direction, and uR values became positive shortly before the sea ice disappeared from the saddle (Fig. a and d). The surface-intensified uR increased in strength during and after the sea ice disappearance, reaching a maximum of 26cms-1 close to the surface (Fig. d). Tb and SST increased to 0.9 and 0.4 °C, respectively, shortly thereafter and remained above the freezing point for more than a month (Fig. c). During this time, the wind direction turned and stayed northerly, and partial sea ice cover was re-established across the saddle (Fig. a and b). Tb and SST, however, remained above the freezing point and reached smaller maxima, while uR further increased in strength, reaching 34cms-1 at the beginning of March (the highest value measured during the whole sampling period; Fig. c and d).

SIC started to decrease in June, and the area was ice-free from mid-July (Fig. b). Most of the reduction in sea ice cover occurred during two pulses of increased uR in the second half of June, which were followed by two pulses of increased Tb and SST (Fig. d, c). After the sea ice cover had disappeared, uR became weaker, less unidirectional, and occasionally more variable in depth (Fig. d). From July to August, the water column steadily warmed. From September until the end of the measurement period in mid-November 2019, SST decreased towards the freezing point, while Tb was generally higher than SST. Sb generally co-varied with Tb during this autumn and decreased overall (Fig. c). By late October 2019, some sea ice had already drifted across the saddle, and SST reached the freezing point, while Tb remained above the freezing point until the mooring was recovered in mid-November (Fig. b and c). Compared to the previous autumn, in the autumn of 2019, the near-bottom waters were both colder and less saline, and sea ice formed earlier.

A wavelet analysis of the current components ur and vr (along -28 and 62°, respectively) reveals a year-round dominance of semidiurnal tidal currents with a clear spring–neap cycle (Fig. a–c). The semidiurnal tidal current was anticyclonic (Fig. f), with variance oriented along 25°, i.e. nearly east-northeast (Sect. ). When the near-bottom temperature Tb was above the freezing point, its semidiurnal variability typically ranged from 0.2 to 0.6 °C, occasionally reaching 0.8 °C.

Weather-band variability (28 h to 10 d) was considerable, particularly during the autumn months (November 2018 to January 2019) (Fig. b, c, and e), with the principal axis oriented along the WNW–ESE direction, approximately aligning with the topography, Sect. ). Tb anomalies were typically within ±0.5 °C, but they doubled from November–December 2018 and from October–November 2019. At the beginning (October 2018) and towards the end of the record (October–November 2019), Tb exhibited elevated variance at the 10 d timescale; however, the edge effects of the wavelet analysis were significant (Fig. e).

The anticyclonic velocity component was generally more energetic than its cyclonic counterpart (Fig. f). The semidiurnal current exhibited almost 10 times more variance in the anticyclonic component, and the diurnal band showed almost all of its variance in the anticyclonic component. At lower frequencies, the currents were anticyclonic and elliptically polarised throughout the entire year (Fig. f), with the exception of winter, when the currents became nearly linearly polarised during periods of 2 weeks or more (not shown).

The strength and direction of the overflow current are influenced by large-scale winds (Fig. ). Regression analysis shows that depth-averaged current anomalies along -28° in the frequency band ranging from 28 h to 10 d (ur′) depend on Ekman transport along the same axis. Geostrophic adjustment to Ekman transport during north-northeasterly winds typically opposes the east-southeastward current across the saddle (Fig. c) and, in cases of anomalously strong winds, even reverses it (Fig. b). Conversely, weaker and/or southerly winds tend to enhance the eastward flow into the Barents Sea (Fig. d).

Anomalously strong north-northeasterly winds, which result in Ekman transport values below the 10th percentile (UEr=-1.84(±0.53) m2s-1 for daily averages), are associated with the strongest current reversal events (ur′=-5.1(±8.8) cms-1 for daily averages, with the current flowing towards the WNW along the topography) (Fig. b). The mean current velocity in this case is 3.1cms-1, oriented towards the west-southwest, with a large spread.

On the other hand, winds with a southerly component, which result in Ekman transport values above the 90th percentile (UEr=0.79(±0.35) m2s-1 for daily averages), are associated with the strongest east-southeastward currents (ur′=4.4(±6.3) cms-1 for daily averages) (Fig. d). This suggests that winds with a southerly component accelerate the normal east-southeastward current through geostrophic adjustment to the Ekman transport set-up. The average mean current in this case flows at 14.1cms-1 towards the east-southeast.

