OSNAP is a sustained, international ocean observing initiative established in 2014 to provide comprehensive measurements of the Atlantic Meridional Overturning Circulation (AMOC) in the subpolar North Atlantic . While the program’s primary objective is to quantify AMOC variability, it also provides integrated meridional heat and freshwater fluxes. The OSNAP array, which consists of moorings, gliders and floats, includes two legs: OSNAP West, spanning from Labrador Shelf to West Greenland, and OSNAP East, extending from East Greenland to Scotland. Figure  illustrates the OSNAP array, including its network of moorings and glider deployments for further details, see.

We compare those observational estimates of oceanic transports to monthly output from a set of global ocean reanalyses, including ORAS5 Ocean ReAnalysis System 5,, CGLORSv7 Centro Euro-Mediterraneo sui Cambiamenti Climatici Global Ocean Reanalysis System,, GLORYS2V4 Global Ocean Reanalysis and Simulation,, GloRanV14 (an improved version of GloSea5, also known as the FOAM, herafter called FOAMv2; ), and GLORYS12V1 . The first four products can be considered as an updated version of the Copernicus Marine Service (CMEMS) Global Reanalysis Ensemble Product (GREP; ), while GLORYS12V1 extends the analysis with its higher resolution. The reanalyses are all based on the NEMO ocean model, with the GREP products being configured at a 0.25° horizontal resolution and 75 vertical levels, while GLORYS12V1 has a finer horizontal resolution of 0.083° and 50 vertical levels. The necessary atmospheric forcing for all ORAs was taken from the ERA-Interim atmospheric reanalysis, after 2019 from ERA5 . While this forcing is similar for all the ORAs, they differ in their data assimilation methods, physical parameterizations, and initial conditions, which contribute to important differences in their transport estimates. Notably, ORAS5 assimilates sea level anomalies (SLA) only between 50° S and 50° N, whereas the other four reanalyses assimilate SLA in ice-free seas globally. All reanalyses assimilate in situ temperature and salinity profiles and SLA, which constrain the large-scale geostrophic circulation, but none assimilate direct velocity observations from current meters or ADCPs. In contrast, OSNAP transport estimates are derived from direct, full-depth observations of velocity, temperature, and salinity obtained from moorings, gliders and hydrographic measurements. As a result, OSNAP and the reanalyses differ fundamentally in how ocean velocities, in particular boundary currents, are constrained.

To assess geostrophic circulation anomalies at the surface, we use satellite-derived sea level anomalies (SLA) from the CMEMS multi-mission product . This dataset provides global, gridded SLA fields referenced to the 1993–2012 mean, with a 0.25° horizontal resolution. It is based on delayed-time altimeter observations from multiple satellite missions and includes standard corrections for tides, atmospheric pressure, and instrumental effects. The resulting fields allow for the estimation of surface geostrophic velocities through sea level gradients.

To reduce short-term variability and highlight interannual changes, all time series are smoothed with a 12-month running mean, unless otherwise indicated.

Velocity and temperature sections reveal oceanic flow and temperature distributions across specific straits or regions, facilitating the understanding of the structure and variability of OHT.

provide gridded sections of velocities, temperatures, and salinities as well as integrated transports of heat, freshwater and volume across the OSNAP line using in situ measurements from moored instruments deployed at multiple depths across the subpolar North Atlantic. These moored observations are supplemented by gliders and Argo floats, which fill spatial gaps between moorings and provide additional data in regions without moored observations. The positions of moorings and glider sections are shown in Fig. . Detailed descriptions of the instruments, as well as calculation and interpolation techniques used, can be found in and .

We aim to compare the observed velocity and temperature sections, as well as integrated transports across the OSNAP section, to those derived from ocean reanalyses. Extracting accurate velocity sections and calculating transports from reanalyses is challenging as ocean models often use complex curvilinear or unstructured grids, necessitating the accurate handling of varying grid geometries. To address this, we use StraitFlux , a Python tool specifically designed to calculate cross-sections and transports in a manner consistent with the discretization schemes of the analyzed models, and to preserve conservation properties of the model's native grids. More details on the calculation methods can be found in .

