Abstract

There is a significant relationship between the preceding winter El Niño–Southern Oscillation (ENSO) and the subsequent East Asian summer rainfall (EASR), and this relationship is helpful for seasonal forecasting in East Asia. This study investigated the relationship between the preceding winter ENSO and EASR in the phase 5 of the Coupled Model Intercomparison Project (CMIP5) models and compared the results with those from the CMIP3 models. In general, the CMIP5 models capture the ENSO–EASR relationship more realistically than the CMIP3 models. For instance, approximately two-thirds of the CMIP5 models capture the ENSO–EASR relationship, whereas fewer than one-third of the CMIP3 models capture the relationship. Further investigation suggests that the improvement could be attributed to simulating the physical processes of ENSO’s impact on the EASR more realistically in the CMIP5 models, particularly the effect of ENSO on tropical Indian Ocean SST and the effect of Indian Ocean SST anomalies on the atmospheric convection over the Philippine Sea. However, there is large diversity in the ENSO–EASR relationship in the CMIP5 models, and most of the models underestimate the relationship. This underestimation comes from the underestimation of the physical processes, particularly from the underestimated impact of the atmospheric convection over the Philippine Sea on the EASR. The CMIP5 models that capture the ENSO–EASR relationship well (badly) also show high (low) skill in representing the physical processes.

1. Introduction

El Niño–Southern Oscillation (ENSO) is one of the most striking processes affecting the interannual climate variations over East Asia and the western North Pacific (WNP), and the preceding winter ENSO events are used as important predictors to forecast East Asian summer rainfall (EASR) anomalies by East Asian meteorologists. Particularly, the 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) (Huang and Wu 1989; Chou et al. 2003; Li and Zhou 2012).

Numerous studies have been conducted to reveal the mechanisms by which the wintertime ENSO affects the EASR. It has been shown that the ENSO-induced tropical Indian Ocean (TIO)–Philippine Sea convection (PSC) teleconnection plays an important role in ENSO’s delayed impact on the EASR (Li et al. 2008; Xie et al. 2009; Song and Zhou 2014b, Xie et al. 2016). ENSO-induced TIO warming persists through the summer and excites a Matsuno–Gill-type response in tropospheric temperature (Matsuno 1966; Gill 1980), with a Kelvin wave wedge penetrating the equatorial western Pacific and suppressing the convection over the Philippine Sea. The suppressed PSC is a part of the Pacific–Japan (PJ) pattern, and also affects the East Asian summer monsoon via the PJ pattern (Nitta 1986; Lu 2004; Kosaka and Nakamura 2006; Xie et al. 2016). Enhanced (suppressed) PSC is typically associated with a lower-tropospheric cyclonic (anticyclonic) anomaly over the WNP (Lu 2001; Kosaka and Nakamura 2006; Lin et al. 2010; Zhu et al. 2011), which leads to less (more) water vapor flux to East Asia and thus causes a below (above) normal EASR anomaly.

The aforementioned mechanisms in most previous studies are revealed or confirmed either by atmosphere general circulation models (AGCMs) or coupled general circulation models (CGCMs). However, the AGCMs have less skill in simulating the climate over the East Asian summer monsoon region (Wang et al. 2005; Wu et al. 2006; Song and Zhou 2014a), and CGCMs may better reproduce atmosphere–ocean interactions over the Indian Ocean and western Pacific, which are crucial for the delayed impact of the ENSO on the EASR (Li et al. 2012, 2014; Ding et al. 2014).

Unfortunately, the individual CGCMs have diverse results in simulating the delayed impact of winter ENSO on the subsequent summer rainfall over East Asia. By analyzing the simulated results of 18 phase 3 of the Coupled Model Intercomparison Project (CMIP3) models, Fu et al. (2013) found that only five models capture the significant ENSO–EASR relationship, and each of these five models seriously overestimates the intensity of the ENSO interannual variability, suggesting that overestimating the ENSO interannual variability may be a prerequisite for reproducing the relationship between the ENSO and EASR. Their further analyses showed that these five models also simulate the strongest TIO sea surface temperature (SST) and PSC interannual variabilities among the 18 CMIP3 CGCMs. These results indicate that representation of the physical processes of ENSO’s impact on the EASR depends on overestimating the ENSO interannual variability, which seems to obtain the right answer for the wrong reason in the CMIP3 CGCMs.

Phase 5 of the Coupled Model Intercomparison Project (CMIP5) has been carried out on a new set of coordinated climate model experiments using a new generation of CGCMs. Compared to the earlier-version CGCMs used in the CMIP3, the CMIP5 models have been improved in many aspects, such as higher horizontal and vertical resolutions in the atmosphere and ocean. Therefore, the output data provide an opportunity to estimate the simulation capacities of these new CGCMs on the ENSO–EASR relationship, and this is the main motivation of the present study.

The CMIP5 models show some improvements over their CMIP3 versions. On the one hand, the ENSO simulation has improved. Bellenger et al. (2014) have systematically evaluated the ENSO simulation in the CMIP5 models and demonstrated that the ENSO simulation improves in some aspects compared to those in the CMIP3 models. For example, the CMIP5 models show a notable improvement in simulating ENSO amplitude, and also an improvement in representing the ENSO seasonal phase locking. Zhou et al. (2014) also noted that the CMIP5 models show reasonable performance in simulating SST mean state and ENSO-related SST anomalies, and some models reasonably reproduce the phase locking of ENSO. On the other hand, the CMIP5 models are better able to simulate the Asian summer monsoon. Sperber et al. (2013) showed that the CMIP5 models make some progress in modeling the time-mean climatological annual cycle, interannual variability, and intraseasonal variability of the Asian summer monsoon. Gao et al. (2015) also suggested that the CMIP5 models are more skillful in simulating the climatological pattern and the dominant mode of summer precipitation in the pan-Asian monsoon region.

