1 Introduction

Because of its impact on global climate variability and its high predictability, the El Niño/Southern Oscillation (ENSO) is regarded as the most important phenomenon for seasonal–interannual climate prediction. From October 2014 to March 2015, the Niño3.4 index, referred to as sea surface temperature anomaly (SSTA) averaged over 170°W–120°W, 5°S–5°N, was in a range of 0.5–0.9 °C. At the same time, except for the February 2015, the Southern Oscillation index (SOI), defined as the standardized surface pressure difference between Tahiti and Darwin (former minus later), was in a range of − 0.6 to − 0.9 (http://www.cpc.ncep.noaa.gov/products/CDB/). The values of the both indices exceeded the thresholds for a weak El Niño condition (Trenberth et al. 1998). However, the atmospheric anomalies during the same period did not show typical ENSO like features, leading to the question why atmospheric circulation deviated from a typical response observed during El Niño conditions?

To illustrate this point further, the observed December–January–February 2014/15 (referred to as DJF 2014/15) and Niño3.4 index based regression patterns for DJF mean SST, precipitation rate (Prate) and 200 hPa stream function (S200) anomalies are shown in Fig. 1. The Niño3.4 index regression patterns represent the spatial patterns that, on average, are seen during ENSO winters. For the Niño3.4 index regressed SSTA pattern (Fig. 1, bottom left), the largest anomalies are in the eastern to central equatorial Pacific, and further, are confined to the east of the date line. In contrast, the observed SST anomalies (SSTAs) for DJF 2014/15 in the tropics had their warm center located over the central Pacific and even extended to the west of the dateline. Also, the warm SSTAs in the tropics extended along a circular arch northeastward towards and along the western coast of North America. The strong positive anomalies along the coast of North America largely persisted during 2014–2016 (Hu et al. 2017) and was referred to as “Warm Blob” (Bond et al. 2015).

Fig. 1
figure 1

Upper row: SST (left), 200 hPa stream function (contours in right) and precipitation rate (shadings in right) anomalies observed in DJF 2014/15. Lower row: regressions of DJF SST, 200 hPa stream function and precipitation rate anomalies onto Niño3.4 SST index for the data period 1949/50–2014/15 for SST and stream function, 1979/80–2014/15 for precipitation. Units are °C for SST, 106 m2 s−1 for stream function, and mm/day for precipitation rate

Full size image

For Prate anomaly, the Niño3.4 index regressed pattern (Fig. 1, bottom right) in the tropics is the familiar east–west dipole pattern, with the positive anomalies extending from the eastern equatorial Pacific to the warm pool region and the negative anomalies covering the Maritime continent region and its vicinity, and extending to the South Pacific convergence zone (Peng et al. 2014). The spatial regression pattern of Prate anomaly corresponds well with the SSTA pattern both in shape and sign, indicating a forced response to SSTAs, a fact that has been validated earlier in atmospheric general circulation model (AGCM) simulations (Peng et al. 2000, 2014). The corresponding observed Prate anomaly pattern in DJF 2014/15 is also an east–west dipole pattern, however, with a reversed polarity: positive anomaly is in the western tropical Pacific and the maritime continent area, and the negative anomaly in the central equatorial Pacific. This contrast in Prate anomaly pattern between the observed and expected from ENSO forcing was also noted by Barnston (2015).

Although the difference in Prate anomaly pattern is striking, the observed and regressed S200 anomaly patterns are surprisingly similar in the tropics. For example, they both have an anti-cyclonic pair straddling the equator over the central Pacific. The difference between them is that the observed anomaly pattern is shifted westward about 15° with respect to the Niño3.4 index regression pattern. Furthermore, its zonal extent in 2014/15 was also narrower and evolved to a cyclonic pair over the eastern Pacific. Differences in anomalous circulation pattern also existed in the extratropics. The most important feature is the pattern orientation over the North America. The Niño3.4 index regression pattern is a north–south dipole structure, with an anti-cyclonic anomaly in the north and a cyclonic anomaly in the south, whereas in DJF 2014/15 observations, the anomaly spatial pattern is an east–west dipole structure, with cyclonic anomaly in the east and anti-cyclonic anomaly in the west. From a global perspective, in both cases the anomaly pattern over the North America is part of a wave train emanating from the tropical Pacific, and therefore, the origins for differences in the anomaly pattern may still be in the tropics.

