Future Slower Reduction of Anthropogenic Aerosols Enhances Extratropical Ocean Surface Warming Trends

Global surface temperature short‐term trends fluctuate between cooling and fast‐warming under the combined action of external forcing and internal variability, significantly influencing the detectability of near‐term climate change. A key driver of these variations is anthropogenic aerosols (AAs), which have undergone a non‐monotonic evolution with rapid reduction in recent decades. However, their reduction is projected to decelerate under a high carbon emission scenario, yet the impact on surface temperature trends remains unknown. Here, using initial‐perturbation large ensembles, we find that future slowdown in AA reduction over Europe and North America expedites the subpolar North Atlantic surface warming by intensifying the Atlantic meridional overturning circulation. Further, it accelerates the South Indian Ocean and Southern Ocean surface warming through positive low‐cloud feedback and oceanic dynamical adjustment, triggered by the poleward migration of westerlies under interhemispheric energy constraint. These AA‐driven warmings exacerbate greenhouse warming, significantly enhancing the detectability of local decadal warming trends.

Notably, the AA emissions have undergone a non-monotonic temporal evolution (Quaas et al., 2022;Wang & Wen, 2022), with a rapid reduction in recent decades due to the implementation of global clean air policies.It pushes surface temperature trends shifting toward fast-warming, resulting in a transient acceleration of the surface warming rate (Jenkins et al., 2022).Under a high carbon emission scenario, however, the reduction of AAs would reach saturation over the extratropical Northern Hemisphere (NH), especially Europe and North America (Figure S1 in Supporting Information S1).The impact of such slower AA reduction over the extratropical NH on global surface temperature trends and their detectability remains unexplored.In particular, as AA reduction approaches saturation, AAs become a diminishing component of the overall anthropogenic forcing in high GHG emission scenarios, which are intuitively considered irrelevant for projecting future warming (Stevens, 2013).
Here based on initial-perturbation large ensembles (LEs) and a novel pattern recognition method (Wills et al., 2020), the emergence timescale (ET) is assessed as an indicator for the detectability of external-driven surface temperature trends across varying temporal spans, which signifies the predominance of externally forced signals over internally generated noise.We find that the projected slower AA reduction over the extratropical NH significantly accelerates surface warming in the subpolar North Atlantic, South Indian Ocean and Southern Ocean through coupled oceanic-atmospheric adjustments.It enhances the detectability of local warming trends within about three decades, manifested as a notable shortening of ET, which is comparable to the impacts caused by GHGs.The findings underscore the necessity of maintaining rigorous AA reduction in future endeavors aimed at achieving effective climate change mitigation, particularly in mitigating severe ocean warming.

Data Sets
Multi-Model LE Archive (Deser, Lehner, et al., 2020) provides initial-condition LEs, with all members in each fully coupled ensemble model generated by the same radiative forcing.The member spread is solely attributed to randomly phased internal variability due to perturbations of the initial conditions.In this way, the same evolving response across members can be regarded as the externally forced signals.To ensure that sufficient data is available to single out external signals on regional scales, we select models with members greater than or equal to 30 (Table S1 in Supporting Information S1), namely CESM1-CAM5, MPI-ESM-LR, CanESM2, GFDL-ESM2M, CSIRO-Mk3.6.Owing to the large deviations of CanESM2 arising from its high warming sensitivity compared to observations (Figure S2 in Supporting Information S1), this model is excluded for further analysis.
To identify the model that best approximates the observation, we utilize several observational data sets for comparison.For surface temperature, we use four data sets: NOAAGlobalTemp5 (Huang et al., 2020) from US National Centers for Environmental Information (NCEI), with a 5°grid resolution beginning in 1880, combined ERSST v5 and GHCN-M v4; HadCRUT5 (Morice et al., 2021) from the UK Meteorological Office Hadley Center, with a 5°grid resolution beginning in 1850, combined HadSST4 and CRUTEM5; GISTEMP v4 Geophysical Research Letters 10.1029/2023GL107772 (GISTEMP Team, 2022;Hansen et al., 2010;Lenssen et al., 2019) from National Aeronautics and Space Administration (NASA), with a 2°grid resolution beginning in 1880, combined ERSST v5 and GHCN v4; BEST (Rohde & Hausfather, 2020) from the Berkeley Earth Group, with a 1°grid resolution beginning in 1850, combined HadSST4 and Berkeley Earth land temperature product.For AOD at 550 nm, we use the Modern-Era Retrospective Analysis for Research and Applications version 2 (MERRA-2) reanalysis data (Buchard et al., 2017) beginning in 1980 provided by NASA.

