Journal of Geophysical Research: Atmospheres

El Niño–Southern Oscillation–related principal interannual variability modes of early and late summer rainfall over East Asia in sea surface temperature-driven atmospheric general circulation model simulations

Authors


Abstract

[1] A large portion of interannual variability during early summer (May–June mean; hereinafter referred to as MJ) and late summer (July–August mean; hereinafter referred to as JA) rainfall over East Asia is dominated by El Niño–Southern Oscillation (ENSO) events. Four ENSO-related modes have been identified by using empirical orthogonal function (EOF) decomposition analysis on East Asian rainfall for the period 1980–1999, with EOF2 for MJ rainfall (hereinafter referred to as MJ-2; similar notations are used for other modes) corresponding to the developing phase of La Niña events, while MJ-3, JA-1, and JA-2 correspond to the decaying phase of El Niño events. The authors investigate the predictability of ENSO-related MJ and JA rainfall modes by analyzing the outputs of 12 atmospheric general circulation models (AGCMs) of the Atmospheric Model Intercomparison Project Phase II (AMIP II), which were run in an AGCM stand-alone mode and were forced by the monthly historical sea surface temperature (SST) from 1980 to 1999. The results show that although the climatological differences between MJ and JA rainfall are reasonably reproduced by most AGCMs, the four ENSO-related interannual variability modes of MJ and JA rainfall show different predictabilities. While both atmospheric circulation and precipitation anomalies associated with MJ-3 (the slow El Niño decaying mode) are reasonably reproduced by nearly all the models, the JA-1 (MJ-2) mode is only partly reproduced by about two thirds (half) of the AMIP II models. All models fail to reproduce the JA-2 mode. The relatively low skills in predicting both JA-1 and JA-2 modes are primarily due to the bias of the AMIP models in simulating both the intensity and the position of the western North Pacific anticyclone. The predictability of the JA-1 mode is slightly higher than that of JA-2, and the difference results from the stronger and longer persistence of SST anomaly (SSTA) forcing associated with the decaying ENSO events. The low skill of the MJ-2 prediction is due to the weak SSTA forcing associated with the developing phase of La Niña. The skills of the AMIP II models in predicting the leading interannual variability modes of East Asian summer rainfall do not depend on the horizontal resolutions of the models.

1. Introduction

[2] East Asian monsoon region is located to the east of the Tibetan Plateau and to the west of the Pacific Ocean. The unique orographic forcing and huge thermal contrasts between the continent and the ocean make the East Asian summer monsoon (EASM) a distinctive component of the global monsoon system [Chen and Chang, 1980; Tao and Chen, 1987; Lau et al., 1988; Ding, 1992]. The EASM is a complicated system that encompasses tropical and subtropical systems and cold air activities over the mid and high latitudes [Zhu et al., 1986]. To predict the EASM variability has been a great challenge to the climate research community.

[3] The conventional seasonal predictions of EASM usually focus on the June–July–August (JJA) anomalies. The current state-of-the-art climate models generally show low skills in predicting or reproducing the subtropical monsoon rainfall anomalies [Wang et al., 2005, 2009a; Zhou et al., 2008, 2009a, 2009b; Li et al., 2010]. Precipitation from a hind-cast experiment involving 14 climate models showed low correlation coefficients between 0.1 and 0.25 with the observation of CMAP (CPC Merged Analysis of Precipitation) [Wang et al., 2009a]. Great efforts have been devoted to understand the origins of variability and to improve the predictability of EASM precipitation [Wang et al., 2009b; Zhou et al., 2009a, 2009b]. With the onset, northward migration and southward withdraw of EASM, the primary rainy season of EASM spans from May to August [Matsumoto and Murakami, 2002; Wang and Lin, 2002; Ding and Sikka, 2006]. Since the EASM exhibits remarkable difference in its mean states of rainfall and circulation between early and late summer, it is suggested that the prediction of bimonthly May–June (MJ) and July–August (JA) rainfall anomalies be more useful [Wang et al., 2009b, hereinafter referred to as WL2009]. In their study, different interannual variability modes of the early and late summer rainfall were identified by applying empirical orthogonal function (EOF) analysis to the observed precipitation (CMAP data set). Correlations between the equatorial and subtropical sea surface temperature anomaly (SSTA) and the leading modes of summer precipitation were also calculated. During the period from 1979 to 2007, El Niño–Southern Oscillation (ENSO)–related modes account for about 35% of MJ variance and about 45% of JA variance [WL2009]. Given the moderate predictability of ENSO, the identification of ENSO-related MJ and JA precipitation modes may provide a potentially high predictability of monsoon rainfall. The current work is an initial effort to improve the predictability of monsoon rainfall, and is an extension of the work by WL2009.