To assess cross-sectional biases and RMSE values between OSNAP and reanalyses, the reanalyses sections are interpolated bilinearly in the along-section and vertical directions onto the common OSNAP gridded section, defined by the OSNAP “along-section” distance coordinate and depth levels.

In addition to cross-sections, we analyze integrated oceanic transports through the OSNAP section. In general, oceanic transports of volume (OVT), heat (OHT), and other tracers through a given cross-section are key metrics for understanding ocean circulation and energy transfer. These transports are defined mathematically as integrals over the cross-sectional area, incorporating the velocity field and additional scalar properties such as temperature or tracer concentration. In this study, we focus on the transport of heat, which can be expressed as: 1OHT=cpρ∫xlxr∫z1(x)z2(x)(θ(x,z)-θref)v(x,z)⋅ndzdx with the velocity vector v and the unit normal vector to the cross-section n. The cross-sectional area of the strait is defined by its width x and depth z. The density of seawater is represented by ρ and the specific heat capacity of seawater by cp. For simplicity, transport calculations based on reanalyses are performed assuming constant ρ (1026 kg m⁻³) and cp (3996 J kgK⁻¹), while for OSNAP observations transports ρ and cp are calculated for each grid cell as a function of temperature, salinity, and pressure . However, the impact of these differences should be minimal, as variations in ρ and cp tend to compensate each other, resulting in negligible changes to the total transports . Additionally, the potential temperature θ and a reference temperature θref are needed for estimating heat transports. Strictly speaking, unambiguous heat transports would require closed mass transports across the examined section, which is generally not the case for partial sections such as OSNAP East and West and only approximately satisfied for the total oceanic transport . Therefore, heat transports depend on the choice of reference temperature. To minimize the ambiguity arising from this choice, θref should be chosen to represent the section-mean temperature of the flow across the considered section . However, to ensure internal consistency across products we calculate all heat transports relative to a constant reference temperature of θref=0 °C, following common practice e.g.,. Throughout this study, the term “heat transport” is therefore used in this conventional, reference-dependent sense and describes the transport of heat referenced to water at 0 °C. To avoid interpolation and preserve the numerical consistency of fluxes on the native model grids, net integrated transports from reanalyses are calculated using StraitFlux's line-integration method, in which transports are integrated directly along the native model grid cell faces that intersect the section, thereby approximating the target section as closely as possible on the native grid .

All transport calculations in this study are based on monthly mean output from the ocean reanalyses. To evaluate the potential influence of temporal resolution, we additionally tested calculations based on daily velocity and temperature fields for GLORYS12V1, which include sub-monthly covariance terms (v′T′) that are not explicitly resolved when using monthly mean fields. The resulting differences in integrated heat transport across the OSNAP section amount to about 2 TW on average over the analysis period (corresponding to approximately 0.5 % of the mean transport, see Fig. ), with negligible impact on variability. This indicates that monthly output provides a sufficiently accurate representation for the purposes of this study.

To complement the ORA-based OHT estimates, a largely independent heat transport estimate was additionally obtained following the oceanic heat budget approach outlined in . 2OHTOSNAPE=OHTGSR/FS+BSO-FS-ρ0cp∂∂t∫0Z(θo-θref)dz-Lfρi∂di∂t-RAdjOSNAPEGSR/FS+BSO

The vertical net surface energy flux, counted positive if downward, is denoted as FS. We define gridpoint ocean temperatures as θo, the reference temperature as θref, the sea ice density ρi (assumed constant at 928 kg m⁻³) and gridpoint average sea ice thickness di. This method infers OHT at OSNAP East by integrating FS, the heat content changes (OHCT; second term inside the brackets of Eq. ) and the melt ice tendencies (MET; third term inside the brackets) over the area between the OSNAP-East section and a nearby section where observations are available.