Variability in near-bottom temperature occurs in the same frequency bands as current variability (Fig. ). We separated the contributions from tides, weather-band processes, and lower-frequency mesoscale activity by applying a band-pass filter to the time series of temperature and current velocity (Sect. ). By analysing temperature anomalies T′ in the velocity anomaly space (u′,v′), we found that positive temperature anomalies were generally located in the southeastern sector of the velocity anomalies, while negative temperature anomalies were found in the northwestern sector, resulting in net heat transport directed southeastwards across the saddle in the identified bands of variability (Fig. a–c). For the weather and mesoscale-activity frequency bands (Fig. b and c, respectively), the largest temperature anomalies were aligned with the principal axis of variability (-28°). In the semidiurnal (Fig. a) and diurnal tidal bands (not shown), however, the largest temperature anomalies were nearly perpendicular to the axis of maximum variance (major axes of the ellipses).

The eddy temperature flux ur′T′‾ along the principal axis (-28°; solid curves in Fig. d–f) was generally positive throughout the year, with values typically between 1 and 2Kcms-1. The largest semidiurnal tidal contribution occurred from October to December in both years, with values around 1Kcms-1 directed towards the southeast. Comparable fluxes were also observed between February and July when warm, saline water intermittently appeared near the saddle. In the weather band, the flux was large from November 2018 to January 2019, with values ranging from 1 to 3Kcms-1, and again during the autumn of 2019, reaching 5Kcms-1 by the end of the measurement period (Fig. e). Between mid-January and August 2019, the flux was negligible. Lower-frequency oscillations on timescales between 1 and 6 weeks exhibited sporadic episodes of large positive eddy temperature fluxes during autumn and winter, reaching 3Kcms-1 in October 2019, but experienced reversals to negative values during January and February 2019 (Fig. f).

To further investigate the driving force behind the eddy temperature flux across the saddle in the weather band, we analysed ensembles of events with high fluxes. In total 20 events with peaks (ur′,T′)>0 and 24 events with peaks (ur′,T′)<0 were detected, both resulting in positive eddy temperature fluxes. Most of these events occurred during October, November, and December (red and blue triangles in Fig. e), when there was no sea ice in the area. The 4 d time series, centred on the selected times for each event, were extracted. The ensemble-averaged records are denoted using angle brackets, e.g. 〈T′〉 (Fig. ).

The ensemble of events that satisfy the first condition is associated with southerly wind stress, resulting in an Ekman transport component that forces a positive current anomaly carrying positive temperature anomalies towards the east-southeast (Fig. a and c). The peak eddy temperature flux of (8.4±1.3)Kcms-1 tends to occur 1 d after the maximum along-isobath Ekman transport. The temperature anomaly turns positive approximately half a day after the current anomaly, and their peaks are 6 h apart.

The second case (Fig. b and d; events marked by blue triangles in Fig. e) corresponds to events where the wind from the north-northeast increases in strength, enhancing Ekman transport opposing the current. This results in a negative current anomaly that caries negative temperature anomalies towards the west-northwest. The peak of the eddy temperature flux reaches 7.8(±1.2)Kcms-1 approximately 1 d after the along-isobath Ekman transport component has reached its largest negative value. The temperature anomaly turns negative 6 h after the current anomaly, and their peaks are 2 h apart.

Several findings from the time series at the mooring position (Fig. ) and the late-autumn hydrographic transects at the saddle (Fig. ) warrant a discussion about the seasonal and interannual variability in the area and whether 2018–2019 was a typical year.

The autumns of 2018 and 2019 differed in terms of hydrography and the onset of the sea ice period. During the autumn of 2018, the water column above the saddle remained warm and free of ice and was only partially covered with ice from December. In contrast, in the autumn of 2019, sea ice drifted to the saddle in late October, and the water cooled to near the freezing point in November. This suggests that the position of the ice cover in the Barents Sea during autumn affects the hydrographic conditions at the saddle at this time of the year. According to , the marginal ice zone had retreated beyond the shelf break and into the Nansen Basin by August 2018, while in August 2019, large parts of the northwestern Barents Sea shelf remained covered by sea ice, resulting in lower temperatures and a fresher surface layer. These differences in large-scale conditions at the end of the melting seasons in 2018 and 2019 must also have affected the subsequent winter conditions at our mooring site. During the same autumns, observations from the Northern Barents Sea Opening showed a large difference in upper-ocean salinity, accompanied by a similar difference in sea ice . Heightened upper-ocean freshening due to increased sea ice melt can create conditions in the area that are conducive to sea ice growth in the following winter .

