Impact of overestimated ENSO variability in the relationship between ENSO and East Asian summer rainfall

Authors

  • Yuanhai Fu,

    Corresponding author
    1. Institute of Atmospheric Physics, Chinese Academy of Sciences, Beijing, China
    2. Climate Change Research Center, Chinese Academy of Sciences, Beijing, China
    • Corresponding author: Y. Fu, Institute of Atmospheric Physics, Chinese Academy of Sciences, PO Box 9804, Beijing 100029, China. (fugreen1981@mail.iap.ac.cn)

    Search for more papers by this author
  • Riyu Lu,

    1. National Key Laboratory of Numerical Modeling for Atmospheric Sciences and Geophysical Fluid Dynamics, Institute of Atmospheric Physics, Chinese Academy of Sciences, Beijing, China
    Search for more papers by this author
  • Huijun Wang,

    1. Institute of Atmospheric Physics, Chinese Academy of Sciences, Beijing, China
    2. Climate Change Research Center, Chinese Academy of Sciences, Beijing, China
    Search for more papers by this author
  • Xiuqun Yang

    1. School of Atmospheric Sciences, Nanjing University, Nanjing, China
    Search for more papers by this author

Abstract

[1] El Niño–Southern Oscillation (ENSO) events in the preceding winter are an important predictor used to forecast the subsequent East Asian summer rainfall (EASR). This study investigates the relationship between the preceding winter ENSO and the EASR in coupled general circulation models, by analyzing the simulated results of 18 Coupled Model Intercomparison Project Phase 3 models. It is found that more than half of these models can approximately reproduce the ENSO's delayed impact on the EASR, and five models can capture the significant ENSO–EASR relationship. All of these five models overestimate the intensity of the ENSO variability, and they are almost the models that most seriously overestimate the ENSO variability, strongly suggesting that overestimated ENSO variability can help coupled models reproduce the relationship between the ENSO and EASR. Further analyses indicate that all of the five “best” models also overestimate the intensity of tropical Indian Ocean sea surface temperature (SST) variability, and they simulate the strongest intensity of Indian Ocean SST variability among the 18 models.

1 Introduction

[2] El Niño–Southern Oscillation (ENSO) is considered to be one of the most important factors that affect the East Asian summer rainfall (EASR), although this influence exhibits instability in various periods [Wang, 2000, 2002]. Particularly, the interannual variation of climate in East Asia and the western North Pacific (WNP) tends to be related to the phases of ENSO, and winter El Niño (La Niña) events generally correspond to heavier (lighter) rainfall in the following summer along the East Asian summer rainband, i.e., along the Yangtze River in China, South Korea, and southern Japan [e.g., Huang and Wu, 1989; Chou et al., 2003]. Therefore, ENSO events in winter are used as a predictor by East Asian meteorologists to forecast summer precipitation anomaly.

[3] So far, many studies have been conducted to reveal the processes of the impact of wintertime ENSO on the EASR. Wang et al. [2000] suggested that El Niño events influence the EASR through an anomalous anticyclone in summer over the WNP. In the decaying years of El Niño (La Niña), the climate anomalies in the WNP persist from winter to the subsequent summer through the positive feedback of atmosphere-ocean interactions associated with local anticyclonic (cyclonic) anomalies [e.g., Wang et al., 2000; Chou et al., 2003; Chen et al., 2012]. Other studies have suggested that the anticyclonic anomaly over the WNP is induced by the ENSO-related Indian Ocean warming anomaly [Li et al., 2008; Xie et al., 2009; Wu et al., 2010]. The tropical Indian Ocean (TIO) sea surface temperature (SST) acts as a capacitor in ENSO affecting atmospheric convection over the Philippine Sea. On the other hand, besides the ENSO-related part, the TIO SST also has its own variations, which can influence the climate over East Asia and the WNP through Hardly cell [Hu, 1997; Yoo et al., 2006]. Enhanced (suppressed) Philippine Sea convection tends to be associated with a lower-tropospheric cyclonic (anticyclonic) anomaly over the WNP [e.g., Lu, 2001; Kosaka and Nakamura, 2006], which leads to less (more) water vapor flux into East Asia and less (more) rainfall along the East Asian summer rainband.

