Volume 49, Issue 4 e2021GL097093
Research Article
Open Access

Contributions of Anthropogenic Aerosol Forcing and Multidecadal Internal Variability to Mid-20th Century Arctic Cooling—CMIP6/DAMIP Multimodel Analysis

Takuro Aizawa

Corresponding Author

Takuro Aizawa

Meteorological Research Institute, Japan Meteorological Agency, Tsukuba, Japan

National Institute of Polar Research, Tachikawa, Japan

Correspondence to:

T. Aizawa,

[email protected]

Contribution: Conceptualization, Methodology, Validation, Formal analysis, ​Investigation, Data curation, Writing - original draft, Writing - review & editing, Visualization

Search for more papers by this author
Naga Oshima

Naga Oshima

Meteorological Research Institute, Japan Meteorological Agency, Tsukuba, Japan

Contribution: Conceptualization, Writing - original draft, Writing - review & editing, Supervision, Project administration, Funding acquisition

Search for more papers by this author
Seiji Yukimoto

Seiji Yukimoto

Meteorological Research Institute, Japan Meteorological Agency, Tsukuba, Japan

Contribution: Conceptualization, Writing - original draft, Writing - review & editing, Supervision

Search for more papers by this author
First published: 28 January 2022
Citations: 5

Abstract

In the Arctic, observed decadal mean surface air temperatures (SATs) were 0.70°C–0.95°C lower around 1970 than those around 1940. The 35-multimodel ensemble mean of historical simulations in the Coupled Model Intercomparison Project Phase 6 (CMIP6) exhibited Arctic surface cooling trend in 1940–1970, which could be attributed to external forcings. Multimodel ensemble means of CMIP6 Detection and Attribution Model Intercomparison Project historical simulations exhibited Arctic surface cooling of −0.22°C (±0.24°C) in 1970 versus 1940 and showed that anthropogenic aerosol forcings contributed to a cooling of −0.65°C (±0.37°C), which was partially offset by a warming of 0.44°C (±0.22°C) due to well-mixed greenhouse gases. In addition to the anthropogenic aerosol forcings, multidecadal internal variability with a magnitude of 0.47°C was the component primarily contributing to the observed Arctic cooling. We identified a spatial pattern of pan-Arctic multidecadal cooling due to the internal variability that resembles the 1940–1970 cooling pattern.

Key Points

  • Most Coupled Model Intercomparison Project Phase 6/Detection and Attribution Model Intercomparison Project models represent the observed multidecadal surface cooling trend during the mid-20th century (1940–1970) in the Arctic

  • Anthropogenic aerosol forcing and multidecadal internal variability are the major components contributing to the 1940–1970 Arctic cooling

  • We identify a spatial pattern of pan-Arctic multidecadal cooling due to internal variability that resembles the 1940–1970 cooling pattern

Plain Language Summary

Instrumental records show Arctic surface cooling during the mid-20th century (1940–1970) followed by ongoing rapid warming since 1970. Long-term global warming has been extensively researched and has been primarily ascribed to anthropogenic greenhouse gas forcing. However, the factors contributing to the mid-20th century Arctic surface cooling remain poorly constrained. In this work, multimodel analyses using state-of-the-art climate models suggest that external factors may have contributed to the high-latitude surface cooling observed in 1940–1970. Further analysis shows that both increased anthropogenic aerosols and multidecadal internal variability provide major contributions to the 1940–1970 Arctic surface cooling. By analyzing surface cooling via unforced long-term climate simulations, we identified a spatial pattern of pan-Arctic multidecadal cooling, similar to the observed cooling pattern in the Arctic during 1940–1970.

