Systematic Comparison of ENSO Teleconnection Patterns between Models and Observations

Xiaosong Yang Center for Ocean–Land–Atmosphere Studies, Calverton, Maryland

Search for other papers by Xiaosong Yang in
Current site
Google Scholar
PubMed
Close
and
Timothy DelSole George Mason University, Fairfax, Virginia, and Center for Ocean–Land–Atmosphere Studies, Calverton, Maryland

Search for other papers by Timothy DelSole in
Current site
Google Scholar
PubMed
Close
Full access

Abstract

This paper applies a new field significance test to establish the existence and consistency of ENSO teleconnection patterns across models and observations. An ENSO teleconnection pattern is defined as a field of regression coefficients between an index of the tropical Pacific sea surface temperature and a field of variables such as surface air temperature or precipitation. The test is applied to boreal winter and summer in six continents using observations and hindcasts from the Development of a European Multimodel Ensemble System for Seasonal-to-Interannual Prediction (DEMETER) and the ENSEMBLE-based predictions of climate changes and their impacts (ENSEMBLES) projects. This comparison represents one of the most comprehensive and up-to-date assessments of the extent to which ENSO teleconnection patterns exist and can be reproduced by coupled models.

Statistically significant ENSO teleconnection patterns are detected in both observations and models and in all continents and in both winter and summer seasons, except in two cases: 1) Europe (both seasons and variables), and 2) North America (both variables in boreal summer). Despite many ENSO teleconnection patterns being significant, however, the patterns do not necessarily agree between observations and models. The degree of agreement between models and observations is characterized as “robust,” “moderate,” or “low.” Only Australia and South America are found to have robust agreement between ENSO teleconnection patterns, and then only for limited seasons and variables. Although many of our conclusions regarding teleconnection patterns conform to previous studies, there are exceptions, including the fact that the teleconnection for boreal winter precipitation is generally accepted to exist in Africa but in fact has only low agreement with model simulations, while that in Asia is not widely recognized to exist but is found to be significant and in moderate agreement with model teleconnections.

Current affiliation: UCAR and NOAA/GFDL, Princeton, New Jersey.

Corresponding author address: Xiaosong Yang, NOAA/Geophysical Fluid Dynamics Laboratory, 201 Forrestal Rd., Princeton, NJ 08540. E-mail: xiaosong.yang@noaa.gov

Abstract

This paper applies a new field significance test to establish the existence and consistency of ENSO teleconnection patterns across models and observations. An ENSO teleconnection pattern is defined as a field of regression coefficients between an index of the tropical Pacific sea surface temperature and a field of variables such as surface air temperature or precipitation. The test is applied to boreal winter and summer in six continents using observations and hindcasts from the Development of a European Multimodel Ensemble System for Seasonal-to-Interannual Prediction (DEMETER) and the ENSEMBLE-based predictions of climate changes and their impacts (ENSEMBLES) projects. This comparison represents one of the most comprehensive and up-to-date assessments of the extent to which ENSO teleconnection patterns exist and can be reproduced by coupled models.

Statistically significant ENSO teleconnection patterns are detected in both observations and models and in all continents and in both winter and summer seasons, except in two cases: 1) Europe (both seasons and variables), and 2) North America (both variables in boreal summer). Despite many ENSO teleconnection patterns being significant, however, the patterns do not necessarily agree between observations and models. The degree of agreement between models and observations is characterized as “robust,” “moderate,” or “low.” Only Australia and South America are found to have robust agreement between ENSO teleconnection patterns, and then only for limited seasons and variables. Although many of our conclusions regarding teleconnection patterns conform to previous studies, there are exceptions, including the fact that the teleconnection for boreal winter precipitation is generally accepted to exist in Africa but in fact has only low agreement with model simulations, while that in Asia is not widely recognized to exist but is found to be significant and in moderate agreement with model teleconnections.

Current affiliation: UCAR and NOAA/GFDL, Princeton, New Jersey.

Corresponding author address: Xiaosong Yang, NOAA/Geophysical Fluid Dynamics Laboratory, 201 Forrestal Rd., Princeton, NJ 08540. E-mail: xiaosong.yang@noaa.gov

1. Introduction

It is well established that the El Niño–Southern Oscillation (ENSO) phenomenon originates from coupled ocean–atmosphere interactions localized in the tropical Pacific but affects the global climate. In fact, ENSO is the primary source of seasonal to interannual climate variability and predictability in many regions of the globe (Shukla and Kinter 2006). ENSO-teleconnected climate variability is described in many papers and details can be obtained in the review papers by Wallace et al. (1998) and Trenberth et al. (1998) and references therein. ENSO and ENSO-teleconnected predictability are reviewed by Latif et al. (1998), Shukla and Kinter (2006), and National Research Council (2010).

The ENSO-teleconnected regression pattern, which is a map of regression coefficients of a climate variable upon the ENSO index, is called the ENSO teleconnection pattern. It provides useful guidance for climate prediction associated with ENSO. For instance, for every unit change in the ENSO index, the regression coefficient gives the expected mean change in the climate variable. A series of studies by Ropelewski and Halpert (1986, 1987, 1989, 1996) and Halpert and Ropelewski (1992) examined and documented large-scale patterns of above- and below-average standardized precipitation and surface temperature associated with both the warm and cold phases of the ENSO. In these studies, seasons and regions of strong, consistent ENSO–precipitation and ENSO–temperature relationships were identified, using surface meteorological data, and a schematic summarizing the significant teleconnections was presented in a global map [e.g., Fig. 1 of Ropelewski and Halpert (1996), Fig. 13 of Halpert and Ropelewski (1992), and Fig. 2 of Trenberth et al. (1998)]. The schematic map, called the R-H map hereafter, is often used to summarize our understanding of consistent ENSO relations with surface temperature and precipitation. However, a systematic comparison between observed teleconnection patterns and the teleconnections simulated by state-of-art coupled models has not yet been documented in a single paper.

One problem with systematically comparing regression patterns between models and observations is deciding whether the particular patterns are statistically significant—it does not make sense to compare patterns if they are not clearly distinguished from natural variability. Local significance tests, that is, testing the significance of a single regression coefficient at every point in a regression map, is often used as part of testing the field significance of a regression map (von Storch and Zwiers 1999, chapter 17.4). However, the fact that local tests are not independent of each other needs to be taken into account, a fact recognized at least since Walker (1923). Also, variables in gridded datasets tend to be correlated with nearby variables, implying that if a particular variable is found to be significant then nearby variables will tend to be significant as well, even if the true regression coefficient vanishes.

Recently, DelSole and Yang (2011) proposed a new method for testing the significance of a regression pattern relative to the null hypothesis that the regression coefficients vanish simultaneously. The method is based on multivariate regression and simultaneously accounts for the multiplicity and interdependence of the tests. The resulting significance test clarifies the connection between field significance, multivariate regression, and canonical correlation analysis (CCA). The method was applied to diagnose long-term trends in annual average sea surface temperature and boreal winter 300-hPa zonal winds.

In this paper, we apply the above field significance test to both models and observations to determine ENSO teleconnection patterns of surface temperature and precipitation globally and in individual continents using the Development of a European Multimodel Ensemble System for Seasonal-to-Interannual Prediction (DEMETER) (Palmer et al. 2004) and the ENSEMBLE-based predictions of climate changes and their impacts (ENSEMBLES) (Weisheimer et al. 2009) data and observations. Our main goals are to establish the existence and consistency of ENSO teleconnection patterns across models and observations. Details of the regression pattern technique are discussed in section 2. The data are described in section 3. In section 4, ENSO teleconnection regression patterns over land in boreal winter and summer are compared between observations and coupled model simulations. Conclusions are given in section 5.

2. Review of field significance based on multivariate regression

In this section we give a brief review of the field significance test proposed by DelSole and Yang (2011). Let y be a field represented by M numbers and x be a climate index. The N observed values of the field can be collected in the N × M matrix and similarly the climate index collected in the N-dimensional vector X. Field significance can be framed as a hypothesis test for the multivariate regression model
e1
where β is an M-dimensional row vector of regression coefficients, called the regression pattern, and ϵ represents random errors. We assume that the time mean has been removed from X and . In this case, field significance is tantamount to testing the hypothesis that all regression coefficients vanish: β = 0. A test for this hypothesis can be derived from the generalized likelihood ratio test. If the errors are independent in time and drawn from a multivariate normal distribution with zero mean, then this test leads to a ratio of determinants of certain covariance matrices. It turns out that this ratio can be written strictly in terms of the canonical correlations between the field and the climate index, thereby revealing a fundamental connection to CCA. Furthermore, the statistic is invariant to switching the x and y variables. This invariance implies that testing the hypothesis β = 0 in (1) is equivalent to testing the hypothesis β′ = 0 in the model
e2
where X and have been switched relative to (1), and β′ is an M-dimensional column vector of coefficients, called the projection pattern. Happily, the latter hypothesis test is standard in regression analysis. The test statistic depends on the multiple correlation coefficient R, which is defined as the maximum correlation between X and a linear combination of . The multiple correlation also is equivalent to the canonical correlation ρ between X and . Specifically, if the null hypothesis β′ = 0 is true, then the statistic
e3
has an F distribution with M and NM − 1 degrees of freedom, where ρ2 is the squared canonical correlation coefficient between X and [or equivalently the square multiple correlation, since there is only one predictand in (2).].

