Influence of High-Latitude Atmospheric Circulation Changes on Summertime

ARTICLES PUBLISHED ONLINE: 13 MARCH 2017 | DOI: 10.1038/NCLIMATE3241

Inﬂuence of high-latitude atmospheric circulation changes on summertime Arctic sea ice

Qinghua Ding1,2,3*, Axel Schweiger3, Michelle L’Heureux4, David S. Battisti5,6, Stephen Po-Chedley5, Nathaniel C. Johnson7, Eduardo Blanchard-Wrigglesworth5, Kirstin Harnos4, Qin Zhang4, Ryan Eastman5 and Eric J. Steig5,6

The Arctic has seen rapid sea-ice decline in the past three decades, whilst warming at about twice the global average rate. Yet the relationship between Arctic warming and sea-ice loss is not well understood. Here, we present evidence that trends in summertime atmospheric circulation may have contributed as much as 60% to the September sea-ice extent decline since 1979. A tendency towards a stronger anticyclonic circulation over Greenland and the Arctic Ocean with a barotropic structure in the troposphere increased the downwelling longwave radiation above the ice by warming and moistening the lower troposphere. Model experiments, with reanalysis data constraining atmospheric circulation, replicate the observed thermodynamic response and indicate that the near-surface changes are dominated by circulation changes rather than feedbacks from the changing sea-ice cover. Internal variability dominates the Arctic summer circulation trend and may be responsible for about 30–50% of the overall decline in September sea ice since 1979.

t is well recognized that recent Arctic sea-ice decline has Observed linkage between circulation and sea ice both natural and anthropogenic drivers1–3, but their relative To examine the physical linkages, we focus on the connection Iimportance is poorly known4–7. This uncertainty arises from between September sea-ice extent and the preceding summer the fact that the contribution of atmospheric internal variability (June–July–August, JJA) atmospheric circulation. We choose this in the Arctic climate system is inadequately understood, and it preceding 3-month window because sea-ice extent anomalies have is unclear how well models reproduce these processes8. Reliably a ∼3-month decorrelation timescale26, and previous studies have distinguishing the natural and anthropogenic contributions of sea- shown a strong link between summer circulation and sea-ice ice loss requires a comprehensive understanding of the mechanisms variability23,24,27,28. We focus on physical mechanisms, analysing that control the variability of sea ice. Although some progress temperature, humidity, and downward longwave radiation (DLR), has been made in this area8–17, the answer is far from clear, and all of which are aﬀected by atmospheric circulation and, in turn, conflicting hypotheses exist18,19. aﬀect sea-ice concentration. A key player is the radiation balance, Earlier work indicated that the decline of sea ice before the which dominates the surface energy balance controlling the growth 1990s was in part owing to an upward trend in the North Atlantic and melt of Arctic sea ice29,30. Oscillation (NAO) index20,21. However, since the early 1990s, the The region with the greatest negative trend in September apparent link between the NAO and sea ice has largely disappeared, sea-ice concentration since 1979 is highlighted in Fig. 1a, and with Arctic sea ice declining further despite the reversal in the NAO includes the Beaufort, Chukchi, and East Siberian Seas, featuring trend22. On the other hand, several studies have argued that the an average decline >10%/decade. Concomitant with the trend in recent NAO trend may still play a key role in the sea-ice retreat23,24. sea ice are trends in the atmospheric circulation: 200 hPa and Alternatively, the Arctic Dipole has been suggested as a driver of 700 hPa geopotential heights have been rising over northeastern sea-ice change in some regions of the Arctic25. Overall, the recent Canada and Greenland, and the surface winds have become more trends in Arctic sea ice cannot be linked to simple indices of anticyclonic (Figs 1b and 2d). To illuminate processes that link climate variability22. the circulation changes to sea-ice changes, we first construct an In this paper, we examine the contribution of the atmospheric index of September sea-ice coverage over the region with the fastest circulation to Arctic sea-ice variability by utilizing an atmospheric sea-ice decline from 1979 to 2014 (Fig. 1c), as well as indices of general circulation model (ECHAM5) coupled with a simple ocean– JJA low-level temperature (surface to 750 hPa), DLR at surface, sea-ice model in which the atmospheric circulation field is nudged and integrated atmospheric water vapour (surface to 750 hPa) to observations. Specifically, we explore how the high-latitude derived from ERA-Interim reanalysis (hereafter ERA-I); indices are summertime atmospheric circulation impacts the September Arctic averaged over the Arctic, poleward of 70◦ N, and shown in Fig. 1c. sea-ice extent, and estimate to what extent changes in atmospheric The decreasing trend in sea-ice concentration is accompanied by circulation explain the observed sea-ice loss of the past few decades. increasing trends in JJA Arctic temperature, DLR and water vapour.

1Department of Geography, University of California, Santa Barbara, California 93106, USA. 2Earth Research Institute, University of California, Santa Barbara, California 93106, USA. 3Polar Science Center, Applied Physics Laboratory, University of Washington, Seattle, Washington 98195, USA. 4NOAA Climate Prediction Center, College Park, Maryland 20740, USA. 5Department of Atmospheric Sciences, University of Washington, Seattle, Washington 98195, USA. 6Department of Earth and Space Sciences, University of Washington, Seattle, Washington 98195, USA. 7Cooperative Institute for Climate Science, Princeton University, Princeton, New Jersey 08540, USA. *e-mail: [email protected]

NATURE CLIMATE CHANGE | ADVANCE ONLINE PUBLICATION | www.nature.com/natureclimatechange 1

© 2017 Macmillan Publishers Limited, part of Springer Nature. All rights reserved. ARTICLES NATURE CLIMATE CHANGE DOI: 10.1038/NCLIMATE3241 a September sea ice bcJJA Z200 Time series 80 2.0 60 8 Sea ice Temp. 60 1.5 6 40 Spe. Hum. LW 0.4 25 40 1.0 4 20 0.2 −10 20 0.5 2 20 −20 0 0.0 0 0.0 0 15 −10 −20 −0.5 −2 −30 −20 −0.2 10 −40 −1.0 −4 −40 −0.4 −60 −1.5 −6 GL-Z200 −80 −2.0 −60 Hum. −8 2 m s−1 per decade 1980 1985 1990 1995 2000 2005 2010 LW Z200 Year Temp. Sea ice d Sea ice related Z200 efgTemp. related Z200 Hum. related Z200 LW related Z200

1 m s−1 20 kg m−1 s−1

0.4 0.5 0.6 0.7

Figure 1 | Relationship between the September Arctic sea ice and summer large-scale circulation. a, Linear trend (% per decade) of September sea-ice concentration from the NSIDC passive microwave monthly sea-ice record (1979–2014). b, Linear trend (m per decade) of JJA Z200 and surface wind − (m s 1 per decade) in ERA-Interim reanalysis. c, Domain-averaged time series for September sea-ice anomaly (%) averaged over the region circled by the ◦ − blue contour in a, lower-level (1,000 hPa to 750 hPa) JJA temperature ( C) and JJA specific humidity anomalies (g kg 1) in the Arctic (averaged over the ◦ − ◦ region north of 70 N), JJA downwelling longwave radiation (LW) anomaly at surface (W m 2) in the Arctic north of 70 N, and JJA Z200 anomaly (m) ◦ ◦ ◦ ◦ over Greenland (66 –80 N, 310 –330 E, indicated by the dot in b, and referred to as GL-Z200). d–g, Correlation of each time series in c with JJA Z200 for the period 1979–2014. In d, regression of JJA surface wind with September sea-ice index is superposed. The sign is reversed in d for simplicity of comparison with other plots. In f, regression between the specific humidity index and vertically integrated water vapour flux is plotted. All linear trends are removed in calculating the correlations in d–g. Black stippling in all plots indicates statistically significant correlation or trend at the 5% level; in d and f, vectors are plotted when regressions are statistically significant at the 5% level.