Superposed on the seasonal cycle were several episodes of warm-water intrusions that affected sea ice cover at the saddle (Fig. b and c). The inflow of Atlantic-origin waters and heat transport across the saddle is important for the onset and duration of typical winter conditions. Strong current velocities directed across the saddle into the Barents Sea (uR) generally preceded the local temperature maxima associated with the warm-water intrusions (Fig. c and d). Sea ice covered the mooring site from mid-January 2019, i.e. before the water column had completely cooled to the freezing point (Fig. b and c). Nonetheless, favourable wind and ice conditions, along with geostrophic adjustment to the wind-driven Ekman transport from Edgeøya to Hopen, resulted in depth-averaged currents reaching 23cms-1, which drove warmer waters from the Storfjordrenna trough to the saddle. A similar response to wind forcing, resulting in deep warm-water overflow, has also been observed on the sill in the inner part of Hornsund in Svalbard .

However, not all warm-water inflow events can be explained by local wind effects. The event with a strong across-saddle current of 30cms-1 and a small increase in temperature, which occurred from mid-February to March, was likely caused by upstream conditions. Accelerating the WSC and the branch entering the Storfjordrenna trough can force the current to follow shallower isobaths to conserve potential vorticity . The small increase in temperature during this period was most likely due to the mixing of warm water en route to the saddle. When the current subsided in early March, the temperature at the saddle returned to the freezing point, and the sea ice concentration increased to 90%. Similarly, events during April involving weaker current pulses transporting cooler waters can be attributed to forcing resulting from the upstream conditions.

Northeasterly winds and relatively strong westward-current reversals in the weeks preceding the November 2019 transect (Figs. a and a, d, and e) likely transported sea ice and cooled the surface waters above the saddle to the freezing point. Large-scale sea ice cover was more extensive in November 2019 than in November 2021. In 2019, the sea ice edge (characterised by 15% sea ice concentration) extended past Hopen, almost reaching the southern tip of Spitsbergen, while in 2021, it only extended south to Edgeøya and did not cover the Olga Basin (not shown). The higher maximum temperature and the alignment of the isohalines and isotherms in the transect from November 2021 (Fig. c) suggest an autumn inflow of Atlantic-origin waters and incomplete cooling of the water column. The SST in the region was substantially higher in 2021 than in 2019, with the largest difference of 3 °C in the Storfjordrenna trough (not shown) and a difference of 1 to 2 °C at the mooring position. This meant that the onset of freezing was delayed by 1 month in 2021 relative to 2019 (Fig. b and d).

The November transects (Fig. a and c) were both cooler and less saline than the October transect (Fig. b). This may be due to a regular seasonal shift from AW influence to Arctic Water influence at the site and/or interannual variability in the region. During the autumn of 2020, SST steadily decreased before rapidly increasing to 1.4 °C in the week leading up to the transect in mid-October (Fig. c). It then decreased to 0 °C before increasing to 2.6 °C in mid-November. Hence, the autumn of 2020 experienced warmer and longer-lasting overflows of Atlantic-origin waters across the saddle and a delayed onset of sea ice compared to other years. AW inflow from north of Svalbard has also been shown to affect seasonal variability in local hydrography in the same way .

The autumn of 2018 was similar to the autumn of 2020 in terms of SST evolution at the mooring position (Fig. a and c). Warm-water inflow from the Storfjordrenna trough kept SST high at the saddle during November 2018 and November 2020, delaying the onset of sea ice in the subsequent winters.

In the context of long-term changes in SST and SIC at the study site, the period from the autumn of 2018 to the autumn of 2019 (mooring year) was an average year compared to the past 2 decades (not shown), and the difference between 2018 and 2019 was not remarkably large. From 2005, interannual variability in SST and SIC at the mooring site has been substantially higher than during the previous 25 years, indicating even larger year-to-year changes than those observed between 2018 and 2019 could be expected.

To summarise, strong interannual variability substantially impacts the seasonal cycle at the study site. Sea ice in the Barents Sea exhibits a high degree of interannual variability e.g. driven by atmospheric temperature, sea surface temperature, and oceanic heat transport . Locally, intrusions of warm water during autumn can delay the onset of winter conditions, inhibit local sea ice growth, or melt imported sea ice, as observed in the autumns of 2018 and 2020 (Figs. b–c, b, and 11a and c). These intrusions, whether driven by local winds or upstream conditions, also frequently reduce the existing partial ice cover above the saddle during winter and spring (Fig. ; earlier years are not shown). This explains the differences in hydrographic conditions observed in November 2019 versus 2021 (Fig. a and c). The more persistent warm-water overflow in 2020 likely delayed the seasonal water mass transformation and contributed to the October 2020 transect being substantially warmer than the November transects.