[4] Prediction of ENSO-related bimonthly rainfall over East Asia heavily relies on the accurate prediction of ENSO evolution. K.-Y. Kim et al. [2008] investigated the SST-forced predictability of the subseasonal modes over East Asia using the outputs of an AGCM in June and July. The simulated first leading EOF mode is an SST-forced mode and thus has a moderate predictability. Up to now, efforts devoted to the predictability of early and late summer rainfall variability modes are quite limited. On the basis of the diagnostic analysis of WL2009, the current study aims to answer the following questions: If the ENSO-related SSTA is perfectly predicted (or prescribed), how predictable (or reproducible) is the summer rainfall anomaly over East Asia by the current state-of-the-art climate models? Are there any differences in the predictabilities of early and late summer rainfall anomalies? We try to answer these questions by analyzing the outputs of 12 AGCMs that participate in the Atmospheric Model Intercomparison Project phase II (AMIP II), which were run in an AGCM stand-alone way mode and were forced by monthly historical SST covering the period of 1980–1999.

[5] The performance of climate models may depend on their horizontal resolutions. Impacts of horizontal resolution on the ability of climate models in simulating global monsoon were evaluated using CMIP3 (the Coupled Model Intercomparison Project phase 3) models for IPCC AR4 (the Fourth Assessment Report of the Intergovernmental Panel on Climate Change) [H.-J. Kim et al., 2008]. Regional aspects of monsoon rainfall simulations show remarkable differences depending on the horizontal resolution of atmospheric models. A comparison between T42 and T106 versions of an AGCM found that the higher-resolution model can better represent both the mean and frequency distribution of precipitation over East Asia [Kimoto et al., 2005]. Low-resolution models tend to yield a larger bias in the simulation of both the location and precipitation rate of Meiyu-Baiu rainfall band in comparison with models of a medium resolution [Ninomiya, 2009]. Kusunoki et al. [2006] showed that an AGCM with 20 km horizontal mesh can reproduce the Meiyu-Baiu rainband in June and July realistically. Thus another motivation of the current study is to identify the resolution dependence of the predictability of interannual variability modes of EASM rainfall. Toward this goal, the AMIP II models are grouped into three categories based on their horizontal resolutions. The performances of multimodel ensemble mean (MMEM) for different resolution groups are then evaluated.

[6] The remainder of the paper is organized as follows. Section 2 describes the data sets and analysis procedures. Section 3 gives a brief description of the climatological EASM rainfall in MJ and JA based on observations and model simulations. Four ENSO-related interannual variability modes of EASM derived from the AMIP simulations are compared with the observations in section 4. Possible reasons for the deficiency of the AGCM simulations are discussed in section 5. Summary and discussion are given in section 6.

2. Model, Data, and Method Description

2.1. Model Outputs and Observational Data

[7] The model outputs used in this study are derived from 12 AGCMs of the AMIP II which are forced by monthly historical SST and sea ice. Brief information of the AMIP II models is shown in Table 1. Documentations of the models can be found at the Website of the U.S. Department of Energy's Program for Climate Model Diagnosis and Intercomparison (PCMDI). The outputs of the AMIP II models have been used in many studies, including discussion on cloud radiation [Potter and Cess, 2004], solar insolation [Raschke et al., 2005], leading modes of the interannual variability of Asian-Australian Monsoon [Zhou et al., 2009a], and the vertical structures of atmospheric temperature anomalies associated with different flavors of El Niño [Zhou and Zhang, 2011].

Table 1. Basic Information of the AMIP II Models Used in the Analysis
ModelTypeApproximate Resolution Longitude × Latitude (deg)Group
CNRM_CM3spectral2.8 × 2.8medium
GFDL_CM2_1grid2.5 × 2.0medium
GISS_MODEL_E_Rgrid5.0 × 4.0low
IAP_FGOALS_0_Ggrid2.8 × 2.8medium
INMCM3_0grid5.0 × 4.0low
IPSL_CM4grid3.8 × 2.5low
MIROC_2_HIRESspectral1.1 × 1.1high
MIROC3_2_MEDRESspectral2.8 × 2.8medium
MPI_ECHAM5spectral1.9 × 1.9high
MRI_CGCM2_3_2Aspectral2.8 × 2.8medium
NCAR_CCSM3_0spectral1.4 × 1.4high
UKMO_HADGEM1grid1.9 × 1.3high

[8] The model-data comparison is done using the monthly averaged precipitation over a 20 year time period from 1980 to 1999. Since the horizontal resolutions of the models differ, all model outputs are first interpolated onto a common grid of 2.5° by 2.5° using distance-weighted interpolation method.

[9] In order to evaluate the models' performances, we use a few observational data sets: (1) the Climate Prediction Center Merged Analysis Prediction (CMAP) data set [Xie and Arkin, 1997]; (2) the circulation field derived from NCEP2 (NCEP-DEO AMIP–II Reanalysis [R-2]) [Kanamitsu et al., 2002]; and (3) the extended reconstructed SST version 2 (ERSSTv2) data set [Smith and Reynolds, 2004].

[10] We also use the Global Precipitation Climatology Project (GPCP) data [Adler et al., 2003] to derive the leading interannual variability modes of summer rainfall over East Asia (figures not shown), and the results are nearly the same as those derived from the CMAP data.

2.2. Analysis Procedure

[11] As in WL2009, to extract the principal interannual variability modes of East Asian summer rainfall, EOF analysis is performed using the bimonthly mean precipitation anomalies over the period from 1980 to 1999. In the following discussions, the first three EOF modes for MJ are referred to as MJ-1, MJ-2 and MJ-3 and the corresponding JA modes are referred to as JA-1, JA-2 and JA-3. To isolate the interannual variations, a high-pass Lanczos filter of 8 years is applied to all data sets before doing EOF and regression/correlation analysis.