1 Introduction

Over the past century, observed surface air temperatures (SATs) in the Arctic have exhibited substantial multidecadal variations: early-20th-century warming, mid-20th-century cooling, and subsequent ongoing enhanced warming (e.g., Aizawa et al., 2021; Bengtsson et al., 2004; Chylek et al., 2009; Gillett et al., 2008; Johannessen et al., 2004; Serreze & Francis, 2006; Shindell & Faluvegi, 2009; M. Wang et al., 2007). Anthropogenic greenhouse gases and aerosols are two major external forcings of global climate change (Bindoff et al., 2013; Boucher et al., 2013). Detection and attribution studies have estimated that anthropogenic aerosol forcings have offset some historical Arctic and global warming induced by increased greenhouse gas forcings (Gillett et al., 2021; Najafi et al., 2015). In the mid-20th century, specifically around 1970, emissions of anthropogenic aerosols and their precursors largely increased over Europe and North America (Deser et al., 2020; Hoesly et al., 2018; Smith et al., 2011; H. Wang and Wen, 2021). Climate model-based studies suggest that these increased anthropogenic aerosols are responsible for most of the observed mid-20th century Arctic cooling (e.g., England, 2021; England et al., 2021; Fyfe et al., 2013; Gagne et al., 2017; Shindell & Faluvegi, 2009). Since the 1980s, aerosol emissions have decreased in Europe, Russia, and North America due to air quality concerns, causing a large decline in sulfate concentration over the Arctic and surrounding regions (Breider et al., 2017). Air pollution reductions in Europe have also contributed to the recent amplification of Arctic warming (Accosta-Navarro et al., 2016). These studies suggest the possible relevance of anthropogenic aerosols in the Arctic cooling observed around 1970.

Internally generated climate variabilities, such as Atlantic multidecadal variability (or Atlantic multidecadal oscillation; AMO) and Pacific decadal variability (or interdecadal Pacific oscillation; IPO), have contributed to historical SAT variations over the past century (Dai et al., 2015; Johnson et al., 2020; Kosaka & Xie, 2016; Meehl et al., 2016; Svendsen et al., 2018; Tokinaga et al., 2017). Observations showed a negative phase of these variabilities around 1970 (e.g., Johnson et al., 2020; Knight et al., 2005), causing a dramatic hiatus of global warming during the mid-20th century. Kosaka and Xie (2016) suggested that tropical Pacific variability may be linked to the observed early 20th century global warming and subsequent dramatic hiatus of warming in the mid-20th century. The peak-to-peak global mean SAT changes due to AMO and IPO have been estimated as approximately 0.15°C (Stolpe et al., 2017). The Arctic region exhibits large internal variability (Yukimoto & Kodera, 2005), which can produce substantial fluctuations in Arctic SATs (Polyakov & Johnson, 2000). England et al. (2021) found substantial contributions from aerosols and internal climate variability to the mid-20th century Arctic cooling using a single model from the Coupled Model Intercomparison Project (CMIP) Phase 5 (CMIP5; Taylor et al., 2012). Recently, CMIP Phase 6 (CMIP6; Eyring et al., 2016) was conducted based on the latest state-of-the-art climate model calculations, and valuable CMIP6 multimodel data are available. Thus, new insights may be gained by investigating the impacts of external forcings and internal variability on the mid-20th century Arctic cooling using CMIP6 multimodel data.

The Detection and Attribution Model Intercomparison Project (DAMIP; Gillett et al., 2016) was conducted as a part of CMIP6. DAMIP was proposed to clarify how external anthropogenic and natural forcings have influenced historical climate changes since the Industrial Revolution. Here, we conducted CMIP6/DAMIP multimodel analyses and quantified contributions to the Arctic cooling during 1940–1970 from greenhouse gases, aerosols, natural forcings, and multidecadal internal variabilities. We also examined the dominant surface cooling pattern in the Arctic associated with multidecadal internal variability via the DAMIP multimodel epoch analysis.

2 Model Experiments and Analysis Methods

2.1 Observation Data and Model Experiments

To examine the evolution of historical Arctic SATs over the 20th century, we utilized historical simulations (total of 352 ensemble members) conducted by 35 global climate models participating in CMIP6. These modeled SAT evolutions were compared with four observational datasets: CRUTEM5 (Osborn et al., 2021), HadCRUT5 (Morice et al., 2021), GISTEMP v4 (Lenssen et al., 2019), and NOAAGlobalTemp version 5 (Huang et al., 2020). In the HadCRUT5 data set, the grids lacking observations throughout a large part of the Arctic have been infilled by statistical methods.

