Decadal changes in Atlantic overturning due to the excessive 1990s Labrador Sea convection

The Atlantic Meridional Overturning Circulation (AMOC) summarizes the basin-scale exchange transporting warm, buoyant waters northward and denser waters southward. How and why the AMOC varies on decadal and longer timescales remains debated, with implications for Northern Hemisphere climate. Prior modeling links decadal AMOC variability to buoyancy-driven density anomalies formed in the subpolar gyre, particularly via deep winter convection in the Labrador Sea. However, observational evidence challenges a simple causal link: weak overturning signals across the Labrador Sea during strong 1990s convection, lack of corresponding changes in the deep western boundary current there, and recent subtropical AMOC declines occurring below the Labrador Sea Water (LSW) layer. OSNAP observations (2014–2018) further indicate that most downwelling occurs in the Irminger and Iceland basins, with only a small Labrador Sea contribution. The purpose of this study is to reassess the role of Labrador Sea buoyancy forcing in interannual-to-decadal AMOC variability and its geographic distribution, with focus on the extended positive NAO phase in the late 1980s–mid-1990s that produced an exceptionally dense vintage of LSW. Using an eddy-rich ocean model and targeted sensitivity experiments, the study tests whether the 1990s Labrador Sea convection event affected AMOC and through what pathways.

Previous work associates decadal AMOC variability with subpolar density anomalies propagating southward along the deep western boundary (e.g., Jackson et al. 2022; Zhang et al. 2019). Modeling studies suggest meridionally coherent, low-frequency AMOC anomalies driven by low-frequency surface buoyancy forcing over the subpolar North Atlantic, with a strengthening through the 1980s–1990s and subsequent decline, broadly consistent with subtropical observations since 2004 and reconstructions. The Labrador Sea has been emphasized as a major conduit through deep convection modulated by the NAO; yet several observations dispute a direct causal link: weak Labrador Sea overturning during the 1990s, no clear convection signal in the Labrador Sea deep western boundary current, and recent AMOC declines occurring below the LSW density range. OSNAP results highlight primary downwelling in the Irminger and Iceland basins. Classical hydrography documents rapid LSW spreading pathways across the subpolar gyre, including into the Irminger Basin. Recent studies also emphasize boundary downwelling along Greenland’s slope and the role of Nordic Seas overflows and entrainment. Coarse-resolution climate models often overproduce LSW and dilute overflows, complicating inference; very fine resolution is suggested to better capture subpolar processes.

Model: VIKING20X, a global ocean–sea ice configuration based on NEMO v3.6 with LIM2, using AGRIF two-way nesting. The Atlantic nest has ~0.05° horizontal resolution (≈3 km at 60°N; 2–3 km near Greenland), embedded within a 0.25° global grid. Vertical grid has 46 z-levels (6 m near-surface to ~250 m in deep layers). Bottom topography from ETOPO1 with partial cells (minimum layer thickness 25 m). Numerical details (advection, diffusion, boundary conditions) follow Biastoch et al. 2021. Forcing and initialization: Atmospheric forcing JRA55-do v1.4 (1958–2019), OMIP-2 protocols, initial T/S from WOA13. Weak sea surface salinity relaxation (piston velocity 50 m/yr; 1-yr timescale for upper 50 m), disabled in sea-ice regions and within ~80 km of Greenland. Experiments: CTRL is a hindcast (1958–2019) with full JRA55-do forcing. SENS branches from CTRL in May 1980 and repeats the Labrador Sea surface heat fluxes (53°–65°N, 45°–64°W) from May 1980–April 1981 each year, while all other forcing fields match CTRL. This suppresses interannual-to-decadal variability of Labrador Sea heat loss and deep convection, isolating its effect on AMOC, especially during the late 1980s–mid-1990s NAO+ phase. Additional hindcasts (CORE, JRA-cr, JRA-short) differ by initial conditions, forcing (CORE v2 vs JRA55-do), runoff and surface freshwater fluxes, and use weaker SSS relaxation (12.2 m/yr; 4.11-yr timescale) to assess robustness. Diagnostics: AMOC transport across sections is defined as the time-varying maximum of the overturning streamfunction in potential density space, integrating cross-section velocity from east to west and from the densest layer upward to the density of interest. Sections analyzed follow OSNAP design: FULL (trans-Atlantic), EAST (its eastern portion in the Irminger/Iceland basins), SILL (Greenland–Iceland–Scotland Ridge overflows), and 48N (transition to subtropics along WOCE AR19/A2). Mixed layer depth is the depth where surface-referenced potential density increases by 0.01 kg m−3 relative to 10 m. LSW formation metrics include volumes binned in potential density (fine bins 26.50–28.0 kg m−3), and layer thickness anomalies of the dense LSW class (σθ 27.83–28.0 kg m−3). Density time series at key locations (1500–2000 m) are low-pass filtered (4-yr running Hamming). Spatial analyses examine mid-depth density anomalies (e.g., 1500 m) and currents to diagnose spreading pathways from the Labrador Sea into the Irminger Basin. Comparisons are made to OSNAP-era observations (2014–2018) and RAPID (26.5°N) where relevant.