[12] The 12 AMIP II models are divided into three categories. The low, medium, and high resolution refers to a resolution of ∼3°–5°, ∼2°–3°, and ∼1°–2°, respectively. The MMEMs for different categories are analyzed.

3. Summer Precipitation in the Observation and AMIP Models

[13] The EASM can be divided into two subseasons, early and late summer, since there are prominent differences in the mean state between MJ and JA [WL2009]. Figure 1 depicts the northward migration of the climatological monthly mean rainfall from May to August. The observed seasonal march of rainfall from May to June and from July to August are gradual, while the change from June to July is pronounced and abrupt (Figure 1a), in other words, the seasonal migration of the monsoon rain belt is not uniform, and thus the summer monsoon season should be divided to early and late summer [WL2009]. As shown in Figure 1a, the prominent rainbands in the early summer are similar: they advance northward slightly with an enhancement in intensity. However, from June to July the major rainy regions along 125°E shift from 28°N to 35°N, and dry season starts in South China. In the late summer, the major rainy center is located to the north of the Yangtze River Valley and adjacent oceans.

Figure 1.

Climatological monthly mean precipitation rate (millimeters per day) from May to August over the period 1980 to 1999 for (a) CMAP, (b) low-resolution models, (c) medium-resolution models, and (d) high-resolution models.

[14] The southwest-northeast orientated rain belt in the subtropical monsoon region is not well reproduced by most of the models (Figures 1b1d). The simulated monthly mean precipitation centers are located to the west of the observation. The spatial correlation coefficients of monthly rainfall between the observation and the simulations are shown in Table 2. No significant differences are seen among the MMEMs. The correlation coefficients of JA are generally higher than that of MJ. Isolated rainfall centers are evident to the east periphery of Tibetan Plateau, but less rainfall is seen over the primary monsoon regions (Figures 1b1d). This bias comes from the relatively low resolution of all the AMIP II models [Yu et al., 2000; Zhou and Li, 2002; Chen et al., 2010]. The differences among the AMIP II models are not evident, since even the 1°–2° resolution in the high-resolution models is not high enough to resolve the small-scale strong rainfall caused by the steep topography on the east periphery of the Tibetan Plateau. The change in rainfall from June to July in the simulations is gradual, not as evident as that in the observation.

Table 2. Correlation Coefficients of Climatological Monthly Mean Rainfall Over East Asia Between the MMEMs of the AMIP II Models and the Observationa
 MayJuneJulyAugust
  • a

    At 20°–50°N, 100°–130°E.

Low0.480.620.680.79
Medium0.570.500.540.70
High0.660.710.720.78

[15] The difference between the early and late summer mean precipitation is shown in Figure 2 as JA minus MJ. In Figure 2a, the observed positive and negative rainfall centers coincide well with the maximum JA and MJ subtropical rainfall centers. The abrupt change of precipitation from early summer to late summer is accompanied with notable changes in the monsoonal circulation [WL2009]. As shown in Figures 2b2d, the AMIP models reasonably simulate the major differences between JA and MJ precipitation. But the negative rainfall centers in the low- and medium-resolution MMEMs are located to the west of the observed. The MMEM of high-resolution models exhibits a better performance, except that the location of rainfall center shifts northward slightly, along with a weaker magnitude of the positive rainfall center.

Figure 2.

Differences of climatological monthly mean precipitation rate between late and early summer, namely JA minus MJ, for (a) CMAP, (b) low-resolution models, (c) medium-resolution models, and (d) high-resolution models. Units are millimeters per day.

[16] The resemblances of model simulations with the observation are quantitatively assessed in Figure 3. A better model performance is measured by a higher pattern correlation coefficient and a normalized standard deviation more close to 1.0. The pattern correlation coefficients of most models with the observation are statistically significant at the 10% level (correlation coefficient larger than 0.11), indicating that most of the AMIP II models have moderate performances in simulating the spatial distribution of climatological summer rainfall difference (late summer minus early summer). The ratios of standard deviation are generally smaller than 1.0 for medium-resolution models, indicating a weaker spatial variation than the observation. The spatial variations simulated by three high-resolution models are larger than the observation. The results of high-resolution models are generally better than low-resolution models in terms of both pattern correlation coefficient and normalized standard deviation. But the medium-resolution models are not always superior to the low-resolution models. These results are consistent with those seen in Figures 1 and 2.

Figure 3.

Comparison of the performance of the AMIP MMEMs and individual models against the observed climatological monthly mean precipitation (CMAP) difference between late and early summer. The abscissa indicates the standard deviation of individual model simulation normalized by the observed standard deviation, while the vertical ordinate indicates the pattern correlation coefficient between each model simulation and the observation. The zero correlation line and the 10% significant level are indicated by thick and thin gray lines, respectively.