To quantify the contributions of each external climate forcing to the mid-20th century Arctic cooling, we analyzed SATs from the following 13 DAMIP participating models (Table S1 in Supporting Information S1): ACCESS-ESM1-5 (Ziehn et al., 2020), BCC-CSM2-MR (Wu et al., 2019), CanESM5 (Swart et al., 2019), CESM2 (Danabasoglu et al., 2020), CNRM-CM6-1 (Voldoire et al., 2019), FGOALS-g3 (Li et al., 2020), GFDL-ESM4 (Dunne et al., 2020), GISS-E2-1-G (Kelley et al., 2020), HadGEM3-GC31-LL (Williams et al., 2018), IPSL-CM6A-LR (Boucher et al., 2020), MIROC6 (Tatebe et al., 2019), MRI-ESM2-0 (Kawai et al., 2019; Oshima et al., 2020; Yukimoto et al., 2019), and NorESM2-LM (Seland et al., 2020). We used outputs from five experiments: preindustrial control (CNTL), historical (HIST), well-mixed greenhouse-gas-only historical (GHG), anthropogenic-aerosol-only historical (AER), and natural solar irradiance forcing- and volcanic forcing-only historical (NAT) simulations. HIST is driven by changes in all anthropogenic and natural forcings, including land use and stratospheric/tropospheric ozone, for 1850–2014. GHG, AER, and NAT are the DAMIP simulations (1850–2020) driven by changes only in the forcing of interest, whereas other forcings are held at the preindustrial level. CNTL was performed for ≧500 years under conditions chosen to be the representative of the period before the Industrial Revolution, in which all forcings are held at the preindustrial level.

2.2 Responses to External Forcings and Internal Variability

The forced SAT responses to external forcings can be highlighted by taking the multimodel ensemble mean (MMM). We defined the forced responses for the MMM in 1970 (ΔTRF) for each experiment (HIST, GHG, AER, and NAT) as follows:
urn:x-wiley:00948276:media:grl63671:grl63671-math-0001(1)
where urn:x-wiley:00948276:media:grl63671:grl63671-math-00031940,j,k and urn:x-wiley:00948276:media:grl63671:grl63671-math-00041970,j,k are the 10-year running mean SATs for 1940 (1935–1944) and 1970 (1965–1974), respectively; j and k indicate the model members (total of 13 models) and ensemble member of each model, respectively; and K is the total number of ensemble members for each model (Table S1). We used all available ensemble members in each model to minimize the internal variability inherent in each DAMIP simulation. The total ensemble numbers for all the models were 188, 107, 93, and 161 for HIST, GHG, AER, and NAT, respectively. We also calculated the linear sum (GHG + AER + NAT) for 13 individual DAMIP models and corresponding MMM. We defined ΔTFR, Arctic for HIST, GHG + AER + NAT, GHG, AER, and NAT as the Arctic SAT response in 1970 relative to 1940 by averaging (Equation 1) over the Arctic region (60°–90°N).
The amplitude of internal variability of SATs on a multidecadal time scale for each DAMIP model (vj) and the corresponding MMM (VI) are defined by applying the unforced CNTL for all DAMIP models (total of 9,953 years) as follows:
urn:x-wiley:00948276:media:grl63671:grl63671-math-0002(2)
where urn:x-wiley:00948276:media:grl63671:grl63671-math-0005j,t and urn:x-wiley:00948276:media:grl63671:grl63671-math-0006j,t−30 are 10-year running mean SATs at t and t−30 for the model j, respectively, and N indicates the total integrated years of CNTL for each model. Each vj value (ranging from 0.28°C to 0.81°C) is calculated for the Arctic region (60°–90°N). VI is interpreted as the multimodel ensemble average of the root mean square for arbitrary 30-year SAT differences in each model.

