Sea ice-air interactions amplify multidecadal variability in the North Atlantic and Arctic region

Arctic Tas exhibits substantial multidecadal variability superimposed on a long-term warming trend, which cannot be explained by monotonic greenhouse gas increases. Previous studies and models suggest internal variability related to Atlantic Multidecadal Variability (AMV) and the Atlantic Meridional Overturning Circulation (AMOC) modulates poleward energy transport, reducing sea-ice cover and amplifying winter Arctic warming via enhanced ocean-atmosphere heat exchange. Sea-ice loss can also weaken AMOC, implying potential two-way coupling. The key questions addressed are whether sea-ice changes can influence AMV/AMOC and whether poleward oceanic heat transport alone can generate large multidecadal Tas variability without sea ice. Understanding these mechanisms is important for explaining recent Arctic warming patterns, associated Eurasian cooling, and for projecting future variability as sea ice declines under anthropogenic forcing.

The study builds on evidence of multidecadal Arctic Tas variability (e.g., Johannessen 2004; Serreze & Barry 2011; Tokinaga et al. 2017) and modeling work attributing low-frequency Arctic variability to internal dynamics and AMV/AMOC (e.g., Beitsch et al. 2014; Jungclaus & Koenigk 2010; Zhang et al. 2019). Mechanisms linking North Atlantic heat transport to Arctic SIC/Tas via surface fluxes have been proposed (e.g., van der Linden et al. 2016). Other studies indicate Arctic sea-ice decline weakens AMOC (Sévellec et al. 2017; Liu et al. 2019) and that sea ice modulates Atlantic multidecadal variability (Escudier et al. 2013; Drews & Greatbatch 2017). The potential for sea-ice loss to affect midlatitude climate via blocking is documented (Yao et al. 2017; Luo et al. 2018). Prior work also highlights projected sea-ice loss and associated changes (Wang & Overland 2012; IPCC AR5 2013; SIMIP Community 2020). Despite suggestions of important sea ice-air interactions, their explicit role in generating/amplifying multidecadal variability of Tas, SIC, NASST and AMOC remained insufficiently demonstrated, motivating this analysis.

Data and models: Observational/reanalysis fields from ERA5 (1950–2020) for Tas, SST, and sea ice on a 1° grid. Comparative checks with NCEP/NCAR were made. CMIP5/CMIP6 historical plus high-emission extensions (RCP8.5/SSP5-8.5) from 12 models, extended to year 2300 for a largely ice-free future. Seven CMIP5/6 models provided 500-year piControl simulations with AMOC. Model outputs remapped to 2.5° for atmospheric and 1° for ocean/ice fields. CESM1 experiments: Community Earth System Model v1.2.1 with CAM4 atmosphere; ocean/ice grid ~1°×0.5°. Two simulation types: (1) fully coupled with dynamic sea ice (two-way ice-air interactions), including a 500-year pre-industrial control (CTL) and a 235-year 1% per year CO2 increase run (1%CO2); (2) FixedIce configuration where sea-ice concentration (SIC) is fixed to the CTL monthly climatology only in the coupler for computing area-weighted surface exchange fluxes (energy, mass, momentum). Sea ice within the ice model and fluxes over ice/water surfaces evolve dynamically; only the weights are fixed, effectively removing two-way ice-air coupling. FixedIce runs include a 500-year control (CTL_FixedIce) and a 1%CO2_FixedIce run. This setup isolates the physical effect of variable SIC on surface fluxes and downstream climate while preserving internal ocean-atmosphere processes. Focus period/season: Boreal winter (DJF), with annual and other seasonal checks showing similar but weaker signals. Signal separation and filtering: Externally forced signals were removed by regression on global-mean Tas (GMT). For reanalysis (1950–2020) and CMIP historical/future (1900–2299) a linear regression at each grid point against multi-model GMT was used. For CESM1 1%CO2 runs, a third-order polynomial fit to each field vs. the model’s own GMT removed nonlinear forced trends (e.g., nonlinear SIC decline). After detrending, a Lanczos 10–90-year band-pass filter (21 weights) isolated decadal–multidecadal variability. Similar results were obtained with more weights, trading larger magnitudes for reduced sample sizes. Indices and regions: AMV defined as detrended, filtered DJF SST anomalies averaged over 0°–60°W, 50°–65°N. AMOC index defined as detrended, filtered DJF Atlantic zonal-mean overturning streamfunction averaged over 1000–2000 m and 40°–55°N. Arctic marginal ice zones: Labrador Sea–Davis Strait (LSDS: 45°–65°W, 55°–70°N), Greenland–Norwegian Seas (GNS: 25°W–5°E, 65°–80°N), Barents–Kara Seas (BKS: 30°–80°E, 70°–80°N). Diagnostics: Standard deviation (SD) maps of filtered anomalies quantified variability. Lead–lag correlations between regional SIC/Tas/LHF/E–P/SSS/MLD/RHO and AMV/AMOC assessed coupling and lags. Power spectra identified common multidecadal peaks. Composite analyses contrasted high vs low LHF years over LSDS and subsequent AMOC responses (3–5 year lags). Statistical significance employed F-tests for SD differences (accounting for effective degrees of freedom), Student’s t-tests for composites, and a resampling method (10,000 reshuffles) for correlations of autocorrelated series.