In practice, the above field significance test cannot be applied if the number of elements in the field exceeds the sample size, which is often the case in climate analysis. This problem is not merely a practical limitation but rather is fundamental. Specifically, if the number of elements in the field exceeds the sample size, then we are effectively testing more hypotheses than samples, which is an ill-posed problem. To overcome this problem, DelSole and Yang (2011) proposed representing the field using a small number of leading principal components (PCs). A key question is how many PCs should be chosen. This question is effectively a model selection problem. A common approach to model selection is to apply cross-validation methods to estimate the independent prediction error and then to choose the model that minimizes the cross-validated error. Unfortunately, the cross-validated error often is a slowly varying function of the number of principal components, implying that the model with the minimum error is not well separated from others with fewer predictors. Following Hastie et al. (2003), DelSole and Yang (2011) proposed estimating a confidence interval for the cross-validated mean square error (CVMSE) and then choosing the most parsimonious model (i.e., the model with the fewest predictors) within the confidence interval of the minimum CVMSE. The full details of this procedure are discussed in DelSole and Yang (2011). A detailed application of this methodology will be discussed in section 3.

It turns out to be instructive to define the cross-validated skill score (CVSS)
e4
which is a measure of fraction of variance explained by the regression model and hence can be compared directly to the squared canonical correlation. A confidence interval for CVSS can be estimated from the cross-validation experiments, after which we choose the simplest model (i.e., the model with the fewest PCs) that is within the confidence interval of the model with maximum CVSS.
Once the number of PCs is selected, it is trivial to determine the squared canonical correlation, the regression pattern β, and the projection pattern β′ in PC space. The projection pattern in PC space is merely the regression coefficient between x and each PC. Then, the regression pattern and the projection pattern are transformed from PC space to data space. The exact solutions of the regression pattern and the projection pattern can be found in DelSole and Yang (2011). In addition, DelSole and Yang (2011) show that the fraction of variance explained by the regression pattern is
e5
We remind the reader that β is a row vector and X is a column vector, so FEV is a scalar.

3. Data

The observational estimates used in this study are the monthly mean 2-m temperature (T2m) over land from the National Centers for Environmental Prediction (NCEP)–National Center for Atmospheric Research (NCAR) reanalysis (Kalnay et al. 1996; Kistler et al. 2001), and the monthly mean precipitation over land from NCEP–Climate Prediction Center (Chen et al. 2002). The time period analyzed in both datasets is from 1948 to 2010, and both datasets are on a 2.5° × 2.5° grid. The observed monthly Niño-3.4 index during the period 1948–2010 is computed from the Extended Reconstruction Sea Surface Temperature analysis version 3b (Smith and Reynolds 2004).

The model simulations used in this paper are the hindcast integrations of the DEMETER and ENSEMBLES projects. The DEMETER dataset, reviewed in Palmer et al. (2004), consists of 6-month hindcasts by seven global coupled ocean–atmosphere models. Each hindcast model produced a 9-member ensemble hindcast, which were averaged to construct ensemble mean hindcasts. We chose two coupled model hindcasts from the European Centre for Medium-Range Weather Forecasts (ECMWF) coupled model (ECM) and Météo France coupled model (MET) since they cover the entire period 1958–2001. The DEMETER hindcasts were initialized at 1 February, 1 May, 1 August, and 1 November and integrated for the subsequent 6 months. We chose hindcasts initialized at 1 May and 1 November. The data are on a 2.5° × 2.5° grid. During the course of this study, the ENSEMBLES dataset, a new generation of DEMETER, was created using a multimodel ensemble of five state-of-art coupled atmosphere–ocean circulation models (Weisheimer et al. 2009). It provides a new 46-yr (1960–2005) hindcast dataset for seasonal-to-annual ensemble predictions. We also chose hindcasts of ENSEMBLES initialized at 1 May and 1 November. The five models in ENSEMBLES are from the ECMWF (M1), the Leibniz Institute of Marine Sciences at Kiel University (M2), Météo France (M3), the Met Office (M4), and the Euro-Mediterranean Centre for Climate Change in Bologna (M5).

Only three-month averages for winter and summer seasons are considered in this study. The boreal winter of a year refers to December–January–February (DJF), and the December refers to the current year while January–February the following year. The boreal summer of a year refers to June–July–August (JJA) or July–August–September (JAS) of that year. The three-month averages were computed from the monthly datasets. Six continental regions were chosen with boundaries specified in Table 1. The climatological mean over the entire period of each variable at each grid point is subtracted prior to analysis.

Table 1.

Domains of six continents.

Table 1.

We are interested in diagnosing multivariate regression patterns associated with ENSO over different land continents with boundaries specified in Table 1. Accordingly, the prespecified time series x is chosen to be the anomalous standard Niño-3.4 SST index (NINO34) in boreal winter or summer. The time series of NINO34 in DJF and JAS are shown in Fig. 1 for observations and for ECM and MET model hindcasts from DEMETER datasets. The NINO34 in DJF is well predicted by the two models, since the correlation coefficients between observed and hindcasted NINO34 are about 0.95 for the two models. In contrast, the correlation coefficients between observations and hindcasts for NINO34 in JAS are smaller, for example, 0.65 for ECM and 0.79 for MET, indicating the seasonally dependent predictability of ENSO (Latif et al. 1998). The weaker correlations for JAS hindcasts presumably arise from the “spring barrier” effect (see Torrence and Webster 1998) since the corresponding initial conditions are 1 May. A similar contrast between correlation skills for DJF and JAS hindcasts also occurs in ENSEMBLES hindcasts (not shown). It should be recognized, however, that these weaker correlations between observations and hindcasts do not imply weaker teleconnection patterns. We emphasize that we are not investigating the predictability of ENSO teleconnection patterns, but rather we are investigating whether the model can reproduce the ENSO teleconnection pattern over land continents given the simultaneous ENSO signals in the tropical Pacific.

Fig. 1.
Fig. 1.

Time series of the Niño-3.4 index for (a) DJF and (d) JAS seasons from observations during the period 1948–2010, (b) DJF and (e) JAS seasons from ECM, and (c) DJF and (f) JAS from MET model hindcasts during the period 1958–2001. The correlation between hindcasts and observations computed over the period 1958–2001 are indicated in the bottom right of the respective panels.

Citation: Journal of Climate 25, 2; 10.1175/JCLI-D-11-00175.1

4. Diagnosing ENSO teleconnection regression patterns over land in boreal winter and summer

a. Detailed example

To illustrate how we apply the new field significance test and identify the significant ENSO teleconnection pattern, we start with an example for DJF T2m in observations and two DEMETER hindcast simulations over North America. The squared canonical correlation ρ2 between NINO34 and the leading PCs of the DJF T2m over North American, as a function of the number of PCs used to represent the data, is shown in the left panels of Fig. 2 for observations and two model simulations. Comparison with the 5% significance curve for ρ2, shown as the dashed line in the left panels of Fig. 2, indicates that the canonical correlation estimated from observations is marginally significant to insignificant for the first few PCs but then becomes statistically significant for 9 or more PCs for the NCEP reanalysis (Fig. 2a) and 6 or more PCs for MET hindcasts (Fig. 2e), while it is significant for all PC truncations for ECM hindcasts (Fig. 2c). Note that the simulated variables are computed from the average of 9 ensemble members, and hence ρ2 is expected to be larger in models. The “jump” in ρ2 around PC9 in Fig. 2a and PC6 in Fig. 2e occur because these PCs are strongly correlated with the Niño-3.4 index. In fact, remarkably, the difference in ρ2 at the jump is precisely equal to the squared correlation between the Niño-3.4 index and the PC at the jump (see DelSole and Yang 2011 for proof). The canonical correlations are significant even for the first PC in Fig. 2c because it happens that in this particular model the pattern that explains the most variance of the ensemble mean forecast also is highly correlated with the Niño-3.4 index.

Fig. 2.
Fig. 2.