The role of tropical diabatic heating (as inferred from the Prate) in influencing global circulation during ENSO winters has been documented in model experiments (Hoerling and Kumar 2002 and references therein) and in diagnostic analyses (Ting and Hoerling 1993; Peng 1995; DeWeaver and Nigam 2004). The anti-cyclonic (cyclonic) pair straddling the diabatic heating (cooling) in the central equatorial Pacific has been inferred as the forced response to the heating–cooling pair over the equatorial Pacific and with the Rossby wave propagation extending this response into extratropical latitudes (Gill 1980; Sardeshmukh and Hoskins 1988). The canonical ENSO heating–circulation relationship, however, does not seem to be at play for the winter of 2014/15 as the anti-cyclonic pair, instead of associated with the heating, straddles the cooling. This leads to the question as to what drove the SST and circulation anomalies, and what was the dynamics behind the tropical circulation anomalies for winter 2014/15?

Based on the resemblance of the North American atmospheric circulation and surface temperature anomalies of recent winters (2013/14 and 2014/15) to the patterns associated with the North Pacific mode (NPM) of SST variability (Deser and Blackmon 1995), Hartmann (2015a, b) suggested that the NPM was a possible cause for the North American climate anomalies in winter 2014/15. By comparing the observed SSTAs shown in Fig. 1 to the NPM pattern (Fig. 1c in Hartmann 2015a), the similarity between them is indeed obvious, but differences in the eastern equatorial Pacific are also apparent. A La Niña-like cold anomaly in the cold tongue area is a part of the NPM pattern, but it was not found in the observed SSTAs of DJF 2014/15. As the relationship between atmospheric circulation and diabatic heating in the tropics has not been examined, the cause and effect of these SSTAs, and significance of their difference, remains unclear. Moreover, the influence from the long term trends of SST was also not explicitly discussed in Hartmann (2015a, b), which is one of the important features of recent changes in the climate and its variability (Deser et al. 2010; Wu et al. 2011; Ji et al. 2014). Thus, a further analysis is necessary for a better understanding of the circulation and climate anomalies that occurred in DJF 2014/15.

A further question is how the oceanic and atmospheric circulations, particularly the progression of ENSO evolved to the patterns of winter 2014/15. In early 2014, most dynamical models predicted a strong El Niño in winter 2014/15 (http://www.cpc.ncep.noaa.gov/products/NMME/archive). Similarly, the empirical lead–lag relationship between NPM and Niño3.4 index revealed in Hartmann (2015a) also indicated an El Niño in the winter. The failure of the El Niño development in 2014 has been attributed to the easterly anomalies in the central and eastern equatorial Pacific during the boreal summer (Menkes et al. 2014; Min et al. 2015; Zhu et al. 2016; Hu and Fedorov 2016). The easterly anomalies are speculated to be related to negative SSTAs in the southeastern subtropical Pacific (Zhu et al. 2016), which may be associated with the negative phase of inter-decadal Pacific Oscillation (Min et al. 2015; Hu and Fedorov 2016). This speculation may imply some predictability of the easterly anomalies, but further evidences are needed to support the point.

In this study we examine the leading modes of wintertime SST variability by using an empirical orthogonal function (EOF) decomposition procedure, and then assess the relative importance of these modes to the observed climate anomalies of DJF 2014/15 in Pacific and North America. We also examine the evolution of the SST and atmospheric circulation anomalies and address the source of the anomalous easterly winds which impeded El Niño development in 2014. The paper is organized as follows: Sect. 2 describes the data used in the study and the analysis procedure; Sect. 3 presents the results; summary and discussions are given in the Sect. 4.

2 Data and analysis procedures

The observational data used in this study include monthly mean SST, 200 hPa stream function (S200) and wind at 1000 hPa from January 1949 to February 2015, including 66 DJF seasons, and DJF mean precipitation from 1979/80 to 2014/15 for total 36 winters. The SST is from Hurrell et al. (2008), the stream function from NCEP/NCAR reanalysis (Kalnay et al. 1996), the wind at 1000 hPa from the ERA-40 of the European Centre for Medium-Range Weather Forecasts (ECMWF) (Uppala et al. 2005), and the Prate from the CPC merged analysis of precipitation (CMAP) (Xie and Akin 1996).