The CMIP5 models have shown some improvements in simulating the relationship between ENSO and the South Asian summer monsoon (e.g., Gao et al. 2015; Li et al. 2015). In this study, we will focus on the ENSO–EASR relationship and evaluate the models’ abilities in simulating this relationship. Another purpose of this study is to evaluate the abilities of current models to represent the physical processes through which the winter ENSO affects the EASR in an attempt to find the limitations of the current models’ abilities in simulating the ENSO–EASR relationship.

The organization of this paper is as follows. In section 2, the datasets and the methods used in this study are described. The comparisons between simulated ENSO–EASR relationship of the CMIP3 and CMIP5 models are shown in section 3. In section 4, we investigated the processes of ENSO’s delayed impact on the EASR in the CMIP5 models to illustrate the possible impacts affecting the simulation of the ENSO–EASR relationship in current models. Section 5 provides conclusions and discussion.

2. Data and methods

We analyzed the results of 22 models in the World Climate Research Programme’s (WCRP’s) CMIP5 multimodel archive for their historical climate simulation. Table 1 lists the detailed features of these models, and further details are documented online (http://cmip-pcmdi.llnl.gov/cmip5/index.html?submenuheader=0).

Table 1.

Descriptions of the models used in this study. (Expansions of acronyms are available online at http://www.ametsoc.org/PubsAcronymList.)

Descriptions of the models used in this study. (Expansions of acronyms are available online at http://www.ametsoc.org/PubsAcronymList.)
Descriptions of the models used in this study. (Expansions of acronyms are available online at http://www.ametsoc.org/PubsAcronymList.)

For the models, 105-yr simulations (1901–2005) of the historical experiment are used, and only one realization is chosen for each model. For the observations, 36-yr Global Precipitation Climatology Project (GPCP) precipitation data (1979–2014) are used. The National Oceanic and Atmospheric Administration (NOAA) Extended Reconstructed SST V3 dataset (1901–2014) is also used. In this study, 36-yr (1979–2014) SST data are used when the calculations involve observed precipitation; otherwise, 105-yr (1901–2005) data are used. All the simulation data have been interpolated onto a common 2.5° × 2.5° grid to enable multimodel ensemble (MME) analysis and to compare between individual simulations and the observations.

The interannual components are obtained by removing trend and interdecadal components from the original time series. Here, the interdecadal components are obtained by applying a 9-yr Gaussian filter on the detrended data. The autocorrelation method is applied to calculate the independent sample size. The interannual standard deviation (StD) is used to depict the intensity of interannual variability. In the study, the correlations and regressions are calculated for individual models first, and then the MMEs are made. Similarly, the variance is calculated for each model first, and then the StD is derived from the variances. The MME is obtained by simply averaging over the available models with equivalent weight. The methodologies are all the same as those in Fu et al. (2013).

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 rainband and identical to that in Lu and Fu (2010). 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). The Philippine Sea convective index (PSCI) is defined as the JJA precipitation averaged over the region 10°–20°N, 110°–160°E, following Lu (2004). In this study, we focused on the relationship between the preceding winter ENSO and the subsequent summer (JJA) EASR, and represented the ENSO by the December–February (DJF) Niño-3 index.

3. ENSO–EASR relationship in the CMIP3 and CMIP5 CGCMs

Figure 1 shows the lead–lag correlations between the monthly Niño-3 index and the JJA EASRI in observations, the MME, and individual models, respectively. In observations, the positive relationship between Niño-3 index and EASRI is strongest from the preceding September to April, becomes weak until August, and turns negative afterward, which is in agreement with previous studies (e.g., Sun et al. 2010; Fu et al. 2013). The MME result successfully captures the temporal evolution of the relationship between ENSO and EASR, with significant positive correlation coefficients during the preceding September–April period and weak correlation during the simultaneous June–August. The correlation coefficient between the curves in the observations and MME is 0.87. It shows that the relationship is stronger during the preceding winter and subsequent spring than that in the CMIP3 MME; all the correlation coefficients during this period are larger than 0.30 in the CMIP5 MME but only 0.20 in the CMIP3 MME (Fu et al. 2013). The ENSO–EASR lead–lag relationship is also reasonably replicated in most CMIP5 models (e.g., CanESM2, CNRM-CM5, FGOALS-g2, GFDL-CM3, GFDL-ESM2G, GFDL-ESM2M, GISS-E2-R, HadGEM2-CC, HadGEM2-ES, MIROC5, and MRI-CGCM3); however, only five CMIP3 models successfully represented the lead–lag relationship (Fu et al. 2013), indicating a large improvement. In addition, several models (e.g., BCC-CSM1, CCSM4, and GISS-E2-H) capture the positive relationship between the ENSO signal in the preceding winter and EASR, but they fail to reproduce the weakened relationship in the simultaneous summer and the negative relationship in the following season. There are even some models (e.g., CSIRO-Mk3.6, IPSL-CM5A-LR, MIROC-ESM, and NorESM1-M) that simulate an inverse relationship between ENSO and EASR during the September–April period in contrast to the observations.

Fig. 1.

Lead–lag correlation coefficients between the monthly Niño-3 index and the JJA EASRI in observations, the MME, and individual models. The correlation coefficient shown in each subfigure is calculated between the observed curve and individual model curve. The horizontal dashed line illustrates the significant value at the 5% level. The left vertical dashed line denotes January in the preceding winter, and the right line denotes July in the subsequent summer.

Fig. 1.

Lead–lag correlation coefficients between the monthly Niño-3 index and the JJA EASRI in observations, the MME, and individual models. The correlation coefficient shown in each subfigure is calculated between the observed curve and individual model curve. The horizontal dashed line illustrates the significant value at the 5% level. The left vertical dashed line denotes January in the preceding winter, and the right line denotes July in the subsequent summer.

