Non-Stationary Effects of the Arctic Oscillation and El Niño–Southern Oscillation on January Temperatures in Korea

: In recent decades, extremely cold winters have occurred repeatedly throughout the Northern Hemisphere, including the Korean Peninsula (hereafter, Korea). Typically, cold winter temperatures in Korea can be linked to the strengthening of the Siberian High (SH). Although previous studies have investigated the typical relationship between the SH and winter temperatures in Korea, this study uniquely focused on a change in the relationship, which reﬂects the inﬂuence of the Arctic Oscillation (AO) and El Niño–Southern Oscillation (ENSO). A signiﬁcant change in the 15-year moving correlation between the SH and the surface air temperature average in Korea (K-tas) was observed in January. The correlation changed from − 0.80 during 1971–1990 to − 0.16 during 1991–2010. The mean sea-level pressure pattern regressed with the temperature, and a singular value decomposition analysis that incorporated the temperature and pressure supports that the negative high correlation during 1971–1990 was largely affected by AO. This connection with AO is substantiated by empirical orthogonal function (EOF) analysis with an upper-level geopotential height at 300 hPa. In the second mode of the EOF, the temperature and pressure patterns were primarily affected by ENSO during 1991–2010. Consequently, the interdecadal change in correlation between K-tas and the SH in January can be attributed to the dominant effect of AO from 1971–1990 and of ENSO from 1991–2010. Our results suggest that the relative importance of these factors in terms of the January climate in Korea has changed on a multidecadal scale.


Introduction
Unlike the globally decreasing long-term trend in the frequency and intensity of cold surges, record-breaking cold events have increased in Northern Hemisphere middle latitudes in recent decades [1]. These cold events, which have concurrently occurred with Arctic warming and sea-ice melting resulting from anthropogenic global warming, have been termed "Warm Arctic, Cold Continents" (WACC) [2]. It is widely accepted that the WACC pattern occurs as a consequence of a decrease in the meridional temperature gradient owing to global warming and polar amplification; the weakening of the polar vortex has induced intensification of the meridional component of large-scale atmospheric flows that bring the cold continental air from high latitudes to middle latitudes [3]. In addition to the WACC pattern, cold extremes under global warming have been investigated in many studies, especially in middle-latitude areas [2,[4][5][6][7].
It has also been suggested that cold continents in the WACC pattern are affected by low-frequency atmospheric variability rather than by Arctic warming [8], which implies that the WACC pattern could be transient instead of a persistent change. This pattern suggested the possibility of affecting the Arctic climate from the solar cycle [9]. Ural blocking associated with Arctic sea ice loss also appeared to affect cold events on the Asian continent [10][11][12]. Oceanic variabilities with a multi-decadal time scale (e.g. Atlantic multidecadal oscillation), not just atmospheric factors, can also play an important role in inducing cold wave events in mid-latitudes of the Northern Hemisphere [13][14][15][16].
The purpose of this study was to understand the mechanism of cold extremes on the Korean Peninsula (hereafter, Korea). Traditionally, cold extremes in East Asia, including Korea, have been related to a change in the strength and degree of extension of the Siberian High (SH) [17,18]. This study investigated the relationship between the SH and the surface air temperature average in Korea in January. However, unlike previous studies that investigated the typical relationship between the SH and the winter temperature in Korea, this study uniquely focused on the change in the relationship and the mechanism behind this change.
In addition to the SH, several other factors have been investigated to explain the mechanisms of cold extremes in East Asia. Variability in Arctic Oscillation (AO) affects the winter temperature change in Eurasia and has been associated with a change in the strength of SH [19,20] in an opposite phase during 1972-1977 and 1987-1995 [21]. A study discussed that there was a shift in Northern Hemisphere circulation pattern before and after 1976. From 1976 to 1996 there were differences in many regional teleconnection patterns [22]. A recent study demonstrates that the relationship between AO and surface air temperature over Korea for the boreal winter season is not static but changes on the decadal time scale [23]. In addition to the AO, the El Niño-Southern Oscillation (ENSO) affects wintertime temperatures in East Asia.