The model data used in the study are wind at 1000 hPa from a set of AGCM simulations. The AGCM used is the atmospheric component of the NCEP climate forecast system version 2 (CFSv2) (Saha et al. 2014). The simulations are with 18 members from the beginning of 1957 to the middle of 2015 and forced with observed SSTs (Hurrell et al. 2008). Each member in the ensemble of simulation starts from different initial conditions that are taken from the days surrounding and including 1 January 1957.

The analysis procedure begins from an EOF analysis for the 66-winter Pacific SSTAs. The spatial domain for the EOF analysis is the north of 30°S and between 120°E and 80°W, including the tropical and northern parts of Pacific, same as that in Hartmann (2015a). The EOF calculation is based on a covariance matrix, and as a consequence, fewer dominant modes explain more variance. After computing the principal components (PCs), which are the time series associated with EOFs, the corresponding spatial patterns of SSTA are obtained with the regression of the global SSTAs at each grid point onto the PCs. Following the same procedure, S200 and Prate anomaly regression patterns associated with the SSTA modes are also obtained. As the Prate is only available since 1979, the PC time series used for regression is from 1979 onward.

After the decomposition procedure, the relative importance of the SST modes in explaining the observed anomalies of the three variables for DJF 2014/15 is assessed through a stepwise reconstruction procedure. The procedure starts from the most dominant mode, and then successive modes are added at each step, until a spatial pattern resembling the observed DJF 2014/15 anomalies is reconstructed. For SSTA, the EOF modes can completely reconstruct the observed anomalies, because they are the modes of SSTA itself. For S200 and Prate anomalies, however, only a part of variance can be explained by the SSTA EOF modes. As a result, the reconstructed S200 or Prate anomalies cannot be as complete as that for SSTA.

The last part of the analysis is to understand how the SSTAs evolved after the winter of 2013/14. This is accomplished by projecting 3-month mean SSTAs onto the EOF patterns in the domain of the SSTA EOF analysis. The time evolution of projection coefficients summarizes the development of each mode. According to Hartmann (2015a), NPM generally precedes ENSO mode by about 1 year. To confirm the finding, lead-lag correlations between the NPM and ENSO mode obtained in our EOF analysis are also calculated. To further understand the relationship between the NPM and ENSO mode, lagged regressions of 3-month mean SST and 1000 hPa wind anomalies to the DJF NPM time series are also done. The regression maps are expected to shed light on the dynamics of the transition from NPM to ENSO mode. For the purpose of understanding why the El Niño in 2014/15 did not fully develop, we further examined the equatorial zonal wind anomalies in early and middle summer, and used the AGCM simulations to examine the impact of southeastern subtropical Pacific SSTAs to the equatorial zonal wind anomalies and ENSO evolution.

3 Results

3.1 Leading modes of SSTAs and their associated precipitation and circulation patterns

Figure 2 shows the patterns of SST, Prate and S200 anomalies associated with the first three EOF modes of SSTA. The corresponding PCs are displayed in Fig. 3. The first mode (EOF1; Fig. 2, top left), explaining 41% of the total variance of SSTA over the domain, is related to ENSO and is referred to as ENSO mode. The SSTA and associated S200 and Prate anomaly patterns (Fig. 2, top right) are almost identical to those from regressions with Niño3.4 index shown in Fig. 1. The PC value of the ENSO mode for DJF 2014/15 is around 1, indicating the ENSO signal was not negligible.

Fig. 2
figure 2

Regressions of DJF mean SST (left), precipitation rate (shading, right) and 200 hPa stream function (contour, right) anomalies onto the first (upper), second (middle) and third (lower) principal components (PCs) of the SSTAs in the Pacific north of 30°S. Units are °C per standard deviation of the corresponding PC for SSTA, 106 m2 s−1 per standard deviation of the corresponding PC for stream function anomaly, and mm/day per standard deviation of the corresponding PC for precipitation rate anomaly

Full size image
Fig. 3
figure 3

PC 1–3 of the DJF SSTAs in the Pacific north of 30°S and the fractions of their explained total variance. PCs are normalized with their own standard deviation