(a) The squared canonical correlation (blue dotted) and associated 5% significance level (red dashed) between the Niño-3.4 SST index and the DJF 2-m temperature in North America, as a function of the number of PCs used to represent the data, for (a) NCEP reanalyses during the period 1948–2010, (c) ECM, and (e) MET during the period 1958–2001. The cross-validated skill score (blue dotted) and associated standard errors (red dotted) of predicting the ENSO regression pattern using the canonical component, as a function of the number of PCs, for (b) NCEP reanalyses during the period 1948–2010, (d) ECM, and (f) MET during the period 1958–2001. The error bar shows the standard error for the maximum cross-validated skill score.

Citation: Journal of Climate 25, 2; 10.1175/JCLI-D-11-00175.1

We perform cross validation as described in section 2. The cross-validated skill score of predicting NINO34 from the leading PCs is shown in the right panels of Fig. 2. The figures show that the maximum skill score occurs at 18 PCs for observation, 4 PCs for ECM, and 11 PCs for MET. Following DelSole and Yang (2011), by the principle of parsimony we chose the simplest model (i.e., the model with the fewest number of PCs) that lies within the confidence interval of the maximum skill score, namely 9 PCs for observations, 3 PCs for EMC, and 7 PCs for MET. After selecting the number of PCs, we compare ρ2 of the selected PCs with the 5% significance curve to verify that it is significant at that truncation. For this case, ρ2 is 0.47 with 9 PCs for observations, 0.78 with 3 PCs for ECM, and 0.63 with 7 PCs for MET, and they all are significant (left panels in Fig. 2). Consistent with this, the confidence intervals for CVSS do not include 0.

The canonical regression patterns derived from the selected PCs are shown in the upper panels of Fig. 3. The canonical patterns in observations and two model simulations all show a south–north dipole, with cold anomalous centers concentrated in the Northern Mexico and Southern United States, and warm anomalous centers covering most of Canada. The pattern derived from the ECM model agrees with that from the NCEP reanalysis, while the magnitudes for the pattern from the MET are smaller than those for ECM as well as NCEP reanalysis, and the cold anomalous centers in the MET pattern (Fig. 3c) shift from the northeastern United States to midwestern United States relative to the NCEP reanalysis and ECM (Figs. 3a,b).

Fig. 3.
Fig. 3.

The (left) observed, (middle) ECM, and (right) MET (top) canonical ENSO regression patterns (shading interval of 0.02°C per NINO34) and (bottom) associated canonical projection patterns for DJF 2-m temperature in the North America. The observed values are from the NCEP reanalyses during the period 1948–2010, and model results are from the DEMETER ECM and MET ensemble mean hindcasts during the period 1958–2001.

Citation: Journal of Climate 25, 2; 10.1175/JCLI-D-11-00175.1

The canonical projection patterns for the T2m in the NCEP reanalysis and two model hindcasts are shown in the lower panels of Fig. 3. This pattern gives the weights to be applied to T2m to construct a predictor of NINO34. The projection patterns show similar negative centers concentrated in the Northern Mexico and Southern United States for observations and model simulations, while they differ in the rest of the domain. Projection vectors tend to be sensitive to components with small variances and to the number of PCs; and so close agreement among projection vectors from different data sources is not expected.

The above procedure was repeated for all model hindcasts in both the DEMETER and ENSEMBLES dataset. We also repeated this procedure for T2m and precipitation in DJF, JJA, and JAS seasons over six continents. Because of the limitations of space, it is not possible to show results from cross-validation experiments and significance tests with selected PCs for all cases; nor even the teleconnection patterns for all seven models. Instead, we choose two DEMETER models for display purposes and summarize the number of selected PCs, the squared canonical correlation, and the cross-validated skill score for all cases in Tables 2 and 3. The CVSS and ρ2 for T2m and precipitation in JAS over six continents are generally larger than those in JJA, so we chose JAS as boreal summer season for display purposes.

Table 2.

The number of selected PCs, CVSS, the squared canonical correlation ρ2, and the FEV of regression patterns over 6 continents for the 2-m temperature with ENSO.

Table 2.
Table 3.

The number of selected PCs, CVSS, the squared canonical correlation ρ2, and the FEV of regression patterns over six continents for precipitation with ENSO.

Table 3.

One might wonder how a regression pattern derived from truncated PCs compares with the corresponding point-by-point regression pattern. In all cases, the two patterns are nearly indistinguishable, consistent with the previous findings of DelSole and Yang (2011). Because the two patterns are nearly indistinguishable, we show only the pattern derived from truncated PCs since this is the pattern specifically identified in the field significance test.

To quantify the agreement between models and observations, we give in Table 4 the spatial correlation between the observed teleconnection pattern and model teleconnection pattern (the spatial mean is not subtracted out when computing the spatial correlation). We use anomaly correlation rather than mean square difference because the anomaly correlation is more easily interpreted and emphasizes overall agreement of the patterns irrespective of the magnitudes. Based on our subjective sense of agreement based on visual inspection of all regression patterns, not all of which are shown in the paper, we have categorized the results into three broad groups. Specifically, we define the ENSO teleconnection patterns to be in “robust agreement” if all seven spatial correlation coefficients are greater than 0.75, “low agreement” if at least one correlation coefficient is less than 0.5, and “moderate agreement” for everything else. We acknowledge that these criteria are arbitrary, but one must “draw the line” somewhere.

Table 4.

The spatial correlation coefficients between observed and modeled ENSO teleconnection patterns for the 2-m temperature and precipitation.

Table 4.

b. The ENSO teleconnection pattern over global land

Applying the significance test to an individual continent will be called a local field test, while applying to over global land will be called a global field test. We applied the global field test for T2m and precipitation in DJF, JJA, and JAS seasons using observations, two hindcasts from DEMETER, and 5 hindcasts from ENSEMBLES. The number of selected PCs, the squared canonical correlation, and the cross-validated skill score for the regression patterns over global land are summarized in Table 5.

Table 5.

The number of selected PCs, CVSS, the squared canonical correlation ρ2, and the fraction of explained variance (FEV) of the regression patterns over global land for the 2-m temperature and precipitation with ENSO.

Table 5.

The squared canonical correlations between NINO34 and DJF T2m over global land are significant at 5% significance level for the NCEP reanalysis and 7 hindcasts. The ENSO teleconnection patterns of DJF T2m over global land are shown in Fig. 4. The patterns generally agree on an anomalous dipole in North America, warm anomalies in the north South America, weak warm anomalies in most Africa, and warm anomalies in the northern Australia.

Fig. 4.
Fig. 4.

The (a) observed and (b)–(g) seven simulated canonical ENSO regression patterns for DJF 2-m temperature. Shading intervals are 0.2°C per NINO34. The observed values are from the NCEP reanalyses during the period 1948–2010, two model results (ECM and MET) from DEMETER ensemble mean hindcasts during the period 1958–2001, and five model results (M1–M5) from ENSEMBLES ensemble mean hindcasts during the period 1960–2005.

Citation: Journal of Climate 25, 2; 10.1175/JCLI-D-11-00175.1

The squared canonical correlations between NINO34 and JAS T2m over global land are significant at 5% significance level for the NCEP reanalysis and 7 hindcasts. The ENSO teleconnection patterns of JAS T2m over global land are shown in Fig. 5. There is relatively little agreement between the patterns, except for a general warming in Africa.

Fig. 5.
Fig. 5.

As in Fig. 4, but for JAS.

Citation: Journal of Climate 25, 2; 10.1175/JCLI-D-11-00175.1

The squared canonical correlations between NINO34 and DJF precipitation over global land are significant at 5% significance level for the NCEP reanalysis and 7 hindcasts. The ENSO teleconnection patterns of DJF precipitation over global land are shown in Fig. 6. The patterns generally agree on dry anomalies in equatorial South America and wet anomalies in eastern Australia.

Fig. 6.
Fig. 6.

As in Fig. 4, but for precipitation. Shading intervals are 0.1 mm day−1 per NINO34.

Citation: Journal of Climate 25, 2; 10.1175/JCLI-D-11-00175.1

The squared canonical correlations between NINO34 and JAS precipitation over global land are significant at 5% significance level for the NCEP reanalysis and 7 hindcasts. The ENSO teleconnection patterns of JAS precipitation over global land are shown in Fig. 7. The patterns generally agree on wet anomalies in Mexico and southern United States, dry anomalies in southern Africa, wet anomalies in eastern China, and wet anomalies in northern Australia.

Fig. 7.
Fig. 7.

As in Fig. 6, but for JAS.

