Accurate climate prediction on timescales ranging from years to decades is crucial for climate adaptation and resilience. Assessing the reliability of such predictions necessitates understanding the mechanisms driving decadal climate variability, including anthropogenic and natural external forcing, and internal variability. The North Atlantic Ocean exhibits significant interannual-to-multidecadal variability influencing regional and global climate (e.g., North American heat waves, North Atlantic tropical cyclone activity, Arctic sea ice extent, Asian monsoon patterns, and worldwide precipitation). While mechanisms governing North Atlantic SST variability are studied extensively, the impact of anthropogenic forcing on internal decadal variability remains an open question. Anthropogenic emissions cause mean state changes in the North Atlantic, such as the projected weakening of the Atlantic meridional overturning circulation (AMOC). This weakening, attributed to increased heat and freshwater flux, reduces or halts North Atlantic deep convection, impacting poleward heat transport and causing phenomena like the North Atlantic warming hole. Understanding how these mean changes under external forcing modulate internal North Atlantic variability is crucial. This study uses the Community Earth System Model Version 2 Large Ensemble (CESM2-LE) to investigate the impact of external forcing on decadal variability in North Atlantic SST. The CESM2-LE's single-model approach, with variations arising solely from initial condition perturbations, allows for the separation of forced and internal variability components, enabling the assessment of changes in internal variability due to external forcing.

Numerous studies have explored North Atlantic SST variability using observations and modeling. These studies have examined the mechanisms governing this variability, including the Atlantic Meridional Overturning Circulation (AMOC) and its connection to phenomena such as the North Atlantic Oscillation (NAO). However, many of these studies were limited in the number of simulations used or focused primarily on historical or pre-industrial climate conditions, leaving gaps in our understanding of how anthropogenic forcing might alter internal decadal variability. Previous research has highlighted the role of ocean dynamics in connecting the NAO to the Atlantic multidecadal oscillation, and the importance of mixed layer depth variability in the emergence of the Atlantic multidecadal SST signal. Studies have also emphasized the persistence and coherence of subpolar sea surface temperature and salinity anomalies associated with this variability. The impact of Arctic sea ice changes on North Atlantic deep convection and the AMOC has also been investigated, suggesting a link between Arctic changes and large-scale ocean circulation. However, a comprehensive understanding of how external forcing interacts with internal variability on decadal timescales within the North Atlantic remains an active area of research.

This study utilizes the Community Earth System Model Version 2 Large Ensemble (CESM2-LE), a 100-member ensemble forced with historical (1850-2014) and SSP 3-7.0 future (2015-2100) radiative forcing scenarios. The ensemble spread arises solely from initial condition perturbations, allowing for the isolation of internal variability. Decadal variability in North Atlantic SST was isolated using an 11-year running mean lowpass filter applied to winter (DJFM) data. An index of northern North Atlantic SST (NNASST) was computed as the area-average over 50-80°N, 90°W-40°E. Ensemble members were classified into warm and cold groups based on their NNASST index during the 5 years (2056-2060) with the highest standard deviation. Composite analyses were then conducted to identify distinct trajectories. The Labrador Sea mixed layer depth (MLD), northward heat transport (NHT) across 50°N, and the AMOC at 50°N were analyzed for both groups. A salinity budget analysis was performed over the upper 295 m of the Labrador Sea to investigate the processes driving density differences between the groups, decomposing the total salinity tendency into contributions from resolved advection, parameterized advection, diabatic vertical mixing, surface flux, and residual terms. Similarly, a heat budget analysis was conducted, decomposing the temperature tendency into similar components. The North Atlantic Oscillation (NAO) index was analyzed to investigate potential triggers for the divergence in trajectories. Two potential positive feedback mechanisms were proposed and explored: one involving the relationship between Labrador Sea deep convection and vertical mixing, and another involving sea ice, wind stress, and vertical mixing. To assess the generalizability of the findings, seven additional large ensembles from CMIP5 and CMIP6 were analyzed, focusing on the ensemble spread of SST in the region of the North Atlantic warming hole. A warming hole index was used to assess the model's representation of the warming hole and its relationship to the increased spread in SST.

