Corresponding author: Prof. Song Yang, Department of Atmospheric Sciences, Sun Yat-sen University, Guangzhou, Guangdong 510275, China. (email@example.com)
 The National Centers for Environmental Prediction (NCEP) Climate Forecast System (CFS) provides important source of information on seasonal climate prediction for many Asian countries that are affected by monsoon. In this study, the authors provide a comprehensive assessment of the prediction of East Asian winter monsoon (EAWM) by the CFS version 2 (CFSv2) using the hindcast for 1983–2010, with a focus on seasonal–interannual time scales. Output from the Atmospheric Model Intercomparison Project (AMIP) and the Coupled Model Intercomparison Project (CMIP) simulations is also analyzed to understand the physical process of monsoon. Several major features of the EAWM are well predicted by the CFSv2. Particularly, the EAWM-related atmospheric circulation and surface climate over oceans are well predicted several months in advance, and the prediction over oceans is better than that over land. While the CFSv2 has low skill in predicting the Arctic Oscillation (AO), it well predicts El Niño-Southern Oscillation (ENSO) and its impact on the EAWM, contributing to the decent prediction of EAWM. Comparisons among hindcast, AMIP, and CMIP indicate that ocean–atmosphere coupling is important for EAWM prediction. While the EAWM in AMIP is weaker, CMIP predicts more realistic monsoon features. The enhanced performance of CMIP is partly attributed to its better simulation of precipitation over the western Maritime Continent (MC). All three types of experiments fail to depict the relationship between EAWM and AO and simulate a stronger-than-observed response of EAWM to ENSO. Improving the simulation of convection over the MC potentially enhances the skill of CFSv2 in predicting the EAWM.
If you can't find a tool you're looking for, please click the link at the top of the page to "Go to old article view". Alternatively, view our Knowledge Base articles for additional help. Your feedback is important to us, so please let us know if you have comments or ideas for improvement.
 The East Asian winter monsoon (EAWM), which is the strongest winter monsoon of the world, exerts significant impacts on weather and climate in and outside East Asia [Chen et al., 2000, 2005; Yang et al., 2002; Yang and Lau, 2006; Chang et al., 2006; Zhou et al., 2007; Li and Yang, 2010]. In the recent decades, there has been an increasing trend of severe weather and climate disasters associated with the anomalies of the EAWM. For example, southern China experienced record-breaking snowstorms in early 2008 [Wen et al., 2009; Zhou et al., 2009; Wu et al., 2011], and prolonged droughts hit southwestern China in winter and spring of 2009–2011 [e.g., Jiang and Li, 2011], causing severe economic and social disasters. However, the EAWM receives less attention by both research and operational community, compared to its summer counterpart – the Asian summer monsoon.
 The EAWM exhibits strong interannual variations. A strong EAWM is characterized by a strong Siberian High (SH), an enhanced East Asian jet stream (EAJS), a deepened East Asian trough (EAT), strong northerly flow over East Asia, and frequent cold surges [Chang et al., 2006; Ding and Sikka, 2006; Wang et al., 2010; Park et al., 2011]. Previous studies have reported that the interannual variations of EAWM are well correlated with El Niño-Southern Oscillation (ENSO) [Zhang et al., 1996; Zhang et al., 1997a; Chen et al., 2000; Wang et al., 2000] and the Arctic Oscillation (AO) [Thompson and Wallace, 1998; Gong et al., 2001; Wu and Wang, 2002; Park et al., 2010]. The EAWM tends to be weak during El Niño, associated with the anticyclonic circulation over the Philippine Sea, which is caused by the warming in the central Pacific [Wang et al., 2000]. The winter AO also affects EAWM circulation components over the middle latitudes, including SH, EAT, and EAJS [Gong et al., 2001; Wu and Wang, 2002; Park et al., 2010; Li and Yang, 2010].
 Most of the previous studies on the EAWM were conducted based on observational data, although numerical models have also been applied in the past decades [Zhang et al., 1997b; Kang et al., 2002; Yang et al., 2008; Wang et al., 2008; Zhou et al., 2010; Li and Wang, 2012]. State-of-the-art models show considerable skill in predicting the Asian summer monsoon [Yang et al., 2008; Wang et al., 2008]. However, how well the EAWM can be predicted by climate forecast systems such as the National Centers for Environmental Prediction (NCEP) Climate Forecast System (CFS) is still not fully known, although a few studies have assessed the predictability of EAWM by different climate models. Zhang et al. [1997b] reported that the Atmospheric Model Intercomparison Project (AMIP) models show a wide range of skills in simulating cold surges and their transient properties. However, most models cannot simulate the relationship between cold surges and tropical convection properly. Most recently, Sohn et al.  showed that the seasonal prediction models that participate in the APEC Climate Center multimodel ensemble seasonal forecast had difficulties in predicting the interannual variability of East Asian climate. On the other hand, Li and Wang  showed that the EAWM can be reasonably predicted by the models from the Development of a European Multimodel Ensemble System for Seasonal to Interannual Prediction [Palmer et al., 2004]. In short, the state-of-the-art models show a wide range of skills in predicting the EAWM.