Full size image

The second mode (EOF2; Fig. 2, middle left), explaining about 10% of the total variance of SSTA over the domain, has its major SST loading in the western and southern tropical Pacific, and also associated with anomalies in the Indian and Atlantic oceans. The corresponding PC2 indicates this mode is related to warming trend in the global oceans, though inter-annual and even multi-decadal variations are also present. The similarity of the EOF2 pattern to the twentieth century SST trend (Deser et al. 2010) confirms its relationship to the warming trends in SSTs. Its associated Prate anomaly pattern (Fig. 2, middle right) is likely a response to the SSTAs with dry (wet) anomalies collocated with cold (warm) SSTAs in the central (western) Pacific. Further, the spatial pattern of the Prate anomaly is similar to that associated with the cold phase of ENSO, but with weaker intensity. The corresponding S200 anomaly pattern in the tropics appears as an east–west pattern, including a cyclonic system towards the south of the negative Prate anomaly and a cross-equator system over the eastern Pacific. According to their location, shape and orientation, the former is likely forced by the diabatic cooling corresponding to the negative Prate anomaly, while the latter seems more complicated. In the northern extratropics, a cyclonic system is centered over Bering sea and with a westward extension towards Mongolia, and a signature of North Atlantic Oscillation pattern is over the North Atlantic and Greenland area. This mode was not part of Hartmann (2015a) analysis, because their data were detrended prior to their EOF analysis.

The third mode (EOF3; Fig. 2, bottom left), explaining about 8% of the variance of SSTA over the domain, is the North Pacific mode (NPM) (Deser and Blackmon 1995). Its SSTA pattern is the same as that in Hartmann (2015a), though our analysis is based on DJF seasonal means alone, not the monthly means over the entire annual cycle. Its large amplitude in the extratropics may suggest a local origin, but it also has tropical loading in the western and central equatorial Pacific. The corresponding Prate anomaly pattern (Fig. 2, bottom right) matches well with the SSTA pattern in the tropics, with positive anomaly over the positive SSTA and negative anomaly over the negative SSTA. Though the SSTAs are much weaker than that in the ENSO mode, for the Prate anomaly, its amplitude reaches almost half of that for the ENSO mode (Fig. 2, shading in right, top and bottom). This may be due to the fact that Prate anomaly is not linearly related to SSTA, and is much more dependent on total SST (Hoerling et al. 1997). In the S200 anomaly pattern (Fig. 2, contours in right and bottom), a wave train starts from the tropical western Pacific, the area of anomalous heating, and then extends across the North Pacific to North America with a ridge along the west coast and a trough over the northeastern part of the continent. The wave train then arches southeast towards the Atlantic, and finally ends at the equator near western Africa. In the tropical eastern Pacific, a cyclonic pair is linked with the diabatic cooling, suggesting that it is forced by the cooling. The time series of this mode (Fig. 3c) is dominated by interannual variability before 1998, but after that, by variations on a lower frequency. It is also noted that the amplitude of the variability becomes stronger in recent decades. The PC values were notably high for DJF 2013/14 (1.75) and 14/15 (2.90), with latter being the highest in the record analyzed.

The fourth mode (EOF4, not shown) is the Pacific decadal oscillation (PDO) mode (Mantua and Hare 2002; Zhang et al. 1997), the same as the second mode in Hartmann (2015a). The reason for the second mode in Hartmann (2015a) to be the fourth mode in our analysis is that it may be due to the difference in data length and removing or retaining the trends, and also because of using monthly vs. DJF data (Wen et al. 2014). The data used in Hartmann (2015a) was from 1900, about 50 years longer than this analysis. However, the PC value for EOF4 in DJF 2014/15 is only 0.25, and therefore, its contribution is small. Other higher modes are either with small PC values for DJF 2014/15 or with patterns having small loading. Therefore, our analysis focuses on the first three leading modes only.

The correlation maps corresponding to the regression maps for each mode were also checked. It is found that most features shown in Fig. 2 are well above the 90% significant level based on a two tailed t-test. We also calculated the PCs and the regression patterns with the data not including winter 2014/15 and compared them with that from the full dataset as shown in Fig. 2, and found differences to be very small.