The study found a significant increase in the ensemble spread of northern North Atlantic SST during the mid-21st century in the CESM2-LE, indicating a much wider range of possible future climate states than in the historical period. This increased spread is concentrated north of 50°N, a region known for its high predictability under current conditions. Distinct trajectories of NNASST emerged, with warm and cold groups diverging significantly starting around 2035. This divergence originates in the Labrador Sea region in the 2030s and spreads across the mid-to-high latitude North Atlantic. The cold group experiences shallower MLDs and an accelerated reduction of Labrador Sea deep convection, leading to earlier deep convection shutdown (2052 vs. 2063 for the warm group). This difference in deep convection leads to a clear separation in AMOC and NHT trajectories, starting in 2032 and 2029 respectively. The cold group has weaker AMOC and reduced NHT, intensifying NNASST differences. Analysis of Labrador Sea density revealed that salinity is the primary driver of density differences, with the cold group having lower density due to lower salinity. The salinity budget analysis showed that weaker vertical mixing in the cold group, limiting the upward transport of higher-salinity subsurface water, is primarily responsible for this fresher upper ocean. Weaker deep convection in the cold group further reduces vertical mixing, creating a positive feedback loop. A second positive feedback involves sea ice, wind stress, and vertical mixing. Colder SST in the cold group reduces sea ice melting, lessening the susceptibility to wind stress and reducing vertical mixing, thereby further weakening convection. The initial trigger for these divergent trajectories appears to be related to differences in the NAO index, which results in either less heat loss from the Labrador Sea through reduced westerlies and/or reduced mechanical stirring through wind stress resulting in reduced vertical salinity mixing. Analysis of seven additional large ensembles showed that the increase in internal SST variability is broadly present in models simulating the formation of a North Atlantic warming hole under global warming, supporting the robustness of the findings. The timing of this increased spread varies across models, likely linked to differences in the timing of crossing a ‘tipping point’ that activates positive feedbacks.

This study demonstrates how future climate change significantly impacts decadal variability in the North Atlantic. The dramatic increase in the range of potential climate states during the mid-21st century, resulting from the interaction of stochastic atmospheric variability and positive feedbacks amplified by a warming-induced shift in the mean ocean state, has significant implications for decadal climate prediction. The identification of two positive feedback mechanisms affecting Labrador Sea deep convection and surface salinity provides a mechanistic explanation for the observed divergent trajectories. The findings highlight the critical role of external forcing in driving increased internal variability, with freshwater forcing acting as a necessary condition for activating the feedbacks. The consistency across multiple models that simulate a North Atlantic warming hole supports the robustness of the results, despite variations in timing. The study's findings have implications for improving decadal predictions and have potential to significantly enhance multi-decadal forecasting, which could be valuable for climate adaptation and mitigation.

This study reveals a substantial increase in the range of possible North Atlantic climate states in the mid-21st century due to complex interactions between external forcing and internal variability. The identification of two key positive feedback mechanisms explains the emergence of distinct, multi-decadal trajectories in SST and associated ocean properties, highlighting the importance of external forcing in triggering and amplifying these changes. Future work should focus on comparing model simulations with observations to determine which trajectory is being taken by the real climate system and on extending dynamical model predictions beyond their typical 10-year horizons, taking advantage of the potentially enhanced predictability once positive feedbacks have activated.

The study relies heavily on the CESM2-LE model. While seven additional large ensembles were analyzed to assess robustness, inter-model differences in the timing of increased variability and the details of processes involved highlight potential model-specific biases. The focus on winter variability may not capture the full complexity of year-round processes. Further research is needed to fully understand the extent to which these findings generalize to other climate models and real-world observations.