Regressing the domain-averaged sea-ice anomaly time series against the following investigation to identify the underlying physical the JJA geopotential height at 200 hPa (Z200) in ERA-I, we find mechanisms. In addition, in this statistical analysis, we find a that decreasing sea ice is accompanied by increasing Z200, with very weak connection between large-scale atmospheric circulation maximum amplitude over Greenland (Supplementary Fig. 1). and surface downwelling shortwave flux in the Arctic during These same interrelationships between the trends in the indices summertime. However, we cannot rule out the possibility that of JJA Z200, temperature, water vapour, DLR and September sea- shortwave changes in the past 30 years, which are not directly linked ice concentration are also apparent using detrended indices: an to the circulation change, may also contribute to the abrupt Arctic Arctic summer with higher than normal Z200 over Greenland, sea-ice retreat in September29. greater DLR at the surface, and increased low-level water vapour The mechanism connecting anomalies in the large-scale upper- and temperature over the Arctic is followed by negative sea- level circulation and lower-level thermodynamics is apparent ice anomalies in September (Fig. 1c); for example, detrended in a zonally averaged vertical cross-section of the atmosphere indices of temperature and specific humidity are correlated at (Fig.2). The recent JJA circulation trend in the Arctic was r = 0.89; detrended indices of sea-ice concentration and DLR are largest in the upper troposphere (200 hPa), while atmospheric correlated at r = −0.75. Importantly, the detrended data shows warming mainly occurred in the lower troposphere below 700 hPa that the summertime anomalies in the indices of near-surface (Fig. 2a). Associated with this trend, significant downward temperature, DLR, and water vapour are simultaneously associated motion below 700 hPa and poleward of ∼75◦ N is observed with a common pattern of atmospheric circulation variability (see (Fig. 2a and Supplementary Fig. 2). Year-to-year variability in Fig. 1d–g), and that this circulation pattern is very similar to the summertime vertical motion in the lower troposphere is also highly circulation pattern that is associated with the sea-ice interannual correlated with similar circulation changes in the upper troposphere variability in September (see Fig. 1d–g). These correlation maps also (Supplementary Fig. 2). Because circulation and vertical velocity compare well with the pan-Arctic Z200 and surface wind trends in are strongly coupled, adiabatic descent forced by the anticyclonic JJA (Fig. 1b), which feature a strong high over northeastern Canada, circulation aloft must be a contributor to the observed low-level Greenland, and the Arctic Ocean. warming. Increased DLR is a direct result of the warming of the The statistical links between atmospheric circulation and lower atmosphere, as well as an increase in water vapour in the lower temperature, water vapour, and DLR anomalies over the Arctic (warm) atmosphere. DLR is also strongly aﬀected by cloud cover. in JJA and subsequent sea-ice anomalies in September, which To examine the relationship between cloudiness and the circulation, are robust at both interannual and decadal timescales, prompt we analyse the relationship between the detrended Z200 heights and

2 NATURE CLIMATE CHANGE | ADVANCE ONLINE PUBLICATION | www.nature.com/natureclimatechange

abcObs. Exp1 Exp2 100 100 100 12 8 10 10 0.8 200 16 200 12 200 8 300 300 300 0.6 8 8 400 14 400 400 12 0.4 500 500 500 600 600 600 0.2 10 700 700 700 −0.2 9 800 800 6 800 6 −0.4 900 6 900 900 1,000 1,000 1,000 −0.6 60 70 80 90 60 70 80 90 60 70 80 90 Latitude (° N) Latitude (° N) Latitude (° N)

d Obs. e Exp1f Exp2 6 6 6 6 0.8 1.2 6 6 6 0.6 0.9 12 0.4 0.6

0.2 0.3

−0.2 −0.3

−0.4 −0.6 6 12 6 12 6 6 −0.6 −0.9 6 6

g Exp2 (sea ice) h Exp2 (time series) 4 0.5 16 Temp. 3 0.4 12 LW 0.3 2 Spe. Hum. 8 0.2 −10 1 0.1 4 −20 0 0.0 0

−30 −1 −0.1 −4 −0.2 −2 −8 −0.3 −3 −0.4 −12 −4 −0.5 −16

1980 1985 1990 1995 2000 2005 2010 LW Temp. Hum. Year

Figure 2 | Simulated impact of atmospheric circulation on Arctic thermodynamic trends. a, Meridional cross-section of the linear trend of zonal ◦ mean JJA temperature (shading, C per decade), geopotential height (black contour, m per decade) and vertical velocity (red/blue contour, interval: − − − 6×10 5 Pa s 1 decade 1) in ERA-I (1979–2014). b,c, Same as a but for Exp-1 and Exp-2 simulations, respectively. d, Linear trend of lower tropospheric ◦ (1,000 hPa to 750 hPa) JJA temperature (shading, C per decade) and geopotential height at 700 hPa (red contour, m/decade) in ERA-I (1979–2014). e,f, Same as d but for Exp-1 and Exp-2 simulations, respectively. g, Linear trend of September sea-ice concentration (% per decade) simulated in Exp-2. ◦ − h, Domain-averaged time series for lower tropospheric (1,000 hPa to 750 hPa) JJA temperature ( C) and specific humidity anomalies (g kg 1), and JJA − ◦ downwelling longwave radiation (LW) anomaly (W m 2) at surface in the Arctic (north of 70 N) simulated in Exp-2. In d to g stippling indicates statistically significant trends at the 5% level. cloud anomalies across the Arctic. An anomalous high pressure over atmospheric circulation itself32,33. To better understand the direction the Arctic Ocean favours a decrease in cloudiness in the upper and of causality, we conduct two model experiments to examine the middle levels of the atmosphere, possibly associated with decreased influence of the observed high-latitude circulation trends on sea storm activity. However, low clouds increase slightly over Greenland ice and other key variables. The first experiment (Exp-1) is a 36-yr and the central Arctic (Supplementary Fig. 3). Because low clouds historical run with the ECHAM5 atmospheric model, in which dominate the Arctic, this circulation-driven shift towards lower the global atmospheric circulation (vorticity and divergence from clouds is qualitatively consistent with the observed increase in DLR. the surface to the top of the atmosphere) is weakly nudged to the corresponding daily reanalysis data (see Methods). For the lower Atmospheric nudging experiments boundary, climatological sea surface temperature (SST), and sea We have provided a plausible mechanism for how circulation ice are imposed everywhere. In the second experiment (Exp-2), all changes can impact Arctic sea ice, but it is difficult to determine settings are the same as those in Exp-1, except that the atmosphere causality with observational evidence alone because of the feedbacks model is coupled to a slab ocean–sea-ice model in the Arctic (north between sea ice and the atmosphere. For example, the removal of 60◦ N), permitting a feedback from the ice–ocean system to the of sea ice can increase surface air temperature, clouds, low-level prescribed atmosphere. Anthropogenic forcings are held constant water vapour and hence DLR31, and may therefore affect the in both experiments. The difference between the two experiments

