A zonal asymmetry in boreal winter surface temperature trend and its recent reversal over the Northern Hemisphere continents

Introduction
Global surface temperature has been gradually increasing over the past few decades. However, there are significant regional differences. While the Arctic has experienced the most significant warming, which is about four times stronger than the rest of the globe since 19791, the mid-latitude regions have experienced a weak warming or even a cooling trend2,3. Fig. 1a displays the linear trend of boreal winter surface temperature from 1979 to 2023. The notable warming trend is observed in high latitudes and most of the Middle East and Africa. However, no significant trend is discernible for most mid-latitude continents. This is due to the large decadal variability4,5,6,7. In fact, the mid-latitude surface temperature trend from the late 1980s to the late 2000s is almost opposite to that observed since the 2010s (Fig. 1b, c). More importantly, they are spatially inhomogeneous with opposite signs between the central Eurasia (CEUR) and the Northeastern North America and Greenland (NNAG).

Linear trends of boreal winter (DJF) surface temperature (a) for the period from 1979 to 2023 (unit: K year-1), for the sub-periods (b) P1 (1988–2009), and (c) P2 (2009–2023). d, e Same as (b, c) but for sea level pressure (SLP) (unit: hPa year-1). Statistically significant trends at the 95% confidence level, based on Student’s t-test, are hatched. The central Eurasia (CEUR; 50–70°N, 0–140°E) and Northeastern North America and Greenland (NNAG; 50–85°N, 10–80°W) domains are indicated by boxes in (b–e). Here, each winter is referred to by its starting year, the winter of 1988 being the seasonal average from December 1988 to February 1989.
The increase in anthropogenic greenhouse gas concentrations has served as an external forcing, driving an overall warming in global mean temperature8. However, internal climate variability also plays a critical role, either amplifying or mitigating surface temperature changes in different regions9,10,11,12. It has been reported that while Arctic warming is primarily attributed to anthropogenic forcing13, mid-latitude surface temperature changes are not only influenced by external forcing but also affected by internal climate variability14,15,16,17,18,19,20.
Several mechanisms have been proposed to explain mid-latitude surface temperature trends. A peculiar cooling trend in the central Eurasia (Fig. 1b) has received particular attention. One representative mechanism is Arctic sea ice decline21,22,23. Arctic sea ice loss, mostly due to anthropogenic warming, enhances upward surface heat and increases lower troposphere temperature21,22,24. It can induce tropospheric circulation changes, leading cold continents22,25,26. Another mechanism involves deep Arctic warming, characterized by the Arctic warming extending from the surface to the upper troposphere27,28,29,30. It has been suggested that deep Arctic warming can generate equatorward propagating Rossby waves that weaken the extratropical jet. It leads to more persistent blocking and associated winter cooling28,29,30. The mid-latitude temperature trend can also be influenced by tropical convection change7. Tropical convection can excite the Rossby wave train to the mid-latitude, affecting the surface temperature trend. An example is convection change in the tropical Pacific due to the Pacific Decadal Oscillation (PDO). An enhanced tropical convection during the negative phase of the PDO can contribute to surface warming over northeastern Canada and Greenland31,32.
Despite these studies, a zonally asymmetric surface temperature trend and its reversal in the late 2000s (Fig. 1b, c) still remain unresolved. The present study investigates the factor(s) that influences the boreal winter (December-January-February; DJF) surface temperature trend in the mid-latitude continents, with an emphasis on CEUR and NNAG temperature trends. By comparing linear trends and interannual variabilities of CEUR and NNAG temperature with those of climate modes, regional and temporal changes in surface temperature trends are examined. The thermodynamic and dynamic processes responsible for surface temperature changes are also identified by comparing the surface energy budget changes.
Results
Decadal change in surface temperature trend
The boreal winter surface temperature trend over the Northern Hemisphere continents is not stationary but varies on a decadal timescale (Fig. 1b, c). The significant CEUR cooling, which was most pronounced from the late 1980s to the late 2000s5,7, has not persisted in recent decades6. The surface temperature trend over NNAG has also changed over time2,33,34,35. Motivated by this, the analysis period is divided into the two sub-periods, i.e., period 1 from 1988 to 2009 (P1) and period 2 from 2009 to 2023 (P2). Here, the starting year 1988 is chosen to depict the maximum cooling trend over CEUR in the recent past. A similar period was used in the previous study when examining the decadal variability of the warm Arctic-cold Eurasian pattern7.