Characterizing Surface Temperature Trend Variations
Global surface temperature exhibits non-monotonic increases, characterized by variations in trends spanning decades, including accelerations of warming and cooling, as well as transitions between colder and warmer periods, a crucial aspect for near-term prediction.To delineate surface temperature trend variations at varying timescales, we employ running trend curves calculated over overlapping sliding-windows of varying lengths.For example, to depict surface temperature trend variations at 5-year timescale, we compute linear trends across all overlapping 5-year sliding-windows of annual-mean surface temperature (i.e., 1950-1954, 1951-1955, …, 2076-2080), for the total (T TOT ), externally forced (T EX ) and internally generated (T IV ) components.Each running trend curve signifies variations in trend at the corresponding timescale, reflecting changes in the rate of warming or cooling.Such that an increase in positive values indicates an acceleration of surface warming.The amplitude of trend variations is quantified by the variance of the running trend curve (σ 2 (T)).Given that the covariance of the external forcing and internal variability is nearly zero, the total variance of surface temperature trend variations can be expressed as σ 2 (T TOT (w)) = σ 2 (T EX (w)) + σ 2 (T IV (w)), where T(w) is the running trend curve at w-year timescale (window length w).Therefore, the combined variance contributions of the externally forced and internally generated components account for 100% of the total variance.

Determining ET
The variance fraction of surface temperature trend variations at w-year certain timescale accounted for by the externally forced component is expressed as σ 2 (T EX (w))/σ 2 (T TOT (w)).With lengthening timescale, this fraction grows continuously (5-year vs. 40-year running trends in Figure 1a).We calculate variance fraction at varying timescales ranging from 5 to 55 years (5 ≤ w ≤ 55).Then ET is identified as the threshold timescale at which the externally forced signals account for 50% of the total variance of surface temperature trend variations, starting to surpass that accounted for by internal variability, that is, p X = σ 2 (T EX (w))/σ 2 (T TOT (w)) ≥ 50% when w ≥ ET (pentagram in Figure 1a).Noting that we compute the running trend curve at each timescale for the entire period of 1950-2080, and then divide it into two epochs of 1950-2020 and 2021-2080 for identification of ET, which circumvents the issue of insufficient time sampling relative to the selected timescale.

Physical Meaning
ET marks the timescale beyond which surface temperature trend variations contain a substantial portion of detectable externally forced signals.It implies that a record with temporal coverage exceeding the ET threshold is likely required to detect external-driven surface temperature trends.A shorter ET at a specific location signifies an increased proportion of external forcing on variations in surface temperature trends, suggesting a higher level of external signal dominance and an increased detectability of external-driven surface temperature trends (Figure 1a).Given heightened potential predictability exhibited by external-forced signals (Boer, 2009;Kushnir et al., 2019), a shorter ET also indicates enhanced potential for near-term prediction to a certain extent.

External-Forced Signal Extraction
The estimation of ET at regional scale requires a sophisticated extraction of externally forced signals, because regional-scale temperature changes are characterized by greater chaotic internal noise than global mean.Here we employed a novel pattern recognition method called 'signal-to-noise-maximizing pattern filtering' (Wills et al., 2020), which builds upon ensemble mean and further accounts for the spatial signal-to-noise ratio, offering a superior alternative to simple ensemble-mean for accurately extracting externally forced signals.Here 'signal' refers to the patterns that members agree on the same temporal evolution in a LE model, regarded as the externally forced response; 'noise' refers to the member spread caused by unforced internal variability.This method singles out purer external-forced signal at regional scale with preserved spatiotemporal covariance, and enables estimation of individual external agents by diagnosing the leading forced signals separately.This is accomplished by the signal-to-noise-maximizing EOFs (more details given in Supporting Information S1).We synthesize the leading forced signals as the total external forcing, with the residual attributed to internal variability.