3.2 Contributions of the leading patterns to the anomalies in DJF 2014/15

Through reconstruction based on the first three leading EOF modes, we assess the contribution of each mode to the observed anomalies in DJF 2014/15. It is noted that SSTA pattern for DJF 2014/15 fits best NPM (Fig. 2, bottom left panel), thus, the reconstruction procedure starts from the NPM (i.e., PC3).

The upper row of Fig. 4 is the reconstructed SST, Prate and S200 anomalies associated with the NPM, that is, the product of PC3 value for DJF 2014/15 and the spatial patterns associated with the EOF3 of SST shown in Fig. 2 (bottom left). Compared to the observed anomalies shown in Fig. 1 (upper row), as expected, the reconstructed SSTA resembles the observation very well, particularly for the warm anomalies over the Pacific domain. However, some differences are also obvious. For example, the strong cold anomaly in the eastern equatorial Pacific in the reconstruction is not found in the DJF 2014/15 observation, and the intensity of the reconstructed warm SSTA in the central tropical Pacific is too weak. For the Prate anomaly, the reconstructed pattern matches well with the observed anomaly not only in the tropics, but also in North Pacific and North America. A major difference is in the intensity in the tropical Pacific, where the reconstructed anomaly is much stronger than the observed. For the S200 anomaly, as described before, the reconstructed pattern is a wave train emanating from the diabatic heating area (associated with the positive Prate anomaly) in the tropical western Pacific, with its features matching well with the observations, particularly over North America. Major differences are that the anti-cyclonic center in the central tropical Pacific is shifted westward, and the intensity of the wave train appears weaker. Pattern correlations of the reconstructed SST, Prate and S200 anomalies to observations are 0.73, 0.55, and 0.69, respectively. Therefore, the contribution of the NPM alone, although prominent, is not adequate to explain the observed anomalies in the tropical and North Pacific in DJF 2014/15.

Fig. 4
figure 4

Reconstructed SST (left), precipitation rate (shading, right) and 200 hPa stream function (contour, right) anomalies for DJF 2014/15 with spatial patterns shown in Fig. 2 and PCs shown in Fig. 3. Upper row is the reconstruction by using NPM (EOF3 and PC3) alone, middle row by using both NPM and ENSO (EOF1 and PC1) mode, and lower row by using all NPM, ENSO and warming trend (EOF2 and PC2) modes. Units are the same as that in Fig. 1

Full size image

Because the explained variance of the ENSO mode is the largest among all EOF modes (Fig. 1) and PC1 value for DJF 2014/15 is around 1.0, the contribution of this mode also needs to be included. The middle row of Fig. 4 presents the reconstructed patterns after adding the ENSO mode. Comparing them with the reconstruction with NPM alone (Fig. 4, upper row) and that from observations (Fig. 1, upper row), we can see that the correspondence with the observed anomalies is improved: (a) the cold SSTAs in the eastern equatorial Pacific disappeared and the intensity of the warm SSTAs increased to match the observations; (b) the intensity of tropical Prate anomalies also reduced to the level in the observation; (c) the westward shift of the anti-cyclonic center also corrected to some extent; and (d) the wave amplitude increased to that in the observation. Improvements, however, were not unanimous, for example, the trough over the northeastern part of North America became weaker, and so was the cyclonic pair in the eastern tropical Pacific. Nevertheless, corresponding pattern correlations of the reconstructed SST, Prate and S200 anomalies to observations improved to 0.86, 0.63, and 0.85, respectively.

The results of the reconstruction by adding the EOF2 are displayed in the lower row of Fig. 4. A discernible improvement for SSTA is in the western tropical Pacific and India Ocean, where the SSTs became a bit warmer. As already indicated by the small PC2 value of 0.5 and the relatively weak circulation and precipitation anomalies, the overall improvement is quite limited. Almost no improvement is reflected in pattern correlations to observations. They are 0.85, 0.65 and 0.83 for SST, Prate and S200 anomalies, respectively.

Overall, the SST, precipitation and circulation anomalies in the winter of 2014/15 can be largely explained by the NPM and ENSO modes. The NPM was the dominant mode in DJF 2014/15 and explains why the atmospheric anomalies did not conform to the typical ENSO response pattern (Barnston 2015). Thus, there is a competition between NPM and ENSO: when ENSO (NPM) is dominant, there will be (not be) strong air-sea coupling in the central and eastern tropical Pacific.

