The East Asian mei-yu (EAMY), which includes the mei-yu over eastern China, baiu over Japan, and changma over Korea, is an important component of the Asia summer monsoon system. The EAMY rain belt jumps northward to the Yangtze and Huaihe River valleys (in China), Japan, and Korea from mid-June to mid-July, with remarkable interannual variability. In this study, the variability and predictability of EAMY are investigated using the retrospective ensemble predictions of the NCEP Climate Forecast System (CFS). The CFS reasonably captures the centers, magnitude, northward jump, and other features of EAMY over most regions. It also reasonably simulates the interannual variations of EAMY and its main influencing factors such as the western Pacific subtropical high, the East Asian monsoon circulation, and El Niño–Southern Oscillation (ENSO). The CFS is skillful in predicting EAMY and related circulation patterns with a lead time of one month. An empirical orthogonal function analysis with maximized signal-to-noise ratio is applied to determine the most predictable patterns of EAMY. Furthermore, experiments in which the CFS is forced by observed sea surface temperature (SST) exhibit lower skill in EAMY simulation, suggesting the importance of ocean–atmosphere coupling in predicting EAMY.
The CFS, which exaggerates the precipitation over the southern–southeastern hills of the Tibetan Plateau, overestimates the relationship between EAMY and tropical–subtropical atmospheric circulation due to the overly strong ENSO signals in the model, whereas the experiments forced by observed SST produce a weaker relationship. On the contrary, the CFS underestimates the link of EAMY to higher-latitude processes. An increase in the horizontal resolution of the CFS is expected to reduce some of these errors.
In East Asia, the northward movement of rainbands during the rainy season, between the onset and withdrawal of the summer monsoon circulation, is characterized by three notable stages. The first stage occurs in late May and early June, with the rainband moving to 18°–25°N, that is, southern China and tropical oceans (Guo and Wang 1981; Lau et al. 1988; Ding 1994, 2004). From mid-June to mid-July, with the northward expansion of the East Asian summer monsoon (EASM) circulation and the western Pacific subtropical high (WPSH), the rainy season comes into its second stage. The rain belt experiences a corresponding northward jump to the Yangtze and Huaihe River valleys (in China), Korea, and southern Japan. The rain belt at this stage is conveniently referred to as the East Asian mei-yu (EAMY) in this study, which includes the Chinese mei-yu, the Japanese baiu, and the Korean changma. Since mid-July, the rainy season turns to the third and last stage, and rainfall occurs mainly over northern China, northern Japan, and the northern Korean peninsula.
As an important component of the Asian summer monsoon system, the EAMY is also characterized by remarkable interannual variability, linked directly to flood and drought events over East Asia. Thus, the EAMY and its societal and economic impacts have been important subjects in both scientific research and meteorological prediction operations (e.g., Oh et al. 1997; Tanaka 1997; Krishnan and Sugi 2001; Zhu et al. 2003; Zhou 1996, 2006; Chen 2004; Ninomiya 2004; Ding and Sikka 2006; Suzuki and Hoskins 2009). Many features of EAMY including its temporal variations, associated atmospheric circulation patterns, and possible influencing factors, mainly on synoptic time scales, have been studied previously (e.g., Ninomiya and Akiyama 1992; Chen 2004; Krishnamurti et al. 2009; Liu and Tan 2009). During the EAMY period, the South Asian high moves eastward from the Tibetan Plateau and affects the Yangtze River valley. In the meantime, the WPSH moves northward with its zonal axis at 20°–25°N, and the southerly monsoon flow reaches 30°N (Zhu et al. 2003). The variability of EAMY is strongly affected by high-latitude blocking over the Eurasian continent (National Climate Center of China 1998; Yang 2001), the western Pacific subtropical high (Wu et al. 2002), and the EASM and its associated moisture transportation (He et al. 2001). As shown by Kosaka and Nakamura (2006), the Pacific–Japan teleconnection pattern (Nitta 1987) affects the climate over the EAMY region through the lower-level southwesterly Asian monsoon flow. The relationships of the Chinese mei-yu with sea surface temperature (SST)—especially El Niño–Southern Oscillation (ENSO) events (Zhao 1999; Zong et al. 2006) and North Atlantic SST (Gu et al. 2009), the Arctic Oscillation (Gong and Ho 2003), and the Antarctic Oscillation (Gao et al. 2003; Nan and Li 2003)—have also been studied extensively.