Bootstrap Test
We use a bootstrap method to test the significance level in LE models.For future change, we sample with replacement of all members (40 members for CESM1-LE) 10,000 times for historical and future period, constructing 10,000 realizations separately.If the difference value is greater than the sum of the standard deviations of the 10,000 bootstrap samples in historical and future period, the change is considered to be significant above 95% confidence level.

ET-Indicated Detectability and Aerosol-Forced Signals
We leverage initial-condition LEs from Multi-Model LE Archive (Table S1 in Supporting Information S1) to extract externally forced signals from chaotic internal variability.Through evaluating performance of LE models in reproducing the observed global-mean surface temperature evolution during 1950-2020, CESM1-LE ensemble average stands out with a high correlation of 0.89 and the smallest root mean square error compared to the observations (Figure S2 in Supporting Information S1), thus we primarily analyze CESM1-LE to present the results, supplemented by other models.The analysis period covers from 1950 to 2080, encompassing historical simulation until 2005 and subsequent future projection under representative concentration pathway 8.5 (RCP8.5)scenario.
Surface temperature trends could be more detectable when externally forced signals predominate over internally generated noise.We establish ET as an indicator for the detectability of external-driven surface temperature trends (Figure 1a; see details in Section 2.2).A projected shortening of ET at a specific location signifies an increased proportion of external forcing on variations in surface temperature trends, suggesting a higher level of external signal dominance and an enhanced detectability of external-driven surface temperature trends.Conversely, a prolongation of ET suggests a decreased detectability.In particular, emphasis should be placed on a qualitative evaluation of the projected changes in ET, as their magnitude may depend on the climate system's sensitivity to external forcings in individual models.
To determine ET and its change driven by projected AAs, it is crucial to precisely extract externally forced signals and distinguish AA-forced responses.A novel pattern recognition method 'signal-to-noise-maximizing pattern filtering' (see details in Section 2.3) was applied to the annual-mean surface temperature under the current and future climates , which differ from Wills et al. (2020) that analyzed the seasonal-mean surface temperature during 1920-2019.The first signal-to-noise pattern (S/NP-1) depicts the GHG-induced global warming (Figure 1b).The second pattern (S/NP-2) signifies surface temperature trend transition due to aerosol source shifting from early industrialized countries (United States, Europe, etc.) to newly industrialized countries (China, India, etc.) (Figure 1c, Figure S3a in Supporting Information S1).The third pattern (S/NP-3) signifies synchronous multidecadal surface temperature variations triggered by global AA co-emissions (Figure 1d, Figure S3b in Supporting Information S1), with the spatial pattern capturing an interhemispheric temperature asymmetry and the temporal evolution closely following the trajectory of global-mean Aerosol Optical Depth (AOD) (pink line in Figure S1 in Supporting Information S1).Here the S/NP-2 and S/NP-3 are recognized as two typical modes of AA-forced responses (Kang et al., 2021;Qin et al., 2020;Shi et al., 2022;Wang & Wen, 2022).They can be detected in the CESM1 industrial-aerosol (AERindus) single forcing experiment (here and Shi et al., 2022), as well as in other LE models.Thus, the S/NP-2 and S/NP-3 are combined as the overall response to aerosols (primarily AAs).Based on empirical test, truncating the leading six S/NPs optimally estimates the forced response which is consistent across subsets of the large ensemble.Therefore, we synthesize the leading six S/NPs as the externally forced signals in the following analysis.

