The roles of global warming and Arctic Oscillation in the winter 2020 extremes in East Asia

The monthly mean AO index (AOI) used is available from the Climate Prediction Center website at www.cpc.ncep.noaa.gov/products/precip/CWlink/daily_ao_index/ao.shtml (accessed 14 June 2021). It is updated monthly following the technology of Thompson and Wallace (2000) with the use of the 1000 hPa GPH (Z1000) fields poleward of 20°N from NCEP-NCAR Reanalysis-1 (Kalnay et al 1996) and the 1979–2000 basic period. Monthly mean indices are estimated by projecting monthly mean anomalies of Z1000 on the loading pattern that was estimated for the basic period as the first EOF of the year-round monthly mean Z1000 anomalies with respect to the corresponding monthly means weighted by square root of the cosine of latitude. All the AO monthly indices have the same loading pattern, and the published indices are not standardized that implies opportunity for estimation of the seasonal mean AOI values. This study focuses on January–March because the extreme seasonal mean AOI was observed during these months in 2020. It is important to note that the linear trend in the JFM AOI during the 41 year period (1979–2019), preceding the analyzed 2020, was negligible, 0.01 σ/year, with 95% confidence intervals being 0.03 σ/year (figure 1).

Zoom In Zoom Out Reset image size

A reasonable estimate of the contribution of the AO and global warming in East Asia would require high resolution long-term data sets covering up to JFM 2020. We selected the ERA5, a fifth-generation ECMWF reanalysis that covers data from 1979 to the present. The ERA5 reanalysis has a horizontal resolution of 0.25° × 0.25° and 37 vertical levels, it is available from the Copernicus Climate Change Service (C3S) Climate Data Store homepage (https://cds.climate.copernicus.eu/cdsapp#!/search?type=dataset&text=era5). The JFM mean temperature variables we use are seasonal means of daily mean (T2m), maximum (T2max), and minimum (T2min) temperature, obtained based on 1 h data (Hersbach et al 2018). Monthly data are additionally used in supporting analyses: SLP, sea surface temperature (SST), wind at 850 hPa surface, surface net shortwave and long-wave radiation (Hersbach et al 2019a, b). The monthly mean snow cover extent (SCE) data were derived from the Rutgers University Global Snow Laboratory data meshed on the irregular 88 × 88 grid.

The local manifestations of global warming as well as the AO impact on temperature essentially differs from each part of East Asia (figure 2). Particularly, the grid-points featuring statistically significant temperature trends are mainly concentrated in the southern part of East Asia and the adjacent seas (figure 2(a)), whereas the AO influence is the strongest in the northern half of the East Asian region (figure 2(b)). Therefore, we divide the whole East Asia region into three subregions allowing more or less spatially homogeneous quantification of the contributions of global warming and the AO to the temperature extremes. Northern East Asia (NEA) is a subregion of comparatively low local temperature trends (LTTs) and of the strongest temperature response to the AO, mainly caused by variations in heat advection associated with the AO. Eastern East Asia (EEA) is a subregion combining both significant temperature trends and strong affect by the AO via its influence on the EAWM. Southern East Asia (SEA) is a subregion of mainly significant LTTs and insignificant correlations of grid-point temperatures with AOI.

Zoom In Zoom Out Reset image size

We examined anomalies of the winter 2020 and the contributions of the AO and global warming to these anomalies with respect to climatology from 1979 to 2019. Global warming is characterized by the trend in globally averaged temperature or, that is the same under linear constraint at least, globally averaged LTTs. However, the LTTs, contributing to the global one, essentially differ from each other (e.g. IPCC 2014, figure 1). So that, in our study, a grid-point manifestation of global warming was represented by a 1979–2019 LTT at this grid-point. Our preliminary analysis of the shape of the trend line for three selected subregions of East Asia with the use of linear, quadratic, and cubic polynomial approximations shows that for the analyzed 41 year period the linear approximation is the most appropriate based on F-value assessments. This result is supported by previous studies demonstrated that the non-linear trend with rapid temperature increase in the 1970s and restrained in the 2000s was the result of overlapping the global trends with natural variability, (Meehl et al 2009, Zhou and Tung 2013). Furthermore, Zhou and Tung (2013) showed that linear trend assessments for different recent time intervals within the past 100 years vary insignificantly when overlapping variability was removed. For East Asia, the temperature trend nonlinear approximations underestimate the 2020 temperature value by several tenths of Kelvin as compared with linear for all three temperature characteristics, T2m, T2max, and T2min (figures 3(a)–(c) and S1 (available online at stacks.iop.org/ERL/17/065010/mmedia)). So that, our assessments could be considered as an upper bound of the global warming contribution for East Asia in JFM 2020.

