Figure 2 presents mean annual cycles and anomaly time series of relevant oceanic transport quantities at the GSR. It is complemented with long-term averages shown in Table 2 (volume fluxes) and Table 3 (heat fluxes). Observations show seasonally varying AW inflow across the GSR (8.0 ± 0.5 Sv; mean ± standard deviation of the mean reported throughout, unless explicitly stated) with a maximum in December and minimum in June–July (Fig. 2a). The AW inflow is largely balanced by PW (Fig. 2c; -1.8± 0.7 Sv) and OW (Fig. 2b; -5.6 ± 0.3 Sv) outflow, yielding a relatively small net volume flux across the GSR of 0.7 Sv (balanced by flows through Bering and Davis straits). The PW outflow exhibits an annual cycle balancing the AW inflow (i.e., maximum outflow in boreal winter), while the OW exchange is more stable throughout the year (i.e., small annual cycle).

All reanalyses but one (CGLORS) underestimate AW inflow across the GSR when compared to observations, but the shape of the annual cycle of all estimates is in good agreement with observations. The AW net flow from high-resolution products does not stand out compared to the 0.25∘ products and is close to the GREP mean. Agreement is also good for the PW outflow, where all products show the observed seasonal maximum flow in boreal winter. The range of reanalysis-based estimates is large in a relative sense, with the observations lying in the middle of the range. There is less coherence across products concerning the OW transports. GLORYS12, GLORYS2V4, and CGLORS are close to observations, with rather persistent overflow on the order of -5.4 Sv and seasonal variations that agree with the observations. Other products tend to have overflows that are too weak (most notably ORAS5), and some also exhibit biases in the representation of the annual cycle (e.g., GloRanV14).

Table 2 additionally includes long-term average volume fluxes in the main AW inflow branches (North Icelandic Irminger Current (NIIC), I-F branch, F-S branch, and European shelf). The reanalysis-based estimates generally agree well with observations. The main discrepancy is the underestimation of I-F inflow and overestimation of F-S inflow by all GREP reanalyses, while the high-resolution products GLORYS12 and GLOB16 are in much better agreement with observations. Direct comparison of GLOB16 to GLOB4 confirms that overestimation of F-S volume flux is reduced when going from 0.25 to 0.0625∘ resolution. This suggests that increased resolution, along with more realistic bathymetry, improves representation of inflow pathways in the reanalyses. We also note that temporal anomaly correlations with observed I-F volume fluxes are low (Pearson correlation coefficients r range from -0.03 to 0.26 and are statistically insignificant) for all reanalyses but are substantially higher for F-S volume fluxes (r ranges from 0.30 to 0.71 and is statistically significant for all products with data assimilation).

Figure 2d shows the mean annual cycle of heat flux across the GSR, i.e., the sum of sensible heat transported by all waters crossing the GSR. The mean annual cycle of GSR heat fluxes generally follows that of AW volume fluxes, with a minimum between boreal spring and early summer and a maximum in fall and early winter. Seasonal minima and maxima in GSR heat flux co-occur with those of AW volume flux; i.e., seasonal variations in heat flux are largely volume flux driven, and the seasonal cycle in volume-weighted temperatures is in phase.

Since net volume flux across the GSR is small, the ambiguity arising from the choice of reference temperature can be considered small as well. However, for the long-term averages we focus on net heat transport into the Arctic Mediterranean, i.e., the sum of heat fluxes across the GSR, plus those through Bering and Davis straits. Values in Table 3 show that all reanalyses exhibit lower net heat transport (by ∼ 16 % for the GREP mean of 256 ± 19 TW) than that observed (306 ± 19 TW; 311 ± 20 TW when including sea ice). GLORYS12 exhibits a mean net heat flux similar to the GREP mean, but we note that all GREP reanalyses have AW influx that is too strong in the F-S branch (Table 2), where the climatologically warmest waters cross the GSR, and thus enhance the heat flux in those products for the wrong reason.