The observed background current (low-pass-filtered over 10 d) tends to flow towards the southeast (Fig. c). This current direction is counteracted by geostrophic adjustment to Ekman transport induced by the large-scale wind pattern from the northeast. The strength and direction of the current are affected by the wind stress, and, in periods with strong winds, the current either reverses to a northwestward flow or is amplified in its normal southeastward direction (Fig. ). Hence, local wind is not the main driver of the background current across the saddle.

The forcing of the slowly varying current is more complex. We found a low and insignificant correlation between spatial differences in sea level and mooring current velocity by analysing the absolute dynamic topography and the associated geostrophic currents (not shown). This means either that the local sea-level slope is not the main driver of the current or that the data coverage and quality of the measurements of the absolute dynamic topography are insufficient for showing a clear relationship. The quality of these measurements may be reduced as a result of seasonal ice coverage and the timing of satellite passages, and the quality of the product north of the Arctic circle has not been thoroughly tested .

On the other hand, the current across the saddle may be dependent on the strength of the upstream current in the Storfjordrenna trough and along the slope west of Spitsbergenbanken (Sect. ). Upstream, along the Norwegian coast, low-pressure systems accelerate the Norwegian Atlantic Current by driving onshore Ekman transport and set-up along the coast . However, it is unclear whether low-pressure systems would accelerate the WSC along the relatively shorter western Spitsbergen coastline . Additionally, high variability in the tracks of low-pressure systems passing Svalbard makes the influence of local wind west of the Storfjordrenna trough and Svalbard highly variable. Current peaks at the mooring site are occasionally related to a strong WSC, which can be quantified using the slope of the absolute dynamic topography between Spitsbergenbanken and the Storfjordrenna trough (not shown), contributing to the current strength in a barotropic manner. Similar to the response of the WSC on the West Spitsbergen Shelf to anomalous wind stress curl , the branch of the WSC flowing through the Storfjordrenna trough can be shifted to shallower isobaths. However, a more detailed analysis is needed to fully understand the effects of upstream forcing.

Another possible driver of the observed mean current is the tidal residual flow on Spitsbergenbanken. In shallow areas with strong tidal currents and non-linear bottom friction, residual currents may influence the circulation pattern . Tidally generated residual currents around Bjørnøya have been documented in buoy data , laboratory experiments, and observations from drifters . Several numerical model studies have shown that tidal rectification occurs in the region . Higher-resolution models which better resolve non-linear terms have reported strong anticyclonic residual tidal currents around Bjørnøya, on Spitsbergenbanken (reaching 15cms-1), and around Hopen . Simulations have reproduced the anticyclonic drift of buoys around Hopen, which is driven by strong tidal currents resulting from the interaction between the semidiurnal tide and the island of Hopen .

Both semidiurnal and diurnal tidal currents contribute to heat transport across the saddle on Hopenbanken, with the semidiurnal contribution being significantly larger. The semi-major axis of the semidiurnal currents is oriented along 20°, i.e. roughly east-northeast, which is approximately perpendicular to the eddy temperature flux in the semidiurnal band (Fig. ). For all frequency bands, the direction of the eddy temperature flux is mainly oriented towards the southeastern sector. This is likely due to the position of the temperature gradient at the saddle and the way the current anomalies are guided by the topography.

Oscillations on timescales between 1 and 6 weeks result in sporadic moderate eddy temperature fluxes in the southeastern sector, comparable in magnitude to those in the semidiurnal band but more variable in direction. The largest contribution towards the east-southeast occurred during the autumn of 2019 (Fig. f), when the wavelet transform of ur (Fig. b), and especially that of Tb (Fig. e), exhibited power in the low-frequency area, indicating that the varying current resulted in even larger temperature variations. These concurrent fluctuations led to a considerable eddy temperature flux during this period. We hypothesise that the large eddy temperature flux during this autumn may have been caused by AW upstream of the saddle being transported across shallower isobaths, possibly due to regional Ekman pumping or an acceleration of the upstream current system, which resulted in higher temperatures on one side of the front at the saddle. Locally, at the mooring site, this was observed as low-frequency oscillations with particularly pronounced temperature variations, leading to a large eddy temperature flux into the Barents Sea.