NATURE CLIMATE CHANGE | ADVANCE ONLINE PUBLICATION | www.nature.com/natureclimatechange 3

a Full forcing b Modiﬁed forcing c September sea ice extent 3 Full forcing Modiﬁed forcing 2

1 −10 0 −20

−30 −1

−2 Obs −3 Unit: 106 km2

1980 1985 1990 1995 2000 2005 2010 Year d Full forcing e Modiﬁed forcing f September sea ice volume 8 Full forcing Modiﬁed forcing 6

4 0.1 2 −0.1 0 −0.3 −2 −0.5 −4 PIOMAS −6 Unit: 1,000 km3 −8 1980 1985 1990 1995 2000 2005 2010 Year

Figure 3 | Simulated impact of atmospheric circulation on summertime Arctic sea-ice trends. a,b,d,e, Linear trend of September sea-ice concentration (a,b, % per decade) and thickness (d,e, m per decade) in Exp-5 (denoted as ‘full forcing’) and Exp-6 (denoted as ‘modified forcing’). c, Anomalous total area of September sea-ice extent (area of ocean with ice concentration of at least 15%) in both simulations and NSIDC observations. f, Anomalous total volume of September sea ice (area of ocean with ice concentration of at least 15%) in both simulations and the Pan-Arctic Ice Ocean Modeling and Assimilation System (PIOMAS; ref. 46). isolates the importance of tropospheric forcing (dynamics) from Is it possible that the observed circulation itself, especially in the amplifying effects of sea-ice–ocean feedbacks. If the mechanism the lower levels, is a response to the sea-ice change? If so, the responsible for the strong correlations between the thermodynamic observed boundary layer warming would be expected as the result variables (temperature, humidity and DLR) and Z200 circulation is of the thermal wind relationship when the circulation is nudged primarily due to a dynamical control of the circulation on radiation to the observed winds as done in both Exp-1 and Exp-2. To test (via the temperature and moisture fields) rather than feedbacks, this idea, in Exp-3, the model’s global winds are nudged to the then the results from Exp-2 should be similar to those in Exp-1. observations only above 700 hPa and a more active interaction Exp-1 and Exp-2 reproduce the spatial distribution in the between sea ice and atmosphere is allowed in the boundary layer observed trends in geopotential height, temperature and vertical of the Arctic. In this new experiment, we still simulate a similar velocity, both in height and latitude (see Fig. 2b,c with Fig. 2a). While but slightly weaker low-level temperature and sea-ice response as the simulated spatial patterns are similar to the reanalysis data, the in Exp-2 (Supplementary Fig. 4). In addition, model experiments height increases are slightly smaller and temperature increases larger that examine the impact of changes in sea-ice conditions in isolation than observed. The simulated lower tropospheric temperature in show no significant impact on the circulation during the summer the experiments is significantly correlated with the ERA-I reanalysis months32,34. Because the impact of sea-ice changes on the circulation data on both interannual and interdecadal timescales. The region may be model dependent, we examine this impact in an additional with the greatest lower tropospheric warming over the Ocean in JJA experiment (Exp-4; see Methods), in which we specify the observed is over the Beaufort, Chukchi, and East Siberian seas (Fig. 2d). The time evolution (1979–2014) of Arctic SST and sea ice (north of results from Exp-2 are very similar to those in Exp-1 and the ERA-I 60◦ N), and impose climatological SST elsewhere. In JJA, the model reanalysis, indicating that the lower-level atmospheric warming is response to the SST warming trend and sea-ice retreat in the Arctic largely due to the changes in atmospheric circulation, while the is very small (Supplementary Fig. 5), in contrast to the responses in influence of sea ice on the atmosphere is small (Fig. 2e,f). This is the other three seasons. This is because, in summer, the temperature not entirely surprising: in JJA, sea-ice extent is at a minimum, the contrast between the open ocean and sea ice is relatively small ice surface is covered with melt water ponds, and the temperature (compared with fall and winter), and the impact of sea-ice loss contrast between the sea ice and the atmosphere is small. Therefore, on the atmosphere is muted. Even in fall and winter, when the changes in sea-ice conditions in JJA cannot have a significant model exhibits the largest response to sea-ice melt, the simulated impact on surface temperature, atmospheric humidity, and DLR. circulation trend is only about 3 m/decade in the upper troposphere, The sea-ice trends simulated in Exp-2 are similar in magnitude but which is only 10 to 20% of the observed change in JJA. Thus, we ‘upstream’ of those observed (see Fig. 2g with Fig. 1c), probably due conclude that the JJA vertical temperature trend in the Arctic in the to a lack of advection of ice by the winds which are not modelled in past 36 yr is mainly controlled by the changes in the mid- and upper the slab ocean, thermodynamic-only sea-ice model used in Exp-1 troposphere (top-down effect), rather than a bottom-up response and Exp-2 (see Methods). due to the sea-ice melt.

4 NATURE CLIMATE CHANGE | ADVANCE ONLINE PUBLICATION | www.nature.com/natureclimatechange

abcObs. (1979−2014) CMIP5 (1979−2014) LENS (1979−2014)

10 JJA Z200 trend JJA 5

−5

2 m s−1 per decade

deObs. (1979−2014) CMIP5 (1979−2014) f LENS (1979−2014)

6 JJA Z700 trend Z700 JJA 3

−3

1 m s−1 per decade

Figure 4 | Observed and estimated radiatively forced trends in upper and lower tropospheric geopotential height and winds. a–f, Linear trends of JJA geopotential height (m per decade) and zonal and meridional winds at 200 hPa (a–c) and 700 hPa (d–f) for the period 1979–2014 from a and d ERA-I (a,d) (repeated from Fig. 1b) the 26-model ensemble mean from the CMIP5 project (b,e) and the 30-member ensemble mean from the LENS project (c,f).