To detect significant Arctic cooling events resulting from multidecadal internal variabilities, we conducted an epoch (composite) analysis using CNTL. Epoch analysis can detect characteristic structures of dominant events in the presence of noise. When a 30-year difference in the 10-year running mean Arctic SAT had a minimum value between t = −30 and 30 years, which accompanied the condition that ΔTArctic,j(=(urn:x-wiley:00948276:media:grl63671:grl63671-math-0007j,turn:x-wiley:00948276:media:grl63671:grl63671-math-0008j,t−30)Arctic) was lower than a threshold  –2vj for model j, we sampled the Arctic SAT time-series and spatial SAT distributions as a cooling epoch from t = −30 to 30 years (61-year window). We composited all cooling events over six decades due to the multidecadal internal variability of all models, with equal weights for each model. This approach enables the detection of spatial patterns in SAT anomalies caused by multidecadal internal variabilities. We also examined the spatial pattern of sea level pressure (SLP) anomalies associated with this SAT variation of the composited cooling event. We then compared the SAT patterns derived from multidecadal internal variabilities to those of the observed mid-20th century Arctic cooling. All DAMIP data used in this study were interpolated to the 5° × 5° HadCRUT5 grid points.

3 Results and Discussion

3.1 Arctic SAT Evolution in CMIP6 and DAMIP Experiments

Figure 1 shows changes in the Arctic SAT anomaly relative to the 19411970 mean, obtained by the 35-model CMIP6 historical simulations and observations. Intermodel differences in past Arctic temperature changes were large among the 35 CMIP6 models. However, 20 of the 35 CMIP6 models showed an Arctic surface cooling stronger than −0.1°C around 1970 relative to around 1940, which produced weak negative SAT anomalies in the MMM at high latitudes around 1970 (Figure 1). The DAMIP models tended to better reproduce the past Arctic temperature changes than the other CMIP6 models. However, we found no potential metrics (e.g., Arctic amplification factor) common to the models that captured the observed SAT changes. Because the MMM is the mean of 352 ensemble members, the internal variability should be substantially removed in the MMM SAT anomaly. The Arctic cooling of 1940–1970 was evaluated as −0.14°C (±0.35°C [intermodel standard deviation]) in the MMM and −0.95°C–−0.70°C in the observations, indicating that the MMM cooling was one-sixth of the observed Arctic cooling. Note that the observations can be regarded as one member. The negative anomalies around 1970 are statistically significant, suggesting that external factors may have contributed to the large cold anomalies shown in the observations (Figure 1). In the observations, the SAT amplitudes were greater at higher latitudes (Figure 1), suggesting that the internal variability was more pronounced in the Arctic than in the mid-latitudes. However, unlike the observations, robust mid-20th cooling was not limited over the Arctic in the MMM. Indeed, the cooling region expanded to lower latitudes.

Details are in the caption following the image

Time–latitude cross-sections of zonal and 10-year running mean surface air temperature (SAT) anomalies (°C) in the twentieth century relative to the base period of 1941–1970 for 35 Coupled Model IntercomparisonProject Phase 6 models and four observations. Shadings indicate the ensemble mean of each model (35 models) and the corresponding multimodel ensemble mean (MMM). Numbers in parentheses denote the ensemble numbers of each model. The MMM is calculated with equal weights given to each model. Values at the upper right of each panel indicate 10-year running mean SAT differences around 1970 relative to around 1940 (ΔT = urn:x-wiley:00948276:media:grl63671:grl63671-math-00991970urn:x-wiley:00948276:media:grl63671:grl63671-math-00981940), averaged over the Arctic region (60°–90° N). Yellow plus marks at the lower left of each panel denote models participating in Detection and Attribution Model Intercomparison Project (13 models). Hatched areas in the MMM indicate results that have not reached the level of statistical significance, which occurs when 60% or more of the models have the same sign.