Broad Shortening of ET in a Warming Climate
In current climate , regions characterized by pronounced short ET ranging from approximately 5 to 20 years are mainly located in the densely populated low-latitude countries and adjacent coastal areas (Figure 2a).These regional features are primarily reproduced in the AERindus single forcing experiment with a spatial correlation of 0.59 at low latitudes 35˚S-35˚N (Figure S4 in Supporting Information S1).Areas with short ET are largely collocated at high AA loadings which are reasonably captured by CESM1 (Figure S5 in Supporting Information S1), suggesting that decadal surface temperature trends therein are greatly modulated by the localized aerosol emissions directly.Massive aerosols govern local surface temperature trajectories by altering solar radiation, and influence adjacent coastal temperature trends through advection.These significant AA-forced signals yield prominent detectability of external-driven surface temperature changes within about two decades.
The external signal dominance in European areas is less apparent (Figure 2a), albeit with historical high AA emissions, because surface temperature changes therein are subject to considerable midlatitude atmospheric internal variability (Ding & Wu, 2020;Mahlstein et al., 2011) that prolongs ET by masking the externally forced responses.Further, the weak external signals in East/South Asia and the Pacific Ocean are likely to be attributed to powerful internal variabilities associated with the El Niño/Southern Oscillation and the Interdecadal Pacific Oscillation.The Southern Ocean exhibits the lowest level of external signal dominance, likely owing to strong initial deep convection (Latif et al., 2013) that damps the sea surface temperature (SST) response.
In a warming climate (2021-2080), the spatial heterogeneity of global ET tends to be reduced compared to current climate (Figure 2b).Future changes in ET are primarily featured by a broad shortening over oceans, indicating an increased detectability of external-driven surface temperature trends within a few decades (Figure 2c).In particular, ET over the tropical and northeastern Pacific, the Chilean coast as well as the subpolar North Atlantic is significantly shortened from around 40-50 to 30 years on average.The shortening is most remarkable over the South Indian Ocean and Southern Ocean, with future ET reaching around 24 years.It implies that detecting external-driven surface temperature trends in those areas is likely to require a temporal span of at least two or three decades in the future.Moderate decrease in detectability is indicated in early industrialized countries, especially the United States and Europe, as evidenced by the slight prolongation of ET.Below we will show that in addition to the expected impacts induced by GHGs, the projected changes in AA emissions play a significant role in modulating surface warming rate and thus causing future ET changes.

AAs Driving Future ET Change by Modulating Surface Warming Rate
We delineate the relative roles of GHGs and AAs in driving future ET changes using S/NP-1 and S/NP-2&3 representative of GHG-and AA-forced signals, respectively.The results show that the projected changes in global ET (Figure 2c) can be regarded as the combination of ET changes driven by both GHGs and AAs (Figures 3a and 3b), which are primarily governed by changes in surface warming rate (Figures 3c and 3d).This determination is supported by other LE models (not shown).Here the surface warming rate change refers to the percentage change of surface temperature trend induced by GHGs or AAs during 2041-2080 relative to 1981-2020, with positive values indicating an acceleration in warming and negative values indicating a deceleration.It suggests that the temperature trend would be more detectable when externally forced warming acceleration is greater.Noting that the focus is on the changes in ET related to the surface warming rate change, rather than future warming trend per se.
Future GHG-forced warming acceleration primarily emerges in the broad ocean surfaces, leading to shortening of ET (Figures 3a and 3c).It is particularly evident in the Chilean coast, South Indian Ocean and Southern Ocean, with a consistency across LE models (Figure S6 in Supporting Information S1).Specifically, strong upwelling and ventilation in those regions historically mute surface warming by bringing up cold thermocline water and promoting deep ocean heat uptake.However, when deep ocean reaches quasi-equilibrium, surface layer tends to warm substantially (Held et al., 2010;Long et al., 2014), which facilitates an increase in surface externally forced signals and thus enhances detectability.A shortening of ET is also evident in the eastern tropical Pacific and northeastern Pacific in CESM1, which is likely related to an El Niño-like warming response (Gan et al., 2017;Vecchi & Soden, 2007).
However, despite the recognized impacts of GHGs, the projected AA changes largely contribute to the extratropical ET changes by modulating surface warming rate (Figures 3b and 3d).Compared to the significant reduction of AAs in the past four decades over the extratropical NH continents, particularly in the United States and Europe, the AA reduction is projected to be largely restrained in future (Figure S7a vs. S7b in Supporting Information S1).Such future relative of AAs in early industrialized regions (Figure S7c in Supporting Information S1) tends to retard local warming through solar radiation scattering, thereby leading to a prolonged ET with a decrease in surface external-forced signals over the United States and Europe (Figure 3b).Notably, in addition to directly influencing terrestrial surface temperature trends, future slower reduction of AAs over the extratropical NH promotes a pronounced acceleration of surface warming in the subpolar North Atlantic, South Indian Ocean and Southern Ocean, consequently leading to a significantly shortened ET therein.Such distinctive ocean surface warming acceleration is reproduced in the AERindus single forcing experiment (Figure S8 in Supporting Information S1), confirming the pivotal role of AAs.The abovementioned ET changes can also be observed in other LE models (Figure S6 in Supporting Information S1).In the following, we elucidate the underlying mechanisms involving the slow oceanic dynamical adjustment and the thermal-dynamical air-sea coupling feedback in the context of interhemispheric asymmetry of AA forcing.