Simulated impact of atmospheric circulation on sea ice anomalies downstream from the Beaufort and Chukchi seas to the Having demonstrated the impact of the circulation on near-surface East Siberian Sea. temperature, water vapour and DLR, we estimate its contribution to A comparison of sea-ice concentration and sea-ice volume the overall downward trend in sea ice. As noted above, the pattern patterns and time series between Exps-5 and 6 supports our of sea-ice decline simulated in Exp-2 is shifted upstream of the hypothesis that trends in the atmospheric circulation have observed sea-ice decline (see Figs 1a and 2g). Without sea-ice drift, contributed substantially to the 1979–2014 reduction of the sea- the strongest reduction in sea ice in Exp-2 occurs over the Canadian ice extent in September. In the absence of GL-Z200 circulation Basin and north of Greenland, where the circulation exerts the anomalies (Exp-6), the decline in sea-ice concentration and largest forcing on lower-level temperature and surface DLR over thickness are only 41% and 40% of the decline in the control run the past 36 yr (Fig. 2h). To extend our results to include the effects (Exp-5), respectively, suggesting that the summertime circulation of ice advection and thus simulate a sea-ice response that can be contributes to as much as 60% of the sea-ice loss since 1979. The more directly compared to observations, we force a global coupled magnitude of this estimate is perhaps not surprising given prior ocean/sea-ice (POP2–CICE4) model with observed atmospheric estimates linking roughly 60% of September sea-ice extent variance fields (ERA-I reanalysis; see Methods) between 1979–2014. Two to sea-level pressure variability in the prior month23. experiments are performed: Exp-5, which uses the complete set of ERA-I forcings, and Exp-6, which removes from the forcing the Estimation of internal and anthropogenic contributions effect of the circulation variability that is linearly related to Z200 Having demonstrated the effect of changes in the atmospheric over Greenland (GL-Z200), identified above (see Methods). circulation on sea-ice trends, we now examine whether the The spatial patterns and time series of sea ice in the control circulation trends are due to internal variability or anthropogenic run (Exp-5) exhibit a close correspondence to the observed sea- forcing. First, to determine the role of anthropogenic warming on ice reduction (see Figs 1a and 3a) further demonstrating the skill the high-latitude circulation, we examine the ensemble average of the POP2–CICE4 model to produce realistic sea-ice variability trends in JJA geopotential height and winds at 200 hPa and 700 hPa when driven with observed forcings. The pattern of the JJA trend over the Arctic region from the historical forcing simulations of in sea-ice melt in Exp-5 is similar to the extent change of sea ice the Coupled Model Intercomparison Project Phase 5 (CMIP5; in Exp-2 (Fig. 2g). Also, an index of the domain-averaged JJA sea- ref. 35) and the Large Ensemble Project (LENS; ref. 36). The average ice melt in Exp-5 (north of 70◦ N) shows a very high correlation over the 26-member (Supplementary Table 1) ensemble in CMIP5 with Z200 over Greenland and the Arctic on both interannual and and the 30-member ensemble in LENS display positive trends in longer timescales (Supplementary Fig. 6). This indicates a similar near-surface temperature and in Z200 and Z700 over our study impact of the atmospheric circulation on thermodynamic anomalies period (1979–2014) (Fig.4 and Supplementary Fig. 7). Unlike in the ECHAM5 and POP2–CICE4 models, and supports the idea observations, which show barotropic, anticyclonic circulation that ice advection transports thermodynamically generated sea-ice trends over Greenland from the surface to the upper troposphere,

NATURE CLIMATE CHANGE | ADVANCE ONLINE PUBLICATION | www.nature.com/natureclimatechange 5

© 2017 Macmillan Publishers Limited, part of Springer Nature. All rights reserved. ARTICLES NATURE CLIMATE CHANGE DOI: 10.1038/NCLIMATE3241 the circulation trends in CMIP5 and LENS ensembles are weak satellite era, particularly in summer41. However, there is evidence and unorganized in the Arctic, and particularly in the lower that the relationship between sea ice and the vertical distribution troposphere (see Fig.4 or Fig. 1b). An amorphous warming with of cloud fraction in reanalysis does not agree well with that in little circulation change is the typical thermodynamic response observations31. Therefore, to the extent that changes in cloudiness in the mid- and high latitudes to increasing CO2 (see also contribute to the DLR trends associated with the trends in Supplementary Fig. 7). The strong gradients in the observed circulation, the connection between large-scale circulation and trend pattern (Fig. 2a) support the idea that internal atmospheric cloud variability is subject to considerable uncertainty and needs to variability is an important contributor to the recent multidecadal be explored in future studies. trends in the high-latitude circulation. Second, to estimate the anthropogenic contribution to the Conclusion observed warming and sea-ice reduction in the Arctic, two Our conclusions are based on experiments involving several models additional experiments are conducted. Exp-7 and 8 are equivalent rather than one integrated model. POP2–CICE4 simulations are to Exp-2 but we remove the effects of global warming on the used to allow for a realistic simulation of sea-ice variability, while high-latitude winds, which are used to constrain the model in ECHAM5 simulations allow for realistic simulations of the linkage Exp-2 (Supplementary Fig. 8). These results show the same strong between circulation and thermodynamic drivers of sea-ice loss, geopotential height increases as in Exp-2, with approximately 70% while permitting nudging to observed wind fields. The close of Arctic low-level warming and sea-ice extent change (north agreement of patterns of thermodynamic forcing on sea ice between of 70◦ N) relative to Exp-2. Hence, these experiments suggest the ECHAM5 and POP2–CICE4 provides confidence in combining that ∼30% of the anomalous thermodynamic sea-ice extent these two tools. reduction is attributable to anthropogenic influences on the Arctic How sea-ice variability and trends can impact the Arctic circulation. Applying this estimate to the overall circulation-driven atmospheric circulation is an area of vigorous research42. Studies sea-ice trend established in Exp-6 (60%), we estimate that about suggest numerous mechanisms in which sea-ice loss modulates ∼42% (70%×60%) of the sea-ice decline observed since 1979 in the large-scale circulation in the lower troposphere in winter43–45. September is due to internal variability. This paper, instead, puts the spotlight on how the high-latitude What are the sources of internal variability on the Arctic circulation impacts sea ice. Although positive feedbacks between circulation? The upper tropospheric circulation in the Arctic has sea ice and the Arctic circulation exist, we find that these are small a strong link to variability originating from the tropics27,37–40. during summer. Instead, circulation variations over the Arctic have Previous research39 identified a relationship between tropical SST been a significant factor in driving sea-ice variability since 1979, and variability and annual mean atmospheric circulation over the Arctic have had a non-trivial contribution to the total surface temperature with a centre of action over Greenland, and model experiments39 trend over Greenland and northeastern Canada39. The potentially show that about 50% of the circulation change and the associated large contribution of internal variability to sea-ice loss over the warming over Greenland are attributable to internal variability next 40 years27 reinforces the importance of natural contributions originating from the tropical Pacific Ocean. Therefore, a substantial to sea-ice trends over the past several decades. The similarity of contribution of tropical Pacific variability on sea ice loss via this high-latitude circulation variability associated with sea-ice loss to teleconnection is to be expected. A further examination of this the teleconnections with the tropical Pacific suggests a contribution question will require a modelling framework that reproduces the of sea-ice losses from SST trends across the tropical Pacific Ocean. tropics–high latitude linkage faithfully and efficiently. Decadal trends in the hemispheric circulation are an important driver of Arctic climate change, and therefore a significant source of Uncertainties in reanalysis data uncertainty in projections of sea ice. Better understanding of these We have used ERA-I as a proxy for observational data. To assess teleconnections and their representation in global models under the robustness of this assumption we examined a suite of currently increasing greenhouse gases may help increase predictability on available reanalyses and station radiosonde data in Greenland. We seasonal to decadal timescales. find that all exhibit a very similar JJA circulation trend in the past 36 yr with a clear anticyclonic maximum over Greenland and Methods northeastern Canada. The strong temporal coherence in GL-Z200 Methods, including statements of data availability and any among all reanalysis data sets (Supplementary Fig. 9) points to associated accession codes and references, are available in the their reliability to capture the circulation variability in and around online version of this paper. Greenland. All data exhibit a similarly strong trend in circulation since 1979 and coherent variations on interannual timescales. Received 26 July 2016; accepted 3 February 2017; The magnitude of trends over Greenland vary, ranging from published online 13 March 2017 15 m/decade to 25 m /decade with ERA-I (19.3 m/decade) in the middle of this range and closest in matching the trend in radiosonde References data (18.8 m/decade). Thus, the circulation trend centred over 1. Min, S. K., Zhang, X. B., Zwiers, F. W. & Agnew, T. Human influence on Greenland seen in all reanalysis produces is a robust feature of the Arctic sea ice detectable from early 1990s onwards. Geophys. Res. Lett. 35, data, and unlikely the result of a changing observation network L21701 (2008). that is common to all reanalyses. The impact of observational 2. Kay, J. E., Holland, M. M. & Jahn, A. Inter-annual to multi-decadal Arctic sea uncertainties on our conclusions can be estimated by linearly ice extent trends in a warming world. Geophys. Res. Lett. 38, L15708 (2011). 3. Notz, D. & Marotzke, J. Observations reveal external driver for Arctic sea-ice scaling the results from Exp-6 by the GL-Z00 trends extracted retreat. Geophys. Res. Lett. 39, L08502 (2012). from different reanalyses products. The atmospheric circulation 4. Stroeve, J., Holland, M. M., Meier, W., Scambos, T. & Serreze, M. Arctic sea ice contribution to sea-ice loss ranges from 48% (NCEP-2) to 75% decline: faster than forecast. Geophys. Res. Lett. 34, L09501 (2007). (MERRA-2). Using the 70% contribution of internal forcing to the 5. Day, J. J., Hargreaves, J. C., Annan, J. D. & Abe-Ouchi, A. Sources of multi- circulation variability established above, we can attribute 30–50% of decadal variability in Arctic sea ice extent. Environ. Res. Lett. 7, 034011 (2012). sea-ice loss to internal variability. 6. Rampal, P., Weiss, J., Dubois, C. & Campin, J.-M. IPCC climate models do not capture Arctic sea ice drift acceleration: consequences in terms of projected sea Reanalysis products are reliable representations of the observed ice thinning and decline. J. Geophys. Res. 116, C00D07 (2011). circulation, humidity and temperature. The reanalysis data probably 7. Swart, N. C., Fyfe, J. C., Hawkins, E., Kay, J. E. & Jahn, A. Influence of internal also reliably represent the variability in total cloud cover in the variability on Arctic sea-ice trends. Nat. Clim. Change 5, 86–89 (2015).

