1. Introduction
To predict climatic phenomena several seasons ahead, it is necessary to have a coupled climate model with various component models (i.e., atmosphere, ocean, ice, land, and so on). To this end, important efforts have been made in recent decades to develop and improve coupled climate models. Specifically, owing to an improvement in physical parameterization as well as remarkable increases in model resolution, high-quality observations, advance initialization techniques, and so on, the predictive performances of climate models have improved (Reichler and Roads 2003; Luo et al. 2005; Guilyardi et al. 2009; Yang et al. 2010; Barnston et al. 2012; Hudson et al. 2013; Bellenger et al. 2014; Saha et al. 2014; Watterson 2015). However, there are still significant errors in simulating the observed climate fields and large biases in climate forecasts. Therefore, the characteristics of predictive data must be comprehensively analyzed from a variety of perspectives to determine the strengths and weaknesses of the forecast models and, thus, effectively utilize predictive information.
The Asia–Pacific Economic Cooperation Climate Center (APCC) issues global monthly and seasonal mean forecasts of various climate variables from the multimodel ensemble (MME) seasonal forecast system (Min et al. 2017; see www.apcc21.org for more information), which is based on forecast data provided by various climate institutes worldwide. Owing to the wealth of model data available, the APCC has conducted a variety of studies using the data to provide reliable forecast information for the Asia–Pacific regions. However, most studies have mainly focused on the regional verification of predictive skill scores for prognostic target variables and major climate modes, including El Niño–Southern Oscillation (ENSO). ENSO is the main modulator of global climate and weather with impacts beyond the tropical Pacific Ocean (McPhaden 1999; McPhaden et al. 2006; Cai et al. 2015). Analyzing ENSO behavior in model simulations is important because ENSO serves as a predictability source of seasonal forecast models (Latif et al. 1998; Peng et al. 2011; Kim et al. 2012; Jia et al. 2015). Moreover, the improved skills of the global winter climate in seasonal forecast models are attributable to the well-resolved ENSO-related ocean forcing and its remote impacts (Quan et al. 2006; Aldrian et al. 2007; Kirtman et al. 2014; Krishnamurthy 2019).
Although considerable progress has been made in predicting ENSO events using coupled climate models (Jin et al. 2008; Barnston et al. 2012; L’Heureux et al. 2020), coupled models still have systematic errors in simulating the tropical Pacific climatological mean states (Jin and Kinter 2009; Manganello and Huang 2009; Magnusson et al. 2013; Vannière et al. 2013; Ham et al. 2014; Kim et al. 2014, 2017). For example, coupled models still demonstrate a systematic bias in the upper tropical Pacific Ocean, which affects the performance of climate models in simulating ENSO behaviors such as strength and teleconnection. A previous study found that equatorial cold tongue bias may affect asymmetric midlatitude teleconnection patterns between El Niño and La Niña (e.g., Bayr et al. 2019). Specifically, the North American winter climate can be influenced by ENSO via forced atmospheric Rossby wave trains, which is known as the Pacific–North America (PNA) teleconnection pattern (Hoskins and Karoly 1981; Wallace and Gutzler 1981; Mo and Livezey 1986; Hoerling et al. 1997; Ropelewski and Halpert 1986; Livezey et al. 1997; L’Heureux et al. 2015). Furthermore, asymmetry exists in the North American winter temperature response between El Niño and La Niña, which is related to asymmetric midtropospheric circulations during ENSO (Zhang et al. 2011, 2014). Zhang et al. (2014) argued that climate models are dominated by the symmetric component in the ENSO-related midlatitude teleconnections rather than the asymmetric component, in contrast to the observations where the asymmetric component in a temperature response to ENSO is comparable to the symmetric component.
Therefore, it is questionable whether seasonal forecast models can reliably predict the asymmetric responses of winter temperatures in North America to the ENSO phases and, as a result, whether the predictive skills of models for local climate may change by ENSO phases. Because the significant role of the ENSO in seasonal predictions is widely known, many studies have analyzed predictive skills with regard to the ENSO-related winter climates of the midlatitudes (e.g., Chen et al. 2017). However, studies connecting the predictive skills of the climate models for the North American climate to errors in ENSO-related atmospheric and oceanic dynamics are lacking. Moreover, the PNA patterns that occur during ENSO in early winter months are different from those in late winter months, and this phenomenon cannot be reproduced in current climate models (Wang and Fu 2000). Therefore, the predictive performance for the North American temperatures of individual winter months (December to February) should be estimated instead of seasonal mean temperatures, linking it to the ENSO-related dynamical performance (i.e., ENSO-related teleconnection patterns) in the models.
This study presents the ability of the seasonal forecast models to replicate the observed teleconnection associated with ENSO over the North American region and the impact of its associated errors on regional forecast performance for winter temperatures over the North American region. The errors in the midlatitude teleconnection associated with ENSO are linked to the mean state in the tropical Pacific; thus, this study provides guidance for future model improvement in more accurately predicting winter-season temperatures over the North American region during two ENSO phases.
The remainder of this paper is organized as follows. Section 2 describes the seasonal forecast models, observational datasets, and analysis methods. Section 3 addresses the model’s ability to predict ENSO-related temperatures in the North American region for the winter months of December, January, and February, as well as how the errors in the remote response of the North American temperatures to ENSO affect the regional forecast performance for the winter surface air temperatures. In section 4, the possible causes of the model errors in the remote response to ENSO are discussed. Section 5 presents a summary and discussion of this study.
2. Data and methods
In this study, seven models are selected to ensure that the hindcast data period was as long as possible among the APCC MME participation models. Table 1 lists the coupled models used in the study, their component models, resolutions, ensemble members, brief initialization methods, and references for detailed descriptions of the models. Over the common period of the hindcast data among the models (1983–2010), various analyses are performed with the combined set of 1-month-lead forecasts for individual winter months of December, January, and February. In this study, 1-month-lead forecasts denote a forecast made from the initial conditions of 1 month ahead. For example, the 1-month-lead forecast from November initial conditions is the forecast for December, the forecast from December initial conditions is for January, and the forecast from the January initial conditions is for February. Based on a simple average of ensemble members of individual models, a series of analyses are performed, and retrospective forecasts are verified against various reanalysis data. For the observed SST, we use the optimum interpolation SST (OISST), version 2 (Reynolds et al. 2007). The observed precipitation is taken from the Global Precipitation Climatology Project (GPCP; Adler et al. 2003). Atmospheric variables, including geopotential height at 500 hPa (Z500) and air temperatures at 2 m, are taken from the National Centers for Environmental Prediction–Department of Energy Reanalysis 2 (NCEP–DOE2; Kanamitsu et al. 2002).