Being closely related to the East Asia winter temperature, the East Asian winter monsoon (EAWM) and its relationship with Ural-Siberia blocking is also influenced by AO and ENSO, especially through their combined effects, which vary with time [24,25]. Some previous studies have mentioned that the relationship of the EAWM with AO and ENSO both experienced significant interdecadal changes (e.g., [26][27][28]). After 1979, the correlation of the first mode of the empirical orthogonal function (EOF) for EAWM with ENSO was weakened, whereas the correlation of the second mode for EAWM with AO was strengthened [29]. A weakening of both EAWM and SH is correlated with a positive phase of North Pacific Oscillation (NPO), and the inverse is true (strengthening correlated with a negative phase) [26,28,30]. A regime shift in EAWM was reported in association with a change in large-scale circulation patterns [30][31][32][33][34]. The interdecadal variability of the EAWM after the mid-1980s and its possible causes has also been examined [29,[34][35][36]. In an analysis using the fifth phase of the Coupled Model Intercomparison Project (CMIP5) simulation data, the relationships between EAWM and each of ENSO and AO were projected to seesaw in the 21st century; this did not appear in historical simulation data [37]. However, most of these previous studies focused on the non-stationary effects of AO and ENSO on a continental scale wintertime climate.
The objective of this study was to examine the multidecadal impact of the nonstationary effects of AO and ENSO on a local scale (the Korean Peninsula) wintertime temperature. The change in the relationship between the surface air temperature averaged over Korea and the strength of SH is explained by the changing influence of large-scale patterns including AO and ENSO. The data and analysis methods used for this study are described in Section 2. In Section 3, we show that the relationship between the cold extremes in Korea and the strength of SH has changed between two different periods during recent decades. Subsequently, to determine the physical reason for the change in the relationship, a composite analysis of the large-scale flow surrounding Korea is presented. Finally, in Section 4, the conclusions are drawn along with a discussion of the results.

Data and Methods
The main dataset used in this study was the National Center for Environmental Prediction/National Center for Atmospheric Research (NCEP/NCAR) reanalysis [38] from 1949-2017 with a horizontal resolution of 2.5 • × 2.5 • . The monthly mean anomaly time series of surface air temperature, mean sea-level pressure and geopotential height at 300 h Pa were used to investigate the relationship between winter cold temperature in Korea and SH and the underlying mechanism for the relationship change between the periods of 1971-1990 (hereafter, P1) and 1991-2010 (hereafter, P2). This interdecadal change in the relationship was determined with a 15-year moving correlation coefficients time series in January between the monthly mean surface air temperature average for Korea (hereafter, K-tas) and an index for the strength of the Siberian High (SHI). The daily mean temperature was exclusively used to formulate a cumulative distribution function to test if a temperature was statistically distinguishable between the two periods.
The focus of the study was K-tas, defined as the surface air temperature in January averaged over the area 34 • -43 • N and 124 • -131 • E, including the Korean Peninsula. Based on the typical relationship of winter temperature in Korea with the strength of the SH, our analysis focused on their correlation over time. The strength of SH was determined from the SHI, which is defined as the mean sea-level pressure anomaly averaged over the region of 40 • -60 • N and 80 • -120 • E [39].
To examine the underlying mechanisms for the change in correlation between K-tas and SHI, the large-scale flow surrounding Korea was analyzed in terms of mean sea-level pressure, which can be characterized in the regions with the AO index as defined by [40], the index of Aleutian Low Intensity (ALI) as defined by [41], and the ENSO index. However, the Nino3.4 index was used to represent ENSO in the subsequent analysis. The Nino3.4 index, defined as the anomaly of sea surface temperature averaged over the area of 5 • S-5 • N and 120 • -170 • W, is computed by using the ERSSTv5 (Extended Reconstructed Sea Surface Temperature version 5) dataset [42]. In addition to the climate variability indices, the climatological means in the two periods were compared for surface air temperature and mean sea-level pressure. A multivariate analysis tool called singular value decomposition (SVD) [43] was also applied to the temperature and pressure. In addition to the large-scale flow at the surface level, the geopotential height at 300 hPa was analyzed using EOF to examine the upper-level change in the flow.