4. ENSO-Related Interannual Modes Simulated by the AMIP II Models

4.1. ENSO-Related Modes in the Observation: Metrics Used in Evaluating Climate Models

[17] The ENSO-related modes identified by the EOF analysis are used as observational metrics to evaluate the AMIP II models. We repeat the analysis of WL2009 but focus on a different time period covering 1980–1999. The first three EOF modes of MJ and JA rainfall anomalies and the corresponding principal components (PCs) derived from the CMAP are shown in Figure 4. The MJ-1 and MJ-2 modes feature anomalous rainfall centers along 24°N and 30°N, respectively (Figure 4a). The positive center of MJ-3 is located south of 43°N, which is different from that in WL2009. The sequence of the three leading EOF modes of JA rainfall (Figure 4b) are different from that of WL2009, with JA-2 (JA-3) in this study corresponds to JA-3 (JA-2) in WL2009. The JA-1 mode features a meridional dipole pattern with its negative center located south of 28°N and the positive center located north of 28°N. The JA-2 mode features an out-of-phase change of rainfall anomalies between the central eastern China along 30°N and the northern China along 38°N. The JA-2 mode in this study is similar to the JA-3 mode in WL2009. The JA-3 mode exhibits a zonal dipole structure with its positive pole located south to the Yangtze River Valley and the negative pole over the marine area along East Asian coast. Note the difference between this study (Figure 4) and WL2009 is due to the different time period considered, with 1980–1999 in this study versus 1979–2007 in WL2009. To validate the fidelity of EOF modes as observational metrics for model evaluation, a similar analysis is applied to the GPCP precipitation, and the results (not shown) are nearly the same as those derived from the CMAP data set.

Figure 4.

The first three leading interannual variability modes of (a) MJ and (b) JA rainfall over the period of 1980–1999 derived from the CMAP data. (top) The regression coefficients between rainfall anomalies and each PC, and (bottom) the corresponding PCs. The fractional variance for each mode is shown. E and L on each PC denote the developing year of El Niño and La Niña events, respectively.

[18] In WL2009, the MJ and JA modes are divided to two categories: MJ-2, MJ-3, JA-1 and JA-3 are ENSO-related modes, while MJ-1 and JA-2 are non-ENSO-related modes. Note that the spatial distribution of each EOF mode depends on the phase of its principal component, so the above discussions assume to the case when the PC time series are positive. In WL2009, among the four ENSO-related modes, the MJ-2 and JA-1 modes are associated with the decaying phase of ENSO events, while the MJ-3 and JA-3 are associated with the developing phase of ENSO events. To verify the relations in our studies, the lead-lag correlation coefficients between the equatorial Indo-Pacific SSTAs and each PC time series associated with the major MJ and JA modes are calculated for the period from 1980 to 1999 (Figure 5). Although we focus on a shorter time period, the results are similar to WL2009: neither the MJ-1 (Figure 5a) nor the JA-3 (Figure 5f) mode is related to ENSO, because there is no significant signal in the central and eastern equatorial Pacific. These two modes are considered as non-ENSO-related modes in this study. The other four modes are ENSO-related modes, as evidenced by the significant lead-lag correlation coefficients with the eastern equatorial Pacific SSTAs. Among the four ENSO-related modes, the MJ-2 mode is associated with the developing phase of La Niña events (Figure 5b), while MJ-3, JA-1 and JA-2 (Figures 5c5e) are associated with the decaying phase of El Niño events. The ENSO-related MJ modes account for about 30% of the total variance, while the ENSO-related JA modes explain about 58% of the total variance.

Figure 5.

Relationships between the MJ and JA principal modes and the equatorial Indian-Pacific SST anomalies averaged between 10°S and 10°N. The relationships are shown by the lead-lag correlation coefficients of SST anomalies with reference to each PC of (a–c) MJ-1 to MJ-3 and (d–f) JA-1 to JA-3. The shading highlights the areas where the correlation coefficient is statistically significant at the 5% level.

[19] The lead-lag correlation coefficients between the PC time series of the first three EOF modes of early and late summer rainfall anomalies and the Niño-3.4 SST index from January–February (JF) of the preceding year [year(−1)] to November–December (ND) of the reference year [year(0)] are shown in Figure 6. The Niño-3.4 index is defined as the regional average of SSTAs within the domain of 5°S–5°N, 120°–170°W. In Figure 6a, the correlation coefficient between PC1 and the Niño-3.4 index is not significant at the 5% level, thus MJ-1 is considered as a non-ENSO-related mode. After the early summer in year (0), PC2 is significantly negatively correlated with the Niño-3.4 index, thus MJ-2 is correlated with the developing phase of La Niña events. From the summer of year (−1) to the early summer of year (0), PC3 is significantly positively correlated with the Niño-3.4 index, thus MJ-3 is accompanied with decaying El Niño. In Figure 6b, the correlation coefficient between JA-3 and the Niño-3.4 index is weak. However, both JA-1 and JA-2 modes are correlated with the decaying phase of El Niño events. Their correlation coefficients are less than those of MJ-3.

Figure 6.

The lead-lag correlation coefficients between the first three EOF principal components of (a) early and (b) late summer and the Niño 3.4 SST index from January–February (JF) of the preceding year (year(−1)) to November–December (ND) of the reference year (year(0)). The dashed gray lines denote the 5% significant level.