Descriptions of seven seasonal forecast models verified in this study. Ens = number of ensemble members.
Observed monthly anomalies are obtained by extracting the climatological mean of each winter month (i.e., December, January, and February) for the period of 1983–2010. In the hindcast data, the December–February monthly data at 1-month lead time from each of the forecast models were reconstructed to obtain monthly datasets covering the period from 1983 to 2010. Anomalies were then obtained by extracting the climatological average from the reconstructed datasets at a 1-month lead time from the individual forecasts.
The temporal correlation coefficient (TCC) is applied to verify the teleconnection patterns related to the simulated ENSO. The remote effect of the climate mode on the surface air temperatures over the North American region can be shown in a TCC map between the ENSO index and temperature at each grid point. The ENSO index is defined as SST anomalies area-averaged in the Niño-3.4 (170°–120°W, 5°S–5°N) region.
To determine the consistency and stability of the ENSO-related SST and rainfall activity in the tropical Pacific and the regional response of the North American winter temperatures to ENSO among the seasonal forecast models, we count the number of models with a significant correlation at the 95% confidence level at each grid point. The number of models is counted separately for each of the positive and negative signs of the significant correlation coefficient, and then the final number is determined through the process of combining them. Thus, the maps of the consistency among models for the ENSO-related tropical Pacific patterns and the ENSO response of the North American winter temperatures range from −7 to 7. When the absolute value is greater, that is, the value is close to either 7 or −7, the models consistently simulate the local and remote responses to ENSO in one direction, whereas a value closer to zero means that the models are inconsistent in replicating the response of the atmospheric and oceanic climate variables to ENSO.
3. Relationship of North America winter temperatures with ENSO
a. Model capability of predicting ENSO
Before discussing the error in the North American temperature response to ENSO, we estimate whether ENSO itself is correctly predicted in the forecast models. Figure 1a shows the TCC between the observed and predicted Niño-3.4 index at 1-month lead time. Most of the models predict ENSO quite accurately because all models except for two have TCC values greater than 0.9 during all winter months. Therefore, the influence of model errors in predicting the ENSO index on the predictive skill for North American winter temperatures is likely negligible.
(a) Temporal correlation coefficients between the observed Niño-3.4 index and the predicted index from seven forecast models during the winter months at 1-month lead time. Also shown are spatial patterns of ENSO-related tropical Pacific (b),(c) sea surface temperature (SST) and (d),(e) rainfall anomalies in January from (center) observations and (right) models. For the observed and modeled ENSO-related patterns, the correlation coefficients of the SST and rainfall anomalies with Niño-3.4 index and the number of models with the correlation coefficients (between the anomalies and Niño-3.4 index) that are significant at 95% confidence level according to the two-tailed Student’s t test, respectively, are shown. In (b) and (d), black hatches indicate that the correlation coefficient at each grid point is significant at 95% confidence level, and in (c) and (e) black hatches indicate that more than 50% of the models have significant correlation coefficients at each grid point.
Citation: Journal of Climate 34, 20; 10.1175/JCLI-D-21-0094.1
To evaluate the spatial patterns of ENSO-related SST and rainfall in the tropical Pacific forecast at a 1-month lead time, SST and rainfall anomalies are correlated with the Niño-3.4 index from observations (Figs. 1b,d) and models (Figs. 1c,e). Most of the models (i.e., four or more of the seven models) reliably capture the main observed spatial features of ENSO-related SST and rainfall anomalies; these results show horseshoe-shaped negative (positive) SST and rainfall anomalies in the western Pacific and positive (negative) SST and rainfall anomalies in the eastern tropical Pacific during El Niño (La Niña), although the meridional width of the ENSO-related rainfall patterns in the eastern Pacific is slightly narrower in the models than in the observations. In the following sections, an additional analysis of the ENSO-related rainfall response in the tropical Pacific is performed and linked with the midlatitude teleconnection errors in the forecast models.
b. Errors in North American winter temperature response to ENSO
Figures 2a and 2b show the spatial pattern of TCCs between North American temperatures and Niño-3.4 index for winter months from the observations and models, respectively. Observations from December show a north–south dipole-like pattern with anomalous warming over northern North America (NNA) and anomalous cooling over southern North America (SNA) during El Niño and vice versa during La Niña, which is a typical ENSO response (Ropelewski and Halpert 1986; Livezey et al. 1997; Infanti and Kirtman 2016). This general feature is predicted very well in most models, although there are some differences in detailed regions with significant correlations between models and observations. For example, in the observations, significant correlations are extended from the east of the U.S. state of Alaska to the Canadian province of Ontario in the NNA and are confined to the northern Mexican region of the SNA. However, more than 70% of the models (i.e., more than five out of seven models) have significant correlation regions that cover the most western part of Canada in the NNA, and from northern Mexico to the southern part of the United States in the SNA. The observed north–south response of the winter air temperatures to ENSO tends to weaken noticeably in January, and the significant correlation region is slightly shifted to the south in February, whereas the strong north–south response features are sustained throughout all winter months in most of the forecast models.
Spatial patterns of temporal correlation coefficients between North American winter temperatures (December–February) and Niño-3.4 index from (a) observations and (b) the number of models with significant correlation coefficients (between the temperature and Niño-3.4 index) at 95% confidence level according to the two-tailed Student’s t test. In (a), black hatches indicate that the correlation coefficient at each grid point is significant at 95% confidence level. In (b), black hatches indicate that more than 50% of the models have significant correlation coefficients. (c) Bar and scatterplots of averaged correlation coefficients in the two blue-outlined regions from observations (orange bars) and models (gray bars and black circles). Area-averaged correlation coefficients for individual models (circles) are averaged for all seven models (gray bars). Horizontal dashed lines indicate the correlation coefficients at 95% confidence level in (c).