Correlation between the Winter Temperature in Korea and SH
The time series of the K-tas and SHI are presented for January along with their correlation coefficients using a 15-year moving window ( Figure 1). The mean of the correlation coefficients for the entire period is −0.55, indicating that an increase in SHI would lead to increasingly severe winter temperatures in Korea [17,[44][45][46].
Notably, the correlation coefficients considerably vary with time, showing a maximum of −0.86 in 1984 and a minimum of −0.04 in 2000. The moving correlation coefficients of the time series appeared to be recurrent, with a period of approximately 40 years; however, the data length was too short to affirm this observation. In contrast, we focused on the period centered approximately on the year 1990, when there was a remarkable change in the correlation: P1 from 1971 to 1990 and P2 from 1991 to 2010. During P1, the correlation coefficient was −0.80, meaning that the January temperature in Korea was closely related to SHI. During P2, the coefficient was −0.16, indicating that the contribution of SH to the temperature in Korea became insignificant. It is worth mentioning the reason why we paid attention only to January. We found that the correlation coefficient between K-tas and SHI is quite different each month in winter. The correlation for January decreases significantly from P1 to P2, whereas the correlation for December and February increase slightly (not shown here). As the result, the correlation for the winter averaged (DJF average) appears to decrease slightly. So, we decided to focus our interest only on January.  Figure 2 presents a map of the correlation coefficient between K-tas and the anomaly of mean sea-level pressure calculated at each grid point. Overall, the correlation was negative over the middle latitude continental area, thereby ensuring a traditional perspective that winter temperatures in Korea are influenced by the dry and cold air divergent from the SH in the north [17,44,47]. Positive correlation over the North Pacific was also notable, and this may have bought warm and humid air to Korea along the border of the Aleutian Low. The correlation pattern during P1 and P2 resembled the negative and positive patterns of the AO, respectively. In the following sub-sections, more in-depth analyses are used to explain this AO-like correlation pattern and the change from P1 to P2. Related to the winter temperatures in Korea, a negative AO event can have a significant influence on SH [19][20][21]. However, when interpreting such a correlation map, we must be cautious that the opposite correlations over SH and the Aleutian Low do not necessarily affect the temperature simultaneously. However, the significant correlations over the regions certainly influenced the winter temperature in Korea during these periods, especially in P1 with higher correlations. The green dots denote statistical significance with a 95% confidence interval. Figure 2 confirms that the considerably lower correlation between K-tas and SHI during P2, as shown in Figure 1, resulted from a weakening of the correlation between K-tas and the pressure anomaly, rather than from a cancellation effect between the positive and negative correlation regions (Figure 2b). The positive and negative peaks in the correlation maps were reduced from −0.82 and 0.72 during P1 to −0.55 and 0.46 during P2.

Time-Mean Anomaly Pattern of Temperature and Mean Sea-Level Pressure
The change in correlation between K-tas and SHI may have also changed the winter climate in Korea. To determine if the time series of the temperature averaged over Korea were distinctive from each other for the two periods, a Kolmogorov-Smirnov test was conducted following [48], as shown in Figure 3, in which the cumulative distribution function of the average daily mean surface air temperature anomalies for Korea are presented. To eliminate global warming effects, the temperature anomalies were detrended beforehand. The test resulted in the distribution function for the period P1 being different from that for the period P2 at a significance level of 0.01; however, the former was statistically identical to that for the entire period, although it did not meet the significance level of 0.05. The distribution functions obtained for the analysis periods indicated that the winter in P1 was certainly colder than the winter in P2, which supports the theory that a higher negative correlation of K-tas with SH may bring increasingly severe cold weather to Korea. The cold winter in P1 with the high correlation between K-tas and SHI was spread throughout large parts of the Asian continent, except for mainland China (Figure 4b), and it became colder than climatology (Figure 4a). During P2 with the low correlation, the winter climate was warmer in most areas of Asia, including Korea, compared with the winter in P1, as well as to the normal state throughout the entire period [6,34]. These temperature changes in Asia during the periods contradicted the global warming rate, as this rate was the highest during P1 in the 20th century but was the lowest during P2, the period also known as "the global warming hiatus" (1998-2012) [5,7]. These contradictory results suggest that the January temperature in Korea during these periods may have been influenced by regional factors, one of which may have been a change in the correlation between K-tas and SHI instead of global warming.