The time series of the boreal winter surface temperature averaged over CEUR (50–70°N, 0–140°E) and NNAG domains (50–85°N, 10–80°W) are illustrated in Fig. 2. A distinct decadal change is evident. A CEUR cooling trend in P1 is switched into a warming trend in P2. A similar decadal trend change, but with an opposite sign is also observed over NNAG. Due to this decadal change, the linear trend from 1979 to 2023 is weak in both CEUR and NNAG. When the Lepage test (see “Methods”) is applied to detect the year of significant trend change36,37,38 (green solid lines in Fig. 2), statistically significant trend changes are identified across the year of 2009 and 2010 for CEUR and NNAG temperature trends, respectively. This result is not sensitive to the length of the time window (not shown), justifying the division of the analysis periods at 2009.

Time series of boreal winter surface temperature over (a) central Eurasia (CEUR; 50–70°N, 0–140°E) and (b) Northeastern North America and Greenland domains (NNAG; 50–85°N, 10–80°W). The analysis domains are indicated by boxes in Fig. 1b, c. The green solid lines denote the Lepage test statistic (HK) for a window length of 10 years. If the HK is greater than 5.99 (green dashed line), the 10-year trend difference between the two time periods is significant at the 95% confidence level.
It is noteworthy that the trend change in CEUR is dominated by Asian temperature (Fig. 1b). When the CEUR domain is split into Europe (50–70°N, 0–60°E) and Asia (50–70°N, 60–140°E), the overall results do not change, with a statistically significant cooling trend in P1 and a warming trend in P2 (Supplementary Fig. 1). However, the trend reversal in 2009 is statistically significant only for the Asian temperature. This indicates that while the decadal variability of surface temperature trend is robust over a broad region of CEUR, its recent reversal is dominated by the Asian surface temperature.
The opposing CEUR and NNAG temperature anomalies are also evident in their interannual variability. When detrended, their interannual co-variability is statistically significant at the 95% confidence level, with correlation coefficients of −0.49 in P1 and −0.42 in P2. This result suggests that CEUR and NNAG temperatures are modulated by one large-scale circulation pattern that controls surface climate in the Northern Hemisphere on interannual to decadal timescales.
One possible factor is the Arctic Oscillation (AO)33,39,40. Large-scale atmospheric circulation in the Northern Hemisphere extratropics and the associated surface temperature anomalies are significantly altered by the AO33,39,40. When the AO is in a negative phase, zonal-mean zonal winds in midlatitudes are significantly decelerated and shifted equatorward. This leads to cold surface temperature anomalies over Eurasia and warm surface temperature anomalies over northern North America41,42,43. In fact, the sea level pressure (SLP) changes in P1 and P2 are remarkably similar to the AO-related SLP anomalies but with opposite signs in the two periods (compare Fig. 1d, e and Supplementary Fig. 2b). This suggests that surface temperature trend and its reversal across the year 2009–2010 may result from the AO-related atmospheric circulation change. The decadal change of the AO index is indeed consistent with that of the surface temperature (compare Fig. 2 and Fig. 3a). The Lepage test also supports their link with a significant AO-index trend change around 2009; i.e., a statistically significant negative trend of the AO index in P1 (−1.04 decade-1) is switched to a positive trend in P2 (0.98 decade-1).

Same as Fig. 2 but for (a) the AO, (b) WACE, and (c) NAM indices, obtained by the normalized leading principal component at 50 hPa. The EOF analysis is conducted with boreal-winter geopotential anomalies poleward of 20°N.
The CEUR temperature can also be influenced by the warm Arctic-cold Eurasia (WACE) pattern44. The WACE index, derived from the second empirical orthogonal function of the Northern Hemisphere surface temperature on an interannual timescale44, is presented in Fig. 3b. Note that the AO is the leading mode of variability in the Northern Hemisphere extratropics. The WACE index shows a statistically significant positive trend in P1, followed by a weak negative trend in P2. However, the trend change across the year 2009–2010 is not discernable. A significant trend change is instead detected in 1990 and 1992, when the Lepage test is applied (green solid line). This result is consistent with the previous studies reporting that the WACE pattern is closely related to the anthropogenic sea ice loss in the Barents-Kara Seas, which has been intensifying since the 1990s45,46,47. This indicates that while the CEUR temperature trend in P1 is likely influenced by both the AO and WACE, its reversal in 2009 is mostly due to the AO.