Citation: Journal of Climate 34, 20; 10.1175/JCLI-D-21-0094.1
To compare the intensity of the linear response to ENSO between observations and models, the correlation coefficients are area averaged over the boxed regions of the NNA and SNA (Fig. 2c). Most models have absolute values of correlation coefficients greater than those from observations except for two models in December. The models’ average of the correlation coefficients averaged in the NNA and SNA boxed regions are more than 0.42 (0.42, 0.42, and 0.50 for December, January, and February, respectively) and less than −0.54 (−0.54, −0.66, and −0.65) respectively. However, the values from observations are much smaller than those from the models (0.30, 0.08, and 0.21 in the NNA region and −0.18, −0.17, and −0.33 in the SNA region). This result implies that the symmetry in the response of the North American winter temperatures between El Niño and La Niña in seasonal forecast models is strong when compared with observations. One thing to note here is that the high correlations in the models are likely due to the analysis of the ensemble means. When the correlations are calculated for members of each model first, and then the same analysis as shown in Fig. 2b is applied to the members, the high correlations over the North American region disappear (not shown). This is because the internal variability is preserved in the model, which is removed by the ensemble average prior to the calculation of correlations. This is the reason probabilistic forecasts are generally utilized to mitigate the weakness of deterministic forecasts. In this study, a series of analyses are performed from the perspective of deterministic forecasts, and further studies on the skills of the probabilistic forecasts by ENSO phases will be documented in a forthcoming study.
To explore the linear or nonlinear features of the winter temperature response in the two boxed regions to ENSO phases, we analyze a box-and-whisker plot, as shown in Fig. 3. First, throughout the winter season, anomalous warming is dominant during the El Niño phases and anomalous cooling during La Niña phases in most of the models in the NNA region. The separation of the response for El Niño and La Niña phases is also clear in the SNA region with the opposite sign. In the observations, however, the strong linear feature shown in the models is absent, except for December over the NNA region and February over the SNA region. In other words, observations tend to have anomalous warming during El Niño phases, whereas they are not dominated by anomalous cooling during La Niña phases, especially in January and February. In the SNA region, there is strong tendency of anomalous cooling during El Niño events and anomalous warming during La Niña in all the models from December to February. The linear features emerge only in observations from February, and strong separation in the models with regard to the ENSO response of the winter temperatures by phases are not observed in December and January.
Box-and-whiskers plots for air temperatures over (left) the northwestern part of North America (NNA; 140°–100°W, 45°–70°N) and (right) the southeastern part (SNA; 110°–80°W, 20°–35°N) from seven individual models and observations in (top) December, (middle) January, and (bottom) February during the El Niño (red) and La Niña (blue) events. Values at all of the grid points over the NNA and SNA regions are used.
Citation: Journal of Climate 34, 20; 10.1175/JCLI-D-21-0094.1
c. Impact on forecast performance
In the previous subsection, the forecast models showed an erroneously strong linear relationship in the response of winter temperatures in the North American region to the ENSO, whereas in the observations, a nonlinear relationship was dominant. Therefore, the response errors in the models may affect the predictive performance for the winter air temperatures during ENSO events, and there would be a difference in the model performance in predicting the winter temperatures between El Niño and La Niña events. To explore how the response error affects the forecast performance of the models, we apply two verification methods, the probability of detection (POD = hits/observed events) and false alarm ratio (FAR = false alarms/forecast events), to the forecasts of the winter warming and cooling events by two ENSO phases. POD measures the fraction of the observed winter warming (cooling) events that can be correctly forecast, and FAR measures what fraction of the predicted warming (cooling) events did not occur during El Niño phases in the NNA (SNA) region and vice versa during La Niña phases. For each winter month, the warming and cooling events are defined as when the surface air temperature anomaly is respectively above or below 0.0°C at each grid point of the North American land region, and El Niño or La Niña events are selected when the Niño-3.4 index is respectively greater than 0.5°C or lower than −0.5°C. The ENSO events are based on observations and are listed in Table 2.
El Niño and La Niña events selected for each winter month.
Figure 4 shows scatterplots of the multimodel-averaged scores at each grid point over the NNA and SNA regions during the El Niño phase versus during the La Niña phase from December to February. The circles for the POD scores tend to be located mainly toward the left side of the diagonal line, whereas those for the FAR scores tend to be located generally toward the right side of the diagonal line. This indicates that the North American winter-season warming events during the El Niño phase in the NNA region are forecast more correctly in the models than the cooling events during La Niña phases and that the North American winter-season cooling events during the El Niño phase in the SNA region are forecast more correctly in the models than the warming events during La Niña phases.
Scatterplots between El Niño and La Niña events for (left) probability of detection (POD, or hit rate) and (right) false-alarm ratio (FAR) scores of winter temperatures over the (a),(b) NNA and (c),(d) SNA regions in December (light-blue circles), January (light-red circles) and February (light-green circles). Horizontal and vertical lines (dark blue for December, dark red for January, and dark green for February) denote standard deviation range against mean values of POD and FAR for La Niña and El Niño events, respectively.
Citation: Journal of Climate 34, 20; 10.1175/JCLI-D-21-0094.1
4. Possible causes of the strong linear relationship with ENSO in the models
a. Asymmetric midlatitude teleconnection patterns
One possible cause of the strong linearity in the models is that the atmospheric teleconnection patterns (e.g., geographical locations) is very symmetric between El Niño and La Niña phases in the models, whereas the observations are characterized by asymmetric midlatitude teleconnection patterns between ENSO phases, as discussed in the introduction. In general, the remote teleconnection dynamics of the North American climate for El Niño and La Niña events can be identified as a spatial pattern of Z500 anomalies (Horel and Wallace 1981; Alexander et al. 2002). A composite analysis of the Z500 during both ENSO phases is performed to explore the midlatitude teleconnection patterns associated with ENSO, as shown in Fig. 5. The observed ENSO events (Table 2) are used for the composite analysis. The use of ENSO events isolated from individual models does not affect the main results (not shown). In Figs. 5a and 5b, we can identify the observed wave train patterns in which one high center is located over Canada and two low centers are around the Aleutians and the southern United States in December of the El Niño phase and one low center is located over Canada and two high centers are around the Aleutians and the southern United States in December of the La Niña phase. The December teleconnection patterns are similar between El Niño and La Niña phases with opposite signs, and the observed feature is simulated quite well in the models. However, in the observations, the disparity of the teleconnection patterns between the ENSO phases increases from January to February, especially in the longitudinal domain of North America, whereas the symmetric teleconnection patterns between ENSO phases continue to be maintained until February in the models. To quantify the degree of asymmetry in the teleconnection patterns from observations and models between ENSO phases, the pattern correlation coefficient (PCC) of the composite fields of Z500 anomalies over 170°–50°W, 20°–75°N is calculated during the winter season of El Niño and La Niña periods, as shown in Fig. 5c. The composite fields of the La Niña–related Z500 anomalies are multiplied by −1 for comparison prior to computing the PCC. Observations show a decline in the PCC (from 0.94 to 0.81) of the Z500 pattern associated with El Niño and La Niña as it transitions into late winter after December. However, in the models, the PCC value does not change much and is greater than 0.94 in multimodel mean throughout the winter months. Therefore, Figs. 5a–c highlight that, in the observation, the asymmetry in the midlatitude teleconnection with regard to the ENSO-related response increases from the early winter to midwinter, which is consistent with previous studies (Hoerling et al. 1997, 2001; Hoerling and Kumar 2002; Lin and Derome 2004; Wu and Hsieh 2004; Kug et al. 2010; Zhang et al. 2014). However, in the models, a strong symmetric pattern persists during the winter season of the ENSO peak. The enhanced forecast error in terms of asymmetry in midlatitude teleconnections are mainly caused by the La Niña–related teleconnection patterns. This is clarified via Fig. 5d, which shows that the PCC scores between observations and models for the Z500 composites around the North American region (170°–50°W, 20°–75°N) during La Niña phases (multimodel mean is 0.68, 0.64, and 0.58, respectively) are smaller than those during the El Niño phases (0.66, 0.78, and 0.81) in general. This result is also consistent with the lower POD and higher FAR scores for winter temperature events during La Niña events in the forecast models than during the El Niño events.