Forecast of the EAMY by numerical models has become an important approach during the past two decades. However, the forecast is mainly for short and medium ranges and the prediction of EAMY on monthly and longer time scales often lacks skill (Chou et al. 1990; Chen et al. 1998; Chien et al. 2002). Large interannual variations of both the amount of rainfall and the dates of EAMY onset and termination have long been a challenge for monthly and seasonal prediction of EAMY by climate models [e.g., see the proceedings of the fifth session of the Forum on Regional Climate Monitoring, Assessment, and Prediction for Asia (FOCRAII) in 2009; see online at http://bcc.cma.gov.cn/Website/index.php?ChannelID=70].
As an atmosphere–ocean coupled model, the National Centers for Environmental Prediction (NCEP) Climate Forecast System (CFS) not only provides the official forecast of the U.S. climate but is also an important source of information for predictions of regional climates outside the United States. It has demonstrated skill in predictions of the climate over Africa (Saha et al. 2006), the United States (Goddard et al. 2006; Higgins et al. 2008), and Asia (e.g., see the proceedings of the second session of FOCRAII in 2006; see online at http://bcc.cma.gov.cn/Website/index.php?ChannelID=89) and the variability of ENSO (Wang et al. 2005) and the tropical Atlantic Ocean (Hu and Huang 2007). In particular, Yang et al. (2008b) analyzed the CFS retrospective ensemble predictions and found that the model successfully simulates many major features of the Asian summer monsoon including the climatology and interannual variability of major precipitation centers and atmospheric circulation systems. It also depicts the interactive oceanic–atmospheric processes associated with precipitation anomalies reasonably well at different time leads. Liang et al. (2009) further applied a maximized signal-to-noise empirical orthogonal function (MSN EOF) analysis to the CFS hindcasts to depict the most predictable patterns of the Asian and Indo-Pacific summer precipitation. The model successfully captures the two most dominant modes of observed climate patterns. The model also successfully captures many features of the East Asian winter monsoon and predicts the winter monsoon with high skill in advance by a few months (Li and Yang 2010). These studies with the CFS motivate this further analysis of the EAMY climate and its prediction. Since how ocean–atmosphere coupling, which is important for tropical monsoon modeling (Wang et al. 2003), affects EAMY simulation and prediction is unknown at present, we will also analyze the output from Atmospheric Model Intercomparison Project (AMIP) experiments in which the atmospheric model of the CFS is used (Jha and Kumar 2009).
The main purposes of this study include the following: to explore the characteristics of the seasonal northward jump of the EAMY and its interannual/intraseasonal variations, to evaluate the simulations of EAMY by the NCEP CFS and AMIP experiments, and to assess the predictability of EAMY by the NCEP CFS at different lead times. The paper is organized as follows. Descriptions of observational datasets and the output from the NCEP CFS and AMIP experiments, along with the MSN EOF analysis method, are presented in section 2. In sections 3 and 4, comparisons of the climatological characteristics and interannual variability between observed and simulated EAMY are provided. In section 5, the predictability of EAMY is discussed. Finally, conclusions are given in section 6.
2. Data and analysis methods
The datasets used in this study include 1) monthly and pentad precipitation from the Climate Prediction Center (CPC) Merged Analysis of Precipitation (CMAP) (Xie and Arkin 1996), 2) 500-hPa geopotential height (GPH) and 850-hPa and 200-hPa winds from the NCEP–National Center for Atmospheric Research (NCAR) reanalysis (Kalnay et al. 1996), and 3) the NOAA Optimum Interpolation (OI) SST analysis (Reynolds et al. 2002). For a case study, daily precipitation at 723 Chinese stations provided by the China Meteorological Administration (CMA) is also analyzed.