 The NCEP CFS, which is a fully coupled forecast system, provides operational prediction of the world's climate including the Asian monsoon climate [Saha et al., 2006]. Li and Yang  reported that a dynamical EAWM index was reasonably predicted by the first version of the CFS (CFSv1), which was replaced by a new version of the CFS (CFS version 2; CFSv2) in March 2011. Compared to CFSv1, the CFSv2 incorporates a number of new physical packages for cloud–aerosol–radiation, land surface, ocean and sea ice processes, and a new atmosphere–ocean–land data assimilation system [Saha et al., 2010]. Improved skill in predicting global land precipitation, surface air temperature, and the Asian summer monsoon has been shown in CFSv2 relative to CFSv1 [e.g., Yuan et al., 2011; Jiang et al., 2012]. Kim et al. [2012a, 2012b] discussed the differences of seasonal predictions of the Asian summer monsoon and Northern Hemisphere winter climate between CFSv2 and European Center for Medium-Range Weather Forecasts System 4. However, many features about the prediction of EAWM by the CFSv2 have not been documented. In this study, we attempt to understand the comprehensive features about the prediction of large-scale aspects of EAWM and the dynamical–physical processes related to the variability and predictability of the winter monsoon in the CFSv2.
 Previous studies have shown that the EAWM interacts with tropical oceanic and atmospheric processes. Monsoon-related cold surges can induce large variability of the convection over the South China Sea, the Philippines, and the Maritime Continent (MC) [Chang and Lau, 1980; Chang et al., 2006]. The latent heat released by the intense, near-equatorial convection in turn affects the upper branch of local meridional circulation, accelerating the upper-tropospheric EAJS and deepening the EAT [Chang and Lau, 1980, 1982; Yang and Webster, 1990; Yang et al., 2002; Chang et al., 2006]. Given the possible interaction between extratropical and tropical systems, ocean–atmosphere coupled models may help improve prediction of the EAWM. Currently, both tier-1 and tier-2 climate prediction systems are applied in operational climate prediction [Bengtsson et al., 1993; Palmer et al., 2004; Sohn et al., 2011]. One of the key differences between the two systems is whether ocean–atmosphere coupling is included in models. It has been shown that the models that include ocean–atmosphere coupling have better skills in predicting the Asian summer monsoon compared to those in which ocean–atmosphere coupling is excluded [Wang et al., 2005; Jiang et al., 2012]. However, the role of ocean–atmosphere coupling in predicting the EAWM is unknown. In this paper, we will examine the differences in simulations of EAWM by the CFSv2 hindcast and AMIP-type simulations, as well as CFSv2 free runs designed as the Coupled Model Intercomparison Project (CMIP). We attempt to offer useful information for understanding how ocean–atmosphere coupling affects the EAWM prediction by the dynamical prediction model.
 The rest of this paper is organized as follows. A description of the CFSv2, three different types of CFSv2 integrations, and observational data sets is given in section 2. Predictions of the climatological features and interannual variations of the circulation patterns and surface climate related to EAWM in the CFSv2 are discussed in section 3. In section 4, we discuss EAWM predictions in different time leads. In section 5, we investigate the role of ocean–atmosphere coupling in monsoon prediction by comparing hindcast, AMIP-type simulation, and CMIP-type simulation. A summary of the results obtained is provided in section 6.
2 Model, Experiments, and Observational Data
 The NCEP CFSv2, a fully coupled dynamical prediction system [Saha et al., 2010], consists of the NCEP Global Forecast System at T126 resolution, the Geophysical Fluid Dynamics Laboratory Modular Ocean Model versions 4.0 at 0.25–0.5° grid spacing coupled with a two-layer sea ice model, and the four-layer Noah land surface model. This study analyzes output from three types of simulations with the CFSv2, including hindcast, AMIP-type simulation, and CMIP-type simulation. For convenience, these simulations are referred to as hindcast, AMIP, and CMIP, respectively. Output from the CFSv2 9 month hindcast is analyzed over a 28 year period of 1983–2010. Initial conditions of hindcast come from the NCEP CFS reanalysis [Saha et al., 2010]. Beginning on 1 January, 9 month hindcast runs were initialized from every 5th day and run from all four analysis cycles of that day. The initial days vary from one month to another. A detail description about the initial time can be found at http://cfs.ncep.noaa.gov/cfsv2.info/ (see file “Retrospective CFSv2 Forecast Data Information”). An ensemble average of the monthly mean values of 24 members is used as the prediction, with the initial dates after the 7th of the particular month are used to construct the ensemble prediction for the subsequent month. The winter of a specific year refers to the December of the previous year and the January and February of the current year (DJF). For DJF 0 month lead forecast, it is an ensemble mean of the runs initialized from 2 and 7 December and 12, 17, 22, and 27 November. The longest 7 month lead forecast for DJF is an ensemble mean of the runs initialized from 1 and 6 May and 11, 16, 21, and 26 April.