It is not clear what determines the AO trend change. The AO trend can be influenced by the external forcing. Feldstein (2002) suggested external forcing as the primary driver of the AO trend since the 1960s, contrasting with internal climate variability during the first half of the 20th century48. It has been argued that an increasing greenhouse gas concentration can exert the AO toward its positive phase by strengthening the meridional temperature gradient in the upper troposphere and lower stratosphere49,50. However, temperature trend in the upper troposphere and lower stratosphere in P1 is not in accordance to the anthropogenic warming (Fig. 4a). It has also been suggested that the negative AO trend is associated with strong Arctic warming extended to the upper troposphere7,27. Fig. 4 indeed shows a deeper Arctic warming in P1 than P2. However, this is not likely due to anthropogenic climate change, as the Arctic mid-tropospheric temperature trend in P2 is negative against anthropogenic warming (Fig. 4b). As shown below, the trend in snow albedo over the high-latitude Eurasian continent in P2 is also opposite to the trend in P1, further contradicting anthropogenic warming. This result suggests that the surface temperature trend change between P1 and P2 is likely due to the internal climate variability.

Linear trends of boreal winter (DJF) zonal-mean temperature (shading, units: K year−1) and zonal wind (contours, at intervals of 0.1 m s−1 year−1) for (a) P1 (1988–2009) and (b) P2 (2009–2023). Statistically significant temperature trends at the 95% confidence level, based on Student’s t-test, are hatched.
While the AO represents the tropospheric climate variability, it can be influenced by the stratospheric circulation. It has been shown that a positive phase of the AO is often preceded by a strong stratospheric polar vortex41,51,52,53. Hu et al. (2018) reported that decadal change in the stratospheric polar vortex intensity is linked to the Eurasian surface temperature trend54. Fig. 3c presents stratospheric Northern Annular Mode (NAM) index, derived from the leading empirical orthogonal function55. Here, the empirical orthogonal function analysis is performed with 50-hPa geopotential height anomalies poleward of 20°N in the boreal winter. The leading mode explains 61.3% of the total interannual variance of the polar vortex. The corresponding principal component, that is the 50 hPa NAM index, shows a statistically significant negative trend in P1 (weakening of the polar vortex; Fig. 4a), followed by a weak positive trend in P2. However, the trend change is not statistically significant when the Lepage test is applied (green solid line). A significant trend change instead appears in 1996, as reported in Hu et al. 54. This result indicates that the negative AO trend and the associated surface temperature trend in P1 are partly influenced by the stratospheric circulation change, although its decadal change across the year 2009–2010 is not.
Surface energy budget
The AO change drives opposing surface temperature trends over CEUR and NNAG. To identify the related thermodynamic and dynamic processes, the surface energy budget is calculated for P1 and P2 (see “Eq. (2)” in the “Methods”). The reconstructed surface temperature trends from the surface energy budget (Fig. 5a, g) are quantitatively similar to the observed surface temperature trends (Fig. 1b, c). This agreement allows the decomposition of surface temperature trend into five thermodynamic contributors, i.e., downward longwave radiation, net shortwave radiation, surface latent heat flux, surface sensible heat flux, and storage term (Fig. 5). For comparison, the linear trend of each component, not divided by climatological surface temperature cubed, is also presented in Supplementary Fig. 3.

a Same as Fig. 1b but for surface temperature trend reconstructed from the surface energy budget (unit: K year-1) for P1 (1988–2009), and contributions of (b) downward longwave radiation (DLW), (c) net shortwave radiation (SW), (d) latent heat flux (LH), (e) sensible heat flux (SH), and f storage term (Storage). g–l Same as (a–f) but for P2 (2009–2023).
The zonal asymmetry between the CEUR and NNAG temperature trends is primarily explained by downward longwave radiation change (Fig. 5b, h). The CEUR temperature trend is also influenced by sensible heat flux (Fig. 5e, k), while other terms have rather minor contributions to CEUR and NNAG temperature trends. Here, it is important to note that the relative importance of each component to the reconstructed surface temperature trend does not change much over time, although the sign is switched from P1 to P2.
The surface energy budgets over CEUR and NNAG are summarized in Fig. 6. All properties are area-averaged over CEUR and NNAG domains. The observed trend (black bars) is well reproduced by surface energy budget (gray bars). The most important contributor to the surface temperature trend is again the downward longwave radiation. Its contribution to the reconstructed CEUR temperature trend is 101.03% in P1 and 98.83% in P2. For NNAG temperature trend, it accounts for 103.64% in P1 and 120.36% in P2 (see also Supplementary Table 1). The second largest contributor varies by region. The CEUR temperature trend is significantly influenced by sensible heat flux with 37.75% in P1 and 21.09% in P2. The net shortwave radiation and the storage term play a minor cancelling role for CEUR temperature trend. The NNAG temperature trend is also partly cancelled by surface heat flux changes. Unlike in CEUR, net shortwave radiation plays a negligible role in NNAG temperature trend.