Citation: Journal of Climate 25, 2; 10.1175/JCLI-D-11-00175.1

Despite all eight patterns over the global land being significant and the general agreement found in some regions, numerous discrepancies exist between observed and simulated patterns in other regions. For instance, the observed pattern of DJF T2m predicts almost nothing in Europe while some models predict warming anomalies and others predict cooling anomalies (Fig. 4). This raises the issue of whether the methodology is sensitive to the choice of domain. To address this issue, we compare the patterns from the local field test and the global field test in six continents. The pattern in the local domain is extracted from the global field test. The spatial correlation coefficients between the two patterns from the same model are larger than 0.95, provided that the local field test is significant, indicating that the methodology is not sensitive to the choice of domain.

c. Africa

The squared canonical correlations between NINO34 and DJF T2m in Africa are significant at 5% significance level for the NCEP reanalysis, ECM and MET hindcasts, with respective correlation coefficients 0.78, 0.66, and 0.79. The observed ENSO teleconnection pattern of DJF T2m in Africa is shown in Fig. 8a. This pattern shows two anomalous warm centers over Northwestern Africa and Southern Africa with amplitudes over 0.3°C per NINO34, which are reproduced by both model simulations (Figs. 8b,c) and by the five model simulations in ENSEMBLES (Fig. 4). However, the observed teleconnection has a cold anomalous area centered in the north of Sudan, which is not reproduced by model simulations. Nevertheless, the spatial correlations between any pair of observed and simulated patterns are greater than 0.5, so we conclude these patterns are in moderate agreement.

Fig. 8.
Fig. 8.

The observed (left) and two simulated (middle) ECM and (right) MET canonical ENSO regression patterns for (top) DJF 2-m temperature (shading interval of 0.2°C) and (bottom) DJF precipitation (shading interval of 0.1 mm day−1) in Africa. The observed values are from the NCEP reanalyses during the period 1948–2010, and model results from ECM and MET ensemble mean simulations during the period 1958–2001.

Citation: Journal of Climate 25, 2; 10.1175/JCLI-D-11-00175.1

The squared canonical correlations between NINO34 and DJF precipitation in Africa are significant at 5% significance level for observations, ECM, and MET hindcasts, with respective correlation coefficients 0.36, 0.53, and 0.75. The observed ENSO teleconnection precipitation pattern for DJF is shown in Fig. 8d. This pattern has a general tripole with dry anomalies over the entire southeastern Africa, wet anomalies in the eastern tropical Africa and the south part of middle Africa, and dry anomalies in the Guinea coast region. While both models and observations agree on the general tripole structure of the Africa teleconnection pattern (Figs. 8e,f), the exact location and transition zones are not the same, which limits the usefulness of this pattern for making predictions. The model patterns have low agreement with observations according to our spatial correlation criterion because two out seven models have spatial correlations less than 0.5.

In JAS season, the squared canonical correlations between NINO34 and T2m in Africa are significant at 5% significance level for the NCEP reanalysis, ECM, and MET hindcasts, with respective correlation coefficients 0.45, 0.69, and 0.54. The ENSO teleconnection patterns of JAS T2m in Africa are shown in the upper panels of Fig. 9. The patterns generally agree with each other in the north of the equator, especially in the anomalous warming centered in the Sudan and cooling centered in the Northwest corner of Africa. In contrast, one model predicts weak cooling over Zambia whereas the other two patterns imply warming in that region. Nevertheless, this pattern is in moderate agreement by our criterion.

Fig. 9.
Fig. 9.

As in Fig. 8, but for JAS in Africa.

Citation: Journal of Climate 25, 2; 10.1175/JCLI-D-11-00175.1

The squared canonical correlations between NINO34 and JAS precipitation in Africa are significant at 5% significance level for observations, ECM, and MET hindcasts, with respective correlation coefficients 0.52, 0.77, and 0.60. The ENSO teleconnection patterns of JAS precipitation in Africa are shown in the lower panels of Fig. 9. The patterns generally agree with each other in the dry anomalous band between equator and the Sahara. The model patterns also agree with the observed pattern in that there is little to no response south of the equator and in northeast Africa (e.g., Egypt). The model patterns disagree strongly over northwest Africa (e.g., Mali), and over central, equatorial Africa (e.g., west of Gulf of Guinea) and have low agreement with observations by our spatial correlation criterion.

The observed precipitation pattern in DJF is consistent with previous findings (Janowiak 1988; Ropelewski and Halpert 1989), although the pattern is in low agreement with the seven coupled model hindcasts. We also detected a significant ENSO teleconnection precipitation pattern in JAS by our method, and this pattern has not been documented in the R-H map. The squared canonical correlation coefficient of JAS precipitation is even higher than that of DJF precipitation, suggesting a stronger teleconnection in JAS. The observed ENSO-teleconnected T2m patterns for both seasons are consistent with previous studies (Halpert and Ropelewski 1992), and the patterns are significant and in moderate agreement.

d. Asia

The squared canonical correlations between NINO34 and DJF T2m in Asia are significant at 5% significance level for the NCEP reanalysis, ECM, and MET hindcasts, with respective correlation coefficients 0.38, 0.83, and 0.70. The ENSO teleconnection patterns of DJF T2m in Asia are shown in the upper panels of Fig. 10. The patterns generally agree in predicting cold anomalies in northeastern Russian and warm anomalies over India and the Indochina Peninsula. However, the model patterns disagree in the sign of the regression pattern over China, western Russia, and in the location of the maximum coefficients. The five ENSEMBLES model patterns also display discrepancy in the sign and magnitudes of the regression pattern (Fig. 4). Not surprisingly, the patterns are in low agreement by our spatial correlation criterion.

Fig. 10.
Fig. 10.

As in Fig. 8, but for Asia.

Citation: Journal of Climate 25, 2; 10.1175/JCLI-D-11-00175.1

The squared canonical correlations between NINO34 and DJF precipitation in Asia are significant at 5% significance level for observations, ECM, and MET hindcasts, with respective correlation coefficients 0.38, 0.85, and 0.83. The observed ENSO teleconnected precipitation patterns in DJF are shown in the lower panels of Fig. 10. The patterns agree well in predicting wet anomalies concentrated in the southern and eastern China with amplitudes over 0.4 mm day−1 per NINO34 and weak dry anomalies in the southeast corner of the continent. The five ENSEMBLES model patterns also agree well in predicting wet anomalies in eastern China, although the magnitudes in M5 are weaker. However, the model patterns disagree in the magnitudes over the India subcontinent. The patterns are in moderate agreement since the seven spatial correlations all exceed 0.50 but do not consistently exceed 0.75.

The squared canonical correlations between NINO34 and JAS T2m in Asia are significant at 5% significance level for the NCEP reanalysis, ECM, and MET hindcasts, with respective correlation coefficients 0.55, 0.60, and 0.45. The observed and simulated ENSO teleconnection patterns of JAS T2m in Asia are shown in Figs. 11a–c. Aside from warming over India and cooling over the southeastern corner of Russia, the regression patterns differ considerably from each other and are in low agreement by our spatial correlation criterion. The numerous discrepancies suggest that models are not able to capture the ENSO teleconnection pattern of T2m over Asia in JAS, despite all three patterns being significant.

Fig. 11.
Fig. 11.

As in Fig. 10, but for JAS in Asia.

Citation: Journal of Climate 25, 2; 10.1175/JCLI-D-11-00175.1

The squared canonical correlations between NINO34 and JAS precipitation in Asia are significant at 5% significance level for observations, ECM, and MET hindcasts, with respective correlation coefficients 0.33, 0.67, and 0.37. The observed and simulated ENSO-teleconnected precipitation patterns in JAS are shown in Figs. 11d–f. We see that the pattern generated by MET is almost opposite of the other two patterns. Ignoring the MET pattern, the other two patterns broadly agree on dry anomalies in the India subcontinent, southeastern Asia, and northern China. Interestingly, the four ENSEMBLES patterns (M2–M5) generally agree on dry anomalies in the India subcontinent and northern China, while the M1 pattern disagrees with other patterns in the India subcontinent. Nevertheless, the fact that two models are “outliers” indicates that this teleconnection pattern cannot be consistently generated by state-of-the-art seasonal forecast models.

The ENSO teleconnected precipitation pattern in boreal summer has strong dry anomalies over India, which is consistent with the summer monsoon precipitation over peninsular India being suppressed during ENSO years (Rasmusson and Carpenter 1983; Shukla and Paolino 1983), although two out of seven models cannot reproduce these dry anomalies. The dry anomalies over Peninsular India were also not well simulated in the National Oceanic and Atmospheric Administration (NOAA)/NCEP Climate Forecast System (CFS) hindcasts (Liang et al. 2009). Note that the ENSO teleconnection precipitation pattern in DJF has not been documented in the R-H map. The observed and simulated ENSO precipitation patterns in DJF agree on the wet anomalies concentrated in the southern and eastern China. The wet tendency associated with ENSO in the southern and eastern China in DJF is consistent with Wu et al. (2003), and it can be explained with the weak (strong) East Asian winter monsoon associated with warm (cold) events in the eastern Pacific through the Pacific–East Asian Teleconnection (Wang et al. 2000). During the warm (cold) events, an anomalous lower-tropospheric anticyclone (cyclone) locates in the western North Pacific, causing anomalously wet (dry) conditions along the East Asian polar front stretching from southern China northeastward to the east of Japan. The observed temperature pattern in DJF has warm anomalies over Southeast Asia, which is consistent with previous findings (Halpert and Ropelewski 1992).

e. Europe

The squared canonical correlations between NINO34 and DJF T2m in Europe are not significant at 5% significance level for the NCEP reanalysis, but they are significant for ECM and MET hindcasts, with respective correlation coefficients 0.43 and 0.18 (Table 2). The ENSO-teleconnected patterns of T2m in two hindcasts differ from each other (Fig. 4). In JAS season, the squared canonical correlations between NINO34 and T2m in Europe are not significant at 5% significance level for the NCEP reanalysis and two hindcasts (Table 2).