 The AMIP is an ensemble mean of 11 integrations by the atmospheric component of CFSv2, which were all initialized from January 1950 with different atmospheric initial conditions. Monthly sea surface temperature (SST) and sea ice are specified from HadISST [Rayner et al., 2003] and optimally interpolated (OI) SST [Reynolds et al., 2007] are used as the boundary conditions for 1950–2008 and 2009–2010, respectively. AMIP runs are forced with time evolving observed monthly CO2 concentration as well. The CMIP integration, initialized on 1 January 1988, is run for 48 years and data from the last 28 years are analyzed.
 The observations used for model verification include the Climate Prediction Center Merged Analysis of Precipitation (CMAP) [Xie and Arkin, 1997], winds and temperatures from the NCEP CFS reanalysis [Saha et al., 2010], and SST from the National Oceanic and Atmospheric Administration OI SST analysis [Reynolds et al., 2007]. With first guess from a coupled atmosphere–ocean–sea ice–land forecast system, the CFS reanalysis has improved the climatological precipitation distribution over various regions and the interannual precipitation correlation with observations over the Indian Ocean (IO), the MC, and the western Pacific compared to several previous reanalyses [Wang et al., 2011].
 In this study, we apply a dynamical EAWM index proposed by Li and Yang  to measure the interannual variability of EAWM. The index is defined as the mean horizontal shear of 200 hPa zonal wind:
IEAWM = (U1–U2 + U1–U3)/2, where
U1 = U200(30–35°N/90–160°E),
U2 = U200(50–60°N/70–170°E), and
U3 = U200(5°S–10°N/90–160°E).
 Compared to previous indices, this index takes into account several influencing factors of the monsoon (e.g., AO and ENSO) and better elucidates the physical processes associated with the EAWM [Li and Yang, 2010; Wang and Chen, 2010, Wang et al., 2010].
 Following Li and Wang , the AO index is defined as the normalized difference in zonally averaged sea level pressure (SLP) anomalies between 35°N and 65°N. The AO index is a good measure of hemisphere-wide fluctuations of air mass between the two annular belts of action over the middle latitudes [Li and Wang, 2003; Li, 2005], and the definition has been used in previous studies [e.g., Wu et al., 2009; Jiang and Li, 2011]. Based on this definition, the AO index can be easily calculated and compared among different datasets. The SH index is defined as averaged SLP over 40–60°N/70–120°E [Gong et al., 2001] and the EAJS index is the areal mean of 200 hPa zonal winds over 30–35°N/130–160°E [Yang et al., 2002]. The EAT is measured by area-averaged 500 hPa geopotential height over 30–45°N/125–145°E [Sun and Li, 1997].
3 Prediction of Climatological Features and Interannual Variations of EAWM
 In this section, we discuss the climatological features and interannual variation of the EAWM. We compare the CFSv2 prediction with 0 month lead against the corresponding observations.
3.1 Climatological Features of Atmospheric Circulation and Surface Climate
 The EAWM is deep system and is associated with various atmospheric features in the entire troposphere. It is associated with the SH and the Aleutian Low at the lower troposphere. These systems result in strong northwesterlies over the eastern flank of SH, which turn to northeasterlies around 25°N and move southward to the tropics (Figure 1a). The EAWM is also related to the EAT along the longitudes of Japan at the middle troposphere and the EAJS maximizing over the southeast of Japan at the upper troposphere (Figure 1d). The CFSv2 hindcast well captures these climatological features related to the EAWM (Figures 1b and 1e). However, the hindcast also shows biases, including the higher-than-observed SLP to the north of 40°N maximizing over northeastern China, the weaker-than-observed cyclonic circulation to the east of Japan, and the subtropical high over the western North Pacific (Figure 1c). In addition, the hindcast has bias in simulating the divergent winds over the MC compared to observations, with easterlies over the western MC and westerlies over the eastern MC. The 500 hPa geopotential height predicted by CFSv2 is lower than observed except that along Japan, resulting in weaker-than-observed EAT (Figure 1f). It is interesting to note that larger bias in 200 hPa zonal wind appears over the tropics than over middle latitudes, with a maximum bias over the western tropical Indian Ocean (IO). The predicted zonal wind along the westerly jet stream is generally weaker than observation except that over the Tibetan Plateau and east of Japan.
 The 2 m air temperature (T2m) over the Asian continent is lower than that over oceans, with minimum over the Tibetan Plateau. This temperature pattern is considered as one of the important driving mechanisms of the EAWM (Figure 2a). The hindcast well reproduces the spatial pattern of T2m, although cold biases are found over north of 45°N and warm biases over the Tibetan Plateau, eastern China, South Korea, and Japan (Figures 2b and 2c). The cold bias is thermally coupled with the higher-than-observed SLP (Figures 1c and 2c), highlighting the importance of surface air temperature simulation in predicting SLP.
 During winter, heavy precipitation bands are located over the tropics and the Southern Hemisphere, with maximum over the MC. Most of East Asia receives light precipitation except the regions from eastern Taiwan to eastern Japan (Figure 2d). A similar precipitation pattern over East Asia can be found in the CFSv2 (Figure 2e). However, the model produces more-than-observed precipitation over most of East Asia, especially the low-precipitation regions (Figures 2d and 2f). In addition, the hindcast has large biases over the MC, with less-than-observed precipitation over the western MC and more-than-observed precipitation over the eastern MC.