Zoom In Zoom Out Reset image size

Son et al (2012) documented nonlinearity in the temperature-AOI relationships. For the high and low AOIs some kind of 'saturation' in temperature response occurs that results in overestimation of the temperature values associated with the high and low AOI values under linear constraints. Therefore, we performed preliminary analysis for East Asia and selected the cubic polynomial function as the most appropriate to characterize the temperature-AOI relationships (figures 3(d)–(f) and S2), with cubic regression coefficients being estimated separately for each grid-point.

Since the linear trend in the AOI was negligible in 1979–2019, we consider the AOI and the grid-point linear temperature trend statistically independent during this period (we discuss this assumption in section 4.1). In our study, the LTT and AO contributions are estimated by regression, with the training period being 1979–2019, with 2020 being the target year. For each grid-point we derive a regression equation for the LTT contribution using the grid-point original temperature time series as predictand and time as predictor. Then, we independently derive the cubic regression equation for the AO contribution on the grid-point linearly detrended time series, with the linear trend, although negligible, being subtracted from the AO indices as well. Significance of the regression coefficients was assessed by Student's t-statistic accounting for the effective series size (Bretherton et al 1999). Since regression coefficients may be positive and negative, the appropriate confidence level was set at the 95% in two-tailed tests.

Following recommendations of Wilks (1995, p 311) we assess similarity between the patterns of observed temperature anomalies in JFM 2020 and patterns of the global warming and AO contributions by spatial correlation coefficients. Accuracy of our statistical model is assessed by the root mean squared error (RMSE) between the JFM 2020 observed and estimated temperature fields. We compare the accuracy of representation of the observed temperature field by different estimated fields (the AO, LTT, AO + LTT contributions) by the mean squared skill score (MSSS), recommended by Murphy (1988):

$$\begin{equation*}{\text{MSSS}} = \left( {{\text{MS}}{{\text{E}}_{{\text{ref}}}} - {\text{MS}}{{\text{E}}_{{\text{anl}}}}} \right)/{\text{MS}}{{\text{E}}_{{\text{ref}}}}\end{equation*}$$

where MSE is the mean squared error of the observed temperature field representation by an estimated field, with the ${\text{MS}}{{\text{E}}_{{\text{anl}}}}$ corresponding to the analyzed estimated field and the ${\text{MS}}{{\text{E}}_{{\text{ref}}}}$ to the reference one. The MSSS is positive if the analyzed estimated field is closer to the observed one than the reference estimated field.

The statistical significance of the spatial correlation coefficients and MSSS values was assessed using a Monte Carlo resampling approach (Wilks 1995). To account for the spatial correlation structure, we use the moving blocks procedure (Wilks 1997). We estimated p-values by computing 1000 coefficients (MSSS values) between fields of randomly scrambled blocks of observations and unchanged contribution fields. We considered the spatial correlation coefficient (MSSS) significant at the 97.5% confidence level in one-tailed test if the p-value was below 2.5%.

As a preliminary analysis to provide a broader understanding of the results for East Asia we examine the AO and LTT contributions to the observed T2m anomalies of JFM 2020 on a global scale. Figure 4(a) shows an exceptionally warm winter in Northern Eurasia (anomalies exceed 5 K), with East Asia anomalies exceeding 2 K, positive anomalies of up to 2 K over south-eastern North America, and negative anomalies of about −3 K over Greenland and Alaska. The AO mainly contributes positively over western (exceeding 5 K) and eastern (up to 5 K) Northern Eurasia and negatively over Greenland (up to −5 K). This spatial pattern of the AO contribution resembles the typical pattern of T2m anomalies associated with the positive AO events, although with stronger anomalies. The contribution of LTT is not spatially uniform. It is the largest over the Eastern Arctic (up to 5 K) while over the continents it is mostly less than 2 K. Areas of the largest underestimation of the JFM 2020 T2m anomalies are Western Siberia (residuals of up to 5 K) and the Labrador Peninsula (residuals of up to 4 K). The largest overestimation is over the Arctic Ocean (residuals are less than −5 K) where both the AO and LTT contribute (incorrectly) positively based on the historical relationships.