Figure 2 shows the JJA precipitation regressed onto the standardized preceding DJF Niño-3 index in observations and CGCMs. In observations, the warm phase of wintertime ENSO corresponds to a significant positive precipitation anomaly along the East Asian summer rainband and a significant negative precipitation anomaly over the Philippine Sea. The MME result simulates a relatively weak ENSO-related positive precipitation anomaly along the East Asian summer rainband with the correlation coefficient between the DJF-Niño-3 index and EASRI being 0.23 and weaker than the observed value (0.47). The ENSO-related negative precipitation anomaly over the Philippine Sea is also relatively weak in the MME result, compared to the observations. The positive precipitation anomalies are very similar between the CMIP3 and CMIP5 MMEs, but the negative precipitation anomaly is stronger in the CMIP5 MME than that in the CMIP3 MME (Fu et al. 2013). Furthermore, more than half of the CMIP5 models (e.g., CanESM2, CCSM4, CNRM-CM5, FGOALS-g2, GFDL-CM3, GFDL-ESM2G, GFDL-ESM2M, GISS-E2-R, HadGEM2-ES, MIROC5, and MRI-CGCM3) simulate the positive precipitation anomaly along the East Asian summer rainband and the negative precipitation anomaly over the Philippine Sea well. These models also simulate the significant ENSO–EASR relationship (Fig. 1). These results suggest that more CMIP5 models realistically represent the ENSO-related precipitation anomaly compared to only five CMIP3 models (Fu et al. 2013). In addition, the model HadCM3 represents the ENSO-related positive precipitation anomaly along the East Asian summer rainband well, but the anomaly is a little southward. The other models (e.g., CSIRO-Mk3.6, FGOALS-s2, IPSL-CM5A-LR, MIROC-ESM-CHEM, and NorESM1-M) could simulate neither the lead–lag relationship nor the precipitation anomaly pattern.

Fig. 2.

The JJA precipitation regressed onto the standardized preceding DJF Niño-3 index in observations, the MME, and individual models. Values significant at the 5% level are shaded (blue, positive; yellow, negative). The contour interval is 0.2, and the zero contour lines are removed. The parallelogram indicates the region used to define the EASRI. Unit: mm day−1.

Fig. 2.

The JJA precipitation regressed onto the standardized preceding DJF Niño-3 index in observations, the MME, and individual models. Values significant at the 5% level are shaded (blue, positive; yellow, negative). The contour interval is 0.2, and the zero contour lines are removed. The parallelogram indicates the region used to define the EASRI. Unit: mm day−1.

The improvements in simulating the ENSO–EASR relationship in the CMIP5 models compared to those in the CMIP3 models are further illustrated in Fig. 3, which shows the distribution of the simulated ENSO–EASR correlation coefficients in the models. Up to 50% of the CMIP5 models have ENSO–EASR correlation coefficients between 0.20 and 0.40, while up to 55% of the CMIP3 models have correlation coefficients between 0.00 and 0.20. In addition, the simulated ENSO–EASR relationship is statistically significant at the 5% level in 14 out of 22 CMIP5 models, and this ratio is much larger than that in the CMIP3 models (only 5 out of 18 models). Furthermore, the ENSO–EASR correlation coefficients show a narrower scope of spread as follows: −0.10 to 0.60 in the CMIP5 models, compared to −0.20 to 0.80 in the CMIP3 models. The simulated correlation coefficients are within the range of 0.20–0.60 in the successful CMIP5 models but are within the range of 0.30–0.80 in the successful CMIP3 models. All these results confirm that the CMIP5 models generally capture the ENSO–EASR relationship better than the CMIP3 models did.

Fig. 3.

Distribution of correlation coefficients between the simulated preceding DJF Niño-3 index and subsequent JJA EASRI in the CMIP3 (blue bar) and CMIP5 (red bar) models. The black bar denotes the observed correlation coefficient.

Fig. 3.

Distribution of correlation coefficients between the simulated preceding DJF Niño-3 index and subsequent JJA EASRI in the CMIP3 (blue bar) and CMIP5 (red bar) models. The black bar denotes the observed correlation coefficient.

Figure 4a shows the scatter diagram of the ENSO–EASR correlation coefficients and the interannual StDs of the DJF-Niño-3 index for individual CMIP5 models. To facilitate the comparison, the results in the CMIP3 models are also given as Fig. 4b, which is a reproduction of Fig. 3 in Fu et al. (2013). Fu et al. (2013) indicated that there are only five CMIP3 models that represent the significant ENSO–EASR relationship, and all of them overestimate the intensity of ENSO interannual variability. That is, the significant ENSO–EASR relationship requires the overestimated ENSO variability in the CMIP3 models. This can also be seen in Fig. 4b. In contrast to the CMIP3 models, the CMIP5 models can capture the reasonable ENSO–EASR relationship within the realistic ENSO variability. Among the 14 CMIP5 models that successfully represent the significant ENSO–EASR relationship, eight models overestimate the ENSO variability and six models underestimate the variability, compared to the observed value (0.79°C) (Fig. 4a). In addition, the ENSO variability is closer to the observations (0.79°C) in the CMIP5 models. The interannual StDs of DJF-Niño-3 index range from 0.35° to 1.33°C in the CMIP5 models, showing a narrower spread than that in the CMIP3 models, which have a spread from 0.28° to 2.25°C. The averaged interannual StD is 0.79°C in the CMIP5 models, which is also more reasonable than that in the CMIP3 models (0.99°C). Furthermore, there is a close relationship between the ENSO–EASR correlations and ENSO variability in the CMIP3 models, indicated by the correlation coefficient between the abscissa and ordinate of scattered blue points shown in Fig. 4b, which is 0.84. This relationship (i.e., higher ENSO intensity tends to correspond to stronger ENSO–EASR relationship) exists in the CMIP5 models, which can be illustrated by the correlation coefficient (0.50) between the abscissa and ordinate of red points in Fig. 4a, indicating that excessively underestimating the ENSO variability impedes successful simulation of the ENSO–EASR relationship.

Fig. 4.