Observed (black) and reconstructed surface temperature trends (gray) over CEUR domain for (a) P1 (1988–2009) and (b) P2 (2009–2023). Surface temperature trend derived from the surface energy budget consists of downward longwave radiation (DLW; coral), net shortwave radiation (SW; orange), latent heat flux (LH; yellow), sensible heat flux (SH; green), and the storage term (Storage; blue). c, d Same as (a, b) but for NNAG domain.
The downward longwave radiation change is closely related to the temperature and moisture advection56,57, accompanied by the AO-related circulation change (Fig. 1d, e), because warming and moistening in the free atmosphere can enhance the downward longwave radiation through the Stefan-Boltzmann law and greenhouse effect, respectively58,59. Figs. 7a, b and 7c, d show the vertically-integrated temperature and moisture advection change, respectively. Strong cooling and drying in P1, followed by warming and moistening in P2, are evident over CEUR. NNAG experiences the opposite trends in the two periods. Their patterns resemble longwave-induced surface temperature trends (Fig. 5b, h). This result suggests that the AO-related circulation changes are largely responsible for the surface temperature trends in the two regions, although the relative importance of temperature and moisture advection changes to the surface temperature trend remains to be determined.

Linear trends of vertically-integrated horizontal temperature advection (unit: K s−1 year−1) from 1000 hPa to 300 hPa levels for a P1 (1988–2009) and b P2 (2009–2023). Same as (a, b) but for (c, d) moisture advection (unit: g kg−1 s−1 year−1), (e, f) near-surface vertical temperature gradient (unit: K m−1 year−1), (g, h) near-surface wind speed (unit: m s−1 year−1), (i, j) near-surface vertical gradient in saturation specific humidity (unit: g kg−1 m−1 year−1), and (k–l) snow albedo (unit: year−1). Significant trends at the 95% confidence level, based on Student’s t-test, are hatched.
The surface sensible heat flux is the second most important term in CEUR temperature trend (Fig. 6a, b). The sensible heat flux is determined by the vertical temperature gradient and near-surface wind. Fig. 7e, f present the near-surface vertical temperature gradient changes. A significant negative trend is observed over CEUR in P1 (Fig. 7e), indicating a warmer surface compared to the atmosphere above (i.e., 2 m temperature). This is mainly due to the AO-induced cold and dry advection in the lower-troposphere (Fig. 7a, c). Meanwhile, a significant positive trend is found over CEUR in P2 (Fig. 7f), consistent with the positive sensible heat flux trend, due to the opposite AO-induced circulation change. The sensible heat flux also depends on near-surface wind speed. In P1, near-surface wind speed over CEUR shows a significant negative trend (Fig. 7g). This indicates that the negative trend of CEUR sensible heat flux in P1 is likely influenced by both decreasing near-surface vertical temperature gradient and wind speed. Their constructive effect also holds in P2, with an opposite sign (Fig. 7f, h).
The surface latent heat flux plays a minor but non-negligible role in NNAG temperature trend, especially in P2 (Fig. 6c, d). The latent heat flux can be understood with surface evaporation; enhanced evaporation from the surface to the air above, due to drying in the atmosphere, can lead to negative latent heat flux trend. The surface evaporation can be enhanced with decreasing near-surface vertical moisture gradient and increasing near-surface wind speed. It turns out that the latent heat flux change in NNAG is more closely linked to moisture gradient change because near-surface wind changes over NNAG are weak and not statistically significant in both P1 and P2 (Fig. 7g, h). Fig. 7i, j show the near-surface vertical gradient changes in saturation specific humidity. Significant positive trends over NNAG are observed in P1 (Fig. 7i), indicating a drier surface compared to the lowermost atmosphere (i.e., 2 m specific humidity). On the other hand, significant negative trends are found in P2 (Fig. 7j), agreeing with the negative latent heat flux trend. Such changes could be influenced by Greenland blocking, as Greenland blocking modulates moisture transport in the NNAG35,60,61,62. Since the Greenland blocking is strongly coupled to the AO63, this result supports that the AO-induced atmospheric circulation change plays a role in the latent heat flux trend through the vertical moisture gradient changes (Supplementary Fig. 2d, f).