FS is estimated indirectly from atmospheric budgets, so these are much better constrained by independent observations than parameterized surface fluxes, which typically are more uncertain and depend on the sea state . Therefore, divergences and tendencies from atmospheric reanalyses ERA5 mass-consistent energy budgets,, MERRA2 and JRA55 are combined with top-of-atmosphere (TOA) fluxes from CERES-EBAF TOA version 4.2 . An updated implementation based on the newer JRA-3Q reanalysis is currently under development and will be addressed in future work once the corresponding energy budgets are fully consolidated.

OHC is taken from the GREP reanalyses mentioned above, the Institute of Atmospheric Physics version 4 IAPv4 and Random Forest Regression Ocean Maps RFROM. The melt ice tendencies MET is calculated from GIOMAS and ORAS5, which both provide ice thickness fields at 1 and 1/4° resolution, respectively.

Heat transport estimates inferred from mass-consistent atmospheric energy budgets see, e.g., are used at two different choke-points: the Greenland-Scotland Ridge (GSR), and the combination of Fram Strait (FS) and the Barents Sea Opening (BSO). While the GSR transports are available continuously from 1993 to 2021 , the Fram Strait and BSO estimates are limited to the ArcGate campaign period from October 2004 to April 2010 . To enable comparisons over the extended OSNAP period, we construct a synthetic ensemble of Fram + BSO transports by extrapolating the limited ArcGate observations. The extrapolation uses a first-order autoregressive [AR(1)] model that preserves the observed seasonal cycle and reproduces realistic variability based on the lag-1 autocorrelation and the standard deviation of monthly anomalies. We generate 1000 synthetic realizations to sample the range of plausible interannual variability. Using the GSR as a boundary introduces larger uncertainties due to the relatively high magnitude and variability of the heat transport across the ridge. In contrast, integrating from Fram Strait and the BSO benefits from lower uncertainties in the upstream OHT, but requires a larger integration domain (potentially leading to larger accumulated errors) and more careful treatment of sea ice processes and storage. By considering both configurations, we leverage the strengths of each approach and account for complementary sources of uncertainty in estimating the subpolar heat transport.

An approach like this would require a closed volume. However, the region bounded by the choke-points and OSNAP East includes an open passage between Scotland and Norway (ScoNo in Fig. ). Since exchanges through this passage are negligibly small (ORAS5 gives a mean transport of 3 ± 8 TW, more than two orders of magnitude smaller than transports across OSNAP East), they are excluded from the calculation. Additionally, to address global inconsistencies between air–sea heat fluxes and ocean heat content tendencies (OHCT), we apply a spatially uniform adjustment RAdj to the net heat fluxes. Specifically, we subtract the monthly difference between the global full-depth OHCT and surface fluxes at every grid point as implemented e.g. in and .

To account for uncertainty in each component of the budget, we combine three estimates for surface fluxes, six estimates for OHC, and two estimates for sea ice in all possible permutations, resulting in 36 estimates for both choke-points. The data used is summarized in Table .

A similar approach has recently been applied and evaluated by , who used surface fluxes and heat storage terms between OSNAP and RAPID to estimate heat transports at the RAPID section. Their results show good agreement with direct in situ estimates in terms of the mean and reasonable agreement in terms of temporal variability, supporting the use of energy budget methods to bridge observational gaps between ocean sections. Our analysis, which additionally incorporates independent ocean reanalyses and explicit uncertainty estimates for all budget components, might provide a useful complement and help to build confidence in this approach.

present a novel Atlantic ocean heat transport dataset derived from a combination of Argo temperature profiles, satellite altimetry and gravimetric constraints. This product provides valuable insight into basin-scale heat transport variability across complete latitude bands. However, because it is defined along zonal sections and provided at 3-monthly resolution, it cannot be directly applied to the OSNAP section geometry or used to distinguish between OSNAP East and West, which is central to the present analysis. Nevertheless, such approaches offer an important complementary perspective on large-scale heat transport variability and are highly valuable for basin-scale assessments.