Scatter diagram of the ENSO–EASRI correlation coefficients (ordinate) and the interannual StDs of the DJF Niño-3 index (abscissa) in the (a) CMIP5 and (b) CMIP3 models. The black rhombuses indicate the observations. The unit is in °C for the interannual StD of DJF-Niño-3 index.

Fig. 4.

Scatter diagram of the ENSO–EASRI correlation coefficients (ordinate) and the interannual StDs of the DJF Niño-3 index (abscissa) in the (a) CMIP5 and (b) CMIP3 models. The black rhombuses indicate the observations. The unit is in °C for the interannual StD of DJF-Niño-3 index.

However, there still exists large diversity in the simulated ENSO–EASR relationship in the CMIP5 models, with the correlation coefficient between ENSO and EASR ranging from −0.10 to 0.60. In addition, almost all the CMIP5 models underestimate the correlation coefficients between the wintertime ENSO and EASR, except for two models (Fig. 4a).

Figure 5 compares the physical processes through which the preceding DJF ENSO affects the subsequent JJA rainfall along the East Asian summer rainband that simulated by the CMIP3 and CMIP5 models. As mentioned in the introduction, these processes are represented by the correlation coefficients between the preceding DJF-Niño-3 index and subsequent JJA TIOI (Fig. 5a), between the JJA TIOI and PSCI (Fig. 5b), and between the JJA PSCI and EASRI (Fig. 5c). At first, Fig. 5a shows that the simulated correlation coefficients between ENSO and TIOI tend to be stronger and closer to the observed value (0.79) in the CMIP5 models than those in the CMIP3 models, suggesting a large improvement. In the largest proportion of the CMIP5 models, which is up to 41%, the correlation coefficients between the DJF-Niño-3 index and TIOI are within 0.70–0.80; in the next largest proportion (23%) of the models, the correlation coefficients range from 0.50 to 0.60. However, in the largest two percentage groups of the CMIP3 models, 28% and 22%, the correlation coefficients are in the range of 0.80–0.90 and 0.20–0.30, respectively. Furthermore, 59% of the CMIP5 models have correlation coefficients within the range of 0.70–0.90, very similar to the observations, and the fraction is much larger than that in the CMIP3 models (39%). In addition, all the CMIP5 models simulate the significant ENSO–TIOI relationship, although 17 out of the 22 models underestimate the relationship.

Fig. 5.

As in Fig. 3, but for the correlation coefficients (a) between the preceding DJF Niño-3 index and subsequent JJA TIOI, (b) between the JJA TIOI and PSCI, and (c) between the JJA PSCI and EASRI.

Fig. 5.

As in Fig. 3, but for the correlation coefficients (a) between the preceding DJF Niño-3 index and subsequent JJA TIOI, (b) between the JJA TIOI and PSCI, and (c) between the JJA PSCI and EASRI.

The CMIP5 models also make large progress in simulating the JJA TIOI–PSCI relationship (Fig. 5b). In the largest proportion of the CMIP5 models, which is 27%, the correlation coefficients change from −0.50 to −0.40 and are statistically significant at the 1% level; in contrast, in the largest proportion of the CMIP3 models, which is 28%, the correlation coefficients are very weak and within the range of only −0.20 to −0.10. Furthermore, the significant TIOI–PSCI relationship is successfully represented in 55% of the CMIP5 models, compared to in only 28% of the CMIP3 models. The CMIP5 models show a large diversity in simulating the TIOI–PSCI relationship; the correlation coefficients spread from −0.60 to 0.40. All the CMIP5 models underestimate the TIOI–PSCI relationship, indicating that current models still have relatively low skill in representing the relationship.

There is almost no appreciable improvement in simulating the JJA PSCI–EASRI relationship in the CMIP5 models compared to those in the CMIP3 models (Fig. 5c). More CMIP5 models simulate stronger correlation coefficients; they are greater than −0.30 in 59% of the CMIP5 models compared to 44% of the CMIP3 models. The significant PSCI–EASRI correlation is successfully simulated in 14 out of 18 CMIP3 (78%) and 17 out of 22 CMIP5 models (77%). The CMIP5 models also show a large diversity in simulating the PSCI–EASRI relationship, with the correlation coefficients between the PSCI and EASRI within the range of −0.70 to 0.10. In addition, the PSCI–EASRI relationship is underestimated in all the CMIP5 models, similar to the results in the CMIP3 models (Fu et al. 2013).

In summary, Fig. 5 suggests that the improvement in simulating the ENSO–EASR relationship in the CMIP5 models could be attributed mainly to the improvements in simulating both the ENSO–TIOI relationship and TIOI–PSCI relationship. However, there are large diversities in the physical processes of ENSO’s impact on the EASR in the CMIP5 models, and most of the models underestimate the physical processes. Therefore, the improving physical processes in the models would be important for realistic simulations of ENSO–EASR relationship.

4. ENSO’s delayed impact on EASR in the “best” and “worst” CMIP5 models

In the former section, we found that the CMIP5 models show diversity in simulating the ENSO–EASR relationship, although they are much improved compared to the CMIP3 models. Some CMIP5 models capture the realistic relationship, but the others fail to capture the relationship or even simulate negative ENSO–EASR relationship. Thus, to further illustrate the possible impacts affecting the simulation of the ENSO–EASR relationship in current CMIP5 models, we executed a composite analysis of the models’ behavior in representing the processes of ENSO’s delayed impact on EASR by comparing the MME results of six CMIP5 models that simulate the highest or most realistic ENSO–EASR correlation coefficients (CNRM-CM5, FGOALS-g2, GFDL-ESM2G, GFDL-ESM2M, GISS-E2-R, and MIROC5—referred to as the “best” models hereafter) to the MME of six CMIP5 models with the lowest correlation coefficients (CSIRO-Mk3.6, FGOALS-s2, IPSL-CM5A-LR, MIROC-ESM-CHEM, MPI-ESM-LR, and NorESM1-M—referred to as the “worst” models hereafter).