[4] The aforementioned mechanisms are obtained either by observations or by atmosphere general circulation models (AGCMs). However, Wang et al. [2005] demonstrated that the AGCM fails to simulate realistic correlations between the SST and rainfall over the East Asian monsoon region in summer. Wu et al. [2006] demonstrated that AGCMs have less skill than coupled general circulation models (CGCMs) in simulating the climate over the East Asian summer monsoon region. Compared to AGCMs, CGCMs may better reproduce the atmosphere-ocean interaction over the Indian Ocean and western Pacific, which is crucial for the delayed impacts of the ENSO on the EASR.

[5] Unfortunately, only a few studies used CGCMs on the response of EASR to wintertime ENSO. Wang [2000] and Jiang et al. [2004] separately studied the relationship between ENSO and East Asian summer monsoon in a CGCM, but they did not discuss the mechanism of ENSO's impact on EASR. Li et al. [2007] used a CGCM to study the relationship between ENSO and WNP anticyclone, but they discussed little about the ENSO's impact on the anomalous WNP anticyclone in the following summer. Therefore, more studies are necessary to assess the simulations of the ENSO-EASR relationship, particularly in light of the moderate ability of the current models in simulating both ENSO and EASR.

[6] In this study, we evaluate the Coupled Model Intercomparison Project Phase 3 (CMIP3) models' ability to capture the ENSO-EASR relationship, and find that some models can represent the relationship, while others cannot. Furthermore, we investigate which kind of models can represent the relationship and how the ENSO-EASR relationship is represented in the “good” models. The organization of this paper is as follows. In section 2, the data sets and the methodologies used in this study are described. The evaluation of models' capacity in simulating the ENSO–EASR relationship is analyzed in section 3. In section 4, the process of ENSO's delayed impact on the EASR is investigated. The conclusions and discussion are presented in section 5.

2 Data and Methodology

[7] We analyzed the results of 18 models in the World Climate Research Programme's CMIP3 multimodel archive, for their 20th century climate (20C3M). Table 1 lists the detailed features of these models, and further details are documented at http://www-pcmdi.llnl.gov/ipcc/about_ipcc.php.

Table 1. Descriptions of the Models Used in This Study
ModelModel I.D.AbbreviationAtmospheric ResolutionEnsemble Members
aBCCR-BCM2.0bcm2.0128 × 64, L171
bCCSM3ccsm256 × 128, L177
cCGCM3.1(T47)cgcm4796 × 48, L175
dCGCM3.1(T63)cgcm63128 × 64, L171
eCNRM-CM3cnrm128 × 64, L171
fCSIRO-MK3.0csiro3.0192 × 96, L172
gCSIRO-MK3.5csiro3.5192 × 96, L173
hECHAM5/MPI-OMecham192 × 96, L164
iFGOALS-G1.0fgoals128 × 60, L173
jGFDL-CM2.0gfdl2.0144 × 90, L173
kGFDL-CM2.1gfdl2.1144 × 90, L173
lGISS-EHgiss72 × 46/45, L175
mUKMO-HadCM3hadcm396 × 73/72, L152
nUKMO-HadGEM1hadgem192 × 145, L162
oMIROC3.2(hires)miroch320 × 160, L171
pMIROC3.2(medres)mirocm128 × 64, L173
qMRI-CGCM2.3.2mricgcm128 × 64, L175
rPCMpcm128 × 64, L174

[8] For the models, 100 year simulations (1901–2000) of the 20C3M experiment are used. For the observations, 30 year Global Precipitation Climatology Project precipitation data (1979–2008) are used. The National Oceanic and Atmospheric Administration Extended Reconstructed SST V3 data (1901–2008) are also used. In this study, 30 year (1979–2008) SST data are used when the calculation is involved with observed precipitation; otherwise, 100 year (1901–2000) data are used.

[9] The interannual components are obtained by removing interdecadal components and trend from original time series. Here, the interdecadal components are obtained by applying a 9 year Gaussian filter on the detrended data. For the interannual components, we applied the autocorrelation method to calculate the independent sample size [Trenberth, 1984].