3.2 Interannual Variations of Atmospheric Circulation and Surface Climate Related to EAWM
 The above analysis of hindcast indicates that the CFSv2 captures most climatological features of EAWM. How well does the model predict the interannual variation of the EAWM in different leads of time? To answer this question, we analyze the dynamical EAWM index recently developed by Li and Yang . Figure 3 shows that both observed and predicted EAWM indices experience strong interannual variability. The CFSv2 predicts the observed values realistically in most years, with a correlation coefficient of 0.55 between the observed and predicted indices, exceeding the 99% confidence level for the Student t-test.
 To assess which components of EAWM are better predicted, we analyze the composite differences in SLP, 850 hPa winds, 500 hPa geopotential height, and 200 hPa zonal wind between strong and weak EAWM years (Figure 4). The strong (weak) EAWM is indicated by the values of EAWM index higher (lower) than 0.5 (−0.5) standard deviation. There are eight high index years (1984–1986, 1996, 1999, 2001, 2006, and 2008) and eight low index years (1983, 1989–1990, 1992–1993, 1998, 2003, and 2007) in observation based on this criterion. However, the CFSv2 predicts an opposite sign for the observed strong EAWM in 1999 and weak EAWM in 1989 (Figure 4). Thus, we exclude these two years in the following composite analysis. It should be note that the composite features of EAWM do not apparently depend on the exclusion of the two years.
 Compared to the low index years, SLP is higher over the Asian continent and lower over oceans in the high index years, with a minimum over eastern Japan and a maximum over the Ural Mountains (Figure 4a). Anomalies of 850 hPa winds with strong EAWM show cyclonic circulation to the east of Japan, anticyclonic circulation over the Ural Mountains, and convergent flow pattern over the MC (Figure 4b). At 500 hPa, the geopotential height with high EAWM index significantly increases over the Ural Mountains and decreases over Northeast Asia and the tropical IO (Figure 4c). Compared to the low index years, the EAJS strengthens in the high index years, accompanied by an increase in the easterlies over the tropics and a decrease in the westerlies over the high latitudes (Figure 4d). All these features indicate that a strong EAWM is characterized by strong SH, EAJS, and lower-tropospheric northerlies over East Asia, a deepened EAT, and a strengthened Ural ridge compared to a weak EAWM (Figures 4a–d) [Li and Yang, 2010; Wang and Chen, 2010, 2010].
 It can be seen from Figure 4 that the hindcast only captures the EAWM-related anomalous circulations over the tropics and the western North Pacific, with a weaker-than-observed magnitude. It fails to reproduce the high SLP and 500 hPa geopotential height over the Ural Mountains and the anomalous easterlies over the Asian continent in the high index years. In the hindcast, the differences in SH, EAT, EAJS, and the northerlies over East Asia between high and low EAWM index years are overall weaker than observations. We also analyze composite features between strong and weak EAWM based on another EAWM index proposed by Gong et al. , which used to measure to variation of the SH. The results indicate that the CFSv2's inability does not depend on the selection of EAWM index.
 A strong EAWM is associated with overall low T2m over north of 35°N, southeastern China, the western Indo-China peninsula, the South China Sea, the western MC, and the tropical IO, but high T2m over the Tibetan Plateau and the subtropical western Pacific compared to a weak EAWM (Figure 5a). The CFSv2 captures the general anomalous pattern of T2m; however, the magnitude of both cold and warm anomalies is smaller than observations (Figure 5c). These biases correspond to the shortcoming of the model in predicting the EAWM-related anomalies over the middle latitudes. A strong EAWM is characterized by excessive precipitation over the MC and the Philippine Sea and deficient precipitation from southern China to south of Japan (Figure 5b) [Wang and Chen, 2010, 2010]. This pattern of anomalous precipitation can be predicted by CFSv2 except a dry bias over the western MC and the eastern tropical IO (Figure 5d).
4 Prediction of EAWM as a Function of Lead Time and Source of Prediction Skill
4.1 Prediction of EAWM as a Function of Lead Time
 Figure 6 presents the correlation of observed EAWM index, AO index, and Niño3.4 SST with those predicted by the CFSv2 (in hindcast) with different lead months. The model shows skills in predicting the EAWM index for most lead months. The skills of EAWM prediction first decrease from 0 month lead to 1 month lead, and then change only slightly with lead time and maintain statistically significant levels for long leads. Because of the significant influences of ENSO and AO on the EAWM, we also examine the skills of CFSv2 in predicting these influencing factors. While the model predicts the variation of Niño3.4 SST very well for all time leads, it can only predict the AO with high fidelity for 0 month lead. Monthly prediction for the differences in 850 hPa winds between two consecutive lead months demonstrates that initial conditions are important for predicting extratropical circulation in the first three months (figure not shown). This feature partly explains why the hindcast only shows skill of seasonal prediction of the AO in 0 month lead. The prediction skill of EAWM does not decay persistently as ENSO, because the link between EAWM and Nino3.4 generally becomes stronger as lead time increases (figure not shown). We also examine the persistence skills of the EAWM index, AO index, and Niño3.4 SST (figure not shown). Compared to the persistence skills of observed Niño3.4, the CFSv2 does not show advantage when lead time is less than 6 months. The observed Niño3.4 exhibits an apparent spring predictability barrier, but the CFSv2 does not show this feature. The CFSv2 has an obvious advantage in predicting EAWM and AO, which shows low persistence in observations.