The temperature distribution along the OSNAP section is shown in Fig. a, while the corresponding temperature biases for the reanalyses relative to the OSNAP observations are presented in Fig. b–f. The OSNAP temperature cross-section reveals a distinct thermal structure across the North Atlantic, with warmer waters concentrated in the Rockall Trough and the upper layers of the Iceland Basin, where the North Atlantic Current transports heat from lower latitudes northward. Below the warm Atlantic Water layer, temperatures gradually decrease and transition to deeper, colder water masses. In contrast, the Irminger Basin and the Labrador Basin are characterized by significantly colder waters. These basins are influenced by waters of subpolar and Arctic origin, with the coldest and densest overflow waters formed in the Nordic Seas. While the reanalysis models broadly reproduce the large-scale thermal structure observed by OSNAP, they exhibit systematic and regionally varying temperature biases. Temperatures are predominantly underestimated in the basin interiors, with cold biases reaching up to -1.0 °C, particularly in ORAS5. In contrast, distinct warm biases appear at the Labrador Shelf as well as the West and East Greenland Shelfs, the highly dynamical Reykjanes Ridge, intermediate waters in Iceland basin, west of Hatton and Rockall Bank, Rockall Trough and deep waters below 2000 m along the continental slopes of the Irminger, Labrador and to a lesser extent also Iceland Basins. While we treat OSNAP observations as a reference, comparisons with the EN4 objective analysis dataset reveal some differences (see Fig. ). EN4 generally shows a cold bias relative to OSNAP, with particularly pronounced discrepancies near the Reykjanes Ridge and along the Labrador Shelf. These differences might reflect EN4’s coarser resolution and the sparser observational coverage in the subpolar North Atlantic, but at the same time, while OSNAP offers more detailed and spatially resolved data, it may also be subject to its own observational and methodological uncertainties .

Figure illustrates the temporal RMSE values relative to OSNAP observations. All reanalyses exhibit similar spatial patterns of error, suggesting a shared underlying structure in their deviations from observations. The strongest RMSE values are found in surface waters, particularly over the continental shelves and along the Reykjanes Ridge, as well as in intermediate waters within the Iceland Basin, where important inflow branches of the North Atlantic Current (NAC) are located. However, no direct mooring observations are available in this region, and ARGO floats provide the primary observational coverage. Additionally, RMSE values are elevated below 2000 m in the Labrador Basin, where a distinct “break” in the error pattern appears. This could be attributed to a lack of observationally constrained variability at these depths in the ORAs as ARGO data are only available down to 2000 m.

Figure a depicts the time-mean currents across the OSNAP-East and OSNAP-West sections, while Fig. b–f illustrate the mean biases of the reanalyses relative to observations. Although the reanalyses generally capture the overall flow structure and main currents, discrepancies exist in both the strength and exact positioning of these currents.

Observations indicate that the Labrador Current (LC) originates close to the shore along the Labrador Shelf and extends along the continental slope down to the seafloor. In contrast, reanalysis data suggest a slightly more offshore and shallower LC position. Additionally, the reanalyses show a weak northward recirculation east of the LC, a feature absent in OSNAP observations, likely due to a lack of mooring coverage in that region. Similarly, all 0.25° reanalyses underestimate the strength of the West Greenland Current (WGC) and displace it slightly towards the shore, while the East Greenland Current (EGC) appears too broadly extended offshore, leading to an underestimation of southward current velocity along the continental slope and an overestimation further into the basin. These biases may contribute to the positive temperature anomalies observed along the continental slopes of the Irminger and Labrador basins. GLORYS12V1 shows very strong but very narrow velocity peaks, it overestimates the WGC and EGC at their peaks along the continental slopes but underestimates their strength farther offshore. Despite this sharp structure, the spatial mean absolute error remains comparable to that of the other reanalyses (see Fig. ). The East Reykjanes Ridge Current is systematically underestimated in all reanalyses, while the North Atlantic Current just west of Hatton Bank is overestimated across all ORAs except GLORYS2V4. Moreover, the reanalyses exhibit pronounced inflows and outflows within the Iceland Basin and, to a lesser extent, the Rockall Trough, regions where no mooring observations are available, making it difficult to assess the accuracy of these features.