[10] The interannual standard deviation (StD) is used to depict the intensity of interannual variability, following a previous study [Fu and Lu, 2010]. In the study, the correlations and regressions are calculated for individual integrations first, and then the average for each model is made, following a previous study [Annamalai et al., 2007]. Similarly, the variance is calculated for each integration first, and then the StD is derived from the variances [Lu and Fu, 2010; Fu, 2012].

[11] The multimodel ensemble (MME) result is obtained by simply averaging over the available models with equivalent weight. This method has been widely adopted in previous studies [e.g., Jiang et al., 2005; Zhou and Yu, 2006; Sun and Ding, 2010]. Because the experiment number differs from model to model (Table 1), the multiexperiment ensemble mean is obtained by averaging all the available integrations in the individual models. Then, all the data are converted to a spectral triangular wave number 42 truncation (T42, approximately 2.8 × 2.8 degrees latitude-longitude) resolution to enable MME analysis.

[12] To facilitate the quantitative estimation of precipitation and circulation, several indices are used in this study. The EASR index (EASRI) is defined as the June–August (JJA) precipitation averaged over the parallelogram region determined by the following points: (25°N, 100°E), (35°N, 100°E), (30°N, 160°E), and (40°N, 160°E), which is used to mimic the East Asian summer rain belt and identical to that in Lu and Fu [2010]. A Philippine Sea convective index (PSCI) is defined as the JJA precipitation averaged over the region (10°–20°N, 110°–160°E) to depict the Philippine Sea convection, which is identical to that in Lu [2004]. Furthermore, the tropical Indian Ocean index (TIOI) is defined as the JJA SST anomalies averaged over the region (20°S–20°N, 40°–110°E), following Xie et al. [2009].

3 Simulation of ENSO-EASR Relationship in CGCMs

[13] Figure 1 shows the lead-lag correlations between the monthly Niño3 index and the JJA EASRI in individual models and observations, respectively. In observations, the positive relationship between Niño3 index and EASRI is strongest from September to April, which indicates the closely linked temporal evolutions of ENSO and EASR. The most significant and the strongest correlations between the winter Niño3 index and EASRI are represented only in five models (cnrm, echam, fgoals, gfdl2.0, and gfdl2.1—referred to as the five “best” models hereafter), all being statistically significant at the 5% level and in agreement with the observations. In addition, the evolution pattern of the ENSO-EASRI lead-lag relation can be reasonably replicated in most of the models (cgcm47, cnrm, csiro3.0, echam, fgoals, gfdl2.0, gfdl2.1, giss, hadgem, mirocm, miroch, and mricgcm) and the MME. The simulated relationships tend to be positive preceding ENSO signal, with the strong relationships appearing from September to April, but weaker simultaneous relationships. However, some models (bcm2.0, cgcm63, csiro3.5, hadcm3, and pcm) fail to simulate the evolution pattern of the ENSO-EASRI lead-lag relation, even simulating inverse correlations between ENSO and EASRI during the September–April period, in contrast to the observations.

Figure 1.

Lead lag correlation coefficients between the monthly Niño3 index and the JJA EASRI in individual models, the MME, and the observations. The vertical line in each subfigure indicates the zero lead time (July (1)).

[14] Figure 2 shows the JJA precipitation regressed onto the standardized December–February (DJF) Niño3 index in the CGCMs and observations, respectively. In observations, the warm phase of wintertime ENSO corresponds to the negative precipitation anomaly over the Philippine Sea and the positive precipitation anomaly over East Asia and the WNP. Most models (ccsm, cnrm, csiro3.0, csiro3.5, echam, fgoals, gfdl2.0, gfdl2.1, hadgem, and miroch) simulate the positive precipitation anomaly in East Asia and negative precipitation anomaly over the Philippine Sea, in agreement with observations. Among them, five models (cnrm, echam, fgoals, gfdl2.0 and gfdl2.1), which happen to be the five “best” models, simulate the significant relationships between ENSO and EASR, which are statistically significant at the 5% level, indicated by the correlation coefficients between the DJF-mean Niño3 index and EASRI (Figure 3). The models csiro3.0, csiro3.5, and miroch also simulate the ENSO-related precipitation pattern, but the ENSO-EASRI relationship is quite weak in these models, with the correlation coefficients being approximately only 0.10. The MME captures the spatial pattern of the ENSO-related precipitation anomalies over East Asia and the WNP, but the relationship is weak, with the correlation coefficient being 0.19. In addition, all the models simulate the positive relationships between the Niño3 index and EASRI, which is consistent with the observed value (0.47), except for the cgcm63, hadcm3, and pcm (Figure 3).