 Figure 7 shows the differences in circulations between the high and low EAWM index years for different lead months. The differences in all variables over oceans are similar to those for 0 month lead and do not change with lead time apparently. Low skill appears in predicting the differences in atmospheric circulation over the northern Asian continent for all time leads. The predicted differences do not show consistent features among various variables and among various leads. For example, the difference in SLP over the Ural Mountains predicted in 5 month lead is more similar to observation compared to the other leads. The predicted differences in 200 hPa winds of long leads are more similar to those of 0 month lead, while the prediction for 1 and 3 month leads is not so good. Following analyses show that the skill in predicting the 200 hPa zonal winds depends on the predicted response of 200 hPa zonal winds to ENSO.
 The predicted differences in both T2m and precipitation do not vary apparently with lead time, especially over oceans (Figure 8). The CFSv2 fails in capturing the EAWM-related variation of T2m over the Asian continent, differing from the prediction of 0 month lead. However, it depicts both T2m and precipitation over oceans for all time leads except some apparent biases over the MC and the tropical eastern IO. The precipitation biases increase over the eastern MC and decrease over the western MC and the eastern IO as lead time increases.
4.2 Source of Skill of EAWM Prediction
 Because of the strong link of EAWM to ENSO, how well the EAWM can be predicted is also determined by the model ability to depict the response of monsoon to ENSO. The regression of 850 hPa winds against Niño3.4 SST (Figure 9a) shows that ENSO affects mainly the tropical components of EAWM, while weak signals are found over northern Asia (Figure 9a). The model captures this pattern for all time leads, except over the western IO (Figures 9b and 9c; figures for lead time longer than 2 months not shown). In contrast, the impact of AO on EAWM is limited to the middle latitudes. The model only captures the AO-related 850 hPa winds over the Ural Mountains and the North Pacific for 0 lead month, and the magnitude of predicted winds is weaker than observation (Figures 9e and 9f). Together with the previous analysis about the predictability of AO, it is concluded that the CFSv2 can only predict the AO and part of its related atmospheric circulation for 0 month lead, probably with a skill from the atmospheric initial conditions.
 To further illustrate the impact of ENSO on EAWM, we analyze the composite patterns of SST and 850 hPa winds between El Niño and La Niña years (Figure 10). According to the criterion of Climate Prediction Center (available at http://www.cpc.ncep.noaa.gov/products/analysis_monitoring/ensostuff/ensoyears.shtml), there are 10 El Niño years (1983, 1987–1988, 1992, 1995, 1998, 2003, 2005, 2007, and 2010) and 9 La Niña years (1984–1985, 1989, 1996, 1999, 2000–2001, 2006, and 2008). As a response to ENSO, SST increases in the tropical IO. As illustrated by Wang et al. , the ENSO-related central Pacific SST warming generates an anticyclonic circulation over the western North Pacific through local air–sea interaction (Figure 10a). Divergent winds are also seen over the MC, with anomalous easterlies over the equatorial IO and anomalous westerlies over the central Pacific. The CFSv2 captures the anomalous SST pattern very well for all time leads. However, the magnitude of SST decreases over the central and eastern Pacific as lead time increases. Compared to observation, the warming over the oceans near East Asia and the tropical IO is exaggerated for all time leads. The model predicts ENSO-related 850 hPa wind anomalies over the Pacific and the tropical eastern IO for all time leads, but fails to capture the wind anomalies over the tropical western IO.
 Corresponding to the observed 850 hPa wind pattern related to ENSO (Figure 10a), in warm years, precipitation decreases over the MC and the western IO and increases from southern China to south of Japan compared to cold years, associated with weak westerlies over the subtropics and weak easterlies over the tropical IO and the western North Pacific (Figure 11a). The differences in precipitation and 200 hPa zonal winds patterns in the hindcast do not vary significantly with lead time (Figures 11b–f). The model captures the general features of anomalous precipitation. However, over the western MC, the difference is negative in observation but positive in the hindcast. This wet bias is also found in other forecast systems such as the JAMSTEC SINETX-F model [Luo et al., 2005]. Besides, exaggerated difference in precipitation is found from southeastern China to south of Japan. Compared to observation, the tropical anomalous westerlies are well predicted, with a maximum over the South China Sea. However, the model depicts weak subtropical easterly anomalies with an eastward maximum over Japan. The weak easterly anomalies can be attributed to the overpredicted convection over the western MC in the hindcast, which strengthens the subtropical westerlies through the local Hadley circulation [Chang and Lau, 1980, 1982]. The magnitudes of the predicted easterlies vary with lead time. The easterlies of 1 and 3 month leads are weaker than those of the other leads, consistent with the prediction of the EAWM-related 200 hPa zonal winds (Figures 7m–p).