The net heat transport is clearly related to the strength of AW inflow, but even CGLORS, which has a higher AW mean volume flux (8.3 Sv) than observations, has a negative net heat flux bias. The reason is that all reanalyses exhibit a warm bias in outflowing OW (not shown) and a cold bias in Davis Strait inflow (see Pietschnig et al., 2018). Based on M. Mayer et al. (2022) and taking oceanic storage into account, the energy-budget-based estimate of the net heat boundary transport suggests even higher values (∼ 348 TW) than observations. This inferred value appears high, but we note that this indirect approach has been applied successfully to infer observation-based oceanic transports in the North Atlantic (Trenberth and Fasullo, 2017; Liu et al., 2020; J. Mayer et al., 2022) and the central Arctic (Mayer et al., 2019), and hence it is deemed credible. This estimate at least adds confidence to the conclusion that the observational estimate of oceanic heat transport is not biased high. Table 3 also confirms that long-term averages for heat flux across the GSR are qualitatively very similar to the net heat transport; i.e., heat fluxes across the GSR are the dominant contributor to oceanic heat transport into the Arctic Mediterranean.

Figure 2e shows deseasonalized anomalies of AW volume flux, with a 12-monthly smoother applied to emphasize interannual variability. Typical variability is similar across observations (temporal standard deviation σ = 0.35 Sv) and reanalyses (σ ranges between 0.22 and 0.47 Sv). Temporal correlations between reanalyzed and observed AW inflow anomalies are moderately high (r ranges between 0.39 and 0.60 and is statistically significant for all products; see the legend of the plot for values).

Figure 2f shows anomalies of total oceanic heat flux across the GSR, which show similar variability to AW volume flux; i.e., the strength of AW inflow modulates not only the seasonal cycle of the total GSR heat flux but also its interannual variations (r ranges from 0.86 to 0.91). GSR total heat flux from reanalyses is in slightly better agreement with observations than AW volume fluxes (r ranges between 0.51 and 0.63 and is statistically significant for all products; see the legend of the plot for values). Figure 2f also shows a prominent negative heat flux anomaly centered around the year 2018, which has already been noted by M. Mayer et al. (2022) for net heat transport into the Arctic Mediterranean.

To set the scene for further investigation, we present climatological temperatures and currents at the GSR based on GLORYS12 in Fig. 3a and b, respectively. Comparison with analogous figures based on the GREP (shown in M. Mayer et al., 2022) allows us to appreciate the benefits of increased resolution (0.08 vs. 0.25∘ resolution), including a more distinct representation of inflow and outflow branches and a more spatially variable bathymetry, especially in the I-F branch.

Next, we investigate the recent reduction in AW volume and GSR heat fluxes in more detail. This is most prominent in the biennial (2-year-long) signal of average anomalies in July 2017–June 2019. Figure 3c shows that during this period strong warm anomalies were present over 0–400 m depth in the eastern Denmark Strait, and warm anomalies are also seen in the F-S branch. The latter suggests a temporary deepening of the AW layer. Velocity anomalies for the July 2017–June 2019 period (Fig. 3d) suggest that the positive temperature anomalies in the eastern Denmark Strait are driven by enhanced NIIC transports. The strongest and deepest velocity anomaly in July 2017–June 2019 is located in the eastern part of the F-S branch, where reduced inflow is present from the surface down to the interface at ∼ 600 m, and hence we focus on this branch next. Negative anomalies are also seen in the overflow from 600 to 1000 m depth. There is some compensation by positive velocity anomalies in the western F-S (meaning reduced southward flow there), but the effect of the eastern F-S anomaly dominates, and the net F-S volume flux anomaly was clearly reduced during this period (see below). We note that these main features are also similar in anomaly sections based on the GREP ensemble mean (not shown).

Observations and all reanalyses show large negative F-S volume inflow anomalies of -0.61, -0.56, -0.34, and -0.24 Sv in the observations, GREP_mean, GLORYS12, and GLOB16, respectively, during July 2017–June 2019 (Fig. 3e). In four out of seven datasets, this is the overall biennial minimum of the 1993–2021 record (not shown). The total AW volume flux anomaly for that period was -0.24, -0.53, -0.37, and +0.08 Sv in the observations, GREP_mean, GLORYS12, and GLOB16, respectively. Thus, F-S volume flux anomalies were partly compensated by other AW branches, with no very clear signal in any of them (not shown). Only GLOB16 exhibits a positive AW volume flux anomaly during that period, which appears to be related to a shift towards generally higher AW volume flux around 2016 (see Fig. 2e). This is not seen in any of the other products.