Figure presents the velocity temporal RMSE values relative to OSNAP observations. All reanalyses display similar spatial RMSE patterns, with GLORYS12V1 exhibiting greater variability due to its higher horizontal resolution. The largest variances are associated with major currents, particularly the North Atlantic Current in the Iceland Basin, where mooring observations are absent. In the basin interiors, the RMSE fields exhibit a streaky appearance, which likely reflects current misplacements but may also partly result from the sparse spatial coverage of the OSNAP moorings in the interior.

Despite broadly reproducing the thermal and dynamic structure along the OSNAP section, the reanalyses exhibit systematic biases with respect to OSNAP in both temperature and velocity fields. The cold biases in basin interiors, warm biases along continental slopes, and discrepancies in current positioning suggest limitations in how these models represent observed oceanic features given the relatively limited observational constraints. Notably, ocean currents are not directly assimilated in these reanalyses. While the general patterns of error are consistent across reanalyses, the absence of mooring observations in key regions, such as the Iceland Basin and Rockall Trough, adds further uncertainty.

Table shows average heat transport values over the August 2014 to June 2020 period. OSNAP observations show a net heat transport of 417 ± 27 TW for OSNAP East. While reanalyses generally yield lower values, all except ORAS5 fall within the observational uncertainty range. For OSNAP West, observations indicate a heat transport of 85 ± 6 TW. Among the reanalyses, only the higher-resolution GLORYS12V1 (92 TW) falls within the observational uncertainty range, whereas ORAS5 significantly overestimates it, and the remaining three reanalyses substantially underestimate the observed values. Figure presents the net integrated heat transports across OSNAP West (a) and East (b), comparing OSNAP-derived heat transports with direct transport estimates derived from the reanalyses using StraitFlux . In addition to individual ORA outputs, the GREP mean (mean of the four 0.25° reanalyses) is shown, along with its associated uncertainty (±1 standard deviation, blue shading). At OSNAP West, heat transport variability is generally lower compared to OSNAP East, though there is a considerable spread among ocean reanalyses. The reanalyses generally capture heat transport variability well, showing high correlations with observations (see Table ), except for ORAS5 (r=0.12, p=0.31). GLORYS12V1, with higher resolution, closely follows observations (r=0.79, p<0.01), while the GREP mean, though biased low, represents variability equally well (r=0.79, p=0.01). In contrast, at OSNAP East, reanalyses generally exhibit weaker MHT variability than the observation, in particular they miss the pronounced 2015 maximum, the subsequent multi-year decline to the 2019 minimum, and the 2019–2020 increase. Consistent with this muted variability, correlations with observed MHT are low for the 0.25° reanalyses, whereas the higher-resolution GLORYS12V1 exhibits a comparatively stronger correlation (r=0.37, p<0.01), suggesting that increased spatial resolution may enhance the representation of heat transport variability.

The lower panel (Fig. c) presents indirect estimates of heat transport using the GSR choke-point (blue lines) and the Fram + BSO choke-point (green lines), as described in Sect. 2.3. Thin lines show all possible combinations using the datasets listed in Table . Uncertainties are estimated by combining the spread (measured as standard deviation) across the datasets involved, the observational uncertainties associated with the GSR, and the spread across the 1000 Fram + BSO realizations. Corresponding mean values are given in Table . They are biased low relative to observations, with larger uncertainties for the Fram + BSO estimate. These uncertainties are dominated by the wider integration area, which leads to larger uncertainties in Fs and OHCT (std = 51 TW), whereas the contribution from the synthetic Fram+BSO ensemble is smaller (std = 21 TW). While these indirect estimates show better agreement with OSNAP observations in the latter half of the record, capturing the 2019 minimum and the subsequent increase, they, like the direct transports from the reanalyses, do not reproduce the pronounced 2015 heat-transport peak and, overall, correlate only weakly with OSNAP over the full period (see Table ). This holds true across all combinations of datasets and both choke-point approaches, despite the use of multiple, independent data sources contributing to the indirect heat transport estimates (see Table ). This discrepancy will be discussed in more detail in Sect. 3.3.