The squared canonical correlations between NINO34 and DJF precipitation in Europe are not significant at 5% significance level for the NCEP reanalysis, but they are significant for ECM and MET hindcasts, with respective correlation coefficients 0.62 and 0.89 (Table 3). The ENSO-teleconnected patterns of precipitation in two hindcasts differ from each other (Fig. 6). In JAS season, the squared canonical correlations between NINO34 and precipitation in Europe are not significant at 5% significance level for the NCEP reanalysis and two hindcasts (Table 3).

Feddersen (2000) found a significant link between simulated temperature anomaly patterns in Europe and ENSO events in boreal winter season using CCA and a weaker relation in boreal summer season. However, he noted that the link that is found in the model is not evident in observations. Our results are consistent with the absence of a significant ENSO teleconnected precipitation–temperature pattern in Europe in the R-H map (Ropelewski and Halpert 1996; Halpert and Ropelewski 1992).

f. Australia

The squared canonical correlations between NINO34 and DJF T2m in Australia are significant at 5% significance level for the NCEP reanalysis, ECM, and MET hindcasts, with respective correlation coefficients 0.57, 0.77, and 0.75. The observed ENSO teleconnection patterns of DJF T2m in Australia are shown in the upper panel of Fig. 12. The patterns are robust and generally agree in predicting warm anomalies in eastern, northern and Western Australia and weak cold anomalies in southeastern Australia.

Fig. 12.
Fig. 12.

As in Fig. 8, but for Australia.

Citation: Journal of Climate 25, 2; 10.1175/JCLI-D-11-00175.1

The squared canonical correlations between NINO34 and DJF precipitation in Australia are significant at 5% significance level for observations, ECM, and MET hindcasts, with respective correlation coefficients 0.49, 0.73, and 0.75. The observed ENSO teleconnection pattern of DJF precipitation in Australia is shown in the lower panel of Fig. 12. The patterns are robust and generally agree in predicting dry anomalies in eastern, northern, and Western Australia and weak wet anomalies in Tasmania.

The squared canonical correlations between NINO34 and JAS T2m in Australia are significant at 5% significance level for the NCEP reanalysis, ECM, and MET hindcasts, with respective correlation coefficients 0.31, 0.62, and 0.52. The ENSO teleconnection patterns of JAS T2m in Australia are shown in the upper panel of Fig. 13. While the patterns agree in predicting cold anomalies in the northern Australia, they differ considerably elsewhere. Since the spatial correlation coefficients in four models are below 0.5, we conclude that there is low agreement in the seven hindcasts (Table 4).

Fig. 13.
Fig. 13.

As in Fig. 12, but for JAS in Australia.

Citation: Journal of Climate 25, 2; 10.1175/JCLI-D-11-00175.1

The squared canonical correlations between NINO34 and JAS precipitation in Australia are significant at 5% significance level for observations, ECM, and MET hindcasts, with respective correlation coefficients 0.28, 0.60, and 0.55. The observed ENSO teleconnection patterns of JAS precipitation in Australia are shown in the lower panel of Fig. 13. The patterns are robust and generally agree on a uniformly dry tendency in the entire continent with maxima concentrated in the eastern Australia.

The observed ENSO teleconnection precipitation patterns in Australia in boreal winter and summer are consistent with the R-H map (Ropelewski and Halpert 1996, 1987) and Mcbridge and Nicholls (1983), and the patterns can be well reproduced in seven coupled model hindcasts. The observed ENSO teleconnected T2m patterns for both seasons are consistent with the R-H map (Halpert and Ropelewski 1992), and the patterns were well simulated in seven hindcasts in boreal winter while the patterns in boreal summer in seven hindcasts differ considerably with observations.

g. South America

The squared canonical correlations between NINO34 and DJF T2m in South America are significant at 5% significance level for the NCEP reanalysis, ECM, and MET hindcasts, with respective correlation coefficients 0.78, 0.79, and 0.87. The ENSO teleconnection patterns of DJF T2m in South America are shown in the upper panel of Fig. 14. The patterns are robust and generally agree on warm anomalies in the northern half of the continent and negative anomalies in Argentina. However, amplitudes of the simulated patterns are considerably larger than those of observed pattern, especially in ECM hindcasts. In addition, the exact location and transition zones of the general dipole in Argentina differ among the three patterns.

Fig. 14.
Fig. 14.

As in Fig. 8, but for South America.

Citation: Journal of Climate 25, 2; 10.1175/JCLI-D-11-00175.1

The squared canonical correlations between NINO34 and DJF precipitation in South America are significant at 5% significance level for observations, ECM, and MET hindcasts, with respective correlation coefficients 0.74, 0.84, and 0.89. The ENSO teleconnection patterns of DJF precipitation in South America are shown in the lower panel of Fig. 14. The patterns display a general “dipole” with dry anomalies in the north and wet anomalies in the south, but the exact location and orientation of the dipole differs between models and observations. The patterns are in low agreement by our spatial correlation criterion.

The squared canonical correlations between NINO34 and JAS T2m in South America are significant at 5% significance level for the NCEP reanalysis, ECM, and MET hindcasts, with respective correlation coefficients 0.83, 0.76, and 0.79. The ENSO teleconnection patterns of JAS T2m in South America are shown in the upper panel of Fig. 15. The model patterns agree fairly well with each other but differ considerably from the observed pattern, especially over Brazil and Argentina (Figs. 15b,c). The patterns in five ENSEMBLES hindcasts also differ considerably from each other (Fig. 5). Consequently, the spatial correlation coefficients in five models are lower than 0.5 (Table 4), indicating low agreement in the patterns of JAS T2m.

Fig. 15.
Fig. 15.

As in Fig. 14, but for JAS in South America.

Citation: Journal of Climate 25, 2; 10.1175/JCLI-D-11-00175.1

The squared canonical correlations between NINO34 and JAS precipitation in South America are significant at 5% significance level for observations, ECM, and MET hindcasts, with respective correlation coefficients 0.47, 0.79, and 0.68. The ENSO teleconnection patterns of JAS precipitation in South America are shown in the lower panel of Fig. 15. The patterns generally agree in predicting dry anomalies in northern South America and weak or no anomalies in the south of the equator. Consistent with the general similarity between modeled and observed patterns, the seven spatial pattern correlation coefficients are higher than 0.57 (Table 4), indicating the pattern of JAS precipitation is in moderate agreement.

The observed ENSO teleconnection precipitation patterns in South America in boreal winter and summer are consistent with the R-H map (Ropelewski and Halpert 1996, 1987), and the patterns can be generally reproduced in seven coupled model hindcasts in boreal summer while not in boreal winter. The observed ENSO teleconnected T2m patterns in both seasons are consistent with the R-H map (Halpert and Ropelewski 1992), and the pattern in boreal winter was well simulated in seven hindcasts while not in boreal summer.

h. North America

The squared canonical correlations between NINO34 and DJF T2m in North America are significant at 5% significance level for the NCEP reanalysis, ECM, and MET hindcasts, with respective correlation coefficients 0.47, 0.78, and 0.63. The ENSO teleconnection patterns of DJF T2m in North America are shown in the upper panel of Fig. 16. The patterns agree on a general “dipole” with cold anomalies in northern Mexico and southern Unites States and warm anomalies in western Canada and Alaska. However, compared with the observed pattern, amplitudes of the ECM pattern are considerably larger especially in western Canada and Alaska, while amplitudes of the MET pattern are smaller. Consistent with the general similarity between modeled and observed patterns, the seven spatial pattern correlation coefficients are higher than 0.57 (Table 4), indicating the pattern of DJF T2m is in moderate agreement.

Fig. 16.
Fig. 16.

As in Fig. 8, but for North America.