[20] To confirm the fidelity of Figures 5 and 6, the patterns of correlation coefficient between the bimonthly mean SSTAs from January–February (JF) through November–December (ND) and each PC time series of individual MJ and JA modes are examined, and the ENSO-related modes are confirmed (figures not shown).

[21] In the AMIP-type simulations, the observational SST is prescribed and SSTA associated with ENSO can be regarded as “perfectly predicted,” thus the ENSO-related MJ and JA modes may be partly reproduced. We examine this hypothesis by analyzing the outputs of the AMIP II models in section 4.2. The simulated precipitation and circulation anomalies are regressed upon the corresponding PC time series associated with the observed EOF modes shown in Figures 46. Then the regression patterns are compared with those derived from the observation. A better resemblance is referred to as higher predictability for convenience of discussion. In the following analysis, we first obtain the MMEM among the group members and then apply the regression analysis to the MMEM. We also calculated regressed fields from each model first and then averaged them for each group, the results are nearly the same with the results above except for the strength of the signals.

4.2. Principal Modes Associated With Decaying El Niño SSTA in AMIP Models

[22] In Figure 5c, the observed MJ-3 mode is associated with the decaying phase of El Niño events. In the early stage of El Niño decaying year, positive SSTA persist in the central and eastern equatorial Pacific until June and July. In the observations (Figure 7a), South China Sea-Philippine Sea and part of South China are dominated by an anticyclone in early summer, excessive rainfall is evident in eastern China-southern Japan along the northern and northwestern flanks of the anticyclone over the Philippine Sea (PSAC) (Figure 7a). In the mean time, an anomalous anticyclone is evident over the northeastern Asia. Along the western flank of the anticyclone, excessive rainfall prevails over North China. In the MMEMs of the AMIP models (Figures 7b7d), the PSAC is well reproduced. The positive rainfall center stretching from the middle and lower reaches of the Yangtze River Valley into the subtropical North Pacific is well reproduced except for the strength, namely the model strength is stronger than the observed. However, the positive rainfall center in North China is not evident in the MMEMs of low- and medium-resolution models which is due to the failure of the models in reproducing the anticyclone over the northeastern Asia.

Figure 7.

Precipitation anomalies (colors in units of millimeters per day) and 850 hPa wind anomalies (vectors in units of meters per second) regressed upon the PC time series of the MJ-3 mode. The EASM domains used for the EOF analysis are also outlined: (a) the observations, (b) MMEM of low-resolution models, (c) MMEM of medium-resolution models, and (d) MMEM of high-resolution models.

[23] The JA-1 mode is also associated with the decaying phase of El Niño. In July and August, positive SSTA almost disappear in the eastern and central equatorial Pacific, indicating a weak simultaneous SSTA forcing (Figure 5d). In the observation, the anticyclone over the western North Pacific (WNPAC) exhibits a northward shift (Figure 8a). Negative rainfall anomalies appear in South China that is under the control of anticyclone, while excessive rainfall along the northern flank of the anticyclone influences the lower reaches of the Yangtze River and the Huaihe River Valleys. In addition, absolute values of correlation coefficients between the PC of JA-1 and SSTA over the western North Pacific are less than 0.4, which is not statistically significant at the 5% level (Figure 5d). Contributions of the local SSTA over the western North Pacific to the maintenance of WNPAC gradually weaken from June to August [Wu et al., 2010]. As a result, in the MMEMs of the AMIP II models, owing to the weak SSTA forcing neither the location nor the strength of the PSAC is well reproduced. The simulated strength of the WNPAC is too weak and exhibits a large bias in its position. Hence the rainfall anomalies are poorly simulated over the East Asia and the western North Pacific (Figures 8b8d).

Figure 8.

Same as Figure 7 except for the JA-1 mode.

[24] The JA-2 mode is also associated with the decaying phase of El Niño events, but the strength of El Niño-related SSTA with the JA-2 mode is weaker than that with the JA-1 mode. In Figure 5e, the positive SSTAs in the central and eastern equatorial Pacific nearly disappear in April and May. By June and July, SSTAs in the central and eastern equatorial Pacific become very weak. In Figure 9, the WNPAC already disappears in JA; and East China between 30°N and 35°N is dominated by a strong anticyclone with suppressed rainfall. To the north of the anticyclone, excessive precipitation is seen over North China. The observational features are not evident in the simulations (Figures 9b9d). The MMEMs of the AMIP II models all fail to reproduce the observed circulation pattern, thus the observed rainfall anomalies are also poorly simulated, which is in contrast to that of MJ-3.

Figure 9.

Same as Figure 7 except for the JA-2 mode.

4.3. Principal Modes Associated With Developing La Niña SSTA in the AMIP II Models

[25] In Figure 5b, MJ-2 is correlated with the developing phase of La Niña events. By May and June, negative SSTA is well established in the central equatorial Pacific. In the observation, East Asia is dominated by strong southwesterly along the northwestern flank of the anticyclone over the Philippine Sea (Figure 10a). Above-normal rainfall extends from East China along 35°N to the southern Japan. The MMEMs of low- and high-resolution models can partly reproduce the Philippine Sea anticyclone; although the simulated anticyclone shifts eastward (westward) in the MMEM of the low-resolution (high-resolution) models (Figures 10b and 10d). The MMEM of the medium-resolution model fails to reproduce the circulation pattern and the associated rainfall anomalies (Figure 10c).