Output from the NCEP CFS retrospective predictions (or hindcasts), available 12-hourly, is also used in this paper. The CFS is a fully coupled operational dynamical seasonal prediction system (Saha et al. 2006). The atmospheric component is the NCEP Atmospheric Global Forecast System (Moorthi et al. 2001), except for a coarser horizontal resolution with a spectral triangular truncation of 62 waves in the horizontal and 64 sigma layers in the vertical. The oceanic component is the Geophysical Fluid Dynamics Laboratory Modular Ocean Mode (MOM) V3.0 (Pacanowski and Griffies 1998). The model is global longitudinally and extends from 74°S to 64°N in latitude. With a zonal resolution of 1°, its meridional resolution is ⅓° between 10°S and 10°N, increasing gradually with latitude before becoming 1° poleward of 30°S and 30°N. In the vertical, the model has 40 layers with 27 layers in the upper 400 m. Its vertical resolution is 10 m from the surface to the 240-m depth, increasing gradually to about 511 m in the lowest layer with a bottom depth of about 4.5 km. The atmospheric and oceanic components are coupled without flux adjustment, and the two components exchange time-averaged quantities once a day. The output from the CFS retrospective predictions is a 9-month integration of 15 members starting from perturbed real-time oceanic and atmospheric initial conditions (ICs) from the NCEP Global Ocean Data Assimilation and the NCEP/Department of Energy Atmospheric Model Intercomparison Project Reanalysis II (Kanamitsu et al. 2002), respectively. For the 15-member runs for a specific starting month (e.g., June), the ICs are days 9–13, 19–23, and the last two days of the previous month (May) and days 1–3 of the concurrent month (June).
Since the EAMY usually starts from mid-June and analysis of it requires daily data, we only use the middle five members with initial conditions of day 19, 20, 21, 22, and 23 in each month. Given the climatological onset date of EAMY, the lead times of 1–5 months (for convenience, labeled LM1, LM2, LM3, LM4, and LM5) correspond to the ICs of 19–23 in May, April, March, February, and January, respectively. Unless specified, the CFS values presented are the ensemble means of the five members, and the climatological means are the time averages from 1981 to 2004.
For the AMIP run, the atmospheric component of the CFS is used, forced by observed monthly SST. The model also has a triangular spectral truncation of 62 waves (T62, about 2° × 2° latitude–longitude resolution) with 64 levels in the vertical. An ensemble size of five, with ICs from the NCEP–NCAR reanalysis and the NOAA OI SST at 0000 UTC for 1–5 December 1980, is used in this study. A detailed review of the AMIP project can be found in Gates (1992).
The MSN EOF analysis method applied in this study was developed by Allen and Smith (1997). As in Venzke et al. (1999), Sutton et al. (2000), Huang (2004), Hu and Huang (2007), and Liang et al. (2009), this tool is used to derive the patterns that optimize the signal-to-noise ratio; that is, the leading modes are the ones that maximize the ratios of the variances of ensemble mean (the signal) to the deviations among ensemble members (the noise). Given that the ensemble mean contains both a common evolution of all ensemble members (presumed to be the signal) and a residual random part related to the unpredictable differences among the ensemble members (internal noise), the MSN EOF technique minimizes the effects of noise. The MSN EOF provides information about the consistency of individual numbers in the hindcast system and, like a conventional EOF mode, a MSN EOF mode provides a spatiotemporal distribution. By definition, the leading MSN EOF mode represents the most predictable pattern in a hindcast system. A larger variance of this mode in the ensemble mean indicates relatively higher predictability. We apply this analysis to document the predictable signals in precipitation over 20°–45°N, 100°–150°E. For the MSN EOF analysis, the output from all 15 ensemble members is used.