6 NATURE CLIMATE CHANGE | ADVANCE ONLINE PUBLICATION | www.nature.com/natureclimatechange

8. Zhang, R. Mechanisms for low frequency variability of summer Arctic sea ice 33. Liu, J. P. et al. Has Arctic sea ice loss contributed to increased surface melting of extent. Proc. Natl Acad. Sci. USA 112, 1422296112 (2015). the Greenland ice sheet? J. Clim. 29, 3373–3386 (2016). 9. Francis, J. A. & Hunter, E. New insight into the disappearing Arctic sea ice. 34. Deser, C., Tomas, R., Alexander, M. & Lawrence, D. The seasonal atmospheric Eos 87, 509–511 (2006). response to projected Arctic sea ice loss in the late 21st century. J. Clim. 23, 10. Graversen, R. G. & Wang, M. Polar amplification in a coupled climate model 333–351 (2010). with locked albedo. Clim. Dynam. 33, 629–643 (2009). 35. Taylor, K. E., Stouffer, R. J. & Meehl, G. A. An overview of CMIP5 and the 11. Chylek, P., Folland, C. K., Lesins, G., Dubey, M. K. & Wang, M.-Y. Arctic air experiment design. Bull. Am. Meteorol. Soc. 93, 485–498 (2012). temperature change amplification and the Atlantic multidecadal oscillation. 36. Kay, J. E. et al. The Community Earth System Model (CESM) large Geophys. Res. Lett. 36, L14801 (2009). ensemble project: a community resource for studying climate change in the 12. Screen, J. A. & Simmonds, I. The central role of diminishing sea ice in recent presence of internal climate variability. Bull. Am. Meteorol. Soc. 96, Arctic temperature amplification. Nature 464, 1334–1337 (2010). 1333–1349 (2015). 13. Serreze, M. C. & Barry, R. G. Processes and impacts of Arctic amplification: 37. Hoerling, M. P., Hurrell, J. W. & Xu, T. Y. Tropical origins for recent North a research synthesis. Glob. Planet. Change 77, 85–96 (2011). Atlantic climate change. Science 292, 90–92 (2001). 14. Bintanja, R., Graversen, R. & Hazeleger, W. Arctic winter warming amplified by 38. Lee, S. Testing of the tropically excited Arctic warming (TEAM) mechanism the thermal inversion and consequent low infrared cooling to space. with traditional El Nino and La Nina. J. Clim. 25, 4015–4022 (2012). Nat. Geosci. 4, 758–761 (2011). 39. Ding, Q. H. et al. Tropical forcing of the recent rapid Arctic warming in 15. Vaughan, D. G. et al. in Climate Change 2013: The Physical Science Basis northeastern Canada and Greenland. Nature 509, 209–212 (2014). (eds Stocker, T. F. et al.) 317–382 (IPCC, Cambridge Univ. Press, 2013). 40. Trenberth, K. E., Fasullo, J. T., Branstator, G. & Phillips, A. S. Seasonal 16. Park, H.-S., Lee, S., Kosaka, Y., Son, S.-W. & Kim, S.-W. The impact of aspects of the recent pause in surface warming. Nat. Clim. Change 4, Arctic winter infrared radiation on early summer sea ice. J. Clim. 28, 911–916 (2014). 6281–6296 (2015). 41. Liu, Y. & Key, J. Assessment of Arctic cloud cover anomalies in atmospheric 17. Park, D.-S., Lee, S. & Feldstein, S. B. Attribution of the recent winter reanalysis products using satellite data. J. Clim. 29, 6065–6083 (2016). sea-ice decline over the Atlantic sector of the Arctic Ocean. J. Clim. 28, 42. Barnes, E. A. & Screen, J. The impact of Arctic warming on the midlatitude 4027–4033 (2015). jetstream: Can it? Has it? Will it? WIREs Clim. Change 6, 277–286 (2015). 18. Screen, J. A., Deser, C. & Simmonds, I. Local and remote controls on observed 43. Overland, J. E. & Wang, M. Increased variability in the early winter subarctic Arctic warming. Geophys. Res. Lett. 39, L10709 (2012). North American atmospheric circulation. J. Clim. 28, 7297–7305 (2015). 19. Perlwitz, J., Hoerling, M. & Dole, R. Arctic tropospheric warming: causes and 44. Francis, J. A. & Skific, N. Evidence linking rapid Arctic warming to linkages to lower latitudes. J. Clim. 28, 2154–2167 (2015). mid-latitude weather patterns. Phil. Trans. R. Soc. A 373, 20140170 (2015). 20. Deser, C., Walsh, J. E. & Timlin, M. S. Arctic sea ice variability in the context of 45. Cohen, J. An observational analysis: tropical relative to Arctic influence on recent atmospheric circulation trends. J. Clim. 13, 617–633 (2000). midlatitude weather in the era of Arctic amplification. Geophys. Res. Lett. 43, 21. Hu, A., Rooth, C., Bleck, R. & Deser, C. NAO influence on sea ice extent in the 5287–5294 (2016). Eurasian coastal region. Geophys. Res. Lett. 29, 2053 (2002). 46. Zhang, J. & Rothrock, D. A. Modeling global sea ice with a thickness and 22. Deser, C. & Teng, H. Evolution of Arctic sea ice concentration trends and the enthalpy distribution model in generalized curvilinear coordinates. role of atmospheric circulation forcing 1979–2007. Geophys. Res. Lett. 35, Mon. Weath. Rev. 131, 845–861 (2003). L02504 (2008). 23. Ogi, M., Rigor, I. G., McPhee, M. G. & Wallace, J. M. Summer retreat of Arctic sea ice: role of summer winds. Geophys. Res. Lett. 35, L24701 (2008). Acknowledgements 24. Ogi, M., Yamazaki, K. & Wallace, J. M. Influence of winter and summer This study was supported by NOAA’s Climate Program Office, Climate Variability and surface wind anomalies on summer Arctic sea ice extent. Geophys. Res. Lett. 37, Predictability Program (NA15OAR4310162). We thank the Max Planck Institute for L07701 (2010). Meteorology and National Center for Atmospheric Research model developers for making the ECHAM5 and CESM available and M. Steele, J. M. Wallace, C. Bitz, Q. Fu, 25. Watanabe, E., Wang, J., Sumi, A. & Hasumi, H. Arctic dipole anomaly and its M. Wang, D. L. Hartmann and D. Frierson for discussions. We acknowledge the CESM contribution to sea ice export from the Arctic Ocean in the 20th century. Large Ensemble Community Project and supercomputing resources provided by Geophys. Res. Lett. 33, L23703 (2006). NSF/CISL/Yellowstone. Q.D. acknowledges support from the University of Washington’s 26. Blanchard-Wrigglesworth, E., Armour, K. C., Bitz, C. M. & DeWeaver, E. Polar Science Center, the UW-Future of Ice Initiative, the Tamaki Foundation and UCSB Persistence and inherent predictability of Arctic sea ice in a GCM ensemble Center for Scientific Computing at CNSI. A.S. is grateful for funding from the National and observations. J. Clim. 24, 231–250 (2011). Science Foundation through grant ARC-1203425. D.S.B. acknowledges support from the 27. Wettstein, J. J. & Deser, C. Internal variability in projections of twenty-first Tamaki Foundation. R.E. acknowledges support from NASA NNXBAQ35G. century Arctic sea ice loss: role of the large-scale atmospheric circulation. J. Clim. 27, 527–550 (2014). 28. Screen, J. A., Simmonds, I. & Keay, K. Dramatic interannual changes of Author contributions perennial Arctic sea ice linked to abnormal summer storm activity. J. Geophys. Q.D. led this work with contributions from all authors. Q.D. made the calculations, Res. 116, D15105 (2011). implemented the general circulation model experiments, created the figures, and led writing of the paper. All authors contributed to the experimental design, interpreting 29. Kay, J. E., L’Ecuyer, T., Gettelman, A., Stephens, G. & O’Dell, C. The results and writing the paper. contribution of cloud and radiation anomalies to the 2007 Arctic sea ice extent minimum. Geophys. Res. Lett. 35, L08503 (2008). 30. Francis, J. A. & Hunter, E. Drivers of declining sea ice in the Arctic winter: Additional information a tale of two seas. Geophys. Res. Lett. 34, L17503 (2007). Supplementary information is available in the online version of the paper. Reprints and 31. Schweiger, A. J., Lindsay, R. W., Vavrus, S. & Francis, J. A. Relationships permissions information is available online at www.nature.com/reprints. between Arctic clouds and sea ice during autumn. J. Clim. 21, Correspondence and requests for materials should be addressed to Q.D. 4799–4810 (2008). 32. Bhatt, U. S. et al. The Atmospheric Response to Realistic Reduced Summer Arctic Sea Ice Anomalies, in Arctic Sea Ice Decline: Observations, Projections, Competing financial interests Mechanisms, and Implications (American Geophysical Union, 2008). The authors declare no competing financial interests.