3.3 Connection of temporal evolution of the NPM and ENSO mode

Having identified the components contributing to DJF 2014/15 climate anomalies, we next analyze their temporal evolution. The analysis of the evolution of SST and atmospheric circulation anomalies after winter 2013/14 further points to the importance of the NPM and ENSO mode. Figure 5 presents the projection coefficients of 3-month mean SST of the tropical and northern part of Pacific on the NPM and ENSO patterns (Fig. 2, left). For NPM, the projection coefficient was 1.75 in DJF 2013/14, strengthened in the late winter and early spring; weakened in the summer; strengthened again in early fall and winter, with the projection coefficient reaching 2.9 in DJF 2014/15. The ENSO mode started from a negative territory in DJF 2013/14, turned positive and gradually strengthened, with the coefficient reaching 0.75 in the early summer. Subsequently, it became weaker in the middle summer but gained some strength in fall, reaching maximum amplitude of 1 in winter 2014/15.

Fig. 5
figure 5

Projection coefficients of SSTAs onto the EOF1 (ENSO mode) and EOF3 (NPM) patterns shown in Fig. 2 from DJF 2013/14 to DJF 2014/15 in the tropical and northern Pacific

Full size image

The interesting characteristic of NPM was its persistently strong intensity throughout 2014, and even reaching the maximum since 1949 in DJF 2014/15. The ENSO mode experienced a development to a weak ENSO, but was overshadowed by NPM, particularly in winter season. Per Hartmann (2015a), NPM tends to be a precursor of ENSO mode, and as a further confirmation, we also calculated the lead–lag correlations between the PC1 and PC3 of DJF SSTAs (Fig. 6), and noted that when PC3 leads PC1 by 1 year, their correlation coefficient reaches maximum 0.38, exceeding the 95% significance level using a two tailed t-test. This is also consistent with other analyses on the connection between anomalies in North Pacific and following year ENSO (e.g., Vimont et al. 2003a, b; Alexander et al. 2010; Wang et al. 2012; Pegion and Alexander 2013; Feng et al. 2014).

Fig. 6
figure 6

Lead–lag correlations between PC1 and PC3 of DJF SSTA shown in Fig. 3. The negative (positive) number in x-axis shows the year number of PC3 leading (lagging) PC1

Full size image

To further demonstrate that the NPM precedes the ENSO mode, Fig. 7 shows the lagged regressions of SST and 1000 hPa wind anomalies onto PC3 for five seasons. For 0-month lagged DJF (Fig. 7a), that is, the simultaneous regression, the NPM SSTA pattern is associated with the warmer SSTA in the western equatorial Pacific and with the southerly anomaly to the south of the equator and the southwesterly anomaly to the north of the equator. The southerly anomaly would bring warmer air to the area and the southwesterly anomaly would cause Ekman transport towards the equator. Similarly, the cooler SST in the eastern equatorial Pacific is likely associated with the anomalous easterly, which may enhance upwelling along the equator (Zhu et al. 2016). The SSTAs over North Pacific are consistent with warmer/cooler air advection and the wind driven Ekman transport. For the 1-season lagged MAM(0) (Fig. 7b), SST and wind anomaly has a similar patterns as that in the 0-month lagged (Fig. 7a), but with reduced amplitude. In the following seasons [JJA(0) to DJF(0)] (Fig. 7c–e), westerly anomalies gradually build up over the central equatorial Pacific, causing the SST in the area to warm up through the Bjerknes feedback (Bjerknes 1969) and leading to a SST pattern resembling the central Pacific type of El Niño (Ashok et al. 2007; Kao and Yu 2009). That is consistent with Hu et al. (2012) that an eastward extended (westward confined) westerly wind anomaly along the equatorial Pacific in early months of a year is more favorable for an eastern (central) El Niño development.