Mechanisms for Future AA-Forced Ocean Surface Warming Acceleration
In the subpolar North Atlantic, a future warming hotspot is exclusively driven by AAs (Figure 3d) associated with S/NP-2&3.Under a high carbon emission scenario, the relative recovery of AAs over North America by the late 21st century (Figure 4a), induces initial surface cooling through shortwave reflection (Figure 4b) and a reduction in precipitation (Figure 4c).This leads to an increase of surface salinity, particularly in the North Atlantic deep water formation region (Figure 4d).Consequently, the stratification is reduced due to the increased upper-ocean density and hence vertical mixing is enhanced, as evidenced by a deepening of boundary layer depth (Figure 4e).This process promotes a stronger Atlantic meridional overturning circulation (AMOC) (Figure 4f), which exhibits robustness across LE models (Figures S9 and S10a, S10b in Supporting Information S1).Previous studies also report the AA-driven strengthening of AMOC via similar density adjustment in CMIP6 historical simulation (Menary et al., 2020;Robson et al., 2022).The intensified AMOC facilitates the acceleration of surface layer warming by transporting warm water poleward, which in part compensates the warming deceleration caused by the GHG-driven weakening of AMOC, injecting increased detectable signals in the subpolar North Atlantic compared to current climate.Consequently, it boosts the detectability of externally driven surface temperature trends over a span of three decades amidst strong internal variability, as evidenced by the notably shortened ET (Figures 2c and 3b; Table S2 in Supporting Information S1).
In the South Indian Ocean and Southern Ocean, future AA changes considerably enhance near-term surface temperature detectability with shortened ET by accelerating surface warming through thermal-dynamical air-sea coupling.It is primarily achieved by S/NP-3 under hemispheric asymmetry of AA emissions, even with local minimal aerosol forcing in the Southern Hemisphere (SH).The high AA emissions over the extratropical NH (Figure S11a in Supporting Information S1) lead to an interhemispheric energy imbalance with strong cooling at the NH midlatitudes (Figure S11b in Supporting Information S1).Such cooling drives an anomalous crossequatorial Hadley circulation that attains energy from the SH to compensate for interhemispheric imbalance (Kang et al., 2008;Wang, Xie, Tokinaga, et al., 2016), thus effectively triggering a southward shift of ITCZ (Hwang et al., 2013;Kang, 2020;Seo et al., 2014).The latter is evident by an increase of convective precipitation to south of the equator and a decrease to north (Figure S11c in Supporting Information S1).Accordingly, the SH westerlies tend to migrate poleward, manifested as a north-negative and south-positive dipole pattern of sea level pressure anomalies over the South Indian Ocean and South Pacific (Figure S11d in Supporting Information S1), due to the barotropic atmospheric circulation change under the constraint of energy rebalance (Ceppi et al., 2013;Lee & Kim, 2003).The above modulations could be converse under a context of decreased emissions of AAs, particularly in recent decades marked by rapid reduction of AAs.However, as the reduction of AAs over the extratropical NH is projected to slow down, the relative leveling-up trend of AAs (Figure S7c in Supporting Information S1) leads to a pronounced cooling tendency at the NH midlatitudes compared to current climate (Figure S8 in Supporting Information S1), favoring the poleward migration of the SH westerlies (Figure 5a).This Geophysical Research Letters 10.1029/2023GL107772 AA-driven wind response is also verified by other LE models (Figures S10c, S10d, and S12a in Supporting Information S1).
The AA-driven poleward migration of the SH westerlies triggers a thermal-dynamical positive feedback loop, amplifying surface warming in the South Indian Ocean and Southern Ocean.The poleward shifted westerlies are conducive to reduced oceanic evaporation and thus anomalous latent heat flux from atmosphere to heating ocean surface around 30˚-60˚S (Figure 5b).Subsequently, warmer surface layer destabilizes the atmosphere, facilitating diminished low clouds (Figure 5c) and hence increased downward net solar radiation (Figure 5d), which forms positive feedback to bolster surface warming.This is particularly evident for the South Indian Ocean warming.In the meantime, the poleward shifted westerlies trigger oceanic adjustments.The weakened midlatitude westerlies reduces oceanic vertical mixing around 30˚-50˚S, as evidenced by a shoaling of boundary layer depth (Figure 5e).It attenuates the inhibition of surface warming by upwelling cold water, promoting SST warming.Moreover, the poleward shifted westerlies induce an anomalous anticlockwise overturning circulation around 30˚-50˚S (Figure 5f), conveying warm water poleward and reinforcing surface warming.A pronounced clockwise circulation is also observed south of 50˚S due to the movement of westerlies (Figure 5f), conveying cold water from high latitudes.This magnifies the temperature contrast between warm air and cold sea surface, and thus ocean gains heat rapidly by sensible heat release from atmosphere in 50˚-60˚S (not shown).The aforementioned coupled oceanic-atmospheric adjustments are also observed to be at work in another LE simulations (Figure S12 in Supporting Information S1).The resulting impact is nearly comparable to that of (Table S2 in Supporting Information S1), collectively enhancing the detectability of human-induced warming trends within about three decades.