Spatial distribution of the composite of (a) El Niño and (b) La Niña events for 500-hPa geopotential height anomalies from observations (shading) and model averages (contours) in (top) December, (middle) January, and (bottom) February. Black dots denote statistical significance at the 90% level for the observed composite pattern, and blue and red contours indicate that four or more of the seven models have a same sign for the composite pattern at each grid point. (c) Pattern correlation coefficient (PCC, with sign reversed) of 500-hPa geopotential heights (Z500) between El Niño and La Niña from observations (orange bars) and models (black circles for individual models, and gray bars for their average). (d) PCC between predicted and observed Z500 from December to February (red bars for El Niño and blue bars for La Niña; black circles indicate individual models).
Citation: Journal of Climate 34, 20; 10.1175/JCLI-D-21-0094.1
b. Longitudinal shift of the tropical Pacific rainfall response to ENSO
To reveal the cause of the strong linear response of the midlatitude atmospheric teleconnection patterns and thus winter temperatures to two ENSO phases in the seasonal prediction models, we will first quantify the degree of diversity among the models with regard to the errors as an index. Quantification of the models’ diversity associated with the erroneous linear response of the North American winter temperatures to ENSO is possible using a principal component analysis (e.g., Li and Xie 2012, 2014). The empirical orthogonal function (EOF) analysis is applied to the eight spatial maps of the TCCs between the winter-season air temperatures and Niño-3.4 index for the seven models and for the observations. Here, the EOF analysis is performed for December, January, and February, separately. Through the EOF analysis, we examine the diversity of the TCCs among eight TCC maps. The first loading vector of the EOF for the TCCs and its associated principal component (PC) values for each model and observation are shown in Figs. 6a and 6b, respectively. The first EOF mode, explaining approximately 50% of the total variance, is characterized by the strong north–south response features with positive TCCs in the NNA region and negative TCCs in the SNA for all three calendar months. This is close to the models’ average of the temperature response pattern previously shown in Fig. 2b and the PC is highly correlated with the strength of the linear winter temperature response to the ENSO at 0.74 for December, 0.83 for January, and 0.95 for February (Fig. 6c). Here, the strength of the linear response is measured by the sum of the absolute values of the TCCs averaged over the boxed regions of NNA and SNA in Fig. 2. Especially, most models have higher PC values and larger intensity of the linear temperature response than the observations. This result indicates that generally in seasonal forecast models the linear response of the North American temperature to the ENSO is dominated over the nonlinear response and the PC values can be utilized as the quantified degree of the response error among models.
(a) The first EOF mode for the TCC patterns of surface air temperatures with Niño-3.4 index from seven models and one observation and (b) their associated principal components (PCs), and (c) scatterplots of the absolute TCC values summed for the two outlined regions of the NNA and SNA (see Fig. 2c) against the PCs for the observation (plus sign) and individual models (circles) in (top) December, (middle) January, and (bottom) February. In (a), the explained variance with the first EOF is shown.
Citation: Journal of Climate 34, 20; 10.1175/JCLI-D-21-0094.1
To link the model’s systematic biases to the ENSO-related temperature response errors, correlation coefficients between the eight PCs and eight climatological SSTs in the tropical Pacific (1983–2010 average) are estimated at each grid point from the seven models and observations (Figs. 7a–c). In this study, we focus on a bias in the tropical Pacific SST because it is directly associated with the tropics-originated forcing that triggers atmospheric teleconnection patterns (Lin and Derome 2004; Bayr et al. 2019; Fang and Yu 2020). In December and January (Figs. 7a,b), when viewing the regions with a significant correlation value at more than 80% confidence level (>0.51) according to a Student’s t test, positive correlation coefficients are observed in the central equatorial Pacific. In February (Fig. 7c), the statistical significance of correlations with SST in the tropics decreases, but the sign of correlation in the central tropical Pacific remains positive. These results indicate that models with a warmer SST climatology over the central tropical Pacific have a greater linearity in the responses of winter temperatures to ENSO. The relationship between the central Pacific SST climatology and the winter temperature response is clarified using scatterplots of the PCs versus SST climatology averaged over the central tropical Pacific region (180°E–130°W, 7°S–2°N), as shown in Fig. 7d. The models with warmer SST climatology in the central tropical Pacific tend to have the larger PC values or response errors, and the correlation coefficient between the two axes is 0.57. For this correlation analysis, we used all values from three calendar months of seven models and observation (when only models are used, the correlation value is increased to 0.83). The models with warm bias with respect to the observations increase from December to February. In December, most models (five out of seven models) have a cold bias and two models with warmer central Pacific climatological SST than the observations have larger positive PC values. In January and February, the observed climatological SST become slight colder and the cold bias in the models is mitigated slightly. In other words, for the models used in this study, the climatological SST in the central tropical Pacific becomes warmer from December to February and the difference between the PC values from the models and those from the observations increases.