Figure 6 shows the ENSO-related precipitation anomalies in observations and the so-called best and worst models. The best models capture well the significant negative precipitation anomaly over the Philippine Sea and the positive precipitation anomaly along the East Asian summer rainband (Fig. 6b), consistent with the observations (Fig. 6a). The correlation coefficient between ENSO and EASR is 0.42 in the best models, which is similar to the observed correlation (0.47). In contrast, the positive precipitation anomaly over the East Asian summer rainband is weakened and almost inappreciable in the worst models (Fig. 6c), leading to a poor ENSO–EASR relationship with the correlation coefficient being only 0.004. In addition, the negative precipitation anomaly over the Philippine Sea is only moderately significant and shifted eastward.

Fig. 6.

The JJA precipitation regressed onto the standardized DJF-Niño-3 index in (a) observations, (b) the “best” models, and (c) the “worst” models. Values significant at the 5% level are shaded (blue, positive; yellow, negative). The contour interval is 0.2, and the zero contour lines are removed. Unit: mm day−1.

Fig. 6.

The JJA precipitation regressed onto the standardized DJF-Niño-3 index in (a) observations, (b) the “best” models, and (c) the “worst” models. Values significant at the 5% level are shaded (blue, positive; yellow, negative). The contour interval is 0.2, and the zero contour lines are removed. Unit: mm day−1.

Figure 7 shows the ENSO-related SST anomalies in observations and the models. Both the best and worst models reproduce the ENSO-related warming anomaly over the basin-scale Indian Ocean (Figs. 7b,c), which is well documented in previous studies (e.g., Xie et al. 2016). The warming anomaly over the northern TIO is especially well simulated by these models, which has been suggested to be more important for the WNP summer climate anomaly than the SST anomalies over the southern Indian Ocean (Xie et al. 2009; Huang et al. 2010). Note that the ENSO-related TIO warming anomaly in the best models is stronger than that in the worst models, although the SST anomaly is statistically significant in both the best and worst models. The correlation coefficient between the DJF-Niño-3 index and TIOI is 0.78 in the best models, and higher than that in the worst models (0.59). In addition, the worst models fail to capture the positive SST anomaly over the marine continent and simulate a false strong positive SST anomaly in the equatorial western Pacific, which is consistent with the result of Jiang et al. (2017). This equatorial SST anomaly may have resulted in the false positive rainfall anomaly in the equatorial western Pacific (Fig. 6c).

Fig. 7.

As in Fig. 6, but for the JJA SSTs (yellow, positive; blue, negative). The contour interval is 0.05. Unit: °C.

Fig. 7.

As in Fig. 6, but for the JJA SSTs (yellow, positive; blue, negative). The contour interval is 0.05. Unit: °C.

Figure 8 shows the TIO-related precipitation anomalies in observations and the best and worst models. The best models successfully reproduce the negative precipitation anomaly over the Philippine Sea and the positive precipitation anomaly along the East Asian summer rainband (Fig. 8b), consistent with the observations (Fig. 8a). The worst models also simulate the negative precipitation anomaly over the Philippine Sea, but the precipitation anomaly is much weaker than the best model result and is only moderately significant (Fig. 8c). In addition, the positive precipitation anomaly along the East Asian summer rainband is much weakened and almost inappreciable. In general, the present result is consistent with the result of Song and Zhou (2014b), who pointed out that the successful reproduction of the PSC anomaly highly depends on the simulation of TIO–PSC teleconnection.

Fig. 8.

As in Fig. 6, but for the JJA precipitation regressed onto the standardized JJA TIOI (blue, positive; yellow, negative).

Fig. 8.

As in Fig. 6, but for the JJA precipitation regressed onto the standardized JJA TIOI (blue, positive; yellow, negative).

Figure 9 shows the PSCI-related precipitation anomalies. In observations, a positive PSCI is highly associated with a negative precipitation anomaly along the East Asian summer rainband (Fig. 9a), which is a manifestation of the PJ pattern (e.g., Nitta 1986; Lu 2004; Kosaka and Nakamura 2006). This significant and negative precipitation anomaly along the East Asian summer rainband is reproduced by the best models (Fig. 9b), although the precipitation anomaly is weaker than the observations (Fig. 9a). In contrast, the worst models simulate very weak precipitation anomaly along the East Asian summer rainband (Fig. 9c). Moreover, the area-averaged regression coefficient is −0.22 over the East Asian summer rainband in the best models, which is much closer to the observations (−0.28) than that in the worst models (−0.11).

Fig. 9.

As in Fig. 6, but for the JJA precipitation regressed onto the standardized JJA PSCI (blue, positive; yellow, negative).

Fig. 9.

As in Fig. 6, but for the JJA precipitation regressed onto the standardized JJA PSCI (blue, positive; yellow, negative).