3. Climatological characteristics of EAMY
The East Asian rainband experiences remarkable northward jumps during the summer monsoon season. As shown in Fig. 1, the rainband is located over southern China and the tropical western Pacific (Fig. 1a). After onset of the monsoon over the South China Sea in late May, the rainband moves northward and covers the Yangtze River valley and southern Japan in June when the tropical monsoon intensifies (Fig. 1b), signaling the onset of EAMY. In particular, over southern Japan the monthly precipitation exceeds 350 mm. In July (Fig. 1c) the rainband moves farther northward, with rainfall above 200 mm per month over the Huaihe River valley (between the Yangtze River and the Yellow River) and even heavier precipitation over Korea and southern Japan. While the EAMY occurs most extensively in July, the rainband over China and southern Japan disappears in August (Fig. 1d). Clearly, the EAMY appears mainly in June and July, which is a period of focus in this study.
The intraseasonal features of the northward movement of rainband can be seen more clearly from the 10-day means of precipitation during June and July. Figure 2 shows that both the Chinese mei-yu and the Japanese Baiu occur from middle June to middle July. Farther to the north, the Korean changma occurs mainly in July. The onset date of mei-yu rainfall over the Yangtze River is about 10 days earlier than that over the Huaihe River. The precipitation amount over the Yangtze River valley is similar to that over southern Japan, and the amount over the Huaihe River valley is close to that over Korea.
The features of EAMY described above are simulated by the CFS. As observed, the major rainband appears over southern Japan, the southern Yangtze River valley, and the tropical western Pacific in June (Fig. 3a). Compared to observations (see Fig. 1b), the amount of simulated precipitation is reasonable over the EAMY region. However, the model overestimates the precipitation over the southern–southeastern hills of the Tibetan Plateau. In July (Fig. 3b), the model performs well in capturing the major rainband over southern Japan. It also produces a rainfall center over the Huaihe River valley. However, the rainband over Korea is predicted too far northward, although the magnitude has been successfully forecast. Similar to the simulation in June, the model overestimates the precipitation over the Tibetan Plateau. This problem is similar to the deficit of the summer monsoon simulation, which is improved in a higher-resolution version of the model (Yang et al. 2008b, see their Fig. 16).
Comparison between Figs. 2 and 4 indicates that the CFS also captures the general features of observed precipitation reasonably well on a 10-day time scale, in spite of the overestimation over the southern Tibetan Plateau. For example, the model captures the locations of major centers of the EAMY in each period. Similar to observations, the EAMY rainband in the CFS exhibits a clear northward jump from June to July. The onset date of the mei-yu over the Yangtze River is about 10 days earlier than that over the Huaihe River, similar to observations. The CFS EAMY begins realistically in mid-June; however, it terminates later than observed, especially over the Huaihe River and southern Japan.
The EAMY in AMIP experiments (Fig. 5) is weaker than that observed. Without ocean–atmosphere coupling, the atmospheric model underestimates the rainbands over eastern China and southern Japan, especially over north of 30°N in July. The underestimation of the mei-yu by atmospheric models is not unique to this study. Indeed, the AMIP experiments with many general circulation models experience substantial difficulties in simulating the EAMY (Kang et al. 2002). As shown by Yoo et al. (2004), an atmosphere-alone model performs apparently worse than an ocean–atmosphere coupled model in simulating the climate anomalies over East Asia for the summers of 1993 and 1994. Comparison of Fig. 5 with Figs. 1b and 1c indicates that the AMIP experiments also significantly underestimate the monsoon precipitation over the South China Sea and the tropical western Pacific. This feature is also consistent with the study of Wang et al. (2003), who have demonstrated the importance of ocean–atmosphere coupling in simulations of monsoon, of which the EAMY is a part.