NATURE CLIMATE CHANGE | ADVANCE ONLINE PUBLICATION | www.nature.com/natureclimatechange 7

Methods run, each with diﬀerent atmospheric initial conditions; the 20-member ensemble Data. We use the monthly circulation, temperature, cloud and radiation data from mean represents the model response to the prescribed history of sea ice and SST. the ERA-Interim reanalysis47 during the period 1979–2015. We analyse the Anthropogenic forcings are held constant in this experiment. observed monthly sea-ice concentration data derived from Goddard edited passive Exp-5: POP2+CICE4 control run. POP2+CICE4 is used to simulate the sea-ice microwave retrievals, which are compiled by the National Snow and Ice Data variability over the past 36 yr using prescribed ERA-Interim atmospheric forcing. Center. Monthly geopotential height records at 200 hPa from six stations along the The forcing fields include daily data of 10 m wind, downwelling shortwave and coast of Greenland in the Integrated Global Radiosonde Archive (IGRA2) were also longwave radiation, 2 m temperature and humidity, sea-level pressure used, except for 1989 and 1994. All stations have suﬃcient daily observations (at and precipitation. least 20 days) to construct reliable monthly averages during the period 1979–2014. Exp-6: Same as Exp-5 except that the atmospheric forcing is modified to excise the We also analyse the output from the thirty realizations, each with slightly perturbed forcing associated with the trends in the Greenland circulation pattern. initial conditions, of historical and future climate projection simulations from the To remove the circulations trend from the observations, we first construct the CESM Large Ensemble Community Project (LENS; ref. 36), and the output from thirty-six-year seasonal (JJA) averaged time series of the Z200 index over

26 climate models (Supplementary Table 1) archived in the CMIP5 (ref. 35) Greenland, Z200GL (GL-Z200 in Fig. 1c). We then linearly regress a key variable B database to estimate the climate response to external forcing in the historical against this time series to obtain spatial pattern β(x,y) of the variable associated record. In both cases, we focus our analysis on the historical satellite era from with the Greenland circulation index. Specifically, for the variable B we have 1979–2014. In both cases, each simulation in the ensemble was forced with identical historical forcing data (greenhouse gases, solar, volcanic and land use) B x,y,t =β x,y ×Z200GL(t) (1) over 1920–2005 and the 21st century representative concentration pathway 8.5 (RCP8.5) scenario through 2100; the experiments that comprise the ensembles where B represents a forcing field (for example, 10 m zonal wind, DLR, temperature, diﬀer only in their atmospheric initial conditions. and so on), x and y indicate the location, t indicates time (JJA), Z200GL is the Greenland Z200 index (GL-Z200 in Fig. 1c), and β is the regression coeﬃcient. Models. The general circulation model used to perform the experiments (Exps-1 to In the second step, the seasonal mean anomalous value of each forcing field is 4, 7 and 8) in this study is the ECHAM5 atmospheric general circulation model48, subtracted from the observed daily (or 6-hourly) forcing data during the with a horizontal resolution of T42 (∼2.8◦ latitude ×2.8◦ longitude) and summer—rendering a modified forcing that does not include variability or trends