Figure 10 further compares the ENSO–EASR relationship and the physical processes in individual CMIP5 models, which is represented by the scatterplots of the ENSO–EASRI correlation coefficients and the ENSO–TIOI correlation coefficients (Fig. 10a), the TIOI–PSCI correlation coefficients (Fig. 10b), and the PSCI–EASRI correlation coefficients (Fig. 10c). Figure 10a shows that the stronger ENSO–EASR relationship tends to be associated with the stronger ENSO–TIOI relationship, which can be illustrated by the correlation coefficient (0.52) between the ENSO–EASR correlations and the ENSO–TIOI correlations. On the other hand, the simulation of the TIOI–PSCI relationship shows a complicated feature (Fig. 10b). There is no clear correspondence between the simulated TIOI–PSCI relationship and ENSO–EASR relationship among the CMIP5 models, indicated by the low correlation coefficient (−0.09) between the abscissa and ordinate of the simulated points in Fig. 10b. In addition, the well-simulated TIOI–PSCI relationship does not guarantee a model to capture the ENSO–EASR correlation. For instance, among the eight models (CCSM4, FGOALS-g2, FGOALS-s2, GFDL-ESM2M, MIROC5, MIROC-ESM, MIROC-ESM-CHEM, and NorESM1-M) that show the TIOI–PSCI correlation coefficients between −0.40 and −0.60, near to the observations, only half of them (CCSM4, FGOALS-g2, GFDL-ESM2M, and MIROC5) can simulate the significant ENSO–EASR relationship. Among the other half of models (which cannot reproduce the significant ENSO–EASR relationship), the model MIROC-ESM-CHEM fails in capturing the ENSO–TIOI relationship (Fig. 10a), and the models FGOALS-s2, MIROC-ESM, and NorESM1-M fail in capturing the PSCI–EASRI relationship (Fig. 10c). In particular, the models FGOALS-s2 and NorESM1-M well capture both the ENSO–TIOI relationship and TIOI–PSCI relationship, but poorly simulate the PSCI–EASRI relationship, which results in the poor ENSO–EASR relationship in these models. Furthermore, all the best models simulate the reliable PSCI–EASRI relationship (Fig. 10c). These results suggest that the PSCI–EASRI relationship may be the most crucial physical process for current models to reproduce the ENSO–EASR relationship. However, unfortunately, capturing the PSCI–EASRI relationship seems to be an enormous challenge, indicated by the fact that all the CMIP5 models analyzed here underestimate this relationship and there is almost no appreciable improvement in this physical process in the CMIP5 models compared with the CMIP3 models (Fig. 5c).

Fig. 10.

The scatter diagrams of the ENSO–EASRI correlations (ordinate) and (a) ENSO–TIOI correlations (abscissa), (b) TIOI–PSCI correlations (abscissa), and (c) PSCI–EASRI correlations (abscissa) in the CMIP5 models. Each dot represents the corresponding value for the model identified by the alphabet (Table 1).

Fig. 10.

The scatter diagrams of the ENSO–EASRI correlations (ordinate) and (a) ENSO–TIOI correlations (abscissa), (b) TIOI–PSCI correlations (abscissa), and (c) PSCI–EASRI correlations (abscissa) in the CMIP5 models. Each dot represents the corresponding value for the model identified by the alphabet (Table 1).

Following Ham and Kug (2015), we identified the dominant intermodel uncertainty in simulating the ENSO response. That is, empirical orthogonal function (EOF) analysis is performed for the deviation of each model’s ENSO-related summer precipitation from the MME. Figure 11 shows the first and second EOF eigenvectors of intermodel differences. The observed ENSO-related summer precipitation is also shown to facilitate the comparison. These two EOFs are statistically well separated by the criterion of North et al. (1982). The first EOF exhibits zonally elongated anomalies in the tropical Pacific, and these anomalies do not correspond well to the observed ENSO-related anomalies (Fig. 11a). In addition, the first mode shows a weak and negative anomaly in East Asia, in contrast with the positive one in the observations. Therefore, we can conclude that this mode may not be associated with the intermodel differences in the ENSO–EASR relationship. This can be further illustrated by the correlation coefficient between the first EOF principal component (PC1) and the simulated ENSO–EASR correlations, which is only 0.03.

Fig. 11.

The (a) first and (b) second EOF eigenvectors of the differences of individual models’ ENSO-related precipitation from the MME, indicated by the shading with the bar scale on the right. The observed ENSO-related precipitation is also shown as contour lines to facilitate the comparison. Here, the ENSO-related precipitation denotes the summer precipitation anomalies regressed onto the preceding DJF-Niño-3 index. Unit: mm day−1.

Fig. 11.

The (a) first and (b) second EOF eigenvectors of the differences of individual models’ ENSO-related precipitation from the MME, indicated by the shading with the bar scale on the right. The observed ENSO-related precipitation is also shown as contour lines to facilitate the comparison. Here, the ENSO-related precipitation denotes the summer precipitation anomalies regressed onto the preceding DJF-Niño-3 index. Unit: mm day−1.

The second EOF is characterized by the negative anomaly in the tropical WNP and positive anomaly in the eastern Indian Ocean (Fig. 11b). These anomalies are consistent with the observed ones. In addition, there also exists a positive anomaly over East Asia, similar to the observations. Furthermore, the correlation coefficient between the PC2 and the ENSO–EASR correlations is 0.55. Therefore, the models with positive PC2 tend to simulate the ENSO-related summer precipitation anomalies more like the observed ones, in comparison with the models with negative PC2.

5. Conclusions and discussions

We examined the relationship between the winter ENSO and subsequent East Asian summer rainfall (EASR) in the CMIP5 models and compared the results to those in the CMIP3 models. In general, the CMIP5 models capture the ENSO–EASR relationship more realistically than the CMIP3 models. For instance, the significant ENSO–EASR relationship is reasonably simulated in approximately two-thirds (64%) of the CMIP5 models but in only less than one-third (28%) of the CMIP3 models. The ENSO–EASR correlation coefficients are within a reasonable range of 0.20–0.40 in most CMIP5 models (up to 50%) but are within an unrealistic range of 0.00–0.20 in most CMIP3 models (up to 55%). The reproduction of the realistic ENSO–EASR relationship does not depend on an overestimation of the ENSO interannual variability in the CMIP5 models, which is an additional indication of clear improvement in comparison with the CMIP3 models. The overestimation of ENSO variability is a prerequisite to capturing the significant relationship in the CMIP3 models, but half of the CMIP5 models that realistically represent the significant ENSO–EASR correlation underestimate the ENSO variability and the other half overestimate the variability.

The improvements could be attributed to the more realistic simulation of the physical processes of ENSO’s delayed impact on the EASR in the CMIP5 models. For instance, in 59% of the CMIP5 models, the simulated ENSO–TIOI correlation coefficients are within the range of 0.70–0.90, very similar to the observed value (0.79); the percentage is much larger than that in the CMIP3 model (39%). The significant TIOI–PSCI correlation is successfully represented in up to 55% of the CMIP5 models but in only 28% of the CMIP3 models. However, the CMIP5 models do not show a clear improvement in simulating the PSCI–EASRI relationship.