- The high-resolution model reproduces the observed concentration of AMOC downwelling in the northeastern subpolar North Atlantic (Irminger and Iceland basins) and a small net Labrador Sea contribution. - Mean AMOC (1980–2019) in CTRL exceeds 18 Sv between 47°–58°N, peaking at ~18.5 Sv near 54°N; at 26°N it is ~18.1 Sv, indicating limited mid-latitude diapycnal exchange downstream of the subpolar gyre. - Most of the lower limb is formed north of 60°N; only ~2 Sv additional lower limb transport is gained between 60°N and 52°N. - Geographical decomposition shows minimal difference between FULL and 48N, demonstrating that nearly all downwelling occurs north of FULL. The EAST section captures the dominant contribution from the northeastern basins. - The Labrador Sea’s net contribution to transatlantic overturning (FULL minus EAST) is modest: typically ~2 Sv before the mid-1980s and after 2005, with an enhancement of roughly 1 Sv during the mid-1990s. - Interdecadal AMOC changes: CTRL shows an increase from the 1970s to a peak during 1994–1998 (by ~4 Sv at FULL/48N) followed by a decline (~5 Sv). Changes at EAST are only slightly smaller (≈3–4 Sv), indicating that variability is dominated by downwelling in the Irminger and Iceland basins. SILL transport is steady, so decadal variability arises downstream of the overflows. - The sensitivity experiment SENS (with repeated Labrador Sea heat fluxes) largely removes the 1980s–mid-1990s AMOC strengthening seen in CTRL, demonstrating that the strong Labrador Sea buoyancy forcing in that era generated the AMOC peak. The subsequent AMOC decline is only weakly reduced in SENS, implying other factors contributed to the post-1990s downturn. - Despite the decadal impact, interannual AMOC variability is insensitive to Labrador Sea heat fluxes: CTRL and SENS are highly correlated on 1–5 year timescales (r=0.95 at EAST; r=0.87 at FULL), pointing to local buoyancy forcing over the Irminger/Iceland basins as the primary driver of interannual variability. - Mechanism: The exceptionally cold 1990–1994 winters produced a very dense LSW class (σθ ~27.83–28.0 kg m−3). These anomalies rapidly spread from the Labrador Sea into the Irminger Basin via mid-depth interior pathways and boundary currents, reaching the southern Irminger Basin within about a year and influencing the deep western boundary current along Greenland. The thick, dense LSW layer increased abyssal density near Cape Farewell, enhancing entrainment into the lower limb and augmenting downwelling in the northeastern basins. - The 1990s event induced a positive AMOC anomaly exceeding 20% relative to background, realized primarily via enhanced diapycnal downwelling between SILL and EAST, with only a small direct Labrador Sea downwelling contribution. - Model–observation comparison for 2014–2018: CTRL transports are somewhat weaker than observations (FULL 15.3 Sv vs 16.6 Sv observed; EAST 12.8 Sv vs 16.8 Sv), but the small FULL–EAST difference (~2 Sv) is consistent with OSNAP’s conclusion of minor Labrador Sea net contribution during that period.

The study resolves the apparent contradiction between modeling emphasis on Labrador Sea convection and observations that locate most subpolar downwelling in the Irminger and Iceland basins. The findings indicate that typical interannual Labrador Sea convection variability does not significantly modulate the AMOC, which is instead governed on those timescales by local buoyancy forcing in the northeastern subpolar basins. However, the extraordinary 1990s Labrador Sea cooling produced an exceptionally dense LSW class that spread rapidly into the Irminger Basin, increased abyssal density along Greenland’s slope, and enhanced diapycnal downwelling into the AMOC lower limb. This remote modulation generated the pronounced mid-1990s AMOC peak. The mechanism implies that Labrador Sea density anomalies can manifest in the transport of the deeper lower limb rather than the LSW density range itself, offering a plausible explanation for observed post-2008 declines in lower deep-water transport at 26.5°N despite stable Nordic Seas overflow transports. It also supports the notion of meridional coherence primarily in the lower limb on multi-year to decadal timescales.

An eddy-rich ocean hindcast reproduces the observed structure of subpolar downwelling and shows that while routine interannual Labrador Sea convection variability has minimal effect on AMOC, the excessive 1990s Labrador Sea cooling generated a >20% AMOC increase by remotely enhancing downwelling in the Irminger/Iceland basins. The decadal AMOC peak arose from rapid spreading of dense LSW into the Irminger Basin and its entrainment into the deep boundary current, modulating abyssal densities and lower limb transports. The AMOC decline after the 1990s appears influenced by factors beyond Labrador Sea heat flux alone. These results reconcile observations with model dynamics and suggest that only large, multi-year Labrador Sea buoyancy anomalies can impact AMOC on decadal scales via remote pathways. Future work should pursue coordinated, multi-model experiments at very fine resolution to robustly resolve subpolar overflows, mesoscale processes, mixing, and entrainment; assess sensitivities to numerical choices and parameterizations; and quantify the roles of different buoyancy-forcing regions and freshwater inputs in setting interdecadal AMOC variability.

- The SENS design uses a repeated 1980/81 Labrador Sea heat flux year; results may be quantitatively sensitive to the chosen repeat year’s convection intensity and the artificial nature of the perturbation. - Simulated transports during OSNAP (2014–2018) are weaker than observed (e.g., EAST and FULL), indicating model biases and/or forcing uncertainties. - Potential density fields show an offset (~−0.06 kg m−3) relative to observations, which may affect absolute thresholds of LSW classes. - Despite eddy-rich resolution, results can be sensitive to numerical schemes, parameterizations, and representation of overflows and mixing; very fine grid sizes may still be required to fully capture small-scale processes. - The study focuses on buoyancy forcing impacts; other contemporaneous factors (e.g., freshwater flux variability, atmospheric patterns outside the Labrador Sea) may contribute to post-1990s AMOC trends and are not fully isolated. - Use of sea surface salinity relaxation and differences in freshwater runoff treatments across experiments may influence the simulated variability.