- Observations (ERA5, 1950–2019) show large multidecadal variability in SIC and Tas concentrated along Arctic sea-ice margins (BKS, GNS, LSDS), with strong anti-correlation between SIC and Tas (r from −0.76 to −0.92, p < 0.05). AMV relates to these variations with small lags; however, AMV-associated NASST anomalies (~0.2 °C) are too small to directly cause the large Arctic Tas swings (~1.5–2 °C), implying a local amplification mechanism. - CESM1 reproduces observed patterns in CTL. When sea ice-air interactions are removed (CTL_FixedIce), multidecadal Tas variability weakens markedly along ice margins, especially over GNS and BKS; NASST variability also declines in the northern North Atlantic and Nordic Seas. - Quantified reductions with FixedIce relative to fully coupled runs: • LSDS: Tas multidecadal variability ↓ ~36%; AMV ↓ ~31% (CTL_FixedIce vs CTL), while SIC variability weakens only slightly at LSDS. • BKS: Tas variability ↓ ~49%; SIC variability ↓ ~16% (CTL_FixedIce vs CTL). Under rising CO2 before year ~150, reductions are larger: Tas ↓ ~70%; SIC ↓ ~19% (1%CO2_FixedIce vs 1%CO2). - Decadal trends: Observed BKS decadal winter warming 1997–2009 is ~2.69 °C/decade. Comparable strongest decadal trends occur in CTL (~2.99 °C/decade for top 5% windows) but are reduced by ~55% to ~1.36 °C/decade when sea ice-air coupling is cut off. Probabilities of recent Arctic/BKS-like decadal warming increase with coupling (Arctic-mean: 2.14% → 5.78%; BKS: 0.00% → 3.64% from CTL_FixedIce to CTL), indicating coupling can amplify multidecadal variability to produce such events. - In future warmer climates: As sea ice retreats/melts (e.g., after ~150 years in CESM1 1%CO2 and by the 23rd century in CMIP5/6), multidecadal variability in SIC, Tas over BKS/GNS/LSDS, and NASST weakens or nearly disappears. Variability centers migrate poleward with ice margins but diminish in amplitude. - Mechanism: A positive feedback loop via surface fluxes. Reduced SIC (triggered by small advected SST anomalies) exposes open water, increasing upward turbulent and longwave fluxes, strongly warming winter near-surface air. Warmer air raises downward LW radiation, further reducing sea ice, amplifying SST/Tas anomalies. In CTL, SIC and upward LW/turbulent fluxes are anti-correlated on multidecadal scales (r ≈ −0.56 to −0.64, p < 0.05). In FixedIce, these relationships weaken or reverse due to decoupling. - Latent heat flux (LHF) is a key mediator: Over LSDS, SIC-associated multidecadal LHF variability decreases by ~74% in CTL_FixedIce relative to CTL, coinciding with reduced SST variability. In future ice-free scenarios, multidecadal LHF variability also declines over the northern North Atlantic/Nordic Seas, collocated with reduced NASST variability. - AMOC variability: CESM1 CTL shows strong multidecadal AMOC variability centered ~1.5 km depth at 40°–50°N. Removing sea ice-air interactions reduces AMOC multidecadal variability by ~20% north of 40°N and weakens mean AMOC. In CTL, LSDS SIC and LHF lead AMOC by ~4–5 years (SIC–AMOC r ≈ −0.4; LHF–AMOC r ≈ −0.5, p < 0.05). SIC decreases drive higher LHF and E–P, increasing SSS and upper-ocean density, deepening MLD and enhancing NADW formation, leading to deeper/stronger AMOC after 3–5 years. FixedIce diminishes multidecadal variability in E–P (~27%) and LHF (~74%), weakening SSS/density/MLD variability and producing shallower, weaker AMOC anomalies. - CMIP multi-model consistency: piControl runs from multiple models show similar lead-lag relationships (AMOC lags SIC/LHF/E–P by ~3–10 years) though with inter-model spread, supporting the amplification role of sea ice-air interactions for AMOC and NASST variability. - Synthesis: Without sea ice-air coupling, large multidecadal Tas variations largely vanish, and SIC, NASST (AMV), and AMOC multidecadal variability weaken substantially. As sea ice retreats under GHG warming, this mechanism wanes, contributing to projected reductions in AMOC variability and impacts on North Atlantic–Arctic climate variability.