Figure 10.

Same as Figure 7 except for the MJ-2 mode.

4.4. Assessment of Individual Models

[26] Since the performance of individual models may be different from the MMEMs, the rainfall anomalies associated with the four ENSO-related modes simulated by all 12 models are compared to the observation in Figure 11. Analysis is done over the region of 5°–50°N, 90°–150°E. In Figure 11a, the MJ-2 mode simulated by five medium-resolution models shows negative correlation coefficients with the observation. In contrast, the simulated MJ-3 mode all has positive correlation with the observation except for the MRI model. For the JA-2 mode (Figure 11b), all models show negative correlations between the simulated and observed rainfall anomalies. Results for the JA-1 mode are better, but still there are about one third of the models showing negative correlation coefficients with the observation. Thus, the MJ-3 mode associated with decaying El Niño has the highest reproducibility in the AMIP II models. Both the MJ-2 and the JA-1 modes have moderate reproducibility. The reproducibility of the JA-2 mode is poor. These results are consistent with those shown in Figures 710.

Figure 11.

Comparison of the performance of the AMIP MMEMs and individual models against the observed (CMAP) ENSO-related precipitation anomalies within the region of (5°–50°N, 90°–150°E) in (a) early and (b) late summer. The ENSO-related rainfall anomalies were calculated by regressing rainfall anomalies upon the PC time series associated with the corresponding EOF modes. The abscissa indicates the standard deviation of individual simulation normalized by the observed standard deviation, while the vertical ordinate indicates the pattern correlation coefficient between the simulation and the observation. The zero correlation line and the 5% significant level are indicated by thick and thin gray lines, respectively.

5. Why MJ-2 and the Two JA ENSO-Related Modes Are Not Well Reproduced

[27] The interannual variability of the EASM rainfall is modulated by the seasonal march of East Asian subtropical front, and the first three MJ modes and the two JA modes are collocated with the climatological position of early and late summer rainbands [WL2009]. Although the MJ-3 mode and the first two JA modes (JA-1 and JA-2) are all associated with the decaying phase of El Niño events in this study, the AMIP II models show different performances in terms of these modes, with a higher reproducibility for MJ-3, but relatively poor skill for JA-1 mode and nearly no skills for JA-2 mode. The MJ-2 mode is associated with the developing phase of La Niña, but the skills of medium-resolution models are low for the MJ-2 mode.

[28] In addition to the four ENSO-related modes, predictability of the two non-ENSO-related modes, namely MJ-1 and JA-3, are also examined. The results show that the reproduction of MJ-1 mode is poor in the AMIP II models, with only three out of 12 models can reproduce it (figures not shown). The results of the three MMEMs are also poor. This is not unexpected since MJ-1 is a non-ENSO-related mode and the driving mechanism (warming over South China in the spring, WL2009) was not included in the AMIP-type simulations. One surprising finding is that the predictability of the JA-3 mode is a little higher than that of the MJ-1 mode; and the results of four AMIP II models and the MMEM of low-resolution models resemble those of the observation. The predictability comes from the SSTA forcing in the Niño-4 region (5°S–5°N, 160°E–150°W; see Figure 5f), indicating that this mode is partly ENSO related. Thus, the definition of ENSO-related modes in WL2009 deserves further discussion.

[29] The positive and negative signs of EOF patterns can be reversed depending on the phase of PC time series. Note that our above discussions all assume that the case when the PC time series are positive. From this point of view, the MJ-3, JA-1, and JA-2 modes can be regarded as the modes associated with the decaying phase of ENSO events, while the MJ-2 mode can be regarded as associated with the developing phase of ENSO events.

[30] Since the two ENSO-related JA modes account for about 58% of the total variance, while the two ENSO-related MJ modes accounts for about 30% of the total variance, we once hoped that the two JA modes would show a higher reproducibility. But the results are contrary to our expectation. In section 5.1, reasons for models' low skill are discussed.

5.1. Bias in Models' Response to the Decaying Stage of ENSO Forcing

[31] In the subsequent year after the mature phase of El Niño events, the anticyclone over the western North Pacific is regarded as the dominant system that affects East Asian climate. Previous studies found that the maintenance time of the WNPAC is mainly determined by the remote forcing of the central and eastern equatorial Pacific SSTAs, but the seasonal phases of time evolution of the WNPAC in AMIP II models are not strictly in accordance with the observation because of the absence of the air-sea interaction in the AMIP-type simulations [Zhou et al., 2009a]. Both the Indian Ocean and local SSTAs contribute to the maintenance time of the WNPAC [Wu et al., 2010]. Is the predictability of the ENSO-related MJ-3, JA-1 and JA-2 modes in the AMIP models related to the reproduction of the position and strength of the WNPAC in the ENSO decaying summer? We further examine the model performance by doing composite analysis of the rainfall and wind anomalies on 850 hPa in the ENSO decaying summers.