Figure 6a shows the June–July climatological patterns of 850-hPa winds and 500-hPa geopotential height of the NCEP–NCAR reanalysis. The major circulation systems associated with EAMY include WPSH, EASM, the lower-level westerly jet stream, the Somalia cross-equatorial flow, and blocking activity at higher latitudes. Atmospheric moisture is transported into and beyond the EAMY region from the Indian Ocean, through the South China Sea, southeastern China, and the East China Sea. All these systems are reproduced by the CFS reasonably well (Fig. 6b). However, compared to observations, the simulated circulation patterns are weaker except for the geopotential height north of the westerly jet stream (Fig. 6c). For example, WPSH and southerly wind over eastern China are weaker, leading to smaller moisture transport, in the CFS than in the reanalysis. This feature is consistent with the lighter CFS precipitation over the Yangtze River basin shown in Fig. 3. Nevertheless, in both observations and the CFS, the northward jumps of the 500-hPa WPSH and the 850-hPa southwesterly flow can be found in the time–latitude cross sections of geopotential height and winds (figures not shown).
As shown in Fig. 7, the WPSH in AMIP experiment is weaker than observations (see Fig. 7b) and weaker than that in the CFS (cf. Figs. 7b and 6c). Compared to observations, the AMIP experiments produce stronger cross-equatorial flow over the central Indian Ocean and stronger southwesterly monsoon flow from the western Bay of Bengal through northern Indo-China peninsula to northern South China Sea. A comparison between Figs. 6c and 7b also indicates a larger model–observation discrepancy in Fig. 7b (for AMIP) over the Arabian Sea and the tropical Indian Ocean.
4. Interannual variations
We now discuss the interannual variations of EAMY and its relationships with WPSH, EASM, the low-level westerly jet stream, and tropical SST, especially ENSO. To facilitate our analysis, we define or adopt several indices to measure the temporal variations of EAMY and its associated systems or phenomena. The Niño-3 SST index is calculated as the averaged SST over 5°S–5°N, 90°–150°W. The EASM index was defined by Zhang et al. (2003) as the difference in 850-hPa zonal winds between 10–20°N, 100°–150°E and 25–35°N, 100°–150°E; the intensity of WPSH is computed as the 500-hPa geopotential height averaged over 15°–30°N, 110°–140°E (Zhao 1999). Our index of EAMY is defined as the mean precipitation over 27.5°–40°N, 110°–140°E, which covers the Chinese mei-yu, the Korean changma, and the Japanese baiu. Model hindcasts are the output based on the ICs of 19–23 May, that is, a lead time of about one month for this section.
Table 1 shows the correlation of EAMY index (June–July) with previous and concurrent Niño-3 SST indices in observations. All values are positive, meaning that warmer (colder) SST over the tropical eastern Pacific is favorable for more (less) EAMY, consistent with the result of Zong et al. (2006). The Niño-3 SST from February to April has the strongest relationship with EAMY and the correlation exceeds the 95% confidence level (t test) in each month. Compared with the correlation between Niño-3 SST and EAMY, the WPSH and the EASM are even more significantly correlated with the EAMY. In particular, the simultaneous correlations of WPSH–EAMY and EASM–EAMY are 0.55 and −0.69, respectively, both exceeding the 99% confidence level.
Figure 8 shows the June–July regressions of 850-hPa winds and 500-hPa geopotential height against EAMY index for observations, CFS, and AMIP. In observations (Fig. 8a), significant positive height–EAMY regression is found over tropical/subtropical regions with a maximum over the western Pacific warm pool and southern Asia. When the WPSH expands westward and covers the coastal regions of southeastern China, moisture is transported from tropical oceans to the EAMY region by the summer monsoon flow along the western flank of WPSH. There is another positive correlation center over North Africa, associated with the North African high, which is also considered an indicator for Chinese mei-yu prediction (Zhao 1999). When this high strengthens, the Somalia cross-equatorial flow is suppressed and the westerly flow over the tropical Indian Ocean weakens. Correspondingly, the EASM weakens, leading to a low-level convergence and an increase in precipitation over the Chinese mei-yu region (He et al. 2001).