However, there are still large diversities in the ENSO–EASR relationship as well as the physical processes of ENSO’s impact on the EASR in the CMIP5 models. Almost all the CMIP5 models underestimate the ENSO–EASR relationship and the physical processes. The underestimation of the ENSO–EASR relationship is related to the underestimation of the physical processes. Given the diversity in simulating the ENSO–EASR relationship, we compared the simulated physical processes of ENSO’s impact on the EASR in the “best” and “worst” CMIP5 models that simulate the highest and lowest ENSO–EASR relationship, respectively. In general, the so-called best CMIP5 models successfully represent the physical processes, but the worst models show low skill in reproducing the processes. Among the physical processes, the effect of PSC on EASR seems to play a crucial role in determining whether the ENSO–EASR relationship can be reproduced or not for the current models. Furthermore, the ENSO–EASR relationship is manifested as the second leading mode of intermodel differences in the ENSO-related precipitation anomalies in the tropics and subtropics.

In this study, we evaluated the ENSO–EASR relationship and the relevant physical processes through which the preceding winter ENSO affects EASR, but did not analyze the impact of simulation of climatological state on this ENSO–EASR relationship. Previous studies suggested that accurately reproducing the climatological state is crucial for models to capture the relationships on the interannual time scale (Zhang et al. 2012; Ham and Kug 2015). Therefore, some questions arise. Does the simulation of ENSO–EASR relationship depend on the simulation of climatological state? If yes, which season (winter, spring, or summer) is the most important for the accurate simulation of climatological state? These issues require to be studied in the future. On the other hand, although the first mode of intermodel differences in ENSO-related precipitation is not related to the simulation of ENSO–EASR relationship (Fig. 11a), it would be interesting to further investigate the features and formation mechanism of this mode.

Acknowledgments

We thank three anonymous reviewers for their various constructive and detailed comments and suggestions, which have greatly helped us to improve the presentation of this paper. We acknowledge the modeling groups for making their simulations available for analysis, the Program for Climate Model Diagnosis and Intercomparison (PCMDI) for collecting and archiving the CMIP5 model output, and the WCRP’s Working Group on Coupled Modeling (WGCM) in making the WCRP CMIP5 multimodel dataset available. This research was supported by the National Natural Science Foundation of China (Grants 41305063, 41320104007, and 41130103).