Figure 2.

The JJA precipitation regressed onto the standardized DJF Niño3 index in individual models, the MME, and the observations. Values significant at the 5% level are shaded (yellow, negative; blue, positive), and the contours are ±0.1, ±0.3, ±0.5, ±0.7, and ±0.9. The parallelogram indicates the region used to define the EASRI. Unit: mm/d.

Figure 3.

Scatter diagram of the correlation coefficients between the DJF Niño3 index and JJA EASRI (ordinate) and the interannual StDs of DJF Niño3 index (abscissa). Each dot represents the corresponding values for the models identified by the alphabets (Table 1). The triangle and alphabet “S” identify the observations, and the square and “T” identify the MME. The dashed line illustrates the significant value at the 5% level. The unit is in °C for the DJF Niño3 interannual StD.

[15] Figure 3 shows the scatter diagram of the DJF Niño3 index StDs and the correlation coefficients between ENSO and EASR. It suggests that the stronger ENSO-EASR correlation tends to be associated with stronger Niño3 interannual variability, and the weaker correlation with weaker ENSO variability, with the correlation coefficient between the ENSO-EASR correlations and DJF Niño3 StDs among the 18 models (samples) being 0.84. All the models that capture the significant ENSO-EASR relationships (i.e., the five “best” models) overestimate the intensity of the ENSO interannual variability. For instance, the model fgoals, which simulates the strongest ENSO interannual variability (the StD being 2.25 °C), simulates the strongest ENSO-EASR relationship (the correlation coefficient being 0.77). The other four models (cnrm, echam, gfdl2.0, and gfdl2.1), which also overestimate the Niño3 index variability than the observations, capture the significant positive correlations between ENSO and EASR, that all being significant at the 5% level. In contrast, all the models that underestimate the intensity of the ENSO variability fail to reproduce the significant relationship. This result suggests that, as to the selected 18 models, the overestimation of the ENSO variability could help CGCMs to represent the correlation between ENSO and EASR.

[16] There also exists a great diversity among the individual models in simulating the ENSO-related precipitation anomalies (Figure 2). Several models (cgcm47, cgcm63, giss, mricgcm, and pcm) simulate very weak ENSO-related precipitation anomalies over East Asia. The positive ENSO-EASR correlations exhibit a wide spread, with the lowest correlation coefficient being 0.03 (hadgem) and the highest being 0.77 (fgoals) (Figure 3).

4 Simulation of Processes of ENSO's Delayed Impacts on EASR

[17] Figure 4 shows the JJA surface temperature regressed onto the standardized DJF-Niño3 index for individual CGCMs and the observations, respectively. More than half of the models (bcm2.0, cnrm, csiro3.0, csiro3.5, echam, fgoals, gfdl2.0, gfdl2.1, hadcm3, and mricgcm) and the MME simulate the ENSO-related warming anomaly over the basin-scale Indian Ocean, which is consistent with the observations. It is worth noting that the five “best” models are among these models. Especially, the ENSO-related SST warming anomalies over the northern TIO can be well simulated by these models, which have been suggested to be more important for the WNP summer climate anomaly than those over the southern Indian Ocean [e.g., Xie et al., 2009].

Figure 4.

The JJA surface temperature regressed onto the standardized DJF Niño3 index in individual models, the MME, and the observations. Values significant at the 0.1% level are shaded (yellow, positive; blue, negative), and the contours are ±0.1, ±0.3, ±0.5, ±0.7, and ±0.9. Unit: °C.

[18] Figure 5 shows the JJA surface temperature regressed onto the standardized EASRI. In observations, the EASRI-related SST anomaly mainly appears in the northern TIO region. It is worth noting that only the five “best” models represent the significant positive EASRI-related TIO SST anomaly. The EASRI-related TIO warming patterns are quite the same as the ENSO-related SST warming patterns in these models (Figure 4). In addition, almost none of the other models represent the anomalous EASRI-related SST pattern in the Indian Ocean. The MME result does not capture the significant SST anomaly in the Indian Ocean, too.