 The above analysis indicates that the source of skill of the EAWM prediction comes mainly from ENSO, with partial contribution from the AO for the 0 month lead prediction. Because the impact of ENSO on EAWM is mostly limited to tropical regions and subtropical East Asia, the EAWM-related components over these regions have a higher prediction skill compared to those over most of the Asian continent. The feature that the correlation between observed and model EAWM indices does not decrease with increasing lead time (Figure 6) can be attributed to the high skill of the model in predicting ENSO and to the model's exaggerated link of the EAWM to ENSO as lead time increases (figure not shown). Although the CFSv2 shows a high skill in predicting ENSO-related features over the Pacific, it has large bias in simulating the ENSO-related precipitation features over the western MC and the eastern IO, which exert a feedback on to the EAWM.
5 Role of Ocean–Atmosphere Coupling in EAWM Prediction
 In this section, we compare the features of EAWM among hindcast, AMIP, and CMIP. The differences between AMIP and hindcast mainly include initial conditions, SST forcing, and ocean–atmosphere coupling. Thus, the difference between hindcast and AMIP can only partially illustrate the role of ocean–atmosphere coupling in EAWM prediction. To understand the possible impact of initial conditions and highlight the importance of ocean–atmosphere coupling, we also analyze the features of EAWM in CMIP simulations.
5.1 Climatological Features
 Figure 12 illustrates the climatological features of EAWM components in AMIP and their differences from observations and hindcast. The AMIP captures the major features of EAWM, including the positions of SH, EAT, and EAJS and the 850 hPa wind pattern. Comparisons among Figures 1c, 1f, 12b, and 12e indicate that the patterns of the differences between AMIP and observation agree with those between hindcast and observation. However, discrepancies are also apparent. Compared to the hindcast, AMIP simulates stronger anticyclonic circulation over Japan and weaker EAT, EAJS, and easterlies over the western MC and the eastern IO. These features demonstrate that AMIP simulates weaker EAWM compared to the hindcast. We note that the differences in atmospheric circulation between AMIP and hindcast are dynamically coupled in the lower troposphere and the upper troposphere, with upper-level convergence and lower-level divergence over the MC.
 Figure 13 shows the climatological features of T2m and precipitation in AMIP and their differences from observations and the hindcast. The AMIP captures the general pattern of T2m, but has large biases in precipitation simulation. Consistent with the biases of atmospheric circulation, the bias patterns of T2m and precipitation of AMIP agree with the bias patterns in the hindcast. Compared to the hindcast, however, AMIP simulates higher T2m over South Asia, lower T2m over the northern Asian continent, deficient precipitation over the MC, and excessive precipitation over the Philippine Sea. Among these differences, the most prominent feature is the deficient precipitation over the MC, which is coupled with the regional atmospheric circulation (Figures 12c, 12f, and 13f). As discussed before, the anomalous convection over the MC can change the intensity of the EAJS and thus the EAWM. Therefore, the weak EAWM simulated by AMIP may be caused by the weak convection over the MC. It should be noted that the AMIP shows a better skill in simulating the T2m over the coast of East Asia compared to the hindcast.
 We next analyze the climatological features of the EAWM simulated by CMIP (Figure 14). The CMIP simulates the major features of atmospheric circulations associated with EAWM (Figures 14a and 14d). The biases of 850 hPa winds over the tropics and East Asia in CMIP are generally small compared to the biases of the hindcast, although the CMIP has large biases in simulating the SLP and 850 hPa winds over most of the high latitudes (Figures 14b and 14c). In AMIP and hindcast, the bias over the MC is a divergent 850 hPa wind pattern; however, it is replaced by a weak convergent pattern in CMIP. The CMIP shows high skills in simulating 500 hPa geopotential height and 200 hPa winds compared to the hindcast, except over the high latitudes (Figure 14e and 14f). The biases of weak EAJS and EAT are apparently improved from AMIP and hindcast to CMIP.
 The CMIP also simulates the major features of T2m and precipitation, although it has a cold bias over the high latitudes and a wet bias over most of Asia (Figures 14a–b and 14d–e). Compared to the hindcast, the cold bias over the high latitude in CMIP becomes worse (Figures 15b and 15c). However, the CMIP shows a high skill in simulating the precipitation over the western MC, in spite of a wet bias over the equatorial western Pacific. Comparison among the hindcast, AMIP, and CMIP demonstrates that the CMIP improves EAWM simulation, attributed likely to the improvement in simulating the precipitation over the western MC.
5.2 Interannual Variations
 The above analyses indicate that the CFSv2 hindcast has a higher skill in simulating the climatological features of EAWM compared to AMIP. It also has a higher skill in simulating the interannual variation of the monsoon, as shown in Table 1. This better performance of the hindcast may be attributed to the more accurate atmospheric initial conditions, because compared with hindcast, the AMIP has little skill in simulating the AO and the EAT, which are sensitive to initial conditions. However, it is interesting to note that AMIP has a higher skill in simulating the interannual variation of the EAJS compared to hindcast. This feature may be attributed to the better response of convection over the western MC to ENSO, which will be discussed later.