19 vertical levels. In Exps-2, 3, 7 and 8, we coupled the ECHAM5 to a slab ocean in in variables that are associated with Z200GL. In the nine non-summer months, the the high latitudes to assess the role of prescribed circulation in driving the SST and forcing is exactly the same as that used in the Exp-5 control experiment. Given a sea ice in the Arctic. The slab ocean is represented as boxes of water of uniform strong correlation between circulation and surface winds, temperature, specific specified depth (50 m). A simple, thermodynamic-only sea-ice model is included humidity, sea-level pressure, and downwelling longwave radiation in the Arctic, when the ocean temperature reaches the freezing point. The ocean temperature or variability and trends in these six variables that are associated with Z200GL are sea-ice condition at each grid point is affected only by heat exchange across the processed and removed from the forcing. The initial states of ocean, sea ice and air–sea interface; there is no direct communication between adjacent ocean grid atmosphere in Exp-5 and Exp-6 are exactly the same. points, nor is there any representation of the deep ocean. In addition, a Exp-7: Same as Exp-2, except that ECHAM5 is nudged to a modified observed cyclo-stationary climatological heat flux is added to the ocean temperature wind patterns in which the long-term trends of simulated winds (zonal and tendency equation to maintain a seasonal cycle of ocean temperature and sea-ice meridional winds) in the ensemble mean (26 members) of CMIP5 during 1979 to conditions that is close to the observed climatology during 1979–1995. 2014 are removed from observation. In Exps-5 and 6, we use the coupled sea-ice–ocean component (POP2+CICE4) Exp-8: Same as Exp-2, except that the wind patterns used to nudge ECHAM5 are that is from the latest version of CESM (ref. 49). This is a model that includes both the residual wind fields resulting from the removal of the long-term wind trends the thermodynamics and dynamics of the ocean and sea ice; the model is run using derived from LENS (30 members) from ERA-I winds. tripole 1-degree ocean/sea-ice grids. Specific details on each model experiment are listed in a later section. Data availability. All reanalysis data used in this study were obtained from publicly available sources: ERA-I reanalysis data can be obtained from the ECMWF public Significance of correlation. The statistical significance of the correlation data sets web interface, http://apps.ecmwf.int/datasets; CFSR can be obtained from coefficient accounts for the autocorrelation in the time series by using an ‘effective NCDC/NOMADS data access, https://nomads.ncdc.noaa.gov/#cfsr; NCEP-2 sample size’ N ∗: reanalysis can be obtained from NOAA/ESRL/PSD data portal, 1−r r ∗ = 1 2 https://www.esrl.noaa.gov/psd/data. MERRA-1 and MERRA-2 data are available at N N the Modelling and Assimilation Data and Information Services Center, 1+r1r2 https://gmao.gsfc.nasa.gov/reanalysis/MERRA. JRA55 can be downloaded from the where N is the number of available time steps and r1 and r2 are lag-one Japan Meteorological Agency (JMA) data dissemination system, autocorrelation coefficients of each variable50. http://jra.kishou.go.jp/JRA-55/index_en.html. IGRA2 radiosonde data can be accessed via the NOAA/NCDC data portal, Experimental design. Exp-1: ECHAM5’s circulation nudged to observed state. https://www.ncdc.noaa.gov/data-access. Simulated global circulation and Because we are interested in the impact of the long-term trend of the observed temperature under anthropogenic forcing were obtained from the CMIP5 and circulation in our model simulations, and to facilitate our computation efforts, we LENS archives accessed through the Earth System Grid Federation data portal, interpolate observed monthly ERA-I data to daily fields for nudging. We use a very http://esgf.llnl.gov. Output data from the model simulations (Exps 1–8) are weak damping term to nudge the three-dimensional (from surface to the top of available from the corresponding author upon request. atmosphere) divergence and vorticity fields of the model to the observed monthly (smoothly interpolated to daily) fields in the past 36 yr; this weak damping allows the model to generate its own day-to-day variability but constrains the model to be References very close to the observed circulation on monthly and longer timescales. In the 47. Dee, D. P. et al. The ERA-Interim reanalysis: configuration and lower boundary, we impose climatological SST/sea-ice everywhere. Anthropogenic performance of the data assimilation system. Q. J. R. Meteorol. Soc. 137, forcings are held constant in this experiment. 553–597 (2011). Exp-2: Same as Exp-1, except that a simple slab ocean/sea-ice model is used in the 48. Roeckner, E. et al. Max Planck Institut für Meteorologie Report Arctic, north of 60◦ N. (Max-Planck-Institut für Meteorologie, 2003). Exp-3: Same as Exp-2, except that we nudge the model’s wind fields above 700 hPa 49. Hurrell, J. W. et al. The community earth system model: a framework for to the observed winds. collaborative research. Bull. Am. Meteorol. Soc. 94, 1339–1360 (2013). Exp-4: The ECHAM5 atmosphere model is forced by observed history of sea-ice 50. Bretherton, C. S., Widmann, M., Dymnidov, V. P., Wallace, J. M. & Blade, I. concentration and SST in the Arctic (north of 60◦ N) and the climatological The effective number of spatial degrees of freedom of a time-varying field. SST/sea-ice everywhere else during the period 1979–2014. Twenty realizations are J. Clim. 12, 1990–2009 (1999).

NATURE CLIMATE CHANGE | www.nature.com/natureclimatechange

Inﬂuence of high-latitude atmospheric circulation changes on summertime Arctic sea ice

t is well recognized that recent Arctic sea-ice decline has Observed linkage between circulation and sea ice both natural and anthropogenic drivers1–3, but their relative To examine the physical linkages, we focus on the connection Iimportance is poorly known4–7. This uncertainty arises from between September sea-ice extent and the preceding summer the fact that the contribution of atmospheric internal variability (June–July–August, JJA) atmospheric circulation. We choose this in the Arctic climate system is inadequately understood, and it preceding 3-month window because sea-ice extent anomalies have is unclear how well models reproduce these processes8. Reliably a 3-month decorrelation timescale26, and previous studies have distinguishing the natural and anthropogenic contributions of sea- shown∼ a strong link between summer circulation and sea-ice ice loss requires a comprehensive understanding of the mechanisms variability23,24,27,28. We focus on physical mechanisms, analysing that control the variability of sea ice. Although some progress temperature, humidity, and downward longwave radiation (DLR), has been made in this area8–17, the answer is far from clear, and all of which are affected by atmospheric circulation and, in turn, conflicting hypotheses exist18,19. affect sea-ice concentration. A key player is the radiation balance, Earlier work indicated that the decline of sea ice beforeSUPPLEMENTARY the which dominates the surface energyINFORMATION balance controlling the growth 29,30 1990s was in part owing to an upward trend in the North Atlantic andmeltofArcticseaice . DOI: 10.1038/NCLIMATE3241 Oscillation (NAO) index20,21. However, since the early 1990s, the The region with the greatest negativeARTICLES trend in September Inapparent the format link between provided the by NAO theand authors sea ice and has unedited. largely disappeared, sea-ice concentration since 1979 is highlighted in Fig. 1a, and with Arctic sea ice declining further despite the reversal in the NAO includesPUBLISHED the ONLINE: Beaufort, 13 MARCH Chukchi, 2017 and | DOI: East10.1038/NCLIMATE3241 Siberian Seas, featuring trend22. On the other hand, several studies have argued that the an average decline >10%/decade. Concomitant with the trend in recent NAO trend may still play a key role in the sea-ice retreat23,24. sea ice are trends in the atmospheric circulation: 200 hPa and Alternatively, the Arctic Dipole has been suggested as a driver of 700 hPa geopotential heights have been rising over northeastern sea-ice change in some regions of the Arctic25. Overall, the recent Canada and Greenland, and the surface winds have become more trends in Arctic sea ice cannot be linked to simple indices of anticyclonic (Figs 1b and 2d). To illuminate processes that link Influenceclimate variability22. of high-latitudethe atmospheric circulation changes to sea-ice circulationchanges, we first construct an In this paper, we examine the contribution of the atmospheric index of September sea-ice coverage over the region with the fastest changescirculation to Arctic sea-ice on variability summertime by utilizing an atmospheric Arcticsea-ice decline from sea 1979 ice to 2014 (Fig. 1c), as well as indices of general circulation model (ECHAM5) coupled with a simple ocean– JJA low-level temperature (surface to 750 hPa), DLR at surface, sea-ice model in which the atmospheric circulation field is nudged and integrated atmospheric water vapour (surface to 750 hPa) Qinghuato observations. Ding Specifically,1,2,3*, Axel we Schweiger explore how3, the Michelle high-latitude L’Heureuxderived4, from David ERA-Interim S. Battisti reanalysis5,6, Stephen (hereafter Po-Chedley ERA-I); indices5, are summertime atmospheric circulation7 impacts the September Arctic averaged5 over the Arctic, poleward4 of 70◦ N, and4 shown in Fig. 1c. Nathanielsea-ice extent, C. and Johnson estimate to, what Eduardo extentchanges Blanchard-Wrigglesworth in atmospheric The decreasing, Kirstin trend Harnos in sea-ice, concentration Qin Zhang is, accompanied by Ryancirculation Eastman explain the5 and observed Eric sea-ice J. Steig loss5,6 of the past few decades. increasing trends in JJA Arctic temperature, DLR and water vapour.