Spatial patterns of correlation coefficients between PCs and SST climatology from models and observation for (a) December, (b) January, and (c) February, and (d) scatterplots of PCs against SST climatology averaged over the outlined region of (a)–(c) from models (colored circles) and observations (colored plus signs). In (a)–(c), the correlation coefficients that are significant at the 80% level according to the two-tailed Student’s t test are color shaded. In (d), the months of December, January, and February are represented with blue, red, and green colors, respectively.
Citation: Journal of Climate 34, 20; 10.1175/JCLI-D-21-0094.1
Previous studies have shown that the eastward shift of the atmospheric (i.e., rainfall) response to the El Niño–related SST forcing relative to the response to the La Niña–related SST forcing may result in a nonlinear atmospheric teleconnection and thus a nonlinear winter temperature response over North America (e.g., Hoerling et al. 1997, 2001). Therefore, the longitudinal shift of the atmospheric response is estimated using the difference between the longitudinal location of the maximum rainfall anomalies averaged along the equator (5°S–5°N) from its composites for El Niño events and that of the minimum rainfall anomalies from the composites for La Niña events. Figure 8c shows the scatter between the longitudinal location difference and PCs. Models with a smaller longitudinal shift between two ENSO phases tend to have larger PC values. In other words, when the difference in the atmospheric response location for the ENSO-related SST forcing is small, the linearity in the response of the winter temperatures to the two ENSO phases is stronger because the nonlinearity in the midlatitude teleconnections becomes weaker. For the seasonal forecast models used in this study, the error in the longitudinal location of the rainfall response to La Niña SST forcing appears to be the main contributor to the strong linear response of the winter temperatures to ENSO because the linear relationship of the PCs with the response location for the La Niña phases is clearer than that for the El Niño phases (Figs. 8a,b). The correlation coefficients between the PCs and longitudinal locations for the El Niño and La Niña phases are 0.11 and 0.57, respectively. Therefore, the overly eastward shift of the rainfall response to La Niña–related SST forcing caused by the warmer SST background in the central tropical Pacific in the models is another possible reason for the overly strong linear relationship between ENSO and winter temperatures over North America.
Scatterplots of (left) PCs and (right) central Pacific climatological SST against the longitudinal location of (a),(d) the maximum rainfall anomalies for El Niño composite and (b),(e) the minimum rainfall anomalies for La Niña composite, along with (c),(f) their difference, from models (colored circles) and observations (colored plus signs). The months of December, January, and February are represented with blue, red, and green colors, respectively.
Citation: Journal of Climate 34, 20; 10.1175/JCLI-D-21-0094.1
The link of the linear (or symmetric) midlatitude atmospheric teleconnection between the El Niño and La Niña phases to the warm climatological SST biases in the central tropical Pacific suggested in this study appears to contradict the findings of Bayr et al. (2019). In their study, models with a stronger equatorial Pacific cold tongue bias tended to simulate weaker nonlinear (or asymmetric) teleconnection patterns over the North American region between ENSO phases. First of all, contradictory results may result from a limited number of models in this study. Furthermore, the relationship between the SST bias and the asymmetric rainfall response between ENSO phases appears to be model-dependent because a weak correlation was observed between them when the various models from phase 5 of the Coupled Model Intercomparison Project (CMIP5) were used as can be identified in Fig. 14 of Bayr et al. (2019). Second, Bayr et al. (2019) did not consider the possibility that the strength of the climatological SST bias in the central Pacific appears to change from December to February as mentioned in Fig. 7d. When viewing a scatterplot of the climatological SST versus the longitudinal center of the tropical rainfall response to El Niño and La Niña (Figs. 8d,e), we can easily recognize that the models with a colder climatological SST show a westward-shifted rainfall response to the El Niño–related SST forcing relative to the observation in December (Fig. 8d). In this time, most of models have a cold tongue bias in the tropical Pacific. When it goes into February, the cold bias is mitigated as previously mentioned and rather several models have a warm bias. This leads to decrease in the westward shift error for El Niño forcing and also an eastward-shifted rainfall response to the La Niña–related SST forcing in some models (Fig. 8e), which results in a small difference in the longitudinal center of the rainfall response between ENSO phases (Fig. 8f). We need an additional study to employ model samples as many as possible for a more robust conclusion.
5. Summary and discussion
This study presents seasonal forecast model errors related to the winter temperature responses to a major oceanic climate variability mode, ENSO, over the North American region, as well as how these errors affect regional temperature forecasts. In most models, a strong linear relationship is maintained in temperature anomalies associated with both ENSO phases during the winter season, which is lacking in observations. This leads to temperature forecast errors during the ENSO mature stages (i.e., December–February). Forecast verification scores of the models indicate that the North American winter-season warming events during the El Niño phase in the NNA region are forecast more accurately in the models than the cooling events during La Niña phases. This is also true for cooling events during the El Niño phase in SNA as compared with warming events during La Niña.
One possible reason for the erroneously strong linearity (or symmetry) of these models in terms of the winter temperature response over North America is that the remote atmospheric teleconnection pattern may be dominantly linear (or symmetric) between the El Niño and La Niña phases. The models cannot realistically simulate the longitudinal shift of the tropical Pacific rainfall response between the El Niño and La Niña phases, which is linked to the midlatitude nonlinear or asymmetric atmospheric teleconnection. Errors in the longitudinal shift between the two phases can be attributed to the warmer central tropical Pacific in the models because of the eastward-shifted rainfall response to La Niña–related SST forcing relative to the observations.
In addition to the asymmetric rainfall response to ENSO forcing, the ENSO-related SST intensity and different ENSO types (i.e., central and eastern Pacific types) can also affect the asymmetric or nonlinear midlatitude teleconnection and temperature response (e.g., Wang and Kumar 2015; Liu et al. 2019). However, the limited ENSO events over a short hindcast period make it difficult to investigate the model errors with respect to the intensity and types of ENSO events and subsequently link the errors in the midlatitude winter temperature responses.
In the scatterplot of Fig. 7d, it can be also recognized that the observed climatological tropical Pacific SST does not change much from December to February; on the contrary, the PC values change significantly. This may indicate that in observations the increased nonlinearity of ENSO teleconnection from December to January is affected by some other factors rather than mean-state changes, as opposed to the strong correlation coefficients in the models. In this study, we do not consider other climatic factors modulating the midlatitude climate, which may offset the ENSO impact and thus yield a nonlinear relationship between the ENSO and North American winter temperatures. Specifically, climate variability originating from the Arctic region is an important factor that has a significant impact on the midlatitude winter climate. Climate models continue to struggle to reliably predict the Arctic climate (Gong et al. 2017; Chen et al. 2019; Block et al. 2020). The role of the Arctic climate in having an erroneously strong linear response of the winter temperatures to ENSO in climate models will be investigated in a future study.