Citation: Journal of Climate 25, 2; 10.1175/JCLI-D-11-00175.1

The squared canonical correlations between NINO34 and DJF precipitation in North America are significant at 5% significance level for observations, ECM, and MET hindcasts, with respective correlation coefficients 0.60, 0.81, and 0.73. The ENSO teleconnection patterns of DJF precipitation in North America are shown in the lower panel of Fig. 16. The patterns agree in predicting wet anomalies in the southern North America (including northern Mexico, southern United States, and southern Rocky Mountain region of the United States). Elsewhere, the comparison is mixed in the sense that most anomalies in the observed pattern are captured by at least one model but not by both models simultaneously (e.g., Great Lakes Region, western coast of Canada). Nevertheless, the patterns are in moderate agreement by our spatial correlation criteria.

In JAS season, The ENSO-teleconnected regression patterns of both T2m and precipitation in North America are not significant at 5% significance level for observations (Tables 2 and 5). The ENSO-teleconnected patterns of T2m and precipitation are significant at 5% significance level in both ECM and MET hindcasts; however, the respective patterns differ from each other (figures not shown).

The observed ENSO teleconnection surface temperature dipole pattern in North America in boreal winter is consistent with Ropelewski and Halpert (1986), and they found the tendency for above normal temperatures in Alaska and western Canada and below normal temperatures in the southeast United States associated with ENSO in the December–March season. The dipole pattern is consistent with the anomalies expected with a Pacific–North America (PNA) circulation pattern Wallace and Gutzler (1981), that is, low pressure in the Gulf of Alaska, a ridge over northwest North America, and a trough in the southeast United States. The observed temperature pattern can be well reproduced in two coupled model hindcasts, especially the ECM hindcasts. The observed ENSO teleconnected precipitation pattern in DJF is consistent with Ropelewski and Halpert (1986, 1989, 1996) in northern Mexico, and southeastern United States. Our regression pattern analysis also indicated wet tendency in the southwestern United States and dry tendency in the Great Lakes Region associated with ENSO in DJF.

The significant dry tendency in the Pacific Northwest associated with ENSO was found by analyzing high-quality stream gauge records in the United Sates (Kahya and Dracup 1993), while the dry tendency in the Pacific Northwest is hardly seen in the observed ENSO teleconnection precipitation pattern in North America (Fig. 16). This discrepancy may arise from the data resolution. To clarify this, we apply the significance test to the DJF precipitation in the United States from the Parameter-elevation Regressions on Independent Slopes Model (PRISM) datasets, which have a very high resolution on a 0.04° × 0.04° grid (Daly et al. 1994) and a long time period from 1895–2010. The squared canonical correlations between NINO34 and observed DJF precipitation in the United States are significant at 5% significance level with the correlation coefficients 0.54. The resulting pattern predicts dry anomalies in the Pacific Northwest besides the dry anomalies in the Great Lakes Region and wet anomalies in the southern United States (figures not shown).

5. Conclusions

This paper investigated the existence and degree of agreement between models and observations of ENSO teleconnection patterns. An ENSO teleconnection pattern is defined as the set of regression coefficients between NINO34 and a field of variables. A rigorous field significance test proposed by DelSole and Yang (2011) was applied to determine whether the regression pattern “as a whole” was significant. The significance test was performed on surface temperature and precipitation in boreal winter and summer in six continents using observations and seven model hindcasts, and the resulting regression patterns were compared between observations and model simulations. To test the sensitivity to the choice of domain, the significance test was performed also over global land and compared to the locally determined pattern. When the field is significant, the local pattern and the global pattern from the same model have spatial correlations exceeding 0.95 in the local domain. The degree of agreement between models and observations was measured with the spatial correlation coefficient and characterized as “robust,” “moderate,” or “low.” Our major conclusions are as follows.

  • In Africa, the observed and modeled ENSO teleconnection patterns are significant at 5% significance level for both T2m and precipitation in DJF and JAS seasons. The T2m patterns for DJF and JAS were in moderate agreement, while the precipitation patterns for both seasons were in low agreement, according to our spatial correlation criteria. The observed precipitation pattern in DJF is consistent with previous findings (Janowiak 1988; Ropelewski and Halpert 1989), and the pattern can be largely reproduced in five coupled model hindcasts. Note that the ENSO teleconnection precipitation pattern in JAS has not been documented in the R-H map. The JAS precipitation pattern can be largely reproduced by the two models, and the squared canonical correlation coefficient of JAS precipitation is even higher than that of DJF precipitation, suggesting a stronger teleconnection in JAS.

  • In Asia, the observed and modeled ENSO teleconnection patterns are significant at 5% significance level for both T2m and precipitation in DJF and JAS seasons. However, only the DJF precipitation pattern was in moderate agreement between observations and simulations, while other patterns were in low agreement. The observed ENSO-teleconnected precipitation pattern in boreal summer has strong dry anomalies over India, which is consistent with the summer monsoon precipitation over peninsular India being suppressed during ENSO years (Rasmusson and Carpenter 1983; Shukla and Paolino 1983). However, the observed precipitation pattern in JAS was not simultaneously reproduced by all models, which limits the usefulness of this pattern for making predictions. Note that the ENSO teleconnection precipitation pattern in DJF has not been documented in the R-H map. The observed and simulated ENSO precipitation patterns in DJF agree on the wet anomalies concentrated in southern and eastern China, and the resulting pattern can be explained with the weak (strong) East Asian winter monsoon associated with warm (cold) events in the eastern Pacific through the Pacific–East Asian Teleconnection (Wang et al. 2000), suggesting the pattern is useful for making predictions.

  • In Australia, the observed and modeled ENSO teleconnection patterns are significant at 5% significance level for both T2m and precipitation in DJF and JAS seasons. The precipitation patterns for DJF and JAS were in robust agreement between observations and simulations, and they are consistent with the R-H map (Ropelewski and Halpert 1996, 1987) and Mcbridge and Nicholls (1983). The ENSO-teleconnected T2m pattern in DJF was in robust agreement between observations and simulations, while the counterpart in JAS was not.

  • In Europe, the observed ENSO teleconnection patterns are not significant at 5% significance level for both T2m and precipitation in DJF and JAS seasons.

  • In South America, the observed and modeled ENSO teleconnection patterns are significant at 5% significance level for both T2m and precipitation in DJF and JAS seasons. The observed ENSO teleconnection precipitation patterns in South America in boreal winter and summer are consistent with the R-H map (Ropelewski and Halpert 1996, 1987). The precipitation pattern in JAS was in moderate agreement between observations and simulations, while the counterpart in DJF was in low agreement. The observed ENSO teleconnected T2m pattern in DJF was in robust agreement between observations and simulations, while the counterpart in JAS was in low agreement.

  • In North America, the observed and modeled ENSO teleconnection patterns are significant at 5% significance level for both T2m and precipitation in DJF but not significant in JAS seasons. The observed ENSO teleconnection surface temperature dipole pattern in North America in boreal winter is consistent with Ropelewski and Halpert (1986). Consistency also was found for the tendency for above normal temperatures in Alaska and western Canada and below normal temperatures in the southeast United States associated with ENSO in the December to March season. The dipole pattern is consistent with the anomalies expected with a PNA circulation pattern (Wallace and Gutzler 1981), that is, low pressure in the Gulf of Alaska, a ridge over northwest North America, and a trough in the southeast United States. The observed temperature pattern can be well reproduced in seven coupled model hindcasts. The observed ENSO teleconnected precipitation pattern in DJF is consistent with Ropelewski and Halpert (1986, 1989, 1996) in northern Mexico and southeastern United States, and it can be largely reproduced by seven models.

Our analysis provides a comprehensive survey of ENSO teleconnection patterns over global land continents using updated long-term climate observational data and state-of-art coupled climate models. The derived ENSO teleconnection patterns of surface temperature and precipitation can be used to monitor and predict land climate for the given ENSO index. It is encouraging that the state-of-art coupled models can well reproduce many of observed ENSO teleconnection patterns over all six continents in boreal summer and winter seasons. The regression maps measure simultaneous relations, not lagged relations, thus they do not contain realistic information of prediction skill. However, the field significance test is not limited to the simultaneous ENSO teleconnection pattern, and it can be applied to the lagged regression patterns associated with ENSO. The regions with significant ENSO teleconnection patterns with no agreement between observations and simulations would seem to need further research. Our assessment is only a snap shot, and future assessments with different models may give different conclusions.

Acknowledgments

We thank two anonymous reviewers for constructive comments, which lead to improvements and clarifications. This research was supported by the National Science Foundation (ATM0332910, ATM0830062, ATM0830068), National Aeronautics and Space Administration (NNG04GG46G, NNX09AN50G), and the National Oceanic and Atmospheric Administration (NA04OAR4310034, NA09OAR4310058, NA05OAR4311004, NA10OAR4310264). The views expressed herein are those of the authors and do not necessarily reflect the views of these agencies.