The study demonstrates that sea ice-air two-way interactions, via surface flux feedbacks, are essential for amplifying small, advected SST anomalies into large multidecadal variations in Tas, SIC, NASST, and AMOC from the subpolar North Atlantic to the Arctic. This addresses the research question by showing that poleward oceanic heat transport alone cannot generate the observed magnitude of multidecadal Arctic Tas variability in winter without the participation of sea ice-air coupling. The mechanism hinges on SIC-modulated upward turbulent and longwave fluxes that warm the lower troposphere and enhance downward LW radiation, reinforcing sea-ice loss and surface temperature anomalies; over the LSDS, increased LHF and E–P elevate SSS and density, strengthening NADW formation and AMOC after several years, which further enhances poleward heat transport and sustains AMV. The findings imply that as sea ice retreats, multidecadal variability in Tas and SIC along current ice margins will diminish and shift poleward, and AMOC/AMV variability will weaken, with potential consequences for European climate and Eurasian winter conditions (e.g., via Ural blocking and cold advection). The work underscores the need for climate models to realistically represent sea-ice extent and ice–air–ocean coupling to simulate AMOC/AMV variability and future changes reliably.

Sea ice-air interactions are a crucial, previously under-quantified amplifier of multidecadal climate variability across the North Atlantic–Arctic system. By modulating winter surface fluxes along ice margins, variable sea ice transforms small SST perturbations into large Tas and SIC anomalies and strengthens or weakens AMOC on multidecadal timescales. When sea ice-air coupling is removed or sea ice vanishes, multidecadal variability in Tas, SIC, NASST (AMV), and AMOC diminishes substantially. As GHG-driven warming reduces subpolar/Arctic sea ice and shifts ice margins poleward, the associated variability will weaken, contributing to reductions in AMOC mean strength and variability. Future work should: (1) quantify model biases linking SIC and AMOC/AMV, (2) investigate how poleward shifts of ice–air interaction centers modulate AMOC/AMV under warming, (3) assess recovery of variability in equilibrium warmer climates without subpolar sea ice, and (4) improve representation of sea ice-air-ocean feedbacks and surface flux processes in models.

- The FixedIce configuration, while isolating the influence of variable SIC on fluxes, is an artificial setup and may not capture all coupled feedbacks; however, it conserves fluxes and allows dynamic sea ice internally. - In fully coupled future-warming simulations, numerous concurrent changes (e.g., stratification, freshwater inputs, circulation shifts) make it difficult to definitively attribute variability reductions solely to loss of sea ice-air interactions. - Inter-model spread is substantial in the SIC–LHF and SIC–AMOC relationships across CMIP models, indicating uncertainty in current model physics and parameterizations. - Some datasets/variables were unavailable for all models (e.g., density RHO and full MLD fields), limiting multi-model comparisons for those metrics. - The analysis focuses on DJF and multidecadal bands (10–90 years); results for other seasons and timescales are qualitatively similar but smaller and not the main emphasis. - The relative contributions of internal variability versus external forcing to recent Arctic/BKS warming trends remain to be further constrained.