Figure 2a shows time evolutions of the Arctic-mean SAT anomalies relative to 18501900 for the MMM of historical simulations by CMIP6, CMIP5, and DAMIP models and for observations. Although the Arctic cooling response for the CMIP6 MMM was estimated at −0.14°C from 1940 to 1970, the CMIP5 MMM showed an Arctic warming of 0.07°C for the same period. The CMIP6 MMM SAT anomalies during the mid-20th century were lower than the CMIP5 MMM SAT anomalies, in agreement with the work of Flynn and Mauritsen (2020), who showed that the CMIP6 global mean SATs during the mid-20th century were lower than the CMIP5 global mean SATs. The discrepancy in the Arctic SATs between the CMIP6 and CMIP5 models may be partially due to a stronger model response to aerosol forcings in the CMIP6 models versus the CMIP5 models or differences in the forcings themselves (Figure S1 in Supporting Information S1; Dittus et al., 2020; Gillett et al., 2021). The MMM SAT evolution derived from the 13 DAMIP models shows almost perfect agreement with that from the 35 CMIP6 models. Therefore, multimodel analysis based on the 13 DAMIP models was considered sufficient for estimating the contributions from internal and external factors to the mid-20th century Arctic cooling.

Details are in the caption following the image

Arctic-mean (60°–90°N) surface air temperature anomaly (°C) evolution relative to 1850–1900. (a) The 10-year running mean of historical simulations for Coupled Model IntercomparisonProject Phase 6 (CMIP6) (green line), Detection and Attribution Model Intercomparison Project (red line), and CMIP5 models (black line) and HadCRUT5 observations (blue line). (b) Annual mean for historical simulations with all anthropogenic and natural forcings (historical, gray line), well-mixed greenhouse gas-only simulations (red line), aerosol-only simulations (green line), and natural forcing simulations (blue line) and observations (HadCRUT5, black line). The modeled values are multimodel means obtained for all ensemble members, with equal weights given to each model. Numbers in parentheses denote the number of models used in the multimodel means. Color shadings indicate 5%–95% confidence intervals based on Student's t-distributions. (b) All model-simulated data were interpolated to the HadCRUT5 grid points and were masked at grid points lacking observations.

The temporal evolutions of the Arctic SAT anomalies for the DAMIP MMM for four experiments—HIST, GHG, AER, and NAT—are shown in Figure 2b. For 19401970, the rate of decrease in the Arctic SAT anomaly due to AER forcings is greater than the rate of increase due to GHG forcings, implying that the aerosol-induced cooling overweighed the GHG-induced warming. The strong cooling for AER during 19401970 is consistent with the increased anthropogenic sulfur emissions during the same period, particularly over the Europe and North America regions (Figure S1 in Supporting Information S1). The responses to anthropogenic aerosol forcings varied substantially depending on the model, ranging from −1.9°C to 0.07°C in 1970. The MMM SAT anomaly for NAT showed fluctuations within ±0.5°C (1850–2014), indicating Arctic warming in the early 20th century (19101940; Aizawa et al., 2021) and volcano-induced cooling around 1883, 1902, 1912, 1963, 1982, and 1991. The MMM responses of the Arctic mean SATs to each external forcing were 2.5-fold larger than those of the global mean SATs (see Figure 1b in Gillett et al., 2021). The intermodel variance in the Arctic mean SATs was much larger than that in the global mean SATs (Figure 1b in Gillett et al., 2021).

3.2 Spatial Distributions of SAT Responses

We analyzed the spatial distributions of forced SAT responses around 1970 relative to around 1940 for each forcing and compared these distributions with HadCRUT5 observations (Figure 3). The observations showed substantial Arctic cooling, especially over the Kara Sea. The observed SAT pattern is indicative of negative AMO and IPO over the North Atlantic and Pacific Oceans, respectively (Figure 3a), as shown by previous studies (Johnson et al., 2020; Knight et al., 2005). Due to these internal variabilities, the MMM and observations may largely disagree.

Details are in the caption following the image

Spatial distributions of the 10-year running mean surface air temperature anomaly (°C) in 1970 (1965–1974) relative to 1940 (1935–1944) for (a) HadCRUT5 observations and multimodel means of (b) historical, (c) linear sum (GHG + AER + NAT), (d) greenhouse gas-only simulations (GHG), (e) aerosol-only simulations (AER), and (f) natural forcing-only simulations (NAT). The multimodel means were obtained for all ensemble members, with equal weights given to each Detection and Attribution Model Intercomparison Project model. All model-simulated data were interpolated to the HadCRUT5 grid points (5° × 5°).

