Identifying a human signal in the North Atlantic warming hole

The study investigates whether the recently observed North Atlantic “warming hole” (reduced or negative SST trends at midlatitudes relative to surrounding regions) is a consequence of human activities or a manifestation of internal climate variability. Given the substantial internal variability of the climate system, detecting anthropogenic fingerprints requires separating forced responses from natural variability. The authors focus on North Atlantic SSTs due to their pronounced influence on Northern Hemisphere climate. Unlike many previous works that linked observed SST patterns primarily to changes in ocean circulation, this paper applies a formal detection–attribution framework to identify and attribute a human fingerprint in North Atlantic SSTs over recent decades.

Prior studies have associated the North Atlantic warming hole with variability and projected changes in the Atlantic Meridional Overturning Circulation (AMOC) and other aspects of ocean–atmosphere coupling (e.g., Drijfhout et al. 2012; Cheng et al. 2013; Marshall et al. 2015; Rahmstorf et al. 2015; Menary & Wood 2018; Gervais et al. 2018). Observational and modeling studies noted an exceptional twentieth-century slowdown in Atlantic overturning and inferred fingerprints of weakening AMOC. A single high-resolution model run under doubled CO2 reproduced a warming-hole-like pattern, but such idealized or single-simulation analyses cannot disentangle internal variability from forced changes or attribute observed trends to specific forcing agents. Detection studies in the atmosphere have successfully used fingerprint and signal-to-noise approaches to identify human influence on temperature structure and seasonality (e.g., Santer et al. 2013, 2018), motivating a similar formal framework for North Atlantic SSTs.

Data and models: Two observational SST datasets combining satellite and in-situ measurements were used: NOAA high-resolution SST (AVHRR-based; daily, 0.25° grid, available since 1982) and HadISST (monthly, 1° grid, since 1870). Forced model responses and internal variability were characterized using: (1) CESM Large Ensemble (CESM-LE; 40 members, 1920–2100; Historical to 2005 and RCP8.5 thereafter), (2) MPI Grand Ensemble (MPI-GE; 100 members, 1850–2100; Historical/RCP8.5), and (3) 35 CMIP5 models (one member per model; Historical/RCP8.5). Long preindustrial control integrations (CESM: 1800 yr; MPI-ESM: 2000 yr; CMIP5: last 200 yr of each control concatenated to ~7000 yr) were used to characterize internal variability. Fingerprint detection and SNR: Over the North Atlantic domain (83°W–8°E, 26°N–67°N), the externally forced fingerprint was defined as the leading EOF (EOF1) of 1982–2017 annual mean SST anomalies (relative to 1982–2017 climatology) computed from the mean of each ensemble (CESM-LE, MPI-GE, CMIP5 multi-model mean). Observed annual SST anomalies were projected onto each fingerprint to obtain a time series of projection coefficients. The signal was defined as trends of these coefficients computed over increasing trend lengths starting from 5 years (1982–1986) and incrementally extended through 2017. The noise was defined as Gaussian distributions of trends of the same lengths obtained by projecting preindustrial control SST anomalies onto the same fingerprints (all overlapping windows for CESM and MPI controls; concatenated CMIP5 controls). Fields were regridded to 1°×1° and area-weighted (sqrt(cos(lat))). Emergence (detection) occurred when the observed signal exceeded the 5% one-sided significance threshold relative to the control-run noise (Student’s t-test). Attribution of forcing agents: Four CESM ensembles analogous to CESM-LE but with one forcing agent held fixed (no time evolution) were analyzed: fixed greenhouse gases (LE-fixGHG; 20 members), fixed aerosols (LE-fixAER; 20), fixed biomass burning (LE-fixBMB; 15), and fixed land use/land cover (LE-fixLUC; 5). EOF1 fingerprints of 1982–2017 SST anomalies were computed for each to assess whether the CESM-LE fingerprint persisted when each forcing agent was fixed. Mechanism and mixed-layer heat budget: To diagnose the mechanism of the fingerprint, the mixed-layer temperature tendency equation was analyzed in CESM-LE over the warming hole region (20°W–40°W, 45°N–55°N). The budget terms included air–sea heat fluxes (radiative, sensible, latent), horizontal heat advection (zonal and meridional, decomposed into geostrophic and Ekman components), and vertical heat transfers (advection, entrainment, turbulent mixing). Due to unavailable explicit vertical terms, vertical heat transfers were computed as the residual between the SST tendency and the sum of surface heat flux and horizontal advection terms. Time series of annual-mean anomalies (relative to 1920–1960) were examined for 1920–2100. Trends in meridional heat advection over 1982–2017 were compared across CESM-LE and each fixed-forcing ensemble to attribute the mechanism.