Figure 5.

Same as Figure 4, but for the JJA surface temperature regressed onto the standardized EASRI, and values significant at the 5% level are shaded.

[19] Figure 6a shows the scatter diagram of the interannual TIOI StDs and the DJF Niño3 StDs. It suggests that the stronger TIOI variability tends to be associated with stronger ENSO variability, and the weaker TIOI variability with weaker ENSO variability, with the correlation coefficient between the Niño3 StDs and TIOI StDs among the 18 models being 0.80. The five “best” models simulate the strongest TIOI variability among the 18 models, and also simulate relatively stronger ENSO variability than most other models and observations. In addition, most models and the MME simulate stronger TIOI variability, compared to the observed value (0.12°C), which is indicated by interannual StDs that ranging from 0.13°C (miroch) to 0.36°C (echam), except for the ccsm, cgcm47, and cgcm63.

Figure 6.

Same as Figure 3, but for the scatter diagrams of (a) the interannual StDs of the DJF Niño3 index and the JJA TIOI, (b) the interannual StDs of the JJA EASRI and TIOI, (c) the ENSO-EASR correlations and the interannual StDs of TIOI. The dashed line illustrates the significant value at the 5% level. The unit is in °C for the interannual StD of the DJF Niño3 index and JJA TIOI, and mm/d for the JJA EASRI interannual StD.

[20] Figure 6b shows the scatter diagram of the interannual TIOI StDs and the EASRI StDs. The stronger EASRI variability tends to be associated with stronger TIOI variability, and weaker EASRI variability with weaker TIOI, with the correlation coefficient between the EASRI StDs and TIOI StDs among the 18 models being 0.61. The five “best” models that simulating the strongest TIOI interannual variability simulate relatively stronger EASRI variability, although almost all the models and the MME underestimate the EASRI variability, compared to the observations (0.56 mm/d).

[21] The scatter diagram of the interannual TIOI StDs and ENSO-EASR correlation (Figure 6c) further indicates that the models reproducing the significant positive ENSO-EASR relationship are the models that overestimate the TIOI interannual variability. The correlation coefficient between the ENSO-EASR correlations and TIOI StDs among the 18 models is 0.77. The five models that simulate the stronger TIOI variability than other models reproduce the significant positive correlations between the ENSO and EASR. Noteworthy is that these five models are the same with those that overestimate the ENSO variability (Figure 3) and simulate significant EASRI-related Indian Ocean SST anomaly (Figure 5), i.e., the five “best” models. Some models (cgcm47, cgcm63, giss, hadgem, miroch, and pcm) fail to represent ENSO's impact on the anomalous TIO SST, simulating too weak correlations between ENSO and TIO SST (Figure 4), although they do overestimate the TIOI variability. These models also fail to simulate the EASRI-related Indian Ocean SST anomaly (Figure 5). These results lead to their failure in reproducing the ENSO-EASR relationship.

[22] The model hadcm3 simulates the strongest interannual StD of EASRI and relatively stronger ENSO and TIOI StDs, which is somewhat similar to the five “best” models (Figure 6). However, this model fails to reproduce the EASRI-related TIO SST anomaly (Figure 5), and possibly due to this, the model hadcm3 simulates negative anomaly of ENSO-related EASR, instead of the positive anomaly in observations (Figure 2). The model hadcm3 also simulates a positive precipitation anomaly in the WNP, but this positive anomaly is located too much southward in comparison with the observed result, and there is mainly a negative precipitation anomaly in the EASR rain belt region.

[23] It is noticed that only the five “best” models can successfully represent the EASRI-related TIO warming anomaly, although more than half the models simulate the ENSO's impact on Indian Ocean SST. These results indicate that some models cannot reproduce the TIO SST's impact on the EASR anomaly. Thus, it raises an important question on why the five “best” models successfully simulate the impact of the TIO SST anomaly on the EASR anomaly.