Conclusion and Discussion
Our findings highlight the significant yet overlooked role of the projected AA changes in amplifying extratropical ocean surface warming trends beyond the widely recognized impacts induced by GHGs.In particular, the intensification of AMOC driven by the projected relative recovery of AAs over North America leads to surface warming acceleration in the subpolar North Atlantic, which is opposed to the GHGs-induced impact (Drijfhout et al., 2012;Orihuela-Pinto et al., 2022).Further, as the reduction of AAs over the extratropical NH slows down, it induces a notable cooling tendency at the NH midlatitudes, thereby promoting a southward migration of the SH westerlies under interhemispheric energy constraint.It subsequently excites coupled oceanic-atmospheric adjustments including positive low-cloud feedback and oceanic dynamical adjustment to accelerate surface warming in the South Indian Ocean and Southern Ocean.This AA-driven warming acceleration exhibits an effect nearly comparable to that of GHGs, significantly enhancing the predominance of external signals over strong internal noise, as evidenced by the significant shortening of ET, indicating the increased detectability of local human-induced warming trends.The impact of NH AAs on Southern Ocean warming suggests that prediction of future surface warming rate in the Southern Ocean may be partially dependent on the rate of NH AA reduction.The AA-induced climatic responses possess higher potential predictability than chaotic internal variability (Boer, 2009;Chikamoto et al., 2013).The shorter ET induced by AA-driven surface warming acceleration also indicates enhanced potential for near-term surface temperature predictability in extratropical oceans.Nevertheless, it should be noted that the estimation of aerosol forcings and its corresponding climatic responses in Earth system models still suffer from significant uncertainties (He et al., 2023;Stier et al., 2013;Watson-Parris & Smith, 2022).Improving the simulation of aerosol-induced global impacts remains a critical challenge, which is pivotal to enhancing the reliability of near-term climate prediction.
Under high carbon emission scenarios, the AA-forced ocean surface warming could escalate a series of climatic impacts including coral reef bleaching, sea level rise, and Antarctic ice shelves melting, and potentially fuel catastrophic events such as large-scale typhoons and hurricanes (Cai et al., 2023;Krishnan et al., 2020).This projected AA-induced regional warming acceleration, in conjunction with the GHG-induced global warming, calls for the urgency of mitigating extreme warming risks and underscores the necessity of implementing climate mitigation strategies that target both types of emissions.In particular, sustained rigorous reduction of AA emissions could be beneficial in alleviating the severe warming of the Southern Ocean.Nevertheless, recent research suggests that by following the carbon neutrality pathway, the impacts of AA reduction on climate and extreme weather events could far outweigh those of GHGs (Wang et al., 2023;Yang et al., 2023).Despite these complexities, several countries are actively developing synergistic pathways toward carbon neutrality and clean air (Lei et al., 2024).Achieving these goals by the mid-21st century could substantially mitigate the extreme ocean warming risks stemming from both GHG and AA.Moving forward, deeper integration of these synergistic governance strategies will be pivotal in guiding national development toward a more sustainable and environmentally friendly future.

subpolar
North Atlantic surface warming by intensifying Atlantic meridional overturning circulation • Slower AA reduction accelerates South Indian and Southern Ocean warming via positive low-cloud feedback and oceanic dynamical adjustment Supporting Information: Supporting Information may be found in the online version of this article.