Volume transport time series are shown in Fig. . OSNAP-derived net volume transport (computed from the gridded sections) across OSNAP East has a mean comparable to the reanalyses but shows substantially reduced variability. This is expected because OSNAP’s velocity reconstruction combines time-varying geostrophic shear with a constrained barotropic component (transport closure/compensation), which dampens section integrated volume-transport variability. In contrast, ocean reanalyses permit time-varying transports in response to atmospheric forcing and freshwater fluxes, leading to larger variability in net volume transport across the open OSNAP section. While the realism of this variability cannot be independently assessed here, its magnitude remains small compared to what would be required to explain the pronounced heat-transport anomaly in 2015. A conservative upper-bound estimate shows that the associated reference-temperature-dependent contribution to heat transport at OSNAP East is of order 1–2 TW on average, which is negligible compared to the mean transport (400 TW), indicating that differences in the thermal structure and distribution of the flow play a dominant role in the 2015 anomaly.

Figure a, b shows heat transports along the OSNAP West and East sections, (panels c, d) the accumulated transports along the sections and (panels e, f) the corresponding variability (standard deviation) along the sections. Figure  shows the same for volume transports. Heat transport estimates reproduce the biases found in the velocity and temperature cross-sections (Figs.  and ): The reanalyses generally agree quite well with observations in terms of the major inflow and outflow branches. However, some discrepancies remain in transport amplitudes and exact positions. At OSNAP West (Fig. a, c, e), all reanalyses except ORAS5, overestimate the heat outflow via the LC. A return current slightly east of the LC partially offsets these biases. However, the eastern inflow branch of OSNAP West, associated with the WGC tends to be slightly too weak in all quarter-degree reanalyses, leading to the negative bias in net heat transport at OSNAP West in CGLORS, GLORYS2V4, and FOAMv2. While GLORYS12V1 also overestimates LC heat transport, this is counterbalanced by stronger heat inflows in the interior of the Labrador Basin. Additionally, heat transport via the WGC, although narrower and stronger than observed, remains in good agreement with observations, resulting in net heat transport estimates close to observed values. In contrast, ORAS5 produces LC heat transports that align well with observations, but stronger heat inflows in the Labrador Basin, combined with slightly underestimated WGC heat transports, lead to a net positive heat transport bias at OSNAP West. At OSNAP East (Fig. b, d, f), heat transports associated with the EGC are slightly displaced in all 0.25° reanalyses and overestimated in GLORYS12V1. Transports in the interior of the Irminger Basin, as well as at the IC and ERRC, are weaker than suggested by the OSNAP observations. In the interior of the Iceland Basin, the reanalyses exhibit high variability and a strong heat inflow and outflow associated with a northward branch of the NAC and a southward recirculation to the east, respectively. This northward-southward circulation feature is not resolved by the OSNAP observations due to OSNAP's array design that relies on end-point dynamic height moorings to capture the total integrated transport and its variability in the Iceland basin. However, because the opposing branches largely compensate each other, the net accumulated heat transport over this segment (from about 2500 to 2750 km) is consistent between the reanalyses and the OSNAP observations (Fig. d). While all reanalyses show positive net heat transport biases when accumulating to the west of Rockall Trough, they tend to underestimate heat transport in the interior of Rockall Trough compared to OSNAP estimates. This results in a generally lower net heat transport for the entire OSNAP East section. More broadly, the reanalyses exhibit high variability (Fig. c) in regions that lack direct mooring observations in OSNAP, underscoring potential uncertainties in the OSNAP transport estimates in these areas.

While differences between observations and reanalyses at OSNAP East persist throughout the OSNAP period (Sect. 3.2), the most striking mismatch occurs during 2015, when OSNAP shows a pronounced heat transport peak that is absent from all reanalyses and indirect transport estimates. We use this event as a case study to characterize and localize the differences between OSNAP and the reanalyses.