The main results of this study suggest that when utilizing predictive information from seasonal forecast models, it is important to determine how the predictive performance of climate models varies according to ENSO phases. In general, we can expect more reliable predictions during ENSO events and attempt to provide seasonal prediction information depending on the general linear response of winter air temperatures to ENSO. However, the main results of this study suggest that the difference in the model prediction performance for the local climate depending on the ENSO phases should be considered. Less skillful model outputs in the La Niña phase should be hedged by other information, such as from statistical models or empirical knowledge, so that the North American region temperature forecast skill can be enhanced.
To ensure the use of predictive information in a more efficient way, a similar analysis should be performed for other climate dominant modes (Indian Ocean dipole, North Atlantic Oscillation, etc.) that also affect the regional climate remotely.
Acknowledgments
The authors acknowledge that the APCC Multi-Model Ensemble (MME) Producing Centers for making their data available for analysis and the APEC Climate Center for collecting and archiving them and for organizing APCC MME prediction. This study was supported by the APEC Climate Center.
REFERENCES
Adler, R. F., and Coauthors, 2003: The version 2 Global Precipitation Climatology Project (GPCP) monthly precipitation analysis (1979–present). J. Hydrometeor., 4, 1147–1167, https://doi.org/10.1175/1525-7541(2003)004<1147:TVGPCP>2.0.CO;2.
Ahn, J. B., and H. J. Kim, 2013: Improvement of one-month lead predictability of the wintertime AO using a realistically varying solar constant for a CGCM. Meteor. Appl., 21, 415–418, https://doi.org/10.1002/met.1372.
Aldrian, E., L. Dümenil Gates, and F. H. Widodo, 2007: Seasonal variability of Indonesian rainfall in ECHAM4 simulations and in the reanalyses: The role of ENSO. Theor. Appl. Climatol., 87, 41–59, https://doi.org/10.1007/s00704-006-0218-8.
Alexander, M. A., I. Bladé, M. Newman, J. R. Lanzante, N.-C. Lau, and J. D. Scott, 2002: The atmospheric bridge: The influence of ENSO teleconnections on air–sea interaction over the global oceans. J. Climate, 15, 2205–2231, https://doi.org/10.1175/1520-0442(2002)015<2205:TABTIO>2.0.CO;2.
Barnston, A. G., M. K. Tippett, M. L. L’Heureux, S. Li, and D. G. DeWitt, 2012: Skill of real-time seasonal ENSO model predictions during 2002–11: Is our capability increasing? Bull. Amer. Meteor. Soc., 93, 631–651, https://doi.org/10.1175/BAMS-D-11-00111.1.
Bayr, T., D. I. Domeisen, and C. Wengel, 2019: The effect of the equatorial Pacific cold SST bias on simulated ENSO teleconnections to the North Pacific and California. Climate Dyn., 53, 3771–3789, https://doi.org/10.1007/s00382-019-04746-9.
Bellenger, H., E. Guilyardi, J. Leloup, M. Lengaigne, and J. Vialard, 2014: ENSO representation in climate models: From CMIP3 to CMIP5. Climate Dyn., 42, 1999–2018, https://doi.org/10.1007/s00382-013-1783-z.
Block, K., F. A. Schneider, J. Mülmenstädt, M. Salzmann, and J. Quaas, 2020: Climate models disagree on the sign of total radiative feedback in the Arctic. Tellus, 72A, 1–14, https://doi.org/10.1080/16000870.2019.1696139.
Cai, W., and Coauthors, 2015: ENSO and greenhouse warming. Nat. Climate Change, 5, 849–859, https://doi.org/10.1038/nclimate2743.
Chen, L.-C., H. van den Dool, E. Becker, and Q. Zhang, 2017: ENSO precipitation and temperature forecasts in the North American Multimodel Ensemble: Composite analysis and validation. J. Climate, 30, 1103–1125, https://doi.org/10.1175/JCLI-D-15-0903.1.
Chen, S., R. Wu, W. Chen, and L. Song, 2019: Performance of the CMIP5 models in simulating the Arctic Oscillation during boreal spring. Climate Dyn., 53, 2083–2101, https://doi.org/10.1007/s00382-019-04792-3.
Fang, S. W., and J. Y. Yu, 2020: A control of ENSO transition complexity by tropical Pacific mean SSTs through tropical–subtropical interaction. Geophys. Res. Lett., 47, e2020, https://doi.org/10.1029/2020GL087933.
Gong, H., L. Wang, W. Chen, X. Chen, and D. Nath, 2017: Biases of the wintertime Arctic Oscillation in CMIP5 models. Environ. Res. Lett., 12, 014001, https://doi.org/10.1088/1748-9326/12/1/014001.
Guilyardi, E., and Coauthors, 2009: Understanding El Niño in ocean–atmosphere general circulation models: Progress and challenges. Bull. Amer. Meteor. Soc., 90, 325–340, https://doi.org/10.1175/2008BAMS2387.1.
Ham, S., A.-Y. Lim, S. Kang, H. Jeong, and Y. Jeong, 2019: A newly developed APCC SCoPS and its prediction of East Asia seasonal climate variability. Climate Dyn., 52, 6391–6410, https://doi.org/10.1007/s00382-018-4516-5.
Ham, Y.-G., S. Schubert, Y. Vikhliaev, and M. J. Suarez, 2014: An assessment of the ENSO forecast skill of GEOS-5 system. Climate Dyn., 43, 2415–2430, https://doi.org/10.1007/s00382-014-2063-2.
Hoerling, M. P., and A. Kumar, 2002: Atmospheric response patterns associated with tropical forcing. J. Climate, 15, 2184–2203, https://doi.org/10.1175/1520-0442(2002)015<2184:ARPAWT>2.0.CO;2.
Hoerling, M. P., A. Kumar, and M. Zhong, 1997: El Niño, La Niña, and the nonlinearity of their teleconnections. J. Climate, 10, 1769–1786, https://doi.org/10.1175/1520-0442(1997)010<1769:ENOLNA>2.0.CO;2.
Hoerling, M. P., A. Kumar, and T. Xu, 2001; Robustness of the nonlinear climate response to ENSO’s extreme phases. J. Climate, 14, 1277–1293, https://doi.org/10.1175/1520-0442(2001)014<1277:ROTNCR>2.0.CO;2.