During P1, the SH was weakened and the Aleutian Low was strengthened, as represented by the negative anomaly in Figure 5b. During P2, the SH was also weakened, but the change was not significant. In contrast, the negative anomaly over the Arctic Ocean peaking over the Barents Sea and the positive anomaly over the Bering Sea displayed the opposite pattern of the temperature anomaly (Figure 4c).  These pressure and temperature anomaly patterns are not in accordance with the correlation pattern in Figure 2. Thus, they are insufficient to explain the correlation change between the two periods. Consequently, the time-mean distributions of temperature and mean sea-level pressure anomalies were not necessarily consistent with the correlation map for the anomalies.

Surface Air Temperature and Mean Sea-Level Pressure
To determine the reason for the correlation change between the two periods, spatial and temporal variations for the surface air temperature in Korea and the mean sea-level pressure in Asia were analyzed simultaneously using a singular value decomposition (SVD); the results are shown in Figure 6. Notably, the pressure pattern of the first mode resembles the correlation map of opposite phases for the two periods. The first mode's contribution to the total variability was 98.5% in P1 and 88.0% in P2. The correlation coefficient between the principal component time series for the temperature anomaly in Korea and the pressure anomaly in Asia was also reduced from 0.82 in P1 to 0.56 in P2. Although a cold tongue of air was elongated to the south during P1 and brought a severe cold event to the Korean Peninsula, this phenomenon was flattened during P2, resulting in a warmer winter. The cold tongue during P1 appeared to have occurred owing to the continental high from the north over the Asian continent being confronted with a marine low that developed over the middle-latitude North Pacific. During P2, the continental high moved to the northwest and the marine low became weaker. Although the continental high and marine low here are decomposed components in the SVD analysis, they contribute to most of the total variability and can be considered as anomaly patterns of the SH and the Aleutian Low, respectively. Subsequently, the reduction in the correlation between K-tas and SHI can be attributed to a shift of the SH to the northwest. However, it is unclear how the shift in the SH was related to the weakening of the Aleutian Low during P2.
Unlike the first mode, the second modes resulting from the SVD analysis for P1 and P2 were similar, and their principal component time series revealed the same correlation coefficient of 0.7 (Figure 7). Although their contribution to the total variability changed from 1.3% in P1 to 9.3% in P2, based on the similarity in the second mode, the first mode was the primary reason for the change in correlation between K-tas and SHI. Tables 1 and 2 compare the correlations between the AO and Nino3.4 indices and the SHI and ALI. The correlation of SHI with AO considerably reduced from −0.43 in P1 to −0.03 in P2, and the correlation of ALI with Nino3.4 significantly intensified from −0.31 in P1 to −0.70 in P2; this confirms that the relative importance of the factors affecting winter climate in Korea changed during these periods. Consequently, the interdecadal change in correlation between the surface air temperature in Korea and the mean sea level pressure in January can be attributed to the stronger effects of AO during 1971-1990 and ENSO during 1991-2010.

Geopotential Height at 300 hPa
In addition to the cold spells in East Asia, including Korea, that resulted from the development and expansion of SHI [17,18], an upper-level wave train and blocking high have recently been attributed to extreme cold weather in Korea [45,46]. Thus, the geopotential height at 300 h Pa was analyzed with EOF.