Differences in OHT variabilities between observations and reanalyses concerning the 2015 peak can be traced back primarily to the eastern OSNAP glider region (indicated by the green line in Fig. ). This region, located in the eastern Iceland Basin and around Hatton Bank, is highly dynamic and influenced by a northward-flowing branch of the NAC. Figure a shows heat transport anomalies for the glider region alone, where OSNAP observations display a distinct 2015 peak, with transports approximately 50 to 130 TW higher than the reanalyses, accounting for much of the offset in the full-section transports (Fig. b). This peak is also evident in volume transport (Fig. b), indicating a link to flow strength or structure rather than temperature alone. In fact, vertical profiles of temperature anomalies (Fig. ) show broad agreement across data sets: both indicate colder than usual conditions in the eastern glider box (blue) and cooler surface waters with warmer intermediate layers in the western box. However, anomalies tend to be stronger in the reanalyses than in OSNAP. Anomalies of currents (Fig. b) reveal a strong but narrow positive anomaly in OSNAP just west of Hatton bank (at x≈2.85 ×106 m) during 2015, suggesting an intensification of the NAC that is not captured to this extent by any of the reanalyses. The combination of this intensified northward flow and slightly warmer temperatures relative to the reanalyses likely explains the anomalously high heat transport observed in OSNAP. As a separate example of mesoscale variability (unrelated to the 2015 heat transport peak), a strong anticyclonic eddy is also present in the western glider region, clearly visible in GLORYS12V1 and, to a lesser extent, in GLORYS2V4. This feature was also sampled by the OSNAP western glider array, as reported by , highlighting the glider system’s capability to capture mesoscale variability. However, no such signal is evident in the velocity field of the publicly available OSNAP dataset used in this study, which is primarily determined using mooring observations that do not resolve the small-scale spatial structures such as eddies between moorings. While this eddy is unrelated to the 2015 heat transport peak, it further illustrates the variability in this region and the importance of consistent observational coverage.

To provide geostrophic context for the 2015 anomaly in the glider region, we next examine along-section sea-level anomalies (SLA) and their gradients. We note a key methodological difference: OSNAP derives time-varying geostrophic shear from T/S and applies a spatially uniform, but time-varying barotropic compensation to obtain absolute velocities, whereas the reanalyses assimilate SLA and thus include time-varying surface geostrophic flow tied to SLA. Accordingly, the SLA–transport comparison below is intended as a diagnostic of reanalysis consistency and as context for OSNAP. In this framework, tight pointwise correlations between SLA gradients and OSNAP transport are not expected. Nevertheless, because along-section SLA gradients set the surface geostrophic shear, we’d expect a coherent relationship after spatial averaging over broader segments (e.g., the glider regions).

Figure presents Hovmöller diagrams of SLA (left), its along-section gradient (middle), and top-to-bottom vertically integrated volume transport (right), based on observations (top row), GLORYS12V1 (middle), and GLORYS2V4 (bottom). SLA in the reanalyses is computed relative to the same 1993–2012 reference period as the observational product. Overall, SLA and its gradients show similar spatial and temporal patterns between the observations and the reanalyses, consistent with the fact that both GLORYS products assimilate SLA data even at these northern latitudes. Fine-scale details are more apparent in the higher-resolution observations and GLORYS12V1 than in the coarser GLORYS2V4.

However, the vertically integrated volume transport fields reveal more pronounced differences. In OSNAP, the strong 2015/16 transport maximum is not accompanied by an equally strong along-section SLA gradient in the eastern glider region, and outside this event the OSNAP transports remain relatively smooth and weak, even during periods of enhanced SLA gradients. This is consistent with OSNAP’s use of a time-mean reference velocity with barotropic compensation, i.e., SLA does not directly set the absolute velocity along the line. The GLORYS reanalyses, which assimilate SLA, exhibit spatially coherent transport anomalies, including pronounced signals in the western glider box, that align more closely with their SLA-gradient fields.

