North Atlantic sea surface temperatures (SSTs) exhibit a multidecadal fluctuation pattern (e. g. , Schlesinger and Ramankutty, 1994; Delworth and Mann, 2000; Zhang and Deluorth, 2005; Ting et al. , 2011) that is referred to as the Atlantic Multidecadal Oscillation (AMO), which has a period of 65-80 years and an amplitude of 0.4°C, ( Kerr, 2000). The amplitude is about half the standard deviation of the annual mean SST. The AMO is primarily the surface signature of the Atlantic Meridional Overturning Circulation (AMOC) and is, therefore, often viewed as an observable fingerprint of AMOC variability over decadal timescales ( Knight et al. , 2005; Yu et al. , 2009; Msadek et al. , 2011). Climate system modeling studies suggest that the AMO is an internal variability mode associated with the AMOC ( Delworth and Mann, 2000; Sutton and Hodson, 2005; Knight et al. , 2006; Knight, 2009), although the relationships between the two have been found to vary in different models ( Medhaug and Furevik, 2011), and external solar and volcanic forcing can play a role in modulating the AMO phase ( Otterå et al. , 2010; Chylek et al. , 2011; Booth et al. , 2012). The AMO has been found to have a considerable role in modulating the global climate. It is associated with the North American and European summer climates ( Sutton and Hodson, 2005), Atlantic hurricanes ( Goldenberg et al. , 2001; McCabe et al. , 2004; Zhang and Delworth, 2006), African Sahel rainfall ( Zhang and Delworth, 2006), and the South Asian summer monsoon ( Goswami et al. , 2006; Lu et al. , 2006; Li et al. , 2008b; Feng and Hu, 2008; Wang et al. , 2009; Luo et al. , 2011).

Recent studies have suggested that the East Asian climate is significantly connected with the AMO (e. g. , Li and Bates, 2007). In winter the AMO is not only positively correlated with East Asian surface air temperatures (EATs), but also acts as a forcing agent by weakening the East Asian winter monsoon ( Li and Bates, 2007; Wang et al. , 2009; Sun et al. , 2011). Considering its quasi-periodicity, the AMO can be used to predict interdecadal variations in East Asian surface air temperatures ( Li et al. , 2009). In summer, the AMO may cause more rainfall in the middle and lower reaches of the Yangtze River valley ( Lu et al. , 2006; Wang et al. , 2009).

Despite the above connection, an inconsistency between the AMO and EATs phases can be seen. From observations, East Asia began to warm in the mid-1980s but the AMO did not enter a positive phase until the early 1990s (cf. Fig. 1a here with Fig. 3a in Wang et al. , 2009, and see Fig. 1 in Li et al. , 2009). A substantial warm period was seen in East Asia between the mid-1920s and the late-1940s, but the AMO was then in a positive phase, with a delay only between the 1930s and 1950s. Therefore, it is necessary to scrutinize their lead-lag connection, and that is the primary motivation for this study. Considering the substantial role of the Pacific Decadal Oscillation (PDO) ( Mantua et al. , 1997) in modulating the East Asian climate ( Zhu and Yang, 2003; Yang et al. , 2004; Li et al. , 2006), as well as its significant association with the AMO ( d’Orgeville and Peltier, 2007), a parallel study of the connection between EATs and the PDO is also made.

The observational SST dataset is the Kaplan Extended monthly SST ( Kaplan et al. , 1998) for the period 1870- 2010. The AMO index is defined as the detrended and nine-year running filtered average annual mean SST anomalies (SSTAs) in the North Atlantic basin (0°N- 60°N, 75°W-7.5°W) ( Enfield and Trimble, 2001). This did not show any obvious differences from using seasonal mean SSTAs. The PDO index is defined as the leading principal component (PC) of monthly SST anomalies in the North Pacific Ocean, poleward of 20°N ( Mantua et al. , 1997). The monthly mean global average SST anomalies were removed to separate this pattern of variability from any global warming signals that may be present in the dataset. The PDO index series used was downloaded from http://jisao. washington. edu/pdo/PDO. The monthly global land surface air temperature for 1901-2006, on a 0.5° × 0.5° grid was obtained from the Climate Research Unit Surface Temperature 3.0 (CRU TS 3.0) dataset ( Mitchell and Jones, 2005). East Asian surface air temperatures were quantified as an index calculated as the mean over the domain (22.5°N-45°N, 100°E-125°E), as was done by Wang et al. (2009). The variable air temperature was filtered and detrended because we were only concerned with decadal variability. Monthly mean air temperatures from the National Centers for Environmental Prediction/National Center for Atmospheric Research (NCEP/ NCAR) reanalysis ( Kalnay et al. , 1996) for 1948-2012, on a 2.5° × 2.5° grid, were used to supplement the data because of a lack of records for recent years in the CRU TS 3.0 dataset. Correlation analysis was the primary analysis method used in this study.