HIST indicated broad, weak cooling over the entire Northern Hemisphere, with a slightly larger signal in the regions showing large aerosol optical depths (Figure 3b and Figure S2 in Supporting Information S1). The SAT cooling distributions in GHG + AER + NAT qualitatively agree with the HIST distributions, although the negative signals tended to be larger in regions with large aerosol optical depths and over the Barents and Greenland Seas (Figure 3c and Figure S2 in Supporting Information S1). The 1940–1970 cooling in HIST is primarily influenced by anthropogenic aerosol forcings (Figure 3e), although the aerosol-induced cooling was partly offset by the GHG-induced warming (Figure 3d). Natural forcing (volcanic aerosols) also contributed to the cooling pattern in HIST (Figure 3f).

The cooling SAT response to the aerosol forcing was larger over the high latitudes than the mid and low latitudes (Figure 3e), indicating an Arctic amplification of surface cooling. For AER, the Arctic amplification was largest over the Barents Sea. GHG also indicated an Arctic warming amplification (Figure 3d). The Arctic amplification factors, defined as the ratio of the Arctic surface temperature change to the global temperature change, were 2.1, 2.6, and 2.3 for GHG, AER, and NAT, respectively. These Arctic amplification factors are similar to those reported by Stjern et al. (2019), who studied the Arctic amplification response to various climate drivers such as carbon dioxide and sulfate aerosols.

3.3 Quantification of Each External Forcing and Internal Variability

We quantified the contributions from greenhouse gases, aerosols, natural forcings, and internal variabilities to the Arctic cooling during 1940–1970. Figure 4 compares the multimodel mean responses (ΔTFR, Arctic) to each external forcing with VI. The ΔTFR, Arctic values for HIST and for the linear sum (GHG + AER + NAT) were estimated to be −0.22°C (±0.24°C) and −0.35°C (±0.35°C), respectively. The difference between HIST and the linear sum was most likely due to nonlinear effects (Deng et al., 2020) or responses to land use forcing (Andrews et al., 2017). The intermodel spread ranged from −0.62°C to 0.34°C for HIST. ΔTFR, Arctic were estimated to be 0.44°C (±0.22°C), −0.65°C (±0.37°C), and −0.14°C (±0.11°C) for GHG, AER, and NAT, respectively. These values are reasonably consistent with the findings of England et al. (2021) based on a single model analysis, who reported a warming trend of 0.60°C/30 years and a cooling trend of −0.81°C/30 years for 1935–1984 due to greenhouse gases and anthropogenic aerosols, respectively. However, the impact of the natural forcings was not evaluated in the work by England et al. (2021). In the current study, forcing by anthropogenic aerosols was considered as the dominant external contributor to the observed mid-20th century Arctic cooling, although the intermodel spread of AER was large. The cooling response to natural forcing was small but not negligible and could be interpreted as remnant effects of the 1963 Agung eruption.

Details are in the caption following the image

Multimodel mean forced responses of Arctic surface air temperatures (SATs) (ΔTRF, Arctic) to each forcing agent in 1970 (1965–1974) relative to 1940 (1935–1944) averaged over the Arctic region (60°–90°N). The multimodel means were obtained for all ensemble members, with equal weights given to each model. Boxes with horizontal bars indicate the ΔTRF, Arctic value for each forcing agent with ranges of ±1 standard deviation for historical (HIST) (gray), linear sum (purple), greenhouse gas-only simulations (light red), aerosol-only simulations (light green), and natural forcing-only simulations (light blue). The whiskers in each box indicate the maxima and minima. The yellow box (Internal) indicates the VI range estimated from the CNTL of 13 models (Equation 2). The red triangle denotes the observed (HadCRUT5) Arctic SAT change from 1940 (1935–1944) to 1970 (1965–1974). The narrow bar at the right side of HIST (HIST + Internal) indicates the sum of the forced SAT response to all forcings (ΔTRF, Arctic for HIST) and VI.