Fig. 7
figure 7

Lagged regression of SST and 1000 hPa wind anomalies onto PC3 of DJF SSTA shown in Fig. 3. a–e Are for zero-lag (DJF), 1-season lag (MAM), 2-season-lag (JJA), 3-season lag (SON) and 4-season lag (DJF), respectively. The units are same as that in Fig. 2

Full size image

Because the wind associated with NPM in North Pacific (Fig. 7a) exhibits a North Pacific Oscillation (NPO) pattern (Rogers 1981; Linkin and Nigam 2008) and the features of SST and wind anomaly patterns in DJF(− 1/0) and MAM(0) (Fig. 7a, b) well resemble the Pacific meridional mode (PMM) (Chang et al. 2007; Di Lorenzo et al. 2015), the lead–lag relationship between NPM and ENSO shown in Fig. 7 actually reflects the same dynamics as the seasonal footprinting mechanism (SFM) which has been extensively studied (Vimont et al. 2001, 2003a, b, 2009, 2014; Anderson 2003, 2007; Chiang and Vimont 2004; Chang et al. 2007; Zhang et al. 2009; Alexander 2010; Yu and Kim 2011; Larson and Kirtman 2013, 2014; Pegion and Alexander 2013). In winter, atmospheric circulation associated with NPO generates SSTAs through changing surface heat flux. As the SSTAs persist into following spring and summer, the subtropical portion of the SSTA, in turn, forces a zonal wind stress anomalies along the equator. Thus, the tropical atmosphere–ocean system responds to the zonal wind stress anomalies through coupled dynamics, leading to development of an ENSO-like SSTA pattern in next winter.

Interestingly, even though there was a strong NPM in winter 2013/14, no El Niño fully materialized in late of 2014. Evolution of equatorial SST and 1000 hPa zonal wind anomalies during November 2013–March 2015 is depicted in Fig. 8 [which is similar to Fig. 1 in Min et al. (2015) and Fig. 2 in Menkes et al. (2014) and Hu and Fedorov (2016)]. Following westerly anomalies in the western Pacific from January to April 2014, warm SSTAs developed in the western and central Pacific from February to May 2014, indicating that the SSTA was associated with the westerly anomalies. Warm SSTAs also occurred in the eastern Pacific, with about 2-month delay with respect to their counter parts in the western and central Pacific. These phenomena suggest an early stage of El Niño development. The process, however, was interrupted by easterly wind anomalies that occurred in June and July 2014 in the central and eastern Pacific (Hu and Fedorov 2016). The easterly anomalies pushed warm SSTAs back to the western Pacific thus impeded El Niño development. Though the westerly anomalies emerged again from September to October 2014, they were confined in the western Pacific, and were only able to give rise to a central Pacific type of weak El Niño near the end of 2014.

Fig. 8
figure 8

Observed time-longitude variation of SST (°C, shading) and 1000 hPa zonal wind (m s−1, contours) anomalies averaged over 5°S–5°N

Full size image

Min et al. (2015) and Zhu et al. (2016) argued that the June–July 2014 easterly anomalies presented in the central and eastern Pacific could be associated with the negative SSTAs in the southeastern subtropical Pacific, which were likely linked to the negative phase of the Inter-decadal Pacific Oscillation (Power et al. 1999; England et al. 2014; Min et al. 2015). To verify this hypothesis, June-July mean wind anomalies at 1000 hPa from the 18-member ensemble mean of the AGCM simulations forced with observed SST are examined and compared with the observation in Fig. 9. The model successfully simulates major features of the observed wind anomalies. The southeasterly anomalies in both observation and model apparently originated from the negative SST area in the southeastern subtropical Pacific. The major difference between the simulation and the observation is in the central equatorial Pacific, where simulated wind is weaker than the observed. The reason for the stronger wind in observation is not clear and could be related to the internal variability in the coupled system. Thus, the suppression of El Niño development in winter 2014/15 could be partially associated with the negative SSTAs in the southeastern subtropical Pacific.

Fig. 9
figure 9

2014 June–July average of SST (°C, shading) and 1000 hPa wind (m s−1) anomalies. a For analyses, and b for AGCM simulations. All the anomalies are with respect to 1981–2010 climatology

Full size image

In support of the above hypothesis, a recent modeling study by Thomas and Vimont (2016) showed large spread in the outcomes of ENSO amplitude associated with positive NPM/PMM forcing. They attributed the uncertainty of the outcomes to the internal variability. This supports the idea that the NPM/PMM is an ENSO precursor, but is not a reliable ENSO predictor.