Figure 1 .
Figure 1.ET identification and externally forced surface temperature responses.(a) Schematic illustration of ET identification.Variance fractions of the externally forced (red) and internally generated (green) components to the total surface temperature trend variations at timescales ranging from 5 to 55 years.ET is determined as the threshold timescale at which externally forced signals account for 50% of the total variance in surface temperature trend variations, starting to surpass internal variability (marked by pentagram).A shorter ET signifies a higher degree of external signal dominance and an enhanced detectability of near-term trends.Here trend variations are depicted using running trends of annual-mean surface temperature, showcasing 5-year and 40-year running windows for global-mean surface temperature.(b)-(d) The leading three S/NPs and corresponding temporal evolutions of global surface temperature during 1950-2080 in CESM1-LE.The temporal evolution of each S/NPs is standardized, with black curve denoting the ensemble average and gray curves denoting individual members.S/NPs are sorted by S k , the ratio of ensembleaverage signal to total variance for each S/NP.

Figure 2 .
Figure 2. A broad shortening of ET in a warming climate.(a), (b) Global ET distribution during 1950-2020 (a) and 2021-2080 (b) under the historical and RCP8.5 radiative forcing in CESM1-LE.(c) Future ET change in a warming climate ((b) minus (a)).A shorter ET at a specific location indicates a higher level of external signal dominance in surface temperature trends.Hatching denotes difference significant above the 95% confidence level based on a bootstrap test.

Figure 3 .
Figure 3. Projected ET change determined by GHG-and AA-induced surface warming rate change.(a), (b) Future ET change driven by GHG (a) and AA (b) in CESM1-LE.Hatching denotes difference significant above the 95% confidence level.(c), (d) Future surface warming rate change induced by GHG (c) and AA (d).Surface warming rate change is measured as percentage change of surface temperature trend between future warming climate (2041-2080) and current climate (1981-2020), with positive values indicating warming acceleration and vice versa.The GHG-induced percentage change of surface temperature trend (%) = (Trend S/NP-1, future Trend S/NP-1,current )/abs (Trend External, current ), likewise for AA (SNP-2&3).Here the 40-year period of current climate ending in 2020 and future warming climate ending in 2080 correspond to the period of rapid and slow reduction of AAs over the extratropical NH, respectively (Figure S1 in Supporting Information S1).

Figure 4 .
Figure 4. AA-driven AMOC intensification accelerates surface warming in the subpolar North Atlantic.(a) Trend change in AOD reconstructed from S/NP-2 (Figure 1c) between 40-year period of future warming climate ending in 2080 and current climate ending in 2020 in CESM1-LE.(b)-(f) Same as (a), but for net solar radiation at surface (positive downward) (b), precipitation (c), sea surface salinity (d), boundary layer depth (e), and AMOC (f) with red/blue shading denoting clockwise/anticlockwise circulation.Hatching denotes difference significant above the 95% confidence level.Noting that similar results are obtained from S/NP-3, albeit with more pronounced responses related to S/NP-2.

Figure 5 .
Figure 5. AA-driven coupled oceanic-atmospheric adjustments conducive to surface warming acceleration in the South Indian Ocean and Southern Ocean.(a) Trend change in sea level pressure (hPa) and surface wind (m s 1 ) reconstructed from S/NP-3 (Figure 1d) between 40-year period of future warming climate ending in 2080 and current climate ending in 2020 in CESM1-LE.Surface wind change significant above the 95% confidence level is displayed.(b)-(f) Same as (a), but for the latent heat flux (positive downward) (b), fraction of vertically integrated low cloud (c), net solar radiation at surface (d), boundary layer depth (e), and Southern Ocean meridional overturning circulation (f) with red/blue shading denoting clockwise/anticlockwise circulation.Hatching denotes difference significant above the 95% confidence level.