REFERENCES

  • Chen, M., P. Xie, J. E. Janowiak, and P. A. Arkin, 2002: Global land precipitation: A 50-yr monthly analysis based on gauge observations. J. Hydrometeor., 3, 249266.

    • Search Google Scholar
    • Export Citation
  • Daly, C., R. P. Neilson, and D. L. Phillips, 1994: A statistical-topographic model for mapping climatological precipitation over mountainous terrain. J. Appl. Meteor., 33, 140158.

    • Search Google Scholar
    • Export Citation
  • DelSole, T., and X. Yang, 2011: Field significance of regression patterns. J. Climate, 24, 50945107.

  • Feddersen, H., 2000: Impact of global sea surface temperature on summer and winter temperatures in Europe in a set of seasonal ensemble simulations. Quart. J. Roy. Meteor. Soc., 126, 20892109.

    • Search Google Scholar
    • Export Citation
  • Halpert, M. S., and C. F. Ropelewski, 1992: Surface temperature patterns associated with the Southern Oscillation. J. Climate, 5, 577593.

    • Search Google Scholar
    • Export Citation
  • Hastie, T., R. Tibshirani, and J. H. Friedman, 2003: Elements of Statistical Learning. Corrected ed. Springer, 552 pp.

  • Janowiak, J. E., 1988: An investigation of interannual rainfall variability in Africa. J. Climate, 1, 240255.

  • Kahya, E., and J. A. Dracup, 1993: U. S. streamflow patterns in relation to the El Niño/Southern Oscillation. Water Resour. Res., 29, 24912503.

    • Search Google Scholar
    • Export Citation
  • Kalnay, E., and Coauthors, 1996: The NCEP/NCAR 40-Year Reanalysis Project. Bull. Amer. Meteor. Soc., 77, 437471.

  • Kistler, R., and Coauthors, 2001: The NCEP–NCAR 50-Year Reanalysis: Monthly means CD-ROM and documentation. Bull. Amer. Meteor. Soc., 82, 247267.

    • Search Google Scholar
    • Export Citation
  • Latif, M., and Coauthors, 1998: A review of the predictability and prediction of ENSO. J. Geophys. Res., 103 (C7), 14 37514 393.

  • Liang, J., S. Yang, Z.-Z. Hu, B. Huang, A. Kumar, and Z. Zhang, 2009: Predictable patterns of the Asian and Indo-Pacific summer precipitation in NCEP CFS. Climate Dyn., 32, 9891001.

    • Search Google Scholar
    • Export Citation
  • Mcbridge, J. L., and N. Nicholls, 1983: Seasonal relationships between Australian rainfall and the Southern Oscillation. Mon. Wea. Rev., 111, 19982004.

    • Search Google Scholar
    • Export Citation
  • National Research Council, 2010: Assessment of Intraseasonal to Interannual Climate Prediction and Predictability. The National Academies Press, 192 pp.

    • Search Google Scholar
    • Export Citation
  • Palmer, T., and Coauthors, 2004: Development of a European Multimodel Ensemble System for Seasonal-to-Interannual Prediction (DEMETER). Bull. Amer. Meteor. Soc., 85, 853872.

    • Search Google Scholar
    • Export Citation
  • Rasmusson, E. M., and T. H. Carpenter, 1983: The relationship between eastern equatorial Pacific sea surface temperatures and rainfall over India and Sri Lanka. Mon. Wea. Rev., 111, 517528.

    • Search Google Scholar
    • Export Citation
  • Ropelewski, C., and M. Halpert, 1986: North American precipitation and temperature patterns associated with the El Niño–Southern Oscillation. Mon. Wea. Rev., 114, 23522362.

    • Search Google Scholar
    • Export Citation
  • Ropelewski, C., and M. Halpert, 1987: Global- and regional-scale precipitation patterns associated with the El Niño–Southern Oscillation. Mon. Wea. Rev., 115, 16061626.

    • Search Google Scholar
    • Export Citation
  • Ropelewski, C., and M. Halpert, 1989: Precipitation patterns associated with the high index phase of the Southern Oscillation. J. Climate, 2, 268284.

    • Search Google Scholar
    • Export Citation
  • Ropelewski, C., and M. Halpert, 1996: Quantifying Southern Oscillation–precipitation relationships. J. Climate, 9, 10431059.

  • Shukla, J., and D. A. Paolino, 1983: The Southern Oscillation and long-range forecasting of the summer monsoon rainfall over India. Mon. Wea. Rev., 111, 18301837.

    • Search Google Scholar
    • Export Citation
  • Shukla, J., and J. L. Kinter, 2006: Predictability of seasonal climate variations: A pedagogical review. Predictability of Weather and Climate, T. Palmer and R. Hagedorn, Eds., Cambridge University Press, 306–341.

    • Search Google Scholar
    • Export Citation
  • Smith, T. M., and R. W. Reynolds, 2004: Improved extended reconstruction of SST (1854–1997). J. Climate, 17, 24662477.

  • Torrence, C., and P. J. Webster, 1998: The annual cycle of persistence in the El Niño/Southern Oscillation. Quart. J. Roy. Meteor. Soc., 124, 19852004.

    • Search Google Scholar
    • Export Citation
  • Trenberth, K. E., G. W. Branstator, D. Karoly, A. Kumar, N.-C. Lau, and C. Ropelewski, 1998: Progress during TOGA in understanding and modeling global teleconnections assocaited with tropical sea surface temperatures. J. Geophys. Res., 103 (C7), 14 29114 324.

    • Search Google Scholar
    • Export Citation
  • von Storch, H., and F. W. Zwiers, 1999: Statistical Analysis in Climate Research. Cambridge University Press, 484 pp.

  • Walker, G. T., 1923: Correlation in seasonal variations of weather. VIII A preliminary study of world weather. Mem. Indian Meteor. Dept., 24, 275332.

    • Search Google Scholar
    • Export Citation
  • Wallace, J. M., and D. S. Gutzler, 1981: Teleconnections in the geopotential height field during Northern Hemisphere winter. Mon. Wea. Rev., 109, 784812.

    • Search Google Scholar
    • Export Citation
  • Wallace, J. M., E. M. Rasmusson, T. P. Mitchell, V. E. Kousky, E. S. Sarachik, and H. von Storch, 1998: On the structure and evolution of ENSO-related climate variability in the tropical Pacific: Lessons from TOGA. J. Geophys. Res., 103 (C7), 14 24114 259.

    • Search Google Scholar
    • Export Citation
  • Wang, B., R. Wu, and X. Fu, 2000: Pacific–East Asia teleconnection: How does ENSO affect East Asian climate. J. Climate, 13, 15171536.

    • Search Google Scholar
    • Export Citation
  • Weisheimer, A., and Coauthors, 2009: ENSEMBLES: A new multi-model ensemble for seasonal-to-annual predictions—Skill and progress beyond DEMETER in forecasting tropical pacific SSTs. Geophys. Res. Lett., 36, L21711, doi:10.1029/2009GL040896.

    • Search Google Scholar
    • Export Citation
  • Wu, R., Z.-Z. Hu, and B. P. Kirtman, 2003: Evolution of ENSO-related rainfall anomalies in East Asia. J. Climate, 16, 37423758.

Save
  • Chen, M., P. Xie, J. E. Janowiak, and P. A. Arkin, 2002: Global land precipitation: A 50-yr monthly analysis based on gauge observations. J. Hydrometeor., 3, 249266.

    • Search Google Scholar
    • Export Citation
  • Daly, C., R. P. Neilson, and D. L. Phillips, 1994: A statistical-topographic model for mapping climatological precipitation over mountainous terrain. J. Appl. Meteor., 33, 140158.

    • Search Google Scholar
    • Export Citation
  • DelSole, T., and X. Yang, 2011: Field significance of regression patterns. J. Climate, 24, 50945107.

  • Feddersen, H., 2000: Impact of global sea surface temperature on summer and winter temperatures in Europe in a set of seasonal ensemble simulations. Quart. J. Roy. Meteor. Soc., 126, 20892109.

    • Search Google Scholar
    • Export Citation
  • Halpert, M. S., and C. F. Ropelewski, 1992: Surface temperature patterns associated with the Southern Oscillation. J. Climate, 5, 577593.

    • Search Google Scholar
    • Export Citation
  • Hastie, T., R. Tibshirani, and J. H. Friedman, 2003: Elements of Statistical Learning. Corrected ed. Springer, 552 pp.

  • Janowiak, J. E., 1988: An investigation of interannual rainfall variability in Africa. J. Climate, 1, 240255.

  • Kahya, E., and J. A. Dracup, 1993: U. S. streamflow patterns in relation to the El Niño/Southern Oscillation. Water Resour. Res., 29, 24912503.

    • Search Google Scholar
    • Export Citation
  • Kalnay, E., and Coauthors, 1996: The NCEP/NCAR 40-Year Reanalysis Project. Bull. Amer. Meteor. Soc., 77, 437471.