Horel, J. D., and J. M. Wallace, 1981: Planetary-scale atmospheric phenomena associated with the Southern Oscillation. Mon. Wea. Rev., 109, 813–829, https://doi.org/10.1175/1520-0493(1981)109<0813:PSAPAW>2.0.CO;2.
Hoskins, B. J., and D. J. Karoly, 1981: The steady linear response of a spherical atmosphere to thermal and orographic forcing. J. Atmos. Sci., 38, 1179–1196, https://doi.org/10.1175/1520-0469(1981)038<1179:TSLROA>2.0.CO;2.
Hudson, D., A. G. Marshall, Y. Yin, O. Alves, and H. H. Hendon, 2013: Improving intraseasonal prediction with a new ensemble generation strategy. Mon. Wea. Rev., 141, 4429–4449, https://doi.org/10.1175/MWR-D-13-00059.1.
Infanti, J. M., and B. P. Kirtman, 2016: North American rainfall and temperature prediction response to the diversity of ENSO. Climate Dyn., 46, 3007–3023, https://doi.org/10.1007/s00382-015-2749-0.
Jia, L., and Coauthors, 2015: Improved seasonal prediction of temperature and precipitation over land in a high-resolution GFDL climate model. J. Climate, 28, 2044–2062, https://doi.org/10.1175/JCLI-D-14-00112.1.
Jin, E. K., and J. L. Kinter III, 2009: Characteristics of tropical Pacific SST predictability in coupled GCM forecasts using the NCEP CFS. Climate Dyn., 32, 675–691, https://doi.org/10.1007/s00382-008-0418-2.
Jin, E. K., and Coauthors, 2008: Current status of ENSO prediction skill in coupled ocean–atmosphere models. Climate Dyn., 31, 647–664, https://doi.org/10.1007/s00382-008-0397-3.
Kanamitsu, M., W. Ebisuzaki, J. Woollen, S. K. Yang, J. J. Hnilo, M. Fiorino, and G. L. Potter, 2002: NCEP–DOE AMIP-II reanalysis (R-2). Bull. Amer. Meteor. Soc., 83, 1631–1644, https://doi.org/10.1175/BAMS-83-11-1631.
Kim, H.-M., P. J. Webster, and J. A. Curry, 2012: Seasonal prediction skill of ECMWF System 4 and NCEP CFSv2 retrospective forecast for the Northern Hemisphere winter. Climate Dyn., 39, 2957–2973, https://doi.org/10.1007/s00382-012-1364-6.
Kim, S. T., W. Cai, F.-F. Jin, and J.-Y. Yu, 2014: ENSO stability in coupled climate models and its association with mean state. Climate Dyn., 42, 3313–3321, https://doi.org/10.1007/s00382-013-1833-6.
Kim, S. T., H.-I. Jeong, and F.-F. Jin, 2017: Mean bias in seasonal forecast model and ENSO prediction error. Sci. Rep., 7, 6029, https://doi.org/10.1038/s41598-017-05221-3.
Kirtman, B. P., and Coauthors, 2014: The North American Multimodel Ensemble (NMME): Phase-1 seasonal to interannual prediction, phase-2 toward developing intraseasonal prediction. Bull. Amer. Meteor. Soc., 95, 585–601, https://doi.org/10.1175/BAMS-D-12-00050.1.
Krishnamurthy, V., 2019: Predictability of weather and climate. Earth Space Sci., 6, 1043–1056, https://doi.org/10.1029/2019EA000586.
Kug, J.-S., S.-I. An, Y.-G. Ham, and I.-S. Kang, 2010: Changes in El Niño and La Niña teleconnections over North Pacific–America in the global warming simulations. Theor. Appl. Climatol., 100, 275–282, https://doi.org/10.1007/s00704-009-0183-0.
Latif, M., and Coauthors, 1998: A review of the predictability and prediction of ENSO. J. Geophys. Res. Oceans, 103, 14 375–14 393, https://doi.org/10.1029/97JC03413.
L’Heureux, M. L., M. K. Tippett, and A. G. Barnston, 2015: Characterizing ENSO coupled variability and its impact on North American seasonal precipitation and temperature. J. Climate, 28, 4231–4245, https://doi.org/10.1175/JCLI-D-14-00508.1.
L’Heureux, M. L., A. F. Z. Levine, M. Newman, C. Ganter, J.-J. Luo, M. K. Tippett, and T. N. Stockdale, 2020: ENSO prediction. El Niño Southern Oscillation in a Changing Climate. Geophys. Monogr., Amer. Geophys. Union, 528 pp.
Li, G., and S.-P. Xie, 2012: Origins of tropical-wide SST biases in CMIP multi-model ensembles. Geophys. Res. Lett., 39, L22703, https://doi.org/10.1029/2012GL053777.
Li, G., and S.-P. Xie, 2014: Tropical biases in CMIP5 multimodel ensemble: The excessive equatorial Pacific cold tongue and double ITCZ problems. J. Climate, 27, 1765–1780, https://doi.org/10.1175/JCLI-D-13-00337.1.
Lim, E. P., H. H. Hendon, S. Langford, and O. Alves, 2012: Improvements in POAMA2 for the prediction of major climate drivers and south eastern Australian rainfall. CAWCR Tech. Rep. 51, 23 pp., http://www.bom.gov.au/research/publications/cawcrreports/CTR_051.pdf.
Lin, H., and J. Derome, 2004: Nonlinearity of the extratropical response to tropical forcing. J. Climate, 17, 2597–2608, https://doi.org/10.1175/1520-0442(2004)017<2597:NOTERT>2.0.CO;2.
Liou, C. S., and Coauthors, 1997: The second-generation global forecast system at the central weather bureau in Taiwan. Wea. Forecasting, 12, 653–663, https://doi.org/10.1175/1520-0434-12.3.653.
Liu, J., H. Wang, E. Lu, and A. Kumar, 2019: Decadal modulation of East China winter precipitation by ENSO. Climate Dyn., 52, 7209–7223, https://doi.org/10.1007/s00382-016-3427-6.
Livezey, R. E., M. Masutani, A. Leetmaa, H. Rui, M. Ji, and A. Kumar, 1997: Teleconnective response of the Pacific–North American region atmosphere to large central equatorial pacific SST anomalies. J. Climate, 10, 1787–1820, https://doi.org/10.1175/1520-0442(1997)010<1787:TROTPN>2.0.CO;2.