Positive regressed values can also be seen at higher latitudes, especially over the eastern Ural Mountain and the Okhotsk Sea, related to blocking activity (Zhao 1999). Long lasting systems at high latitudes, such as blocking highs, bring cold and dry air into the mei-yu region and cause convergence when it meets the locally warm and wet air. A center of significantly negative regression is located over Korea and adjacent regions. The regression of zonal wind against EAMY is also negative in the westerly jet core. The regression pattern along the East Asian coast is similar to the so-called Pacific–Japan pattern, which may be forced by the convective activity near the Philippines (Nitta 1987; Nitta and Hu 1996).
Most of the features discussed above are reproduced by the CFS. As seen from Fig. 8b, the model captures the significant relationship between EAMY and the geopotential height in the tropical/subtropical belts, especially over the western Pacific, southern Asia, and North Africa. It also displays positive regression over the Ural Mountains and negative regression over the subtropical westerly jet stream. However, compared to observations, these features are exaggerated. On the contrary, the model underestimates the relationship at higher latitudes, related to the Okhotsk Sea blocking. The overestimation problem at low and middle latitudes may be partially caused by the overly strong ENSO signals in Asian climate produced by the model (Yang et al. 2008b; Liang et al. 2009). Indeed, in the AMIP experiments in which the model is forced by observed SST, the areas of significant positive and negative regression in the lower and middle latitudes decreases remarkably (Fig. 8c). In addition, the southwesterly monsoon flow over Southeast Asia that provides moisture to the EAMY region shifts eastward. The anomalous convergence over the EAMY region in AMIP is also weaker than in observations.
Figure 9 illustrates the anomalies of EAMY, WPSH, EASM, and Niño-3 SST in observations, the CFS, and AMIP. For EAMY, the observed and CFS curves exhibit similar interannual variations, especially in the strong EAMY years (1983, 1993, 1998, and 2003) and weak EAMY years (1981, 1994, and 2004). The correlation of EAMY between observation and CFS is 0.54, exceeding the 99% confidence level. On the contrary, the correlation of EAMY between observation and AMIP is only 0.29. The WPSH and the EASM also show consistent variations between reanalysis and the CFS simulation, with correlation coefficients of 0.45 and 0.47 (exceeding the 95% confidence level), respectively. Furthermore, the CFS Niño-3 SST highly resembles the observed, with a correlation coefficient of 0.91. These features indicate that the CFS is quite skillful in simulating the EAMY and its associated large-scale processes. However, the skill of AMIP simulation is lower than that of the CFS. This feature and the well-captured interactive oceanic–atmospheric processes associated with the Asian monsoon in the CFS (Yang et al. 2008b) suggest the importance of air–sea interaction in EAMY simulation and prediction.
5. Predictions of EAMY
As shown in section 4, the CFS LM1 simulation of EAMY has a significant correlation with observation (R = 0.54, also see Fig. 9a), meaning that the forecast issued in May has a high skill for June–July EAMY prediction. As expected, this skill of prediction decreases with lead time. For example, the correlation of EAMY between observation and the CFS is, respectively, 0.36 for LM2, 0.18 for LM3, and −0.09 for LM4 (see Fig. 10). In the following discussion, we further substantiate the feature shown in Fig. 10 by examining the atmospheric features in the CFS at different lead times.
Since the atmospheric circulation patterns associated with WPSH and EASM are also reasonably simulated by the CFS, we further analyze these patterns and SST features for the various leads. Figure 11a shows the regressions of CFS LM1 500-hPa geopotential height and 850-hPa winds against the observed EAMY. The figure shows several features similar to those observed from Fig. 8a (for observations) and Fig. 8b (for CFS LM1). For example, the observed EAMY is associated with southwesterly flow from the South China Sea, southeastern China, and the western Pacific in the CFS, which provides water vapor supply. Relatively lower height appears over the EAMY region and its northeast, with relatively higher values to the south. The features shown above for LM1 are apparently weaker for the regressions of CFS LM2 variables against the observed EAMY (Fig. 11b). Not surprisingly, no meaningful features can be found for CFS LM3 and LM4 (figures not shown).