4 Summary and discussion

In an effort to explain why the atmospheric circulation and SST anomalies of winter 2014/15 in the central equatorial Pacific lacked ocean–atmosphere coupling seen in a typical ENSO event, this study decomposed DJF SST, precipitation rate and 200 hPa stream function anomalies onto the patterns related to the principal components of DJF SSTA. Following this procedure, the relative importance of these patterns to the observed DJF 2014/15 anomalies was identified. It is found that the anomalies of the three variables in winter 2014/15 were determined by the patterns related to the two SST modes, the NPM and the ENSO mode. The contribution from the NPM dominated and resulted in the seemingly lack of air-sea coupling in the central equatorial Pacific and the east–west structure of atmospheric circulation anomalies over the North America. The ENSO mode accounted for the warm SSTAs in the eastern equatorial Pacific and part of atmospheric circulation anomalies in the central equatorial Pacific. The ENSO mode also enhanced the intensity of SST, precipitation rate and circulation anomalies to reach the levels in observation. The results indicates a competition between NPM and ENSO: when ENSO (NPM) is dominant, there will be (not be) strong air–sea coupling in the central and eastern tropical Pacific.

Based on the 1-year lead-lag statistical relationship between NPM and ENSO and the fact that a strong NPM was in its positive phase in winter 2013/14, a strong El Niño was to be anticipated for winter 2014/15. The same was also expected from most dynamical seasonal forecast models in early 2014. The anticipated El Niño development, however, was impeded by the easterly anomalies that occurred in June and July 2014 in the central equatorial Pacific. Model simulations suggested that the easterly anomalies were associated with the negative SSTAs in the southeastern subtropical Pacific. A recent study (Zhu et al. 2016) also attributed the failure of dynamical model predictions partially to the inability of models in predicting the persistently negative SSTAs in the southeastern Pacific Ocean. Their model experiments suggest that 40% of amplitude errors biases of SSTA at its peak phase (i.e., November or December 2014) could be attributed to lack of the negative SSTAs in the southeastern Pacific.

The results of this study also imply that Niño3.4 SSTA index alone may not be sufficient to adequately characterize an ENSO. As highlighted, based on the analysis of DJF 2014/15, the basic shortcoming of Niño3.4 SSTA index is that it can’t distinguish whether the SSTA in the Niño3.4 region is associated with the ENSO mode or NPM, or both. NPM is partially associated with the convective anomalies over the western Pacific, and the ENSO mode connects with the convective anomalies in the central and eastern tropical Pacific. In 2014/15 case, although fingerprint of SSTAs associated with ENSO was present, the corresponding atmospheric anomalies were not typical to an ENSO pattern due to a dominant role played by NPM instead of the ENSO mode. Nevertheless, such anomalies in the tropical Pacific in winter 2014/15 still helped to push the 2015/16 El Niño event to extreme magnitude (Levine and McPhaden 2016).

A remaining question is why NPM pattern was persistently strong in 2014? As the part of NPM pattern in the extratropical North Pacific, like other SSTA patterns in extratropical oceans, is basically wind driven (Davis 1976; Cayan 1992; Kumar et al. 2013; Hartmann 2015a), this question is equivalent to asking why the atmospheric circulation pattern, which gives rise to NPM SSTA pattern, was persistently strong in the year. As shown in Fig. 2 (lower and right), the atmospheric wave train associated with NPM looks like it originated from the diabatic heating (corresponds to precipitation) anomalies in the western tropical Pacific, and the diabatic heating anomalies were apparently due to the underlying warm SSTA (compare to lower and left in Fig. 2). Thus, the persistency and the intensity of the NPM pattern could be due to the persistency and intensity of warm SSTA in the western tropical Pacific. The evolution of equatorial Pacific SST (Fig. 8) shows that warm SSTAs did persist in the western Pacific from winter 2013/14 to winter 2014/15 and strengthened after the August 2014. Though having an eastward shift from February to May due to the westerly anomalies, they were pushed back by the easterly anomalies in June and July. Based on our analyses, one could be tempted to partially attribute the persistently strong NPM in 2014 to the warm SSTA in the tropical western Pacific, which in fact is the tropical part of NPM pattern itself. However, a recent study by Xie and Zhang (2017) showed that an AGCM has low probability in simulating the North American climate anomalies in DJF 2014/15, even though when the AGCM is forced with observed SST. They emphasized the importance of the unpredictable atmospheric internal variability in driving the climate anomalies of the winter 2014/15.