The multimodel mean responses in HIST showed only one-quarter of the observed mid-20th century Arctic cooling, indicating that the external forcings alone could not reproduce the observed Arctic cooling. VI was estimated to be 0.47°C, which is comparable in magnitude to AER ΔTFR, Arctic value of −0.65°C (Figure 4). The range of the multidecadal internal variability (−0.47°C–0.47°C) is similar to the range of −0.42°C to 0.33°C reported by England et al. (2021). When the multidecadal internal variability was combined with the cooling response to all forcings (i.e., HIST + Internal in Figure 4), its value reached −0.69°C (−0.93°C–−0.45°C). The observed cooling of −0.81°C is in the range obtained for the HIST response combined with the multidecadal internal variability. This finding implies that either the observed Arctic SAT change was a relatively low-probability event or the MMM for HIST underestimates the Arctic SAT decrease for 1940–1970. This result highlights the major contributions of multidecadal internal variability and anthropogenic aerosol forcings to the observed Arctic cooling during 1940–1970.

3.4 Pan-Arctic Multidecadal Cooling Due To Internal Variability

Using epoch analysis, we identified significant, low-probability Arctic surface cooling events due to multidecadal internal variability. Thirty-eight events were sampled from 9,953 years of the CNTL for the 13 DAMIP models (Table S1 in Supporting Information S1) as substantial Arctic surface cooling events. The composited SAT anomaly time-series of these 38 events traced the observed surface cooling and subsequent warming well (Figure 5a). A 30-year Arctic surface cooling event comparable to the observed cooling event could occur once per 100–300 years (Figure S3 in Supporting Information S1). The ranges of 30-year Arctic SAT trends for weak and strong cooling fluctuations were estimated to be −0.6°C and −1.2°C with reemergence periods of approximately 70 and 2000 years, respectively (Figure S3 in Supporting Information S1). We note that the ongoing Arctic warming since 1980 (Figure 2) is comparable to the very rare (2000-year reemergence period) strong fluctuation of 1.2°C. As anthropogenic sulfur emissions and sulfate aerosols will decrease in any future scenarios of shared socioeconomic pathways (Gidden et al., 2019), Arctic warming will continue over the near-term future even under strong cooling fluctuations generated by internal variability.

Details are in the caption following the image

(a) Time series of the Arctic mean surface air temperature (SAT) anomaly (°C) relative to climate values from t = −30 to 30 years. The climate values were calculated using all integrated years for each model. The thick black line indicates the multimodel mean composited for all cooling epochs. The multimodel means were calculated using all Detection and Attribution Model Intercomparison Project models, with equal weights given to each model. The light blue shading indicates the range of maxima and minima in each cooling epoch. The green line indicates the Arctic SAT deviations in the HadCRUT5 observations from 1938 to 1998 relative to 1968 (defined as t = 0 year), when the observed SAT exhibits a minimum. The observed SAT anomaly in 1968 was set to −0.68°C to match the minimum value of the modeled SAT anomaly. Spatial distributions of the multimodel mean 30-year differences in decadal mean (b) SAT (°C) and (c) sea level pressure (hPa) between t = 0 and −30 years.

The composited spatial distribution of 30-year SAT differences for this multidecadal internal variability showed pronounced cooling over the pan-Arctic, specifically over the Barents-Kara Seas, Greenland Sea, and southern coasts of Greenland (Figure 5b). The cooling regions were confined in the Arctic, suggesting a multidecadal internal variability inherent in the Arctic, where the center of action locates in the Barents-Kara Seas. Over the Arctic, the SAT anomaly pattern of this pan-Arctic multidecadal cooling (hereafter, referred to as PAMC) is similar to the Arctic cooling pattern observed in the mid-20th century (Figures 3a and 5b). The large amplitude of the observed SAT change concentrated in the Arctic (Figures 1 and 2) is a manifestation of the PAMC.