[32] ENSO index derived from the CPC (Climate Prediction Center) in the National Oceanic and Atmospheric Administration are used to identify ENSO events. El Niño and La Niña episodes are determined on the basis of the threshold of ±0.5°C for the 3 month running mean of SSTAs in the Niño-3.4 region (5°S–5°N, 120°–170°W). From 1980 to 1999, there are five El Niño decaying summers, which are in 1983, 1988, 1992, 1995, and 1998; and 4 La Niña decaying years, which are in 1984, 1985, 1989, and 1996. Composite fields are calculated as differences between El Niño decaying summers and La Niña decaying summers. In the observation, in early summer the western North Pacific is dominated by a strong anticyclone (Figure 12a). In the northwestern flank of the anticyclone, there are positive rainfall anomalies. On the southern and southeastern flanks of the anticyclone, there are negative rainfall anomalies. The middle and lower reaches of the Yangtze River Valley along 30°N are dominated by excessive rainfall anomalies. In late summer, the observed WNPAC shifts northward and westward. The center of positive rainfall anomalies moves into the extratropical Pacific. Rainfall in the western North Pacific monsoon region is suppressed because of the weakening of monsoon trough [Wang et al., 2000].

Figure 12.

(left) Early and (right) late summer rainfall anomalies (shaded, in units of millimeters per day) and 850 hPa anomalous winds (vectors in units of meters per second) based on the composite of the ENSO decaying years during 1980 and 1999 for (a) observation, (b) low-resolution, (c) medium-resolution, and (d) high-resolution MMEM.

[33] In early summer, both the location of the WNPAC and the associated rainfall anomalies are partly reproduced by the MMEMs of different resolution models (left column of Figures 12b12d). The simulated rainfall anomalies are generally weaker than the observation, which is consistent with the weaker strength of the WNPAC in the models. This limitation is the most evident in the MMEM of medium-resolution models (Figure 12c). The relatively better rainfall simulations in the MMEM of high-resolution models over the western Pacific indicates a reasonable heating field, which leads to a better simulation of circulation anomalies (Figure 12d). The correlation coefficients of rainfall anomalies over East Asia and the western North Pacific between the observation and model simulations range from 0.43 to 0.62 (Table 3): all are statistically significant at the 5% level.

Table 3. Correlation Coefficients of Early and Late Summer Rainfall Anomalies Over East Asia and the Western North Pacific at 0°–50°N, 100°–180°Ea
 LowMediumHigh
  • a

    Based on the composite of the ENSO decaying years during 1980 and 1999 between the MMEMs of the AMIP II models and the observation.

MJ0.590.430.62
JA0.370.370.54

[34] In late summer, the simulated WNPAC is almost absent in the MMEM of low-resolution models (Figure 12b). The simulated WNPAC in the MMEM of medium-resolution models is weaker and shifts eastward relative to the observation (Figure 12c). Thus large biases are seen in the simulated rainfall anomalies over the western North Pacific. In the MMEM of the high-resolution models, the simulated WNPAC shifts a little eastward and southward, leading to a large bias in rainfall anomalies over the western Pacific (Figure 12d). Thus, the bias of the AMIP II models in simulating both the intensity and the position of the WNPAC is the main reason for the low skill of rainfall simulation. Correlation coefficients of rainfall anomalies in JA over East Asia and the western North Pacific between the observation and model simulations range from 0.37 to 0.54 (Table 3), which are generally lower than those in early summer.

[35] The extratropical rainfall anomalies over East Asia are affected by anomalous climate over the western Pacific via a meridional teleconnection pattern termed as Pacific-Japan (PJ) pattern [Huang and Li, 1987; Nitta, 1987; Huang and Sun, 1992; Kripalani and Singh, 1993; Kosaka and Nakamura, 2006]. Influence of the PJ pattern on East Asian climate is the strongest in August [Kripalani and Singh, 1993]. The deficiency of the AMIP II models in simulating rainfall anomalies over the western Pacific indicates a bias in reproducing the local heating fields, which further impacts the simulation of extratropical East Asian climate anomalies through the PJ teleconnection pattern. Thus, the predictability of the JA-1 and the JA-2 modes is lower than that of MJ-3, although all of these three modes are forced by the decaying phase of ENSO-related SSTAs.

5.2. Influence of SSTA Intensity Associated With ENSO Events

[36] Why do the AMIP II models fail to reproduce the MJ-2 mode? The MJ-2 mode is associated with the developing phase of ENSO events. In May and June, the simultaneous SSTA over the central and eastern equatorial Pacific associated with the developing La Niña is weak (Figure 5b), which is not strong enough to effectively force the AGCMs, hence about half of the AMIP II models show nearly no skills in simulating the MJ-2 mode.

[37] Although both JA-1 and JA-2 are associated with the decaying phase of El Niño events, the JA-2 mode is more difficult to simulate than JA-1. The difference is partly due to the intensity of SSTA associated with ENSO events. As shown in Figure 5e, in April and May the SSTAs in the central and eastern equatorial Pacific associated with JA-2 is weak, indicating a weaker SSTA forcing than that with JA-1. On the contrary, the Niño-3 (5°S–5°N,150°–90°W) SSTA associated with the JA-1 mode is stronger than that associated with JA-2, and the persistence of positive SSTA in the central equatorial Pacific is about two months longer. The stronger and longer persistent forcing of SSTA leads to a higher predictability of the JA-1 mode relative to JA-2.