Zoom In Zoom Out Reset image size

Table 1 shows the portion of the JFM 2020 observed anomalies congruent with the AO and global warming. For whole East Asia, the average AO contribution accounts for 58% of T2m anomalies, while the LTTs contribute 32%, together they explain 90% of the observed anomalies. Over NEA, the AO portion increases up to 78%, while the global warming contribution accounts for 32%. Summation of these two contributions leads to 10% overestimation of the observed anomalies and essential negative residuals. EEA demonstrates results closest to the observations, 94% of the observed anomalies are congruent with the joint AO and global warming contribution, with the AO accounting for 66% and LTTs for 28%. The poorest results are for SEA, 45% of jointly explained portion of the observed anomaly, with the AO portion being 3% and the global warming portion being 42%. In general, our statistical model, implying contributions by only the AO and global warming, is appropriate for East Asia on regional scale and on the subregional scale for NEA and EEA, for which residuals are about 10% of the observed anomaly. For SEA, where the AO influence is uncertain (figure 3), the only LTTs underestimate anomalies and remain large residual discussed in section 4.2. Results for T2max and T2min resemble those for T2m and are shown in supplementary material (table S1).

The positive T2m anomalies exceeding 2 K over the land and 1 K over the seas span almost whole East Asia (figure 5(a)), with new records being established in the northernmost NEA (anomalies exceed 5 K in the Lena-Aldan interfluve), most of EEA (anomalies of up to 4 K), and coastal areas of SEA, (anomalies of up to 3 K). The AO contributes positively (1 to 5 K) over NEA and EEA (figure 5(b)). However, the AO positive contribution over SEA is negligible and even locally negative as a result of the low coefficient of determination (<0.1) between observed and estimated T2m (figure 2(b)). The LTTs contribute positively over whole East Asia, with contribution of 1 to 2 K spanning SEA and parts of NEA and EEA (figure 5(c)). Regionally, the AO contribution is the strongest over NEA and weakest over the SEA, while LTTs' is relatively strong over SEA.

Zoom In Zoom Out Reset image size

As shown in figure 5(d), the statistical model underestimates observed anomalies over SEA (residuals of up to 2 K) and over the Lena-Aldan interfluve and Manchuria in NEA (residuals of up to 3 and 2 K). Meanwhile, in the rest of NEA, residuals, mainly negative, are quite small and randomly dispersed over land while the main area of the negative residuals (overestimation) is over the adjacent seas. Over EEA, residuals do not exceed 1 K, with residuals less than 0.5 K spanning the Korean Peninsula and Japanese Archipelago where the observed anomaly extremes achieve 4 K being the highest on record. The detail discussion on the residuals is posted in the section 4.2. For T2min and T2max, the results are similar to those of T2m. Please refer to the supplementary material for details (figures S3 and S4).

Spatial correlation between the observed T2m field and the field estimated using the LTT is 0.38 while that estimated using AOI is 0.72, that indicates that the AO contribution pattern is closer to the observed anomaly pattern than the LTT contribution.

The RMSE for the T2m field estimated using the LTT (1.95 K) is larger than that associated with the AO (1.34 K) that supports the result from the consistency assessment on the superiority of the AO contribution. Superiority of the AO contribution also results from the significant positive MSSS value (0.53) assessing consistency between the observed field and the AO contribution in respect to the LTT contribution. However, the lowest RMSE (1.00 K) is between the observed field and the field combining both the AO and LTT contributions, meanwhile, the significant positive MSSS value (0.45) proves that it is global warming that essentially improves the field estimated based on the AO alone. The field consistency scores for T2min and T2max are similar to those obtained for T2m, see tables S2 and S3 in supplementary material for detailed values. Consequently, the observed strong positive temperature anomalies including extreme anomalies over East Asia could have been achieved only as a combined effect of the extreme positive AO event and global warming.

Our study is based on assumption of statistical independence between AOI and global warming trend over the 41 year training period during which the AOI trend was negligible. Meanwhile, statistical independence during a certain period does not imply physical independence between the AO and global warming. Particularly, at least two physically plausible mechanisms linking the AO and global warming that offset each other have been suggested. The first mechanism has been detailed and supported in the model studies by Shindell et al (2001). Increase of well mixed greenhouse gas concentration in atmosphere results in increase in horizontal temperature gradient between the upper tropical troposphere and lower polar stratosphere that results in enhancement of the polar vortex, that is, the positive AOI polarity. The second suggested mechanism is based on the Arctic amplification, the enhanced Arctic warming as compared with the middle and lower latitudes caused by the positive feedback between the sea ice retreat and global warming. It must result in decrease of the temperature gradient between middle and polar latitudes, enhancement of meridional circulation and weakening of zonal one, that is, in the negative AOI polarity (Cohen et al 2020, and references therein). However, it should be noted that there is no consensus on the second mechanism yet, with model studies disagreeing each other and observed periods being too short for reliable statistical confirmation (Cohen et al 2020). The first mechanism prevailed in the 1960s–1990s, with the AOI trend during 1959–1997 being 0.041 ± 0.032σ/year. Associated significant contribution to the global warming signals during the 1960s–1990s was demonstrated by Thompson et al (2000) and numerous following studies. However, since the 1990s, the AOI trend has been decreasing down to the negative values (Kryzhov and Gorelits 2015), and the AO signal became distinguishable from that of global warming (Cohen and Barlow 2005). This AOI decrease was possibly caused by the second mechanism.