Figure 5c shows the patterns of 30-year SLP differences associated with the PAMC. The high-pressure anomalies appear to correspond to negative SAT anomalies, due to the higher air density in the colder lowest atmosphere. The spatial patterns in SAT and SLP for the PAMC from each model showed large intermodel variations, indicating that no common pattern is found among the DAMIP models (Figures S4 and S5 in Supporting Information S1). The spatial patterns associated with each PAMC event vary, in agreement with the findings of Beitsch et al. (2014). Note that the PAMC is not tied to any specific modes of variability, such as the Arctic Oscillation (Thompson & Wallace, 1998) or IPO, and likely results from a combination of various internal variabilities. The Arctic cooling pattern observed in the mid-20th century (Figure 3a) may have arisen from a superposition of internal variabilities, for example, from a superposition of parts of the AO and IPO, on the cooling induced by anthropogenic aerosol increases (Figure 3e and Figure S1 in Supporting Information S1).

4 Conclusions

Most CMIP6 models reproduced a portion of the multidecadal surface cooling trend observed during 1940–1970 in the Arctic, which can be attributed to external forcings. The SAT spatial patterns in 1970 relative to 1940 for GHG and AER indicate Arctic amplification, with corresponding factors of 2.1 and 2.6, respectively. Multimodel analysis of the 13 DAMIP models indicates an Arctic surface cooling of −0.22°C (±0.24°C) for HIST, which arises from −0.65°C (±0.37°C) for AER, 0.44°C (±0.22°C) for GHG, −0.14°C (±0.11°C) for NAT, and other residual effects. The range of multidecadal internal variability obtained from the Arctic-mean SATs is estimated to be ±0.47°C. The multidecadal internal variability combined with the range of Arctic SAT responses to all forcings (HIST, −0.93°C–−0.45°C) marginally covers the mid-20th century Arctic cooling of −0.81°C observed by HadCRUT5, implying that the observed Arctic SAT change is a relatively low-probability manifestation of multidecadal internal variability during 1940–1970. Thus, anthropogenic aerosol forcing and multidecadal internal variability are the two major factors contributing to the mid-20th century Arctic cooling around 1970.

Epoch analysis based on CNTL for the 13 DAMIP models reveals the spatial pattern of the PAMC due to internal variabilities, which resembles the observed Arctic cooling pattern in the mid-20th century. The PAMC is attributed to a combination of various internal variabilities rather than a specific mode, such as the AO or IPO. The ongoing warming signal will override the fluctuations due to internal variabilities in the Arctic, leading to acceleration of the near-term future Arctic warming.

Acknowledgments

This study was supported by the Arctic Challenge for Sustainability II (ArCS II) project (Program Grant No. JPMXD1420318865) and the Integrated Research Program for Advancing Climate Models (TOUGOU, Program Grant No. JPMXD0717935561) from the Ministry of Education, Culture, Sports, Science and Technology (MEXT), Japan, and in part by the Japan Society for the Promotion of Science (JSPS) KAKENHI (grant numbers: JP18H03363, JP18H05292, and JP21H03582), the Environment Research and Technology Development Fund (JPMEERF20202003 and JPMEERF20205001) of the Environmental Restoration and Conservation Agency of Japan, and a grant for the Global Environmental Research Coordination System from the Ministry of the Environment, Japan (MLIT1753). We also thank anonymous referees for careful reading the manuscript and for helpful suggestions.

    Data Availability Statement

    The CMIP6 and DAMIP data used in this study are available on the Earth System Grid Federation (ESGF) website (https://esgf-node.llnl.gov/search/cmip6/) with digital object identifiers (dois). If readers want to access the data, please register with ESGF and search by selecting the appropriate activity (CMIP6 or DAMIP), source ID (model name), and experiment ID (piControl, historical, hist-aer, hist-GHG, or hist-nat) on the website. The observational datasets, CRUTEM5, HadCRUT5, GISTEMP, and NOAAGlobalTemp are available at https://crudata.uea.ac.uk/cru/data/temperature/CRUTEM.5.0.1.0.anomalies.nc, https://crudata.uea.ac.uk/cru/data/temperature/HadCRUT.5.0.1.0.analysis.anomalies.ensemble_mean.nc, https://data.giss.nasa.gov/pub/gistemp/gistemp1200_GHCNv4_ERSSTv5.nc.gz, and ftp://ftp.ncdc.noaa.gov/pub/data/noaaglobaltemp, respectively.