Integrated over the measurement period, the weather-band eddy temperature flux was approximately twice that of the low-frequency and semidiurnal bands, which had comparable magnitudes. During the period of relatively high sea ice coverage (from February to July), the semidiurnal tide transported twice as much heat across the saddle as the weather band and 1.5 times as much as the band associated with mesoscale activity.

The eddy temperature fluxes calculated in Sect.  contain divergent and rotational components. Only the divergent component plays a part in the actual transport of heat across the saddle . Using our data set, we cannot separate the contributions of the divergent and rotational eddy temperature fluxes; hence, the values must be considered an upper bound.

In recent years, the presence of AW in the Storfjordrenna trough upstream of the study site has increased. The Storfjordrenna trough is one of the areas in the Barents Sea that has experienced the highest increase in sea surface temperature over the past 2 decades . The climatology in the Storfjordrenna trough (Sect. ) also shows that the water column has become warmer since 2000 compared to previous years (Fig. ) and that Atlantic-origin waters have shoaled and now reach further east towards the saddle during summer (Fig. ). Shoaling has also been observed in the Eurasian Arctic as well as along the coast and in the fjords of western Spitsbergen . Additionally, AW advected into the fjords of western Spitsbergen has warmed in the past 2 decades . At the mooring site, the years following 2005–2006 saw, on average, higher mean annual SSTs and fewer days of sea ice coverage compared to the years between 1981–1982 and 2004–2005 (not shown), which may be due to increased AW inflows into the Barents Sea after 2006 compared to the Fram Strait . The warmest year in this period was 2015–2016, which saw abnormally high mean SSTs over the saddle, only 106 d with any sea ice, and no SIC value above 80%. This was also the year when the Storfjordrenna trough, along with the Barents Sea in general , experienced record-low sea ice cover, a warm saline surface layer , and an exceptionally warm and long-lasting marine heatwave .

As AW circulates over shallower isobaths, the physical processes in the channel between the Storfjordrenna trough and the Olga Basin may more easily transport heat eastwards in the future. Anomalous wind stress curl may affect the circulation depth of AW in the Storfjordrenna trough, potentially causing AW or Atlantic-origin waters to move closer to the saddle between the Storfjordrenna trough and the Olga Basin (Sect. ).

Several factors determine the fate of warm waters crossing the saddle. The circulation pattern in the channel and downstream influences whether Atlantic-origin waters are transported towards the Arctic or the Atlantic side of the polar front, which follows the 200m isobath separating the Olga Basin from Hopendjupet (Fig. ). Although the time-averaged flow observed by the mooring was directed towards the southeast, with most non-tidal variability contributing to heat exchange along the west-northwest–east-southeast axis (Sect. , , and ), the region's complex bathymetry and limitations in our observations prevent a definitive conclusion regarding the specific path of AW and the proportion of heat contributing to the heat budget on the northern side of the polar front. Nonetheless, AW transport through the channel between the Storfjordrenna trough and the Olga Basin has the potential to impact the Olga Basin, which still experiences a seasonal ice cover and maintains a thick cold layer of Arctic Water e.g..

Given the possibility that Atlantic-origin waters from the Storfjordrenna trough may be transported to the Arctic side of the polar front east of Hopenbanken, the density stratification within the Olga Basin becomes an important factor in determining the impact of this transport on local conditions. Denser waters on the eastern side would cause Atlantic-origin waters to penetrate at shallower depths, impacting the winter and spring sea ice cover and generally depositing more heat into the upper part of the water column, which could influence sea ice growth in the subsequent season. Conversely, less dense waters on the eastern side might lead to the subduction of overflow waters. If pure AW, or water with properties more similar to those of AW than those observed in this study, flows over the saddle, its density may exceed that of local water in the Olga Basin (e.g. Arctic Water and East Spitsbergen Water) and thus contribute to deeper stratification. We observed a deepening of the 34.4gkg-1 isobath and a small decrease in temperature in parts of the Olga Basin (Fig. b and c) over the past 2 decades. However, due to the particularly sparse data coverage in the Olga Basin, we are cautious about interpreting these changes.

Atlantic-origin waters reaching the saddle between the Storfjordrenna trough and the Barents Sea will affect the properties of the East Spitsbergen Current as it flows into Storfjorden. An increase in AW mixed into the East Spitsbergen Current across the saddle will have repercussions for the coastal current in Storfjorden and along the west coast of Spitsbergen, as well as for the exchange between the shelf and the fjords. It will also affect dense-water production in Storfjorden and subsequent deep-water export to the Fram Strait, as the initial salinity of the source water sets the conditions for the density of the overflow .