In Figure 8, the first mode in P1 and P2 resembles the AO pattern known as the opposite phases of the air pressure anomaly between the Arctic region and the middle latitudes [49]. The first mode's contribution to the total variability was 36% in P1 and 26% in P2, which was similar to the contribution change of the first mode of mean sea-level pressure in the SVD analysis with surface air temperature ( Figure 6). The second mode in P1 appears to be related to the NPO [30], known as the second EOF mode of the mean sea-level pressure. During P2, the Aleutian Low was dominant. This change in the pattern over the North Pacific between the two periods is consistent with a previous study, in that the southern lobe of the NPO pattern has shifted eastward since 1990 in association with ENSO [50]. These upper-level patterns in P1 and P2 shown in the second EOF mode reveal two distinct features in the North Pacific: mean sea-level pressure anomalies forced by North Pacific Gyre Oscillation (NPGO), resulting in NPO and Pacific Decadal Oscillation (PDO) [51] associated with variability in ENSO [52], resulting in a strengthening of the Aleutian Low.
To confirm that the results of EOF analysis with the geopotential height at 300 h Pa can be explained by surface-level features, the correlations of the geopotential height with AO and Nino3.4 indices are shown in Figure 9. Remarkably, the correlation map with the AO index is similar to the first mode eigenvector for both periods (Figure 8a), whereas the correlation map with the Nino3.4 index is similar to the second mode eigenvector. The highest positive correlation occurred over the Northwest Pacific coast near the Korean Peninsula, with a correlation coefficient of 0.74 for P1 and 0.73 for P2 (Figure 9a,b). Based on the correlation between the surface air temperature in Korea and the mean sea level pressure displaying a typical AO pattern (Figure 2), this result suggests that the winter temperature in Korea is highly affected by AO from the surface level to this upper level at 300 h Pa. The pressure pattern of negative AO in P1 is especially known to cause the intensification of a blocking high over the Ural Mountains in January [7], followed by the development of an atmospheric trough over East Asia [19] that can bring cold air inflow from the polar region to middle-latitude areas including Korea.
The correlation with the Nino3.4 index peaked with a negative maximum at the Aleutian Low over the North Pacific showing correlation coefficients at −0.61 in P1 and −0.77 in P2 (Figure 9c,d). The stronger correlation over the region in P2 was consistent with the Aleutian Low having been strengthened during El Niño events in boreal winter [53].

Summary and Conclusions
In this study, the correlation between the average surface air temperature in Korea and the SHI considerably changed from −0.8 during 1971-1990 to −0.16 during 1991-2010. Both the mean sea-level pressure pattern regressed to the temperature and SVD analysis with the temperature and the pressure supports that the high negative correlation during 1971-1990 was largely affected by AO through its interdecadal change. During this period, the SH and Aleutian Low were strengthened with an expansion to middle latitudes. Cold air from the north along the strengthened SH brought severe cold to Korea more frequently; this is a typical causality of extreme cold events in this region. However, even if the Siberian High does not develop, there is a possibility that extremely cold winters have occurred repeatedly throughout the Northern Hemisphere, including the Korean Peninsula. This connection with AO was validated by the EOF analysis with an upper-level geopotential height at 300 h Pa.
EOF analysis results provide additional insights into winter temperature and pressure patterns in Korea during 1991-2010; specifically, they were most affected by the ENSO through teleconnection from low latitudes. During this period, the January temperature was warmer than that during P1.
Although there have been studies that have investigated the influences of AO and ENSO on the winter climate in Korea, this study uniquely focused on the change in the factors determining the winter temperature by means of moving correlation. The changes in the two periods may have resulted from multidecadal variations related to AO and ENSO or their combination. In the future, it will be necessary to reveal more underlying mechanisms of the non-stationary effects of AO and ENSO. To do so, we require simulations using a sophisticated climate model with proper experimental design, along with the analysis of more long-term observational datasets.