  • Kistler, R., and Coauthors, 2001: The NCEP–NCAR 50-Year Reanalysis: Monthly means CD-ROM and documentation. Bull. Amer. Meteor. Soc., 82, 247267.

    • Search Google Scholar
    • Export Citation
  • Latif, M., and Coauthors, 1998: A review of the predictability and prediction of ENSO. J. Geophys. Res., 103 (C7), 14 37514 393.

  • Liang, J., S. Yang, Z.-Z. Hu, B. Huang, A. Kumar, and Z. Zhang, 2009: Predictable patterns of the Asian and Indo-Pacific summer precipitation in NCEP CFS. Climate Dyn., 32, 9891001.

    • Search Google Scholar
    • Export Citation
  • Mcbridge, J. L., and N. Nicholls, 1983: Seasonal relationships between Australian rainfall and the Southern Oscillation. Mon. Wea. Rev., 111, 19982004.

    • Search Google Scholar
    • Export Citation
  • National Research Council, 2010: Assessment of Intraseasonal to Interannual Climate Prediction and Predictability. The National Academies Press, 192 pp.

    • Search Google Scholar
    • Export Citation
  • Palmer, T., and Coauthors, 2004: Development of a European Multimodel Ensemble System for Seasonal-to-Interannual Prediction (DEMETER). Bull. Amer. Meteor. Soc., 85, 853872.

    • Search Google Scholar
    • Export Citation
  • Rasmusson, E. M., and T. H. Carpenter, 1983: The relationship between eastern equatorial Pacific sea surface temperatures and rainfall over India and Sri Lanka. Mon. Wea. Rev., 111, 517528.

    • Search Google Scholar
    • Export Citation
  • Ropelewski, C., and M. Halpert, 1986: North American precipitation and temperature patterns associated with the El Niño–Southern Oscillation. Mon. Wea. Rev., 114, 23522362.

    • Search Google Scholar
    • Export Citation
  • Ropelewski, C., and M. Halpert, 1987: Global- and regional-scale precipitation patterns associated with the El Niño–Southern Oscillation. Mon. Wea. Rev., 115, 16061626.

    • Search Google Scholar
    • Export Citation
  • Ropelewski, C., and M. Halpert, 1989: Precipitation patterns associated with the high index phase of the Southern Oscillation. J. Climate, 2, 268284.

    • Search Google Scholar
    • Export Citation
  • Ropelewski, C., and M. Halpert, 1996: Quantifying Southern Oscillation–precipitation relationships. J. Climate, 9, 10431059.

  • Shukla, J., and D. A. Paolino, 1983: The Southern Oscillation and long-range forecasting of the summer monsoon rainfall over India. Mon. Wea. Rev., 111, 18301837.

    • Search Google Scholar
    • Export Citation
  • Shukla, J., and J. L. Kinter, 2006: Predictability of seasonal climate variations: A pedagogical review. Predictability of Weather and Climate, T. Palmer and R. Hagedorn, Eds., Cambridge University Press, 306–341.

    • Search Google Scholar
    • Export Citation
  • Smith, T. M., and R. W. Reynolds, 2004: Improved extended reconstruction of SST (1854–1997). J. Climate, 17, 24662477.

  • Torrence, C., and P. J. Webster, 1998: The annual cycle of persistence in the El Niño/Southern Oscillation. Quart. J. Roy. Meteor. Soc., 124, 19852004.

    • Search Google Scholar
    • Export Citation
  • Trenberth, K. E., G. W. Branstator, D. Karoly, A. Kumar, N.-C. Lau, and C. Ropelewski, 1998: Progress during TOGA in understanding and modeling global teleconnections assocaited with tropical sea surface temperatures. J. Geophys. Res., 103 (C7), 14 29114 324.

    • Search Google Scholar
    • Export Citation
  • von Storch, H., and F. W. Zwiers, 1999: Statistical Analysis in Climate Research. Cambridge University Press, 484 pp.

  • Walker, G. T., 1923: Correlation in seasonal variations of weather. VIII A preliminary study of world weather. Mem. Indian Meteor. Dept., 24, 275332.

    • Search Google Scholar
    • Export Citation
  • Wallace, J. M., and D. S. Gutzler, 1981: Teleconnections in the geopotential height field during Northern Hemisphere winter. Mon. Wea. Rev., 109, 784812.

    • Search Google Scholar
    • Export Citation
  • Wallace, J. M., E. M. Rasmusson, T. P. Mitchell, V. E. Kousky, E. S. Sarachik, and H. von Storch, 1998: On the structure and evolution of ENSO-related climate variability in the tropical Pacific: Lessons from TOGA. J. Geophys. Res., 103 (C7), 14 24114 259.

    • Search Google Scholar
    • Export Citation
  • Wang, B., R. Wu, and X. Fu, 2000: Pacific–East Asia teleconnection: How does ENSO affect East Asian climate. J. Climate, 13, 15171536.

    • Search Google Scholar
    • Export Citation
  • Weisheimer, A., and Coauthors, 2009: ENSEMBLES: A new multi-model ensemble for seasonal-to-annual predictions—Skill and progress beyond DEMETER in forecasting tropical pacific SSTs. Geophys. Res. Lett., 36, L21711, doi:10.1029/2009GL040896.

    • Search Google Scholar
    • Export Citation
  • Wu, R., Z.-Z. Hu, and B. P. Kirtman, 2003: Evolution of ENSO-related rainfall anomalies in East Asia. J. Climate, 16, 37423758.

  • Fig. 1.

    Time series of the Niño-3.4 index for (a) DJF and (d) JAS seasons from observations during the period 1948–2010, (b) DJF and (e) JAS seasons from ECM, and (c) DJF and (f) JAS from MET model hindcasts during the period 1958–2001. The correlation between hindcasts and observations computed over the period 1958–2001 are indicated in the bottom right of the respective panels.

  • Fig. 2.

    (a) The squared canonical correlation (blue dotted) and associated 5% significance level (red dashed) between the Niño-3.4 SST index and the DJF 2-m temperature in North America, as a function of the number of PCs used to represent the data, for (a) NCEP reanalyses during the period 1948–2010, (c) ECM, and (e) MET during the period 1958–2001. The cross-validated skill score (blue dotted) and associated standard errors (red dotted) of predicting the ENSO regression pattern using the canonical component, as a function of the number of PCs, for (b) NCEP reanalyses during the period 1948–2010, (d) ECM, and (f) MET during the period 1958–2001. The error bar shows the standard error for the maximum cross-validated skill score.

  • Fig. 3.

    The (left) observed, (middle) ECM, and (right) MET (top) canonical ENSO regression patterns (shading interval of 0.02°C per NINO34) and (bottom) associated canonical projection patterns for DJF 2-m temperature in the North America. The observed values are from the NCEP reanalyses during the period 1948–2010, and model results are from the DEMETER ECM and MET ensemble mean hindcasts during the period 1958–2001.

  • Fig. 4.

    The (a) observed and (b)–(g) seven simulated canonical ENSO regression patterns for DJF 2-m temperature. Shading intervals are 0.2°C per NINO34. The observed values are from the NCEP reanalyses during the period 1948–2010, two model results (ECM and MET) from DEMETER ensemble mean hindcasts during the period 1958–2001, and five model results (M1–M5) from ENSEMBLES ensemble mean hindcasts during the period 1960–2005.

  • Fig. 5.

    As in Fig. 4, but for JAS.

  • Fig. 6.

    As in Fig. 4, but for precipitation. Shading intervals are 0.1 mm day−1 per NINO34.

  • Fig. 7.

    As in Fig. 6, but for JAS.

  • Fig. 8.

    The observed (left) and two simulated (middle) ECM and (right) MET canonical ENSO regression patterns for (top) DJF 2-m temperature (shading interval of 0.2°C) and (bottom) DJF precipitation (shading interval of 0.1 mm day−1) in Africa. The observed values are from the NCEP reanalyses during the period 1948–2010, and model results from ECM and MET ensemble mean simulations during the period 1958–2001.

  • Fig. 9.

    As in Fig. 8, but for JAS in Africa.

  • Fig. 10.

    As in Fig. 8, but for Asia.

  • Fig. 11.

    As in Fig. 10, but for JAS in Asia.

  • Fig. 12.

    As in Fig. 8, but for Australia.

  • Fig. 13.

    As in Fig. 12, but for JAS in Australia.

  • Fig. 14.

    As in Fig. 8, but for South America.

  • Fig. 15.

    As in Fig. 14, but for JAS in South America.

  • Fig. 16.

    As in Fig. 8, but for North America.

All Time Past Year Past 30 Days
Abstract Views 0 0 0
Full Text Views 799 349 6
PDF Downloads 411 101 3