The Arctic has seen rapid sea-ice decline in the past three decades, whilst warming at about twice the global average rate. 1 2 YetDepartment the relationship of Geography, between University Arctic of California, warming Santa and Barbara, sea-ice California loss is93106, not USA.well understood.Earth Research Here, Institute, we University present of evidence California, that trends 3 inSanta summertime Barbara, California atmospheric 93106, USA. circulationPolar Science may Center, have Applied contributed Physics Laboratory, as much University as 60% of Washington, to the September Seattle, Washington sea-ice extent98195, USA. decline 4NOAA Climate Prediction Center, College Park, Maryland 20740, USA. 5Department of Atmospheric Sciences, University of Washington, Seattle, since 1979. A tendency6 towards a stronger anticyclonic circulation over Greenland and the Arctic Ocean7 with a barotropic structureWashington in 98195, the USA. troposphereDepartment increased of Earth and the Space downwelling Sciences, University longwave of Washington, radiation Seattle, above Washington the ice by 98195, warming USA. andCooperative moistening Institute the for lowerClimate troposphere. Science, Princeton Model University, experiments, Princeton, New with Jersey reanalysis 08540, USA. data*e-mail: [email protected] atmospheric circulation, replicate the observed thermodynamic response and indicate that the near-surface changes are dominated by circulation changes rather than feedbacksNATURE CLIMATE from CHANGE the changing| ADVANCE ONLINE sea-ice PUBLICATION cover. Internal | www.nature.com/natureclimatechange variability dominates the Arctic summer circulation trend and may be1 responsible for about 30–50% of the© 2017 overall Macmillan decline Publishers in SeptemberLimited, part of Springer sea ice Nature. since All rights 1979. reserved.

t is well recognized that recent Arctic sea-ice decline has Observed linkage between circulation and sea ice both natural and anthropogenic drivers1–3, but their relative To examine the physical linkages, we focus on the connection Iimportance is poorly known4–7. This uncertainty arises from between September sea-ice extent and the preceding summer the fact that the contribution of atmospheric internal variability (June–July–August, JJA) atmospheric circulation. We choose this in the Arctic climate system is inadequately understood, and it preceding 3-month window because sea-ice extent anomalies have is unclear how well models reproduce these processes8. Reliably a 3-month decorrelation timescale26, and previous studies have distinguishing the natural and anthropogenic contributions of sea- shown∼ a strong link between summer circulation and sea-ice ice loss requires a comprehensive understanding of the mechanisms variability23,24,27,28. We focus on physical mechanisms, analysing that control the variability of sea ice. Although some progress temperature, humidity, and downward longwave radiation (DLR), has been made in this area8–17, the answer is far from clear, and all of which are aﬀected by atmospheric circulation and, in turn, conflicting hypotheses exist18,19. aﬀect sea-ice concentration. A key player is the radiation balance, Earlier work indicated that the decline of sea ice before the which dominates the surface energy balance controlling the growth 1990s was in part owing to an upward trend in the North Atlantic andmeltofArcticseaice29,30. Oscillation (NAO) index20,21. However, since the early 1990s, the The region with the greatest negative trend in September apparent link between the NAO and sea ice has largely disappeared, sea-ice concentration since 1979 is highlighted in Fig. 1a, and with Arctic sea ice declining further despite the reversal in the NAO includes the Beaufort, Chukchi, and East Siberian Seas, featuring trend22. On the other hand, several studies have argued that the an average decline >10%/decade. Concomitant with the trend in recent NAO trend may still play a key role in the sea-ice retreat23,24. sea ice are trends in the atmospheric circulation: 200 hPa and Alternatively, the Arctic Dipole has been suggested as a driver of 700 hPa geopotential heights have been rising over northeastern sea-ice change in some regions of the Arctic25. Overall, the recent Canada and Greenland, and the surface winds have become more trends in Arctic sea ice cannot be linked to simple indices of anticyclonic (Figs 1b and 2d). To illuminate processes that link climate variability22. the circulation changes to sea-ice changes, we first construct an In this paper, we examine the contribution of the atmospheric index of September sea-ice coverage over the region with the fastest circulation to Arctic sea-ice variability by utilizing an atmospheric sea-ice decline from 1979 to 2014 (Fig. 1c), as well as indices of general circulation model (ECHAM5) coupled with a simple ocean– JJA low-level temperature (surface to 750 hPa), DLR at surface, sea-ice model in which the atmospheric circulation field is nudged and integrated atmospheric water vapour (surface to 750 hPa) to observations. Specifically, we explore how the high-latitude derived from ERA-Interim reanalysis (hereafter ERA-I); indices are summertime atmospheric circulation impacts the September Arctic averaged over the Arctic, poleward of 70◦ N, and shown in Fig. 1c. sea-ice extent, and estimate to what extent changes in atmospheric The decreasing trend in sea-ice concentration is accompanied by circulation explain the observed sea-ice loss of the past few decades. increasing trends in JJA Arctic temperature, DLR and water vapour.

1Department of Geography, University of California, Santa Barbara, California 93106, USA. 2Earth Research Institute, University of California, Santa Barbara, California 93106, USA. 3Polar Science Center, Applied Physics Laboratory, University of Washington, Seattle, Washington 98195, USA. 4NOAA Climate Prediction Center, College Park, Maryland 20740, USA. 5Department of Atmospheric Sciences, University of Washington, Seattle, Washington 98195, USA. 6Department of Earth and Space Sciences, University of Washington, Seattle, Washington 98195, USA. 7Cooperative Institute for NATUREClimate CLIMATE Science, CHANGE Princeton | www.nature.com/natureclimatechange University, Princeton, New Jersey 08540, USA. *e-mail: [email protected] 1 © 2017 Macmillan Publishers Limited, part of Springer Nature. All rights reserved. NATURE CLIMATE CHANGE | ADVANCE ONLINE PUBLICATION | www.nature.com/natureclimatechange 1