Luo, J.-J., S. Masson, E. Roeckner, G. Madec, and T. Yamagata, 2005: Reducing climatology bias in an ocean–atmosphere CGCM with improved coupling physics. J. Climate, 18, 2344–2360, https://doi.org/10.1175/JCLI3404.1.
Magnusson, L., M. Alonso-Balmaseda, and F. Molteni, 2013: On the dependence of ENSO simulation on the coupled model mean state. Climate Dyn., 41, 1509–1525, https://doi.org/10.1007/s00382-012-1574-y.
Manganello, J. V., and B. Huang, 2009: The influence of systematic errors in the southeast Pacific on ENSO variability and prediction in a coupled GCM. Climate Dyn., 32, 1015–1034, https://doi.org/10.1007/s00382-008-0407-5.
McPhaden, M. J., 1999: Genesis and evolution of the 1997-98 El Niño. Science, 283, 950–954, https://doi.org/10.1126/science.283.5404.950.
McPhaden, M. J., S. E. Zebiak, and M. H. Glantz, 2006: ENSO as an integrating concept in Earth science. Science, 314, 1740–1745, https://doi.org/10.1126/science.1132588.
Merryfield, W. J., and Coauthors, 2013: The Canadian seasonal to interannual prediction system. Part I: Models and initialization. Mon. Wea. Rev., 141, 2910–2945, https://doi.org/10.1175/MWR-D-12-00216.1.
Min, Y.-M., V. N. Kryjov, S. M. Oh, and H. Lee, 2017: Skill of real-time operational forecasts with the APCC multi-model ensemble prediction system during the period 2008–2015. Climate Dyn., 49, 4141–4156, https://doi.org/10.1007/s00382-017-3576-2.
Mo, K. C., and R. E. Livezey, 1986: Tropical–extratropical geopotential height teleconnections during the Northern Hemisphere winter. Mon. Wea. Rev., 114, 2488–2515, https://doi.org/10.1175/1520-0493(1986)114<2488:TEGHTD>2.0.CO;2.
Molod, A., L. Takacs, M. Suarez, J. Bacmeister, I.-S. Song, and A. Eichmann, 2012: The GEOS-5 atmospheric general circulation model: Mean climate and development from MERRA to Fortuna. NASA Tech. Rep. Series on Global Modeling and Data Assimilation, Vol. 28 (NASA/TM-2012-104606/Vol 28), 115 pp., http://gmao.gsfc.nasa.gov/pubs/docs/tm28.pdf.
Peng, P., A. Kumar, and W. Wang, 2011: An analysis of seasonal predictability in coupled model forecasts. Climate Dyn., 36, 637–648, https://doi.org/10.1007/s00382-009-0711-8.
Quan, X., M. P. Hoerling, J. Whitaker, G. Bates, and T. Xu, 2006: Diagnosing sources of U.S. seasonal forecast skill. J. Climate, 19, 3279–3293, https://doi.org/10.1175/JCLI3789.1.
Reichler, T. J., and J. O. Roads, 2003: The role of boundary and initial conditions for dynamical seasonal predictability. Nonlinear Processes Geophys., 10, 211–232, https://doi.org/10.5194/npg-10-211-2003.
Reynolds, R. W., T. M. Smith, C. Liu, D. B. Chelton, K. S. Casey, and M. G. Schlax, 2007: Daily high-resolution-blended analyses for sea surface temperature. J. Climate, 20, 5473–5496, https://doi.org/10.1175/2007JCLI1824.1.
Ropelewski, C. F., and M. S. Halpert, 1986: North American precipitation and temperature patterns associated with the El Niño/Southern Oscillation (ENSO). Mon. Wea. Rev., 114, 2352–2362, https://doi.org/10.1175/1520-0493(1986)114<2352:NAPATP>2.0.CO;2.
Saha, S., and Coauthors, 2014: The NCEP Climate Forecast System version 2. J. Climate, 27, 2185–2208, https://doi.org/10.1175/JCLI-D-12-00823.1.
Vannière, B., E. Guilyardi, G. Madec, F. J. Doblas-Reyes, and S. Woolnough, 2013: Using seasonal hindcasts to understand the origin of the equatorial cold tongue bias in CGCMs and its impact on ENSO. Climate Dyn., 40, 963–981, https://doi.org/10.1007/s00382-012-1429-6.
Wallace, J. M., and D. S. Gutzler, 1981: Teleconnections in the geopotential height field during the Northern Hemisphere winter. Mon. Wea. Rev., 109, 784–812, https://doi.org/10.1175/1520-0493(1981)109<0784:TITGHF>2.0.CO;2.
Wang, H., and R. Fu, 2000: Winter monthly mean atmospheric anomalies over the North Pacific and North America associated with El Niño SSTs. J. Climate, 13, 3435–3447, https://doi.org/10.1175/1520-0442(2000)013<3435:WMMAAO>2.0.CO;2.
Wang, H., and A. Kumar, 2015: Assessing the impact of ENSO on drought in the U.S. Southwest with NCEP climate model simulations. J. Hydrol., 526, 30–41, https://doi.org/10.1016/j.jhydrol.2014.12.012.
Watterson, I. G., 2015: Improved simulation of regional climate by global models with higher resolution: Skill scores correlated with grid length. J. Climate, 28, 5985–6000, https://doi.org/10.1175/JCLI-D-14-00702.1.
Wu, A., and W. W. Hsieh, 2004: The nonlinear Northern Hemisphere winter atmospheric response to ENSO. Geophys. Res. Lett., 31, L02203, https://doi.org/10.1029/2003GL018885.
Yang, S.-C., M. Rienecker, and C. Keppenne, 2010: The impact of ocean data assimilation on seasonal-to-interannual forecasts: A case study of the 2006 El Niño event. J. Climate, 23, 4080–4095, https://doi.org/10.1175/2010JCLI3319.1.
Zhang, T., M. P. Hoerling, J. Perlwitz, D.-Z. Sun, and D. Murray, 2011: Physics of U.S. surface temperature response to ENSO. J. Climate, 24, 4874–4887, https://doi.org/10.1175/2011JCLI3944.1.
Zhang, T., J. Perlwitz, and M. P. Hoerling, 2014: What is responsible for the strong observed asymmetry in teleconnections between El Niño and La Niña? Geophys. Res. Lett., 41, 1019–1025, https://doi.org/10.1002/2013GL058964.