Table 1. Correlations of Observation With Hindcast and With AMIP for EAWM, Arctic Oscillation (AO), Siberia High (SH), East Asian trough (EAT), and East Asian Jet Stream (EAJS). Values in Italic, Bold, and Underline Font Exceed the Confidence Levels of 90%, 95%, and 99%
 Table 2 shows the links of the EAWM to its various components. Both the hindcast and the AMIP simulate a stronger response of the EAWM to Niño3.4 SST and EAJS compared to observations, while the CMIP exhibits a weaker response. All the three types of runs by CFSv2 fail to capture the relationships between EAWM and AO. Compared to observation, weaker links of SH and EAT to the EAWM are found in hindcast and AMIP, especially in AMIP. The relationships between the EAWM and its components are all underestimated by CMIP. Compared to the links of the EAWM to SH and EAT in AMIP, the hindcast depicts the vertical coupling among the different EAWM components more realistically.
Table 2. Correlation of EAWM Index With Niño3.4, Arctic Oscillation (AO), Siberia High (SH), East Asian Trough (EAT), and East Asian Jet Stream (EAJS) in Observation, Hindcast, AMIP, and CMIP. Values in Italic, Bold, and Underline Fonts Exceed the Confidence Levels of 90%, 95%, and 99%
5.3 Response of EAWM to ENSO
 In section 'Source of Skill of EAWM Prediction', we have shown that ENSO is the main source of predictability of the EAWM in the hindcast. Wang et al.  reported that the warming over the central Pacific affected EAWM through local air–sea interaction. Here, we further investigate the influence of ocean–atmosphere coupling on the relationship between ENSO and EAWM. In addition to hindcast and AMIP, we also analyze the differences between the warm and cold years of Niño3.4 SST in the CMIP to illustrate the possible role of ocean–atmosphere coupling in predicting the EAWM. The El Niño (La Niña) years in CMIP are defined as the years in which the normalized Niño3.4 SST is higher (lower) than 0.5. There are eight events for El Niño and La Niña, respectively.
 Figure 16 shows the composite differences in SST and 850 hPa winds between the warm and cold years in the model and their differences from observations. The three types of runs capture the major features of ENSO-related SST and 850 hPa winds. Compared to observations, they all reproduce stronger anticyclonic circulations over the western North Pacific, although the hindcast has a cold bias over the central Pacific. Compared to the cold years, these strong anticyclonic circulations are associated with deficient precipitation over the western North Pacific and excessive precipitation from southern China to south of Japan in the warm years (Figure 17). Compared to hindcast, the AMIP depicts a stronger anticyclonic circulation and larger difference in precipitation over the western North Pacific (Figures 16 and 17). The three types of runs have biases in simulating the response of 850 hPa winds over the tropical IO to ENSO, with anomalous westerlies over the west of MC, which together with the easterlies over the tropical western Pacific forms a convergent pattern over the western MC (Figures 16d–f). These wind differences are associated with excessive precipitation over the western MC in the warm years compared to the cold years (Figure 17). However, the difference in 850 hPa winds in the hindcast is larger than that in AMIP, accompanied by larger difference in precipitation over the western MC. Both the hindcast and the CMIP produce a warm bias in the difference between the warm and cold years over the tropical southeastern Pacific. Both the hindcast and the AMIP reproduce the difference pattern of 200 hPa wind between the warm and cold events, but with an eastward shift of the maximum anomalous subtropical easterlies compared to observation (Figures 17a–c). In contrast, the CMIP simulates westward shifts of maximum anomalous tropical westerlies and subtropical easterlies (Figures 17a and 17d). Compared to observations, the AMIP simulates strong westerly anomalies over South Asia, while weak easterly anomalies over East Asia are found in the hindcast.
 The above analyses of the differences between warm and cold years of ENSO indicate that the CFSv2 tends to simulate a strong response of EAWM to ENSO. The AMIP depicts an even stronger response of EAWM to ENSO compared to the hindcast. The model has biases in predicting the ENSO-related winds and precipitation over the western MC and the tropical IO. However, the bias in AMIP is smaller than that in the hindcast.
 The NCEP CFSv2, which became operational in March 2011, provides important source of information about the seasonal climate prediction for many East Asia countries where monsoon climate prevails. In this study, we have provided a comprehensive assessment of the prediction of EAWM by CFSv2, focusing on seasonal-to-interannual time scales. We also have investigated the importance of ocean–atmosphere coupling for predicting the EAWM and various monsoon-related physical processes by the CFSv2.