The features discussed above can be shown more clearly in Fig. 12a, which shows the correlation between the observed EAMY and the predicted WPSH from LM1 to LM4. Obviously, the correlation is larger in the shorter leads. Similar features can also be seen in the correlation between the observed EAMY and the CFS EASM (Fig. 12b). The correlation is −0.16, −0.30, −0.28, and −0.46 for the initial condition in February, March, April, and May, respectively. That is, only the EASM index in LM1 provides useful information for EAMY prediction. Since the EAMY shows the strongest relationship with the Niño-3 SST of February–April in observations (Table 1), we also analyze the feature in CFS simulations from LM1 to LM4. It is found that the most significant correlation occurs in the one-month lead forecast (using the ICs on 19–23 May) and the weakest correlation occurs in the four-month lead forecast (Fig. 12c).
We further apply the MSN EOF analysis (see section 2) to depict the most predictable patterns of EAMY in the CFS, and compare it with the conventional EOF patterns. Figure 13a shows the first conventional EOF mode and the corresponding time series of observed precipitation for the period from 20 June to 10 July. Clearly, this leading mode, explaining 21% of the total variance, depicts an EAMY pattern with positive values from the Yangtze and Huaihe River valleys to South Korea and southern Japan. A similar pattern can also be found in the first EOF mode of the 15-member mean CFS LM1 precipitation (Fig. 13b), which explains 28% of the total variance. There is a similarity between the two time series, especially in some strong EAMY years (1983 and 1992) and weak EAMY years (1990 and 1994). However, apparent discrepancies occur since 1999 with a prevailing interannual variation of EAMY in observation but not in the CFS. It should also be pointed out that the first EOF mode of June and July precipitation in the AMIP experiments (not shown) exhibits features very different from those discussed above.
Figure 14 shows the most predictable patterns (the first MSN EOF mode) of precipitation and the principal components (PCs) for LM1 and LM2, which are not apparently sensitive to the size of analysis domain. The most striking features of Fig. 14a are the positive values (increase in precipitation with positive PC, and vice versa) over the regions from southern China to southern Japan, which include the main EAMY area. Negative values exist over northern China and Korea, meaning that the EAMY rain belt in the CFS is more southward than observed. In the model, the first mode explains 29% of the total variance, higher than its counterpart of observation. The time series of both observation and ensemble mean show consistent variations before 1997 and are different afterward. The correlation between the two time series is 0.48, exceeding the 95% confidence level.
The predictability of EAMY decreases dramatically when the lead time changes from one month to two months (Fig. 14b), in agreement with the result explored in Fig. 10. Although the most predictable pattern in LM2 is similar to that in LM1, the PCs of ensemble means (red line) varies inconsistently with those of the observations (black line). In the LM2 forecast, the variance of the first mode decreases to 24% of the total variance and the correlation coefficient between the time series of observation and CFS ensemble mean is only 0.14.
The above results indicate that the NCEP CFS is able to predict EAMY reasonably well with a lead time of one month. Moreover, the Niño-3 SST index, the WPSH, and the EASM predicted in the previous month also possess predictive potential for EAMY in the CFS. It should be pointed out that, in this section, we have only demonstrated the overall skill of the NCEP CFS in predicting the intensity of the June–July averaged EAMY rainband. To predict the more detailed features of the rainband such as the onset date, its location, and associated extreme rainfall, a higher resolution of the model is needed. As shown by Kawatani and Takahashi (2003), high resolution is an important factor for mei-yu or Baiu prediction. In this context, the next generation of the NCEP CFS with a higher resolution of T126 is expected to perform better than the current version in predicting the EAMY.
As an important component of the Asian summer monsoon system, the East Asian mei-yu exhibits remarkable variability on various time scales. The EAMY rain belt jumps northward to the Yangtze and Huaihe River valleys (in China), southern Japan, and Korea from mid-June to mid-July. The onset dates of Yangtze mei-yu and Japanese baiu are about 10 days earlier than the Huaihe mei-yu and Korean changma. While previous studies have shown some skill in short- to medium-range forecasts of EAMY by numerical (weather) prediction models, prediction of EAMY for longer time scales is a challenge to many climate models. Recent studies have indicated that the NCEP CFS simulates many major features (Yang et al. 2008a,b) and, in particular, the most dominant modes of the Asian summer monsoon reasonably well. These results are encouraging for analyzing the forecast skill of EAMY by the CFS. In this paper, we have investigated the climate features of EAMY and its prediction by the NCEP CFS. We have also depicted the most predictable patterns of EAMY and discussed the importance of ocean–atmosphere coupling for EAMY modeling.