Figure 1a shows a comparison of the EATs, AMO, and PDO indices, based on annual means. The EATs evolved, overall, in accordance with the AMO, with a positive simultaneous correlation (with a correlation coefficient of 0.61). Despite this correlation, close scrutiny suggested that their variations were not perfectly simultaneous. For example, the AMO warm phase over the 1930s-1950s occurred just after a period of warm EATs, with a lag of some years. Also, East Asia began warming from the mid-1980s, but the AMO did not enter a positive phase until the early-1990s. The EATs-AMO correlation (with a correlation coefficient of 0.61) was much stronger than the EATs-PDO correlation (with a correlation coefficient of 0.38), and the simultaneous correlation between the PDO and the AMO was much less significant (with a correlation coefficient of 0.01). There are also lead-lag connections between EATs and the PDO and between the AMO and the PDO, which can clearly be seen in Fig. 1b.

Figure 1b shows the evolution of the correlation coefficients and the different lead-lag times of up to 30 years. This 30-year period corresponds to almost half of the AMO period. When the AMO led EATs by 17 to 30 years (the solid line curve in Fig. 1b), their correlation was significant and negative (with a correlation coefficient of less than -0.58). However, when the AMO lagged EATs by up to 12 years, the correlation was positive. In particular, when the AMO lead period was 25 years, the negative correlation reached a maximum and when the AMO lag period was six years the positive correlation reached a minimum.

A similar lead/lag correlation existed between the AMO and PDO indices (the dashed curve in Fig. 1b). When the AMO led the PDO by 10 to 20 years their correlation was negative, and while the AMO lagged the PDO by 15-25 years their correlation was positive. A similar result was obtained by d’Orgeville and Peltier (2007). The correlation between the PDO and EATs (the dotted curve in Fig. 1b) reached a positive maximum when the PDO led EATs.

All of the above correlations are summarized in Table 1. The AMO was most negatively correlated (with a correlation coefficient of -0.91) with EATs when it lagged by a period of 24-28 years, and EATs strongly positively correlated with the AMO (with a correlation coefficient of 0.72) when it led by five-seven years. The PDO was strongly positively correlated with the AMO (with a correlation coefficient of 0.71) when it led by 19-21 years, but strongly negatively correlated (with a correlation coefficient of ~ -0.84) when it lagged by 16-18 years. The PDO was positively correlated with EATs (with a correlation coefficient of ~ 0.67) when it led by 13-15 years, but negatively correlated (with a correlation coefficient of ~ -0.76) when it lagged by 24-26 years.

These lead-lag correlations are illustrated further in Fig. 2, in which each index is shifted to give a direct phase correspondence or anti-correspondence with its counterpart. The EATs phase almost anti-corresponded with the AMO when it was shifted back by 26 years. Similarly, the AMO phase almost anti-corresponded with EATs when it was shifted back by 15 years (not shown). However, the PDO and the AMO phases evolved in a consistent manner when the PDO was shifted back by 17 years, particularly during recent decades.

The possibility that the lead-lag correlation derived from the annual mean evolved with a seasonal cycle was studied because the East Asian climate exhibits significant seasonality. Figure 3 shows a comparison over the four seasons. An overall similarity with the annual mean could be seen (cf. Fig. 3 with Fig. 1b), suggesting that there was little seasonality in the above EATS-AMO-PDO lead-lag correlations. Little seasonality was evident even in the AMO-PDO relationship. This can be seen from the almost homogeneous evolution in all four panels in Fig. 3 (dash line), which is consistent with the season-independent features of the two oceanic modes. From the simultaneous correlations (lag 0), it appeared that EATs positively correlated with the PDO or AMO over a whole year. This is in agreement with a previous finding that the positive phase AMO corresponds with the warmer Asian Continental cycle, giving a weaker winter monsoon but a stronger summer monsoon ( Li and Bates, 2007; Lu et al. , 2006; Wang et al. , 2009).

Table 1

Table 1

Table 1 The maximum lead-lag correlation coefficients with different lead-lag terms. Lagging Leading AMO PDO EATs AMO -0.84 (16-18 yr) -0.91 (24-28 yr) PDO 0.71 (19-21 yr) 0.67 (13-15 yr) EATs 0.72 (5-7 yr) -0.76 (24-26 yr)

Table 1 The maximum lead-lag correlation coefficients with different lead-lag terms.

Figure 1

Figure 1 A comparison of (a) the evolution of EATs, AMO, and PDO indices and (b) their lead-lag correlations.