[24] Figure 7 shows the precipitation anomaly regressed onto the standardized TIOI in models and observations, respectively. In observations, the positive TIOI index corresponds to the negative precipitation anomaly over the Philippine Sea, and the positive precipitation anomaly in East Asia and the WNP. It seems that most of the model can simulate the positive TIOI-related precipitation anomaly over the Philippine Sea and negative anomaly over East Asia and the WNP. However, there are only five models (the five “best” models) that can successfully reproduce the broad significant negative precipitation anomaly over the Philippine Sea and the significant positive anomaly over East Asia and the WNP. On the other hand, only three of the five “best” models (fgoals, gfdl2.0, and gfdl2.1) and the other two models (miroch and mirocm) represent the significant relationships between the TIOI and PSCI, as indicated by the significant correlation coefficients, which are statistically significant at the 5% level (Figure 8), being consistent with observations (−0.65). As to the other two models of the five “best” models (cnrm and echam), they well reproduce the TIO-related precipitation anomaly over East Asia and the WNP, but display strong and northward shifted positive precipitation anomalies in the equatorial western Pacific (Figure 7), which lead to positive PSCIs, and finally causing positive TIOI-PSCI correlation coefficients. The MME simulate a very weak TIOI-related precipitation anomaly over East Asia and the WNP region.

Figure 7.

Same as Figure 2, but for the JJA precipitation regressed onto the standardized JJA TIOI.

Figure 8.

Same as Figure 3, but for the scatter diagram of the TIOI-PSCI correlations and ENSO-PSCI correlations. The dashed line illustrates the significant value at the 5% level.

[25] Figure 8 shows the scatter diagram of the ENSO-PSCI correlations and the TIOI-PSCI correlations. It further suggests that the ENSO-PSCI and TIOI-PSCI correlation coefficients are extremely consistent with each other: The models with a stronger ENSO-PSCI correlation are those with a stronger TIOI-PSCI correlation, and vice versa, with the correlation coefficient between the TIOI-PSCI correlations and ENSO-PSCI correlations among the 18 models being 0.93. For instance, the models (fgoals, gfdl2.0, gfdl2.1, and miroch) representing significant negative ENSO-PSCI correlations also represent the significant negative TIOI-PSCI correlations, whereas the models (csiro3.5, echam, hadcm3, and mricgcm) simulating positive ENSO-PSCI correlations also simulate positive TIOI-PSCI correlations. In addition, both the ENSO-PSCI and TIOI-PSCI relationships are not significant in the MME result.

[26] Figure 9 shows the precipitation regressed onto the standardized EASRI in the simulations and observations, respectively. In observations, the EASR is highly correlated with the negative precipitation anomaly over the Philippine Sea, the positive precipitation anomaly over East Asia and the WNP, and the negative anomaly north of 40°N over Northeast Asia. All the models, except cgcm47 and giss, simulate the negative EASRI-related precipitation anomaly over the Philippine Sea and the positive precipitation anomaly over East Asia and the WNP. The result suggests that the inherent relationships of the East Asian summer monsoon can be well reproduced in almost all the models, which can be further illustrated by Figure 10b.

Figure 9.

Same as Figure 2, but for the JJA precipitation regressed onto the standardized EASRI.

Figure 10.

Same as Figure 3, but for (a) the scatter diagrams of the ENSO-EASR correlations and ENSO-PSCI correlations, (b) the ENSO-EASRI correlations and PSCI-EASRI correlations, and (c) the interannual StDs of EASRI and PSCI. The dashed line illustrates the significant value at the 5% level. The unit is in mm/d for the interannual StD of the PSCI and EASRI.

[27] Figure 10a shows the scatter diagram of the ENSO-PSCI correlations and the ENSO-EASR relationships. It indicates that the models simulating significant negative ENSO-PSCI correlation tend to simulate the significant positive ENSO-EASR relationship, whereas the models simulating weaker ENSO-PSCI correlations tend to simulate weaker ENSO-EASR relationships, with the correlation coefficient between the ENSO-EASR correlations and ENSO-PSCI correlations among the 18 models being −0.60. Four of the five “best” models (cnrm, fgoals, gfdl2.0, and gfdl2.1), which represent the statistically significant relationships between the ENSO and PSCI, simulate the strongest ENSO-EASR correlations. One of the five “best” models (echam) simulates a significant ENSO-EASR relationship but a positive ENSO-PSCI correlation coefficient, because the positive ENSO-related precipitation anomaly in the equatorial western Pacific is too strong and shifted northward into the extent of the Philippine Sea (Figure 2), which happens to be the PSCI's defined region. In addition, almost all the other models and the MME, which simulate insignificant ENSO-PSCI correlations, simulate insignificant relationships between the ENSO and EASR.