Results indicate that the CFS is reasonably skillful in simulating the climatological features of EAMY and its associated atmospheric circulation patterns by a one-month lead. It captures the EAMY rainbands and precipitation amount over most regions in June–July, the main EAMY period, including the rainbands over the Yangtze River valley, the tropical western Pacific, and southern Japan in June and those over the Huaihe River valley and Korea in July. The CFS reproduces the northward jump of EAMY rainband from June to July. Compared to the simulations over China and Japan, the EAMY rainband over Korea is predicted too far north. The model also overestimates the precipitation over the southern–southeastern hills of the Tibetan Plateau. Results also indicate that the CFS captures the major influencing circulation systems during the EAMY period, including the WPSH, the EASM, the lower-level westerly jet stream, the Somalia cross-equatorial flow, and the blocking systems at higher latitudes. The northward jumps of the 500-hPa WPSH and the 850-hPa southwesterly are also simulated reasonably well. However, the EAMY associated circulation patterns in the CFS are overall weaker than observed.
The CFS captures the interannual variations of EAMY and associated large-scale patterns, especially the relationships of EAMY with WPSH, EASM, and Niño-3 SST in LM1. Owing to the overly strong signals of ENSO in the model, the relationships of EAMY with the atmospheric circulation systems are overestimated in the tropics and the subtropics, whereas they are underestimated at higher latitudes.
A MSN EOF analysis shows that the most predictable pattern EAMY of the CFS in LM1 is similar to the leading mode of observations, suggesting again that the EAMY is predictable by one month in advance. Besides, the simulated Niño-3 SST index, WPSH, and EASM of one-month lead also provide antecedent information for the EAMY forecast.
Finally, we have compared the EAMY and its associated large-scale features in the CFS LM1 and AMIP experiments. It is found that the skill of the AMIP experiments in simulating both EAMY and tropical Asian summer monsoon is lower than that of the CFS. This difference suggests the importance of air–sea coupling for EAMY and Asian monsoon forecasts. The other factor for this difference is initial conditions. In this study, the CFS results that are used to compare with AMIP features are from the output with atmospheric ICs of about 1-month lead. In addition, while monthly SST forcing is used in the AMIP experiments, the atmosphere and oceans are interactive and exchange time-averaged quantities once a day.
In short, the NCEP CFS of T62 resolution is able to simulate and predict many large-scale and low-frequency features of the East Asian mei-yu. However, it also faces difficulties in capturing the regional features of EAMY. It is interesting to analyze the variability and predictability of EAMY in the next version of the NCEP CFS, which has a higher resolution, among many other differences from the current version of the forecast system. With the higher resolution, which is important for studying the Asian monsoon, we may also be able to better understand the relationship between EAMY and the tropical Madden–Julian oscillation.
We thank editor James Renwick and four anonymous reviewers who have provided thorough reviews of this paper. We also thank Drs. Rongqing Han and Changzheng Liu of the CMA National Climate Center and Dr. Kingtse Mo of the NOAA Climate Prediction Center for a number of helpful discussions. This work was jointly supported by the Ministry of Science and Technology of China (2009DFA23010) and the NOAA-CMA bilateral program. Hu was partially supported by the NOAA CVP Program (NA07OAR4310310), NSF ATM-0830068, NOAA NA09OAR4310058, and NASA NNX09AN50G.
Corresponding author address: Dr. Song Yang, NOAA Climate Prediction Center, 5200 Auth Road, Room 605, Camp Springs, MD 20746. Email: firstname.lastname@example.org