 Many major features of the EAWM are well predicted by the CFSv2, which includes the SH, the EAT, the East Asian jet stream, and the low-tropospheric winds, surface temperature, and precipitation over East Asia. A recently defined dynamical EAWM index is also well predicted. The EAWM-related atmospheric circulation, T2m, and precipitation over oceans can be predicted several months in advance. However, the model shows low skill in predicting the extratropical atmospheric circulation over land, which can be partly attributed to its inability in reproducing the AO and its impact on atmospheric circulation. The model's higher skill in predicting the EAWM components over oceans is attributed to its better performance in ENSO prediction. However, bias also exists in the hindcast for depicting the response of EAWM to ENSO.
 Comparisons among hindcast, AMIP, and CMIP experiments indicate that ocean–atmosphere coupling is important for EAWM prediction by the CFSv2. The AMIP reproduces overall weaker-than-observed EAWM, associated with deficient precipitation over the MC. Compared to the AMIP and the hindcast, the CMIP reproduces more realistic EAWM, which is partly attributed to its better performance in predicting the precipitation over the western MC. The hindcast has overall higher skill in predicting the interannual variation of EAWM compared to the AMIP except for the prediction of EAJS. The CFSv2 fails to capture the relationship between the EAWM and the AO. Compared to observation, the CFSv2 tends to simulate a stronger response of EAWM to ENSO. The AMIP depicts an even stronger response of EAWM to ENSO compared to the hindcast. The CFSv2 has a bias in predicting the ENSO-related winds and precipitation over the western MC and the tropical IO. Interestingly, the bias in AMIP is less than that in the hindcast.
 It should be pointed out that, in this study, we have just preliminarily shown that ocean–atmosphere coupling is important for predicting the EAWM based on an analysis of output from hindcast, AMIP, and CMIP and an in-depth investigation is needed to distinguish the specific contribution of air–sea interaction clearly. However, the predictions of climatological precipitation and the EAWM among hindcast, AMIP, and CMIP indicate that the convection over the MC exerts a significant impact on EAWM prediction. Nevertheless, the CFSv2 has biases in predicting the convection over the MC, for both climatology and the response to ENSO. Is the unrealistic response of convection over the MC to ENSO caused by the CFSv2 bias in predicting climatological precipitation? Climatologically, over the eastern MC, wet biases are found in hindcast and CMIP but a dry bias is seen in AMIP (Figures 2f, 13e, and 15e). However, during El Niño years, wet biases are found over the eastern MC in all three types of model runs (Figure 17). On the other hand, climatologically, the CMIP has a small precipitation bias over the western MC, while hindcast and AMIP have dry biases over the western MC especially the AMIP (Figures 2f, 13e, and 15e). However, the AMIP does not have apparent bias over the western MC during El Niño, while hindcast and CMIP show apparent wet biases (Figure 17). It is thus difficult to find an impact of climatological bias of precipitation over the MC on the prediction of interannual variation of precipitation only based on a comparison among hindcast, AMIP, and CMIP. How the climatological precipitation bias over the MC affects the prediction of interannual variation of precipitation over the MC and the EAWM deserves more sensitive experiments with climate models. In addition, the EAWM demonstrates strong interaction between the extratropics and the tropics. The convection over the MC is partially affected by cold surges, the higher-frequency variations of the EAWM. Further investigations are needed to understand how strongly the convection over the MC is affected by cold surges in the NCEP CFSv2.
 The EAWM affects a broad region from the high latitudes to the tropics. The interannual variability of EAWM shows large differences between the southern portion and the northern portion of East Asia [e.g., Wang et al., 2010]. This study shows that the predictability of the EAWM is also different between the two portions. The southern EAWM component, whose variability is mainly affected by ENSO, exhibits larger predictability. However, smaller predictability is found for the northern EAWM component, which is mostly governed by the extratropical atmospheric circulation such as AO. It is challenging to have a decent prediction of the AO by the CFSv2. Large bias of surface air temperature is found over the northern Asian continent, suggesting that the CFSv2 may have bias in predicting the snow cover in this region. Previous studies have reported that the interannual variability of both the AO and the northern EAWM component is partly forced by Eurasian snow cover [Saito and Cohen, 2003; Wang et al., 2010]. Using Community Atmosphere Model version 3, Allen and Zender  showed that realistic variations of the AO could be simulated when the model was forced by satellite-based snow cover fraction data. Thus, more realistic simulations of Eurasian snow cover may be important for improving the prediction of EAWM.
 The authors thank the three anonymous reviewers for their constructive comments, which improved the overall quality of the paper. This study was jointly supported by the National Natural Science Foundation of China (Grant 41105061), the National Basic Research Program of China (Grant 2012CB417202), the Basic Research and Operation Program of the Institute of Plateau Meteorology, CMA (Grant BROP201215), Sun Yat-sen University “985 Project” Phase 3, the Open Research Fund Program of Plateau Atmosphere and Environment Key Laboratory of Sichuan Province (Grant PAEKL-2011-C2), and the R&D Special Fund for Public Welfare Industry (Meteorology) (Grant GYHY201106015). Xingwen Jiang, who was partially supported by the U.S. National Oceanic and Atmospheric Administration and China Meteorological Administration Bilateral Program, thanks NOAA's Climate Prediction Center for hosting his visit while this study was conducted.