The net shortwave radiation plays a minor role over CEUR. A significant decreasing trend in western Europe and increasing trend in the high-latitude Eurasian continent are evident in P1, whereas they switch the sign in P2. The net shortwave radiation trend change over CEUR is closely related to the snow albedo change (Fig. 7k, l), as reduced snow cover less reflects incoming shortwave radiation. The pattern correlations between net shortwave radiation and snow albedo changes over CEUR are -0.55 in P1 and -0.62 in P2, respectively. High-latitude snow cover is known to be modulated by the AO39,40,64. During the negative AO, anticyclonic anomalies over the Barents-Kara Seas become dominant, leading to dry and cold advection and less snowfall in high-latitude CEUR. Cyclonic anomalies over the North Atlantic also become stronger, leading to moisture advection and more snowfall in western Europe (Supplementary Fig. 2g).
Summary and Discussion
This study examines the boreal winter surface temperature trend and its decadal change in the Northern Hemisphere continents. Unlike a persistent anthropogenic warming, surface temperature trends in mid- to high-latitude continents exhibit a time-varying and regionally dependent structure. Specifically, a zonally-asymmetric temperature trend, with a cooling trend over central Eurasia (CEUR) and a warming trend over Northeastern North America and Greenland (NNAG), is observed during the period 1988–2009 (P1). These surface temperature trends are reversed in the recent period 2009–2023 (P2). While the surface temperature trends in P1 are partly associated with the stratospheric polar vortex and the warm Arctic-cold Eurasia pattern, their reversal in 2009–2010 is not driven by either of these factors. The zonal asymmetry in the surface temperature trend and its recent reversal are instead closely related to AO-related atmospheric circulation changes. Since the AO is a representative internal climate variability in the extratropical atmosphere, it suggests that the decadal changes of the Northern Hemisphere temperature trends are mainly determined by the internal climate variability.
The surface energy budget analysis reveals that the surface temperature trend is primarily determined by the downward longwave radiation changes due to the AO-related temperature and moisture advection. The secondary factor varies by region with a constructive contribution of sensible heat flux change to the CEUR temperature trend and a destructive effect of latent heat flux change to the NNAG temperature trend. This result suggests that surface temperature trends over CEUR and NNAG are determined by slightly different thermodynamic and dynamic processes.
Methods
Surface energy budget
To evaluate the contribution of different thermodynamic processes to the surface temperature (more precisely skin temperature) trend, we adopt the surface energy budget and decompose it into the downward longwave radiation, the net shortwave radiation, the latent heat flux, the sensible heat flux, and a storage term57,65. The surface energy budget can be written as follows:
where ({{rm{F}}}_{{rm{DLW}}}) is the downward longwave radiation at the surface, ({{rm{F}}}_{{rm{SW}}}) is the net shortwave radiation, ({{rm{F}}}_{{rm{LH}}}) is the surface latent heat flux, ({{rm{F}}}_{{rm{SH}}}) is the surface sensible heat flux, ({rm{varepsilon }}) is the emissivity (({rm{varepsilon }}) = 1 in this study), ({rm{sigma }}) is the Stefan-Boltzmann constant, and ({{rm{T}}}_{{rm{s}}}) is the surface temperature. R is a storage, which represents the heat conduction from below the surface. The downward flux is set to positive. The daily surface temperature and first four vertical energy fluxes in Eq. (1) are obtained from the European Centre for Medium-Range Weather Forecast (ERA5) reanalysis dataset66. All variables are detrended by removing their long-term linear trends for the period 1979–2023. Note that the zonal-mean trend is not removed.
The ({{rm{T}}}_{{rm{s}}}) trend can be obtained by Taylor expansion with the upward longwave flux term to first order and rearranging terms. The reconstructed ({{rm{T}}}_{{rm{s}}}) trend can be written as follows:
where (Delta {{rm{T}}}_{{rm{s}}}) is the linear trend of surface temperature, (Delta {{rm{F}}}_{{rm{i}}}) is the linear trend of ({{rm{F}}}_{{rm{i}}}), and (bar{{{rm{T}}}_{{rm{s}}}}) is taken to be the climatology of surface temperature.
Lepage test statistic
The Lepage test statistic67 (HK) is employed to detect the change points in adjacent data on a decadal timescale. The HK has the advantage of evaluating the statistical significance between two samples, even when the parent populations are unknown36. It has been used to detect various types of change in climate time series, including linear trends, discontinuous changes, and step-like changes36,37,38,67,68. The HK is calculated as a combination of the squares of the standardized Wilcoxon (W) and Ansari–Bradley (A) statistics.
When the HK exceeds 5.99 (4.21), two samples are statistically different at the 95% (90%) confidence level. In this study, the HK is used to identify the surface temperature trend change. The annual-mean surface temperature trend for N years before a given year Y is compared with the trend for N years with a given year Y and N-1 years after a given year Y. The year Y is continuously shifted by one year along the time series (Y (to) Y + 1). A 10-year moving window (N = 10) is used.
Responses