Temperature transport anomalies in the F-S branch are strongly correlated with volume flux anomalies (compare Fig. 3e and f), and there is a clear reduction in F-S branch temperature flux in July 2017–June 2019 of -21.9, -24.1, and -16.1 TW in the observations, GREP_mean, and GLORYS12, respectively. The contribution of temperature anomalies in the F-S branch during that time was small, with biennial anomalies of observed volume-weighted temperatures between -0.2 and -0.1 K (similar for all products). The total AW heat flux anomaly for July 2017–June 2019 was -11.0, -22.5, and -19.4 TW in observations, GREP_mean, and GLORYS12, respectively. Thus, the reduction in GSR heat transports during that period was mainly driven by a reduction in volume inflow through the F-S branch. It was only partly offset by compensating transport anomalies in other branches. Very similar biennial heat transport anomalies for the total GSR (-11.1, -20.9, and -19.3 TW for observations, GREP_mean, and GLORYS12, respectively) confirm AW as the main driver of heat flux variability across the GSR.

For a better understanding of mechanisms contributing to the GSR transport anomaly around 2018, we first consider basic statistical relationships between SLA and oceanic quantities. We find a statistically significant temporal correlation of the zonal SLA gradient at GSR with observed AW volume flux anomalies (Fig. 4a), which is plausible in terms of geostrophic balance. The correlation pattern looks very similar when performed with the F-S branch volume flux alone (not shown). The pattern of temporal correlation between the SLA field and observed anomalies of volume-weighted temperature of AW transports (Fig. 4b) is distinct from the relationship with AW volume flux (compare Fig. 4a). It emphasizes SLA in the North Atlantic subpolar gyre (SPG), with higher SLA in the SPG (i.e., a weaker gyre) associated with higher volume-weighted temperature and vice versa. Although correlations are high (r up to 0.69) in the SPG region, they are not statistically significant. The cause may be the low number of degrees of freedom, as the SLA in the SPG exhibits high temporal auto-correlation (see Fig. 4d discussed below).

Actual SLA anomalies averaged over July 2017–June 2019 (Fig. 4c) indicate a weakened zonal SLA gradient at the GSR, albeit not very strongly pronounced, and anomalously low SLA in the SPG region. According to the correlation patterns discussed above, these two features suggest reduced AW volume flux (as suggested by patterns in Fig. 4a) and anomalously low AW volume-weighted temperatures (as suggested by patterns in Fig. 4b). We also note the positive SLA anomalies north of the GSR with a maximum in the central Nordic Seas.

To put these results in context, we define two SLA-based indices (shown in Fig. 4d) from the correlation patterns found in Fig. 3a and b. The similarity of the correlations in Fig. 4a to the correlations between total OHT at the GSR and SLA shown in M. Mayer et al. (2022) reinforces use of their gradient-based index (i.e., standardized SLA difference between 58–60∘ N, 2–0∘ W and 63–67∘ N, 20–15∘ W). This index is correlated with AW volume flux anomalies (r = 0.62 for observed transports, ranging from 0.47 to 0.70 for reanalyses, which is statistically significant in all cases). The second index uses spatial SLA averages in the North Atlantic region (55–60∘ N and 40–15∘ W) as an inverse proxy for SPG strength. This index is correlated with anomalies of volume-weighted temperatures of AW transports. The two indices in Fig. 4d show different characteristics, with the SPG index varying on decadal timescales, while the gradient index shows stronger interannual variations. The SPG index has been negative since ∼ 2014, which suggests a strong SPG and lower inflow temperatures in recent years, in agreement with results by Hátún and Chafik (2018). This SPG index exhibits two minima during the July 2017–June 2019 period, although these are not extreme relative to the entire time series.

Figure 4e shows volume-weighted temperatures of AW waters from different products and confirms their overall decrease in recent years. Comparison with the SPG index in Fig. 4d suggests a generally delayed response of volume-weighted temperature in Atlantic inflow water at the GSR to SPG strength. Lagged correlation analysis indeed suggests positive correlations peaking (at values around 0.3 to 0.5, depending on the dataset) when temperatures are lagging the SPG index ∼ 2–3 years (not shown), but the correlations are not significant due to the high auto-correlation of time series.