- Observed SST trends (1982–2017) show a North Atlantic warming hole with enhanced warming at high and low latitudes and weak cooling (or reduced warming) at midlatitudes. - The forced responses in CESM-LE and MPI-GE ensemble means reproduce a similar warming-hole pattern; the CMIP5 multi-model mean also shows the feature, albeit weaker and displaced poleward due to inter-model differences. - Fingerprints (EOF1 of 1982–2017 SST anomalies) explain most variance: CESM-LE 91.7%, MPI-GE 90.7%, CMIP5 88.6%, and resemble the modeled forced trends. - Signal-to-noise analysis shows the externally forced fingerprint emerges from internal variability in observations around the year 2000. A temporary dip in SNR occurs after 1991, likely due to Mt. Pinatubo’s volcanic cooling, followed by a monotonic SNR increase. - The fingerprint is detectable in nearly all individual members of CESM-LE and MPI-GE after ~2000, indicating models capture the observed pattern changes. Noise (one standard deviation) from the MPI-GE control is ~35% larger than that from CESM-LE, delaying emergence in MPI-GE. - Attribution: When greenhouse gases are fixed (LE-fixGHG), the CESM-LE fingerprint disappears; when aerosols, biomass burning, or land-use/land-cover are fixed, the fingerprint persists. Thus, the detectable SST fingerprint is attributable to greenhouse gas emissions. - Future evolution: The warming-hole pattern persists through 2100 under RCP8.5 in CESM-LE, MPI-GE, and CMIP5 means, consistent with ongoing anthropogenic forcing. - Mechanism: In CESM-LE, the mixed-layer temperature budget shows a primary balance between increasing air–sea heat fluxes (warming; largely reduced latent cooling) and a cooling tendency from decreasing meridional heat advection since the early 1990s. The decline in northward meridional heat advection (oceanic heat transport to midlatitudes) underlies the warming hole; zonal advection and vertical transfers do not drive its formation. - Mechanism attribution: Only fixing greenhouse gases removes the forced decline in meridional heat advection (1982–2017 trends), linking the mechanism to GHG-driven changes, consistent with a weakening ocean circulation in projections.

The analysis formally detects an externally forced SST fingerprint in the North Atlantic that matches observed pattern changes and emerges from internal variability around 2000. The detection across multiple ensembles and observational datasets, alongside the persistence of the fingerprint in fixed-aerosol, fixed-biomass burning, and fixed-land-use ensembles but its disappearance when greenhouse gases are fixed, demonstrates that the observed warming-hole pattern is anthropogenic and specifically attributable to greenhouse gas emissions. Mechanistically, the pattern arises from a GHG-driven reduction in northward meridional heat advection (declining oceanic heat transport), which offsets and locally overcomes the surface warming from increased air–sea heat fluxes at midlatitudes. This mechanistic attribution is consistent with independent evidence of a weakening Atlantic overturning circulation in observations and projections. These findings directly address the research question by separating forced signals from natural variability and linking both the pattern and its underlying oceanic processes to human influence, with implications for Northern Hemisphere climate impacts over the coming decades.

This paper provides formal detection and attribution of an anthropogenic fingerprint in the recent North Atlantic warming hole. Using observations, large-ensemble simulations, and CMIP5 models, the authors show that the SST fingerprint emerges around 2000 and can be attributed specifically to greenhouse gas emissions. A diagnosed decline in northward meridional heat advection drives the pattern, and this mechanism is also attributable to greenhouse gases. The warming-hole pattern is projected to persist under continued GHG increases, implying continued influence on regional climates across the Northern Hemisphere. Future research could: refine ocean circulation and heat transport diagnostics with higher-resolution and better-constrained ocean observations; quantify contributions from distinct GHG species and aerosol–cloud interactions; assess scenario dependence beyond RCP8.5; and evaluate model biases affecting the location and amplitude of the warming hole.

- Model dependence and biases: Differences among CESM-LE, MPI-GE, and CMIP5 (e.g., location/amplitude of the warming hole) indicate sensitivity to model formulation, external forcing implementations, and ocean circulation biases. - Detection period and datasets: Detection focused on 1982–2017 to include high-resolution satellite SST products; earlier periods without satellite coverage were not analyzed for detection. Volcanic events (e.g., Mt. Pinatubo) affect SNR time evolution. - Noise characterization: The MPI-GE control exhibits ~35% larger variability than CESM-LE, affecting emergence timing; concatenated CMIP5 controls assume stationarity and comparable variability across models. - Forcing attribution: Fixed-forcing CESM experiments isolate single forcing agents but remain subject to CESM-specific physics; aerosol and land-use representations may differ from other models. - Mechanism diagnosis: The mixed-layer budget analysis was only feasible in CESM-LE due to data availability; vertical heat transfer terms were inferred as a residual, introducing potential uncertainty. Regional focus (20°W–40°W, 45°N–55°N) may not capture all basin-scale processes. - Scenario limitation: Projections and persistence assessments rely on RCP8.5; other forcing scenarios may yield different magnitudes or timings of emergence.