Correlations between SLA gradients and volume transport across the glider region are shown in Fig. . Solid lines represent correlations between each reanalysis and its own SLA gradient field, while dashed lines represent correlations to the observational SLA gradient. As expected given OSNAP’s time-mean reference velocity, OSNAP shows weak pointwise correlations with observed SLA gradients (mean ≈ 0.08 along the section). In contrast, GLORYS12V1 and GLORYS2V4 exhibit higher correlations with observed SLA gradients (0.40 and 0.39, respectively) and even larger values when compared to their own SLA fields (0.79 and 0.51). Both products assimilate the same multi-mission satellite altimetry-derived SLA observations but differ in horizontal resolution and assimilation methodology (see Table 1). Notably, the higher-resolution GLORYS12V1 shows the strongest correlations overall, consistent with its improved spatial representation of circulation features. Correlations are weaker for the ORAS5 reanalysis (not shown), with a correlation of just 0.25 against observed SLA gradients, likely a result of it not assimilating sea level anomalies north of 50° N. To reduce small-scale noise and facilitate visualization, the along-section correlation curves shown in Fig.  are smoothed to 1° resolution. This smoothing is applied only for plotting purposes and does not affect the calculation of the correlations themselves, which are performed on the native-resolution fields along the OSNAP section. This allows us to focus on mesoscale dynamics while suppressing small-scale variability and potential sampling mismatches. Despite this visual smoothing, pointwise correlations across the section remain relatively low for the OSNAP dataset. However, when SLA gradients and volume transport are averaged over the two full glider regions before computing the correlation, substantially higher values are found for OSNAP (0.55 for the western, 0.47 for the eastern glider region) and GLORYS12V1 (0.72 for the western, 0.89 for the eastern), while the correlation for GLORYS2V4 decreases slightly (0.32 for the western, 0.37 for the eastern). These reanalysis-based values refer to correlations with their respective SLA gradient fields. Overall, these results suggest that, while local correlations are sensitive to methodology, sampling, and the limited ability of observations and reanalyses to resolve small-scale processes, the large-scale geostrophic relationship emerges clearly after spatial averaging over the glider regions in both the observations and the high-resolution reanalyses.

Correlations computed using full geostrophic velocity fields, calculated from reanalysis temperature and salinity profiles via pressure gradients and referenced to surface altimetry show nearly identical patterns (not shown), confirming that SLA gradients dominate the transport signal.

To complement the heat transport analysis, we also assess the Meridional Overturning Circulation (AMOC) in density space at OSNAP East (Fig. ). The reanalyses reproduce the mean overturning streamfunction reasonably well, consistent with earlier findings by , with a maximum around 27.5–27.6 kg m⁻³ and peak strengths of about 16 Sv, slightly higher in the reanalyses, particularly in ORAS5. In terms of variability, OSNAP observations show a clear peak in 2015, which coincides with the observed heat transport variability (see Fig. ). Interestingly, the reanalyses also display a peak in overturning strength, though shifted slightly later, to early 2016. However, and as discussed in Sect. 3.2, this overturning peak does not coincide with a heat transport peak in the reanalyses. Looking at transports of the in- and outflow branches of the overturning, we find that in GLORYS12V1 both inflow and outflow heat transports are stronger than in OSNAP, but the outflow has a greater net effect than in OSNAP. This suppresses the overall heat transport peak despite intensified overturning. The mismatch may result from temperature biases, such as colder inflow or warmer outflow, and suggests that the overturning in the reanalyses may involve different water mass properties or pathways than observed.

To diagnose why the reanalysis MOC peak does not translate into a total OHT peak, we decompose MHT into overturning and gyre components (Fig. ), as done e.g. by . In the OSNAP observations, the 2015 maximum is overturning-driven and followed by a multi-year decline. The 2019 minimum is likewise set by the overturning term and the consecutive increase is reinforced by a strong gyre contribution. In the reanalyses, the overturning term is biased high in the mean but its 2015 anomaly is weaker and lacks the subsequent decline, while the gyre term is systematically lower than in OSNAP (even negative in ORAS5). The 2019 minimum in the overturning term is weaker in the reanalyses and a contrasting maximum in the gyre term flattens the 2019 minimum in the net transports (see Fig. ). Therefore, the good correspondence in MOC yet poor correspondence in total OHT might be due to the temperature contrast associated with the overturning being too low or shifted in the reanalyses and additionally the gyre term dampening the total OHT anomalies.