[28] Figure 10b shows the scatter diagram of the PSCI-EASRI correlations and the ENSO-EASRI correlations. Almost all the models reproduce the significant negative PSCI-EASRI correlations, except four models (cgcm47, echam, giss, and pcm). Moreover, four of the five “best” models (cnrm, fgoals, gfdl2.0, and gfdl2.1) that represent significant negative PSCI-EASRI correlations simulate significant ENSO-EASR relationships. In contrast, all the models that fail to capture the significant correlations between the PSCI and EASRI do not capture the significant relationships between the ENSO and EASR as well. However, the model echam reproduces a significant ENSO-EASR correlation and an insignificant PSCI-EASRI correlation, which is due to the strong and northward shifted precipitation anomaly (Figure 9).

[29] In addition, the scatter diagram of the interannual PSCI and EASRI StDs shows that there exists an obvious linear trend that the stronger intensity of the EASRI interannual variability closely relate to the stronger intensity of the PSCI interannual variability (Figure 10c), with the correlation coefficient between the EASRI StDs and PSCI StDs among the 18 models being 0.69, suggesting that the EASR variability is directly affected by the interannual variability of the Philippine Sea convection.

5 Conclusions and Discussion

[30] In this study, the delayed impact of winter ENSO on the subsequent summer rainfall over East Asia and the WNP is investigated, by analyzing the outputs of 18 CMIP3 coupled models.

[31] It is found that out of the 18 models, there are five models (cnrm, echam, fgoals, gfdl2.0, and gfdl2.1) that successfully capture the significant positive ENSO-EASR relationships. All these five models overestimate the intensity of ENSO variability and simulate the strongest ENSO variability among the selected models. Thus, the overestimated ENSO variability could help the CGCMs reproduce the true relationship between the ENSO and EASR.

[32] All of the five “best” models also overestimate the intensity of TIO SST variability and the ENSO variability. Actually, these five models simulate the strongest intensity of TIO SST variability among the 18 models. Overestimating the ENSO and TIO SST variability and reproducing the TIOI-PSCI relationship seems to be a prerequisite for reasonable simulation of the physical processes of the ENSO's delayed impact on EASR in current models.

[33] It should be noted that well simulating EASR is still a challenge for the climate model community. One of the major model defaults is that majority of state-of-the-art climate models (both AGCM and CGCM) are unable to capture the climatology of the East Asian summer monsoon, including the summer-mean location/intensity and seasonal migration of the western North Pacific subtropical high and the EASR. This is mainly due to the fact that the EASR is affected by various factors, including middle-high latitude weather disturbances, land-sea contrast, thermal and dynamical impacts of Tibetan Plateau, and tropical SSTs, and no one has a dominant impact on EASR. ENSO is only one of the factors that affect EASR anomalies. Therefore, the ENSO-EASR relationship is modest in observations [e.g., Wu et al., 2003], and the prediction skill for EASR variability is low [Gao et al., 2011; Liang et al., 2009]. In this sense, it is encouraging that some current models, although simulating unrealistically strong ENSO variability, capture the delayed impact of the ENSO on EASR.

[34] Recently, Zhang et al. [2012] suggested that accurate simulations of tropical background circulation in AGCMs play an important role in capturing the ENSO's delayed impact on the EASR. Similarly, Turner et al. [2005] also suggested that more accurate simulation of the basic state may help a model better represent the ENSO-South Asian monsoon relationship. We examined the climatological features of circulation and precipitation reproduced by the 18 CMIP3 models, but failed to find evidence for significant differences in these simulated basic states between the models that capture the ENSO-EASR relationship and the other models. This implies that the overestimated ENSO variability might help the CGCMs reproduce the ENSO-EASR relationship through other mechanism(s), rather than through improving the simulation of basic state.

Acknowledgments

[35] We thank three anonymous reviewers for their various constructive and detailed comments and suggestions, which have greatly helped us improve the presentation of this paper. This research was supported by the National Basic Research Program of China (973 Program) under grant 2010CB950304 and the CAS Strategic Priority Research Program under grant XDA05110203.