6. Summary and Discussion

6.1. Summary

[38] With an intension to predict successfully the bimonthly MJ and JA rainfall anomalies over East Asia, recent studies have identified four ENSO-related MJ and JA rainfall modes over East Asia. The predictability or reproducibility of the ENSO-related modes are investigated by analyzing outputs of the AMIP II models, which were run in an AGCM stand-alone mode and were forced by monthly historical SST for the period from 1980 to 1999. Resolution dependence of model performance is also discussed. The major findings are summarized as follows:

[39] 1. The multimodel ensemble means of the models with different resolutions reasonably simulate the observed differences between early and late summer precipitation over East Asia, indicating a moderate performances of the AMIP II models in reproducing the climatological mean difference of early and late summer precipitation.

[40] 2. Although focusing on different time periods, with 1980–1999 in this study versus 1979–2007 in WL2009, the EOF analysis on precipitation have revealed similar ENSO-related modes of early and late summer rainfall in East Asia, albeit with different sequence in the EOF modes. For the period of 1980 to 1999, the MJ-2 mode is associated with the developing phase of ENSO events, while the MJ-3 mode and the first two leading modes of JA are associated with the decaying phase of ENSO events. The two ENSO-related MJ modes account for 30% of the total variance, while the two JA ENSO-related modes explain about 58% of the total variance.

[41] 3. The four ENSO-related interannual variability modes show different predictability. While both the atmospheric circulation and precipitation anomalies associated with MJ-3 are reasonably reproduced in all models, the JA-1 and MJ-2 modes are reproduced in only two thirds and one half of AMIP models, respectively. Surprisingly, nearly all models fail to reproduce the JA-2 mode.

[42] 4. The low skills of the AMIP II models in predicting JA-1 and JA-2 modes are primarily due to the bias of the AMIP II models in simulating the intensity and position of the WNPAC. Since the amplitude and persistence of SSTA driving JA-1 is stronger and longer than that driving JA-2, the predictability of JA-1 is slightly better than JA-2. The low skill of MJ-2 prediction is mainly due to the weak SSTA forcing in the developing phase of ENSO.

[43] In addition, the skills of the AMIP II models in predicting the ENSO-related modes are not resolution dependent. This may be due to the fact that none of the AMIP II models has actually high horizontal resolution.

6.2. Discussion

[44] The identification of ENSO-related modes for early and late summer precipitation was expected to be useful in monsoon rainfall prediction [WL2009]. Given the high predictability of ENSO-related SST anomalies, we hoped that the MMEM of the AMIP II models would show encouraging skills in predicting the ENSO-related MJ and JA precipitation modes, since the observed SST, which can be regarded as perfectly predicted, were used to drive the models. For modes associated with the decaying phase of ENSO events, while our study enriches previous study (WL2009) in showing evidence that the early summer precipitation anomalies are highly predictable, the low skill for the late summer modes cast a shadow upon the ability to improve monsoon rainfall prediction.

[45] Why do the AMIP II models show low skills in simulating the late summer monsoon rainfall variability modes? Our analysis shows evidence that the relatively weaker SSTA in late summer of ENSO decaying year is one reason. The limitation of the AMIP-type experiment design may be another reason, because it lacks active air-sea feedback. Both data diagnosis and numerical experiments indicate that air-sea coupling is crucial for summer monsoon rainfall simulation [Wang et al., 2005, 2009a; Wu et al., 2006; Zhou et al., 2009a, 2009b]. The skill for rainfall reproduction in SST-driven simulations over the western North Pacific, the South China Sea and the Bay of Bengal is low because of the neglect of active air-sea feedback [Zhou et al., 2009a]. The evolution phase of the anticyclone over the western North Pacific is not accurately consistent with the observation, which may come from the absence of active air-sea feedback, leading to the absence of the response to remote oceanic forcing [Zhou et al., 2009a]. Results from numerical experiments indicate that the increase of monsoon convection in mid-June and the onset of subtropical western North Pacific monsoon in mid-July are dominated by the land memory and atmospheric transient effects, while SST effect is secondary and is a significant negative feedback for these subseasonal changes [Ueda et al., 2009]. Whether air-sea coupling over the western North Pacific and an involvement of land forcing can improve the simulation of early and late summer monsoon precipitation variability modes deserves further study. Analysis in section 5.1 shows that the reasonable reproduction of both the position and the strength of the WNPAC in models, which require air-sea coupling imperatively, is a determining factor for the predictability of JA-1 and JA-2. As a result, improvements for the first two modes of late summer are much needed.

[46] In addition, the SST data used to drive the AMIP II models is at monthly interval, thus the observed high-frequency SST forcing is not included. Whether daily SST forcing can improve the simulation warrants future study.

Acknowledgments

[47] This work is jointly supported by the National Program on Key Basic Research Project (2010CB951904), the National Natural Science Foundation of China (40890054), and China-UK-Swiss Adapting to Climate Change in China Project (ACCC)-Climate Science. Helpful comments from three anonymous reviewers and the Editor, Steven Ghan, are highly appreciated.