Our training period comprises both the AOI positive and negative trend periods, with the AOI trend for the whole 41 year training period turning out negligible. We should note that external forcings not related to global warming that may also affect the AO have been also suggested, particularly, variations in ultra-violet solar radiation and volcano eruptions (e.g. Shindell et al 2001). Therefore, if the AOI trend were significant, uncertain would be the causes of the trend, and the AO signal would become undistinguishable from that of global warming in an empirical study. Meanwhile, for the period 1979–2019 when the AO trend is negligible, the AO and global warming signals become distinguishable, so do their contributions to the anomalies of JFM 2020.

We have also analyzed status of the SCE of JFM 2020 which may influence on the AO and East Asia temperature at the same time. As has been shown by Juzbašić et al (2021) development of the extreme positive JFM AO event was abnormal in the autumn and winter of 2019/20. It was preceded by enlarged October and November SCE in Eastern Eurasia that tends to result in the negative AO phase rather than positive (Cohen et al 2007, Cohen and Jones 2011, Han and Sun 2018). However, in spite of the extreme positive JFM AO event, because of its abnormal development, the autumn 2019 area averaged SCE anomaly standardized in respect to 1979–2019 was 0.95 in East Asia. The positive SCE anomaly persisted into winter being 0.56 in JFM 2020, weakening the AO-induced warming impact. Meanwhile, the East Asia JFM SCE negatively correlates with the AO with coefficient −0.44 on the detrended 41 year series.

It is also worth noting that the positive SST anomalies in the equatorial Pacific, with the JFM 2020 Nino3.4 index being 0.5, similarly to the autumn positive SCE anomalies in Eastern Eurasia, were favorable for the negative phase of the AO (Fletcher and Cassou 2015) rather than the extreme positive AO event.

Temperature anomalies are overestimated (negative residuals) by 10% as average over NEA, with the largest negative residuals spanning the seas as a result of underestimation of enhanced cold advection from the continent with anomalous westerlies and prevailing negative SST anomalies (figures 6(a) and S5). The negative residuals in the continental area may be a result from the positive SCE anomaly in NEA, with standardized in respect to 1979–2019 area average SCE anomaly in JFM 2020 being 0.75. On the background of average overestimation, there are two areas of essential underestimation of the observed anomalies those are the northernmost NEA part and the central NEA part over Manchuria. The northernmost area of NEA in JFM 2020 was affected by anomalous westerlies in the western part and southerlies in the eastern part (figure S5), which caused additional heat advection from warmer western domain and from the East Sea (figures 4(d) and 6(a)), not inherent in the classical AO pattern, and could result in the underestimation of anomalies by the model. It should be noted that SST of the East Asia adjacent seas was extremely high in JFM 2020 (figure 6(a)) in contrast to the negative SST anomaly in 1989 when the previous extreme AO event occurred (Juzbašić et al 2021). The area of the positive residuals in Manchuria is a comparatively flat valley bordered in the north with an arc of the trans-Baikal ridges opened to anomalous south-easterlies of JFM 2020 (figure S6). Also, these positive residuals could result from an anomalous radiative forcing during JFM 2020 (figure 6(b)) that corresponds to the enhanced role of radiative forcing in temperature variability in the dryland belt (Groisman et al 2018).

Zoom In Zoom Out Reset image size

Underestimation by 40%–69% of the observed anomalies over SEA, at least in its eastern part where anomalies were extreme, was caused by the anomalous heat advection to the region from the adjacent seas with extremely high SST (figures 6(a) and S5). Given that the tendency of a weaker winter monsoon marked by warmer winters to occur in El Niño years (Ha et al 2012), the extremely high SST could be related to a weak El Niño during JFM 2020. However, both low intensity of El Niño (Nino3.4 index was 0.5) and rather low mosaic distributed correlation between SST in the adjacent seas and the Nino3.4 index, slightly exceeding the 95% confidence level (figure S7), do not imply a strong El Niño impact on East Asia T2m in JFM 2020. Meanwhile, in the north-western continental part anomalous radiative forcing (figure 6(b)) could explain essential underestimation of the observed anomalies by the LTT alone with no additional contribution by the AO.

Analysis of residuals, that is, the portion of the observed anomalies not dominated by global warming and the AO, with both impacts being parameterized on the 1979–2019 series, reveals that the residuals were mainly caused by extremely high SST of the seas adjacent to East Asia, anomalous south-easterlies, and locally anomalous radiative forcing.