We conducted a simple lead-lag correlation analysis toinvestigate the connection between the AMO and EATs using instrumental records from the past 100 years or so, and found the strongest positive correlation when EATs led the AMO by a period of five-seven years. This result not only supports a previous finding that the AMO is positively simultaneous correlated with EATs ( Li and Bates, 2007; Wang et al. , 2009; Li et al. , 2009) , but also provides a new insight on their connection.

The positive AMO-EATs correlation can be used to predict the East Asian climate. Considering the quasi- periodicity of the AMO and its dominant role in modulating the decadal variability in the East Asian climate, Li et al. (2009) projected that around the early 2020s the AMO-related cooling caused by the internal variability of the climate system will offset a portion of the warming induced by anthropogenic greenhouse gases. Therefore, the East Asian warming trend that began in the 1980s will slow down along with the expected transition of the AMO phase from positive to negative around the 2020s. If the result presented here, that EATs lead the AMO by five-seven years, is realistic the slowing of East Asian warming will occur five-seven years earlier than Li et al. (2009) predicted, i. e. , in the early 2010s, regardless of the underlying cause-effect relationship. It is worthwhile noting that cooler years have recently been frequent, including the winters of 2007/08, 2008/09, and 2011/12, in East Asia (Fig. 4) and in the northern hemisphere in general (not shown). Whether this represents a precursor of the AMO entering a cold phase is intriguing but unclear.

We also compared the connection between EATs and the PDO and found a maximum correlation between EATs and the PDO when the latter led the former by 13-15 years. Considering that the PDO entered a cold phase around the late 1990s ( Zhu et al. , 2010), a cooling tendency in EATs is also expected around the early 2010s (based on the EATs-PDO lag correlation). This will overlap with the above AMO-based projected tendency. Nevertheless, attention needs to be paid to this tendency.

Previous atmospheric general circulation model (AGCM) or coupled model studies suggested that the AMO acts as a forcing agent on the East Asian climate, and this has been used to explain the observed AMO- EATs simultaneous correlations ( Li and Bates, 2007; Lu et al. , 2006; Wang et al. , 2009). However, the results presented here reveal a maximum correlation when EATs led AMO by five-seven years. This poses important questions about the cause-effect relationship involved and the underlying mechanisms. There is a weakness in this study caused by the instrumental records used covering just about 100 years, which is less than two AMO cycles. This is insufficient for an investigation into the robustness of the EATs-AMO correlation to be made. Therefore, the lead-lag connection revealed here does not mean that there is a cause-effect relationship.

Even taking the limitations of this study into account, several recent studies have provided indications to explain the connection between EATs and the AMO. Warmer EATs are usually accompanied by anomalous southwesterly winds from the South China Sea and the western North Pacific, and the El Niño event in the tropical eastern Pacific Ocean, particularly during seasons other than summer ( Wang et al. , 2000; Yang et al. , 2007; Li et al. , 2008a). The El Niño-like SSTA tends to shift the Aleutian low to the east and cause low-level winds over the Bering Sea to blow from the east and north off Alaska ( Niebauer, 1998). This results in the intensification of fresh water outflow from the Arctic across the Bering Strait entering the northern North Pacific, causing less fresh water to enter the northern North Atlantic, and favoring anomalous intensification of the AMOC and a positive phase AMO SSTA ( Hu et al. , 2012). Warmer EATs usually correspond with a warmer Indian Ocean ( Wang et al. , 2000; Li et al. , 2008a). The warmth in the Indian Ocean tends to induce a lower-level anticyclone over northern Africa, and, subsequently, a lower-level southerly anomaly over the tropical Atlantic ( Hoerling et al. , 2006), favoring intensification of the AMOC. In brief, tropical air-sea interactions, an atmospheric bridge, and the basic basin connection in oceanic flow may explain the EATs-AMO lead-lag connection to some extent. Similar mechanisms have been proposed in previous studies ( Zhang et al. , 2006; Lu and Dong, 2008). Nonetheless, this issue deserves further study. We will conduct a comprehensive analysis using proxy datasets, such as stalagmite oxygen isotopes ( Tan et al. , 2004), and investigate the underlying mechanisms using coupled model simulations in CMIP5 (the fifth phase of Coupled Model Intercomparison Project) experiments.

Figure 2

Figure 2 A comparison of the EATs, AMO, and PDO indices when one of them is shifted. In (a) and (c) EATs are shifted back by 25 and 15 years, respectively, and in (b) the PDO is shifted back by 17 years.

Figure 3

Figure 3 As Fig. 1b, but for four seasons: (a) winter, (b) spring, (c) summer, and (d) fall.

Figure 4

Figure 4 The temporal evolution of EATs since 1948: (a) annual mean, (b) winter mean, and (c) spring mean.