REFERENCES

REFERENCES
Bellenger
,
H.
,
E.
Guilyardi
,
J.
Leloup
,
M.
Lengaigne
, and
J.
Vialard
,
2014
:
ENSO representation in climate models: From CMIP3 to CMIP5
.
Climate Dyn.
,
42
,
1999
2018
, doi:.
Chou
,
C.
,
J.-Y.
Tu
, and
J.-Y.
Yu
,
2003
:
Interannual variability of the western North Pacific summer monsoon: Differences between ENSO and non-ENSO years
.
J. Climate
,
16
,
2275
2287
, doi:.
Ding
,
H.
,
R.
Greatbatch
,
W.
Park
,
M.
Latif
,
V.
Semenov
, and
X.
Sun
,
2014
:
The variability of the East Asian summer monsoon and its relationship to ENSO in a partially coupled climate model
.
Climate Dyn.
,
42
,
367
379
, doi:.
Fu
,
Y.
,
R.
Lu
,
H.
Wang
, and
X.
Yang
,
2013
:
Impact of overestimated ENSO variability in the relationship between ENSO and East Asian summer rainfall
.
J. Geophys. Res. Atmos.
,
118
,
6200
6211
, doi:.
Gao
,
Y.
,
H.
Wang
, and
D.
Jiang
,
2015
:
An intercomparison of CMIP5 and CMIP3 models for interannual variability of summer precipitation in Pan-Asian monsoon region
.
Int. J. Climatol.
,
35
,
3770
3780
, doi:.
Gill
,
A. E.
,
1980
:
Some simple solutions for heat-induced tropical circulation
.
Quart. J. Roy. Meteor. Soc.
,
106
,
447
462
, doi:.
Ham
,
Y.-G.
, and
J.-S.
Kug
,
2015
:
Improvement of ENSO simulation based on intermodel diversity
.
J. Climate
,
28
,
998
1015
, doi:.
Huang
,
G.
,
K.
Hu
, and
S.-P.
Xie
,
2010
:
Strengthening of tropical Indian Ocean teleconnection to the northwest Pacific since the mid-1970s: An atmospheric GCM study
.
J. Climate
,
23
,
5294
5304
, doi:.
Huang
,
R.
, and
Y.
Wu
,
1989
:
The influence of ENSO on the summer climate change in China and its mechanism
.
Adv. Atmos. Sci.
,
6
,
21
32
, doi:.
Jiang
,
W.
,
G.
Huang
,
K.
Hu
,
R.
Wu
,
H.
Gong
,
X.
Chen
, and
W.
Tao
,
2017
:
Diverse relationship between ENSO and the northwest Pacific summer climate among CMIP5 models: Dependence on the ENSO decay pace
.
J. Climate
,
30
,
109
127
, doi:.
Kosaka
,
Y.
, and
H.
Nakamura
,
2006
:
Structure and dynamics of the summertime Pacific–Japan teleconnection pattern
.
Quart. J. Roy. Meteor. Soc.
,
132
,
2009
2030
, doi:.
Li
,
C.
,
R.
Lu
, and
B.
Dong
,
2012
:
Predictability of the western North Pacific summer climate demonstrated by the coupled models of ENSEMBLES
.
Climate Dyn.
,
39
,
329
346
, doi:.
Li
,
C.
,
R.
Lu
, and
B.
Dong
,
2014
:
Predictability of the western North Pacific summer climate associated with different ENSO phases by ENSEMBLES multi-model seasonal forecasts
.
Climate Dyn.
,
43
,
1829
1845
, doi:.
Li
,
R. Q.
,
S. H.
,
B.
Han
, and
Y. H.
Gao
,
2015
:
Connections between the South Asian summer monsoon and the tropical sea surface temperature in CMIP5
.
J. Meteor. Res.
,
29
,
106
118
, doi:.
Li
,
S.
,
J.
Lu
,
G.
Huang
, and
K.
Hu
,
2008
:
Tropical Indian Ocean basin warming and East Asian summer monsoon: A multiple AGCM study
.
J. Climate
,
21
,
6080
6088
, doi:.
Li
,
X.
, and
W.
Zhou
,
2012
:
Quasi-4-yr coupling between El Niño–Southern Oscillation and water vapor transport over East Asia–WNP
.
J. Climate
,
25
,
5879
5891
, doi:.
Lin
,
Z.
,
R.
Lu
, and
W.
Zhou
,
2010
:
Change in early-summer meridional teleconnection over the western North Pacific and East Asia around the late 1970s
.
Int. J. Climatol.
,
30
,
2195
2204
, doi:.
Lu
,
R.
,
2001
:
Atmospheric circulations and sea surface temperatures related to the convection over the western Pacific warm pool on the interannual scale
.
Adv. Atmos. Sci.
,
18
,
270
282
, doi:.
Lu
,
R.
,
2004
:
Association among the components of the East Asian summer monsoon system in the meridional direction
.
J. Meteor. Soc. Japan
,
82
,
155
165
, doi:.
Lu
,
R.
, and
Y.
Fu
,
2010
:
Intensification of East Asian summer rainfall interannual variability in the twenty-first century simulated by 12 CMIP3 coupled models
.
J. Climate
,
23
,
3316
3331
, doi:.
Matsuno
,
T.
,
1966
:
Quasi-geostrophic motions in the equatorial area
.
J. Meteor. Soc. Japan
,
44
,
25
43
.
Nitta
,
T.
,
1986
:
Long-term variations of cloud amount in the western Pacific region
.
J. Meteor. Soc. Japan
,
64
,
373
390
.
North
,
G. R.
,
T. L.
Bell
, and
R. F.
Cahalan
,
1982
:
Sampling errors in the estimation of empirical orthogonal functions
.
Mon. Wea. Rev.
,
110
,
699
706
, doi:.
Song
,
F.
, and
T.
Zhou
,
2014a
:
The climatology and interannual variability of East Asian summer monsoon in CMIP5 coupled models: Does air–sea coupling improve the simulations?
J. Climate
,
27
,
8761
8777
, doi:.
Song
,
F.
, and
T.
Zhou
,
2014b
:
Interannual variability of East Asian summer monsoon simulated by CMIP3 and CMIP5 AGCMs: Skill dependence on Indian Ocean–western Pacific anticyclone teleconnection
.
J. Climate
,
27
,
1679
1697
, doi:.
Sperber
,
K. R.
,
H.
Annamalai
,
I.-S.
Kang
,
A.
Kitoh
,
A.
Moise
,
A.
Turner
,
B.
Wang
, and
T.
Zhou
,
2013
:
The Asian summer monsoon: An intercomparison of CMIP5 vs. CMIP3 simulations of the late 20th century
.
Climate Dyn.
,
41
,
2711
2744
, doi:.
Sun
,
X.
,
R. J.
Greatbatch
,
W.
Park
, and
M.
Latif
,
2010
:
Two major modes of variability of the East Asian summer monsoon
.
Quart. J. Roy. Meteor. Soc.
,
136
,
829
841
, doi:.
Wang
,
B.
,
Q.
Ding
,
X.
Fu
,
I.-S.
Kang
,
K.
Jin
,
J.
Shukla
, and
F.
Doblas-Reyes
,
2005
:
Fundamental challenge in simulation and prediction of summer monsoon rainfall
.
Geophys. Res. Lett.
,
32
,
L15711
, doi:.
Wu
,
R.
,
B. P.
Kirtman
, and
K.
Pegion
,
2006
:
Local air–sea relationship in observations and model simulations
.
J. Climate
,
19
,
4914
4932
, doi:.
Xie
,
S.-P.
,
K. M.
Hu
,
J.
Hafner
,
H.
Tokinaga
,
Y.
Du
,
G.
Huang
, and
T.
Sampe
,
2009
:
Indian Ocean capacitor effect on Indo–western Pacific climate during the summer following El Niño
.
J. Climate
,
22
,
730
747
, doi:.
Xie
,
S.-P.
,
Y.
Kosaka
,
Y.
Du
,
K. M.
Hu
,
J. S.
Chowdary
, and
G.
Huang
,
2016
:
Indo-western Pacific Ocean capacitor and coherent climate anomalies in post-ENSO summer: A review
.
Adv. Atmos. Sci.
,
33
,
411
432
, doi:.
Zhang
,
M.
,
S.
Li
,
J.
Lu
, and
R.
Wu
,
2012
:
Comparison of the northwestern Pacific summer climate simulated by AMIP II AGCMs
.
J. Climate
,
25
,
6036
6056
, doi:.
Zhou
,
T.
, and Coauthors
,
2014
:
Chinese contribution to CMIP5: An overview of five Chinese models’ performances
.
J. Meteor. Res.
,
28
,
481
509
, doi:.
Zhu
,
Y.
,
H.
Wang
,
W.
Zhou
, and
J.
Ma
,
2011
:
Recent changes in the summer precipitation pattern in east China and the background circulation
.
Climate Dyn.
,
36
,
1463
1473
, doi:.

Footnotes

© 2017 American Meteorological Society. For information regarding reuse of this content and general copyright information, consult the AMS Copyright Policy (www.ametsoc.org/PUBSReuseLicenses).