Abstract

The prediction of mei-yu onset date (MOD) in the middle and lower reaches of the Yangtze River valley (MLYRV) is an important and challenging task for those making seasonal climate predictions in China. In this paper, the atmospheric and oceanic conditions in the preceding winter and spring related to MOD are analyzed. It is found that the MOD is associated with the intensity of the Ural high and the East Asian trough in high latitudes, with the intensity of the upper-level westerly jet in middle latitudes, and with the contrast of land–sea temperature and pressure in the preceding winter and spring, which are proxies for the intensity of the East Asian winter monsoon (EAWM). It is suggested that the intensity of the EAWM is the most crucial factor affecting the MOD. Years with an early MOD usually correspond to strong EAWMs in the preceding winter, and vice versa. The EAWM can affect the MOD by influencing the East Asian summer monsoon (EASM) through tropical ocean–atmosphere and tropical–extratropical interactions. Based on the above analysis, a physics-based statistical forecast model is established using multivariable linear regression techniques. The hindcast of MOD during the 13 yr from 1998 to 2010 is carried out to evaluate the performance of this forecast model. The MOD can be predicted successfully in 8 out of the 13 yr. The forecast model predicts the MOD in the years with strong mei-yu intensity more accurately than in those with weak mei-yu intensity, especially for cases of extreme flooding. This is useful in the prevention of flooding disasters.

1. Introduction

The mei-yu pattern is a unique phenomenon in the seasonal transition process of the general circulation in East Asia and is an important component of the monsoon system. The mei-yu rainband extends eastward from the middle and lower reaches of the Yangtze River valley (MLYRV) in China to the Korean Peninsula and Japan, stretching for thousands of kilometers (e.g., Tao and Chen 1987; Ninomiya and Murakami 1987; Ding 1992, 2004). In the MLYRV, mei-yu usually begins in mid-June, and ends in mid-July (Ding et al. 2007; Sampe and Xie 2010). During this period, there is a long duration of precipitation with a great amount of rainfall, which can easily lead to flooding disasters. Thus, staff at meteorological offices in China have tried to predict the amount of mei-yu rainfall, mei-yu onset/ending date, and mei-yu season length and intensity, among other factors, before the onset of the mei-yu each year. Among these variables, people are particularly concerned with the mei-yu onset date (MOD). Accurate prediction of MOD will be beneficial for the prevention and mitigation of flood disasters. Thus, the prediction of MOD before the mei-yu season is one of the main foci of the meteorological offices and local governments in China.

It is well known that the mei-yu onset in the MLYRV is the result of a rapid and significant adjustment of the atmospheric circulation in East Asia. The mei-yu onset not only indicates the beginning of the rainy season in the MLYRV but also is accompanied by the arrival of the leading edge of the East Asian summer monsoon (EASM) in the region, the jump of the western Pacific subtropical high (WPSH) axis to the north of 20°N, the stabilized WPSH axis within 20°–25°N, and the northward migration of the upper-level westerly jet over Eurasia to the north of the Tibetan Plateau (Tao and Chen, 1987; Li and Yanai, 1996; Ding 1992; Ding et al. 2007). Therefore, the atmospheric circulation patterns in the years with early MODs exhibit different characteristics from those in the years with late MODs (Tomita et al. 2011).

To predict the MOD in the MLYRV, many researchers have studied the precursory signals affecting the MOD. Xu et al. (2001) found that the MOD of the MLYRV was mainly associated with the North Atlantic Oscillation (NAO) in the preceding winter, and with the sea surface temperature anomaly (SSTA) in the North Atlantic in winter and spring. Chen and Zhao (2000) considered that the El Niño–Southern Oscillation (ENSO) pattern has evident impacts on MOD in the MLYRV; that is, the MOD is late in most El Niño years, while it is early in most La Niña years. Wang et al. (2009) showed that the ENSO events that occurred in the central Pacific (CP-ENSO) during February of the preceding winter and spring had significant effects on the interannual variation of MOD. During the CP-ENSO warm (cold) phase, the MOD is likely to be late (early). Tomita et al. (2011) showed that the MOD in Japan had no significant correlation with the typical large-scale interannual variation (e.g., ENSO, the tropospheric biennial oscillation of the Asian monsoon, and the NAO).

Up to now, the MOD in seasonal climate prediction has usually been based on the qualitative analysis of the factors in the preceding winter and spring, and the prediction is usually given as “early,” “normal,” or “late.” Such predictions of MOD are subjective and of poor skill. There have been many studies on the prediction of the mei-yu rainfall amount in the MLYRV (Fan et al. 2008; Ying et al. 2013). However, the predictors of MOD in the MLYRV and the internal physical mechanisms have not been understood comprehensively. Few models can predict the MOD in the MLYRV quantitatively and successfully. Therefore, a physics-based statistical forecast model is established to quantitatively predict the MOD in the MLYRV. In addition, because the MOD has a great interannual variability with periodicity from quasi-biennial to 4 yr (Tomita et al. 2011), the method proposed by Wang et al. (2000), Fan et al. (2008), and Fan and Wang (2009, 2010a,b) is employed, which forecasts the year-to-year increment of a variable instead of the variable itself. The physical base for this method lies in the fact that the tropospheric biennial oscillation (TBO) has been revealed to exist in the East Asian monsoon, ENSO, and summer rainfall in eastern China (Meehl 1997; Chang and Li 2000; Li et al. 2001).

2. Data and methods

The National Centers for Environmental Prediction–National Center for Atmospheric Research (NCEP–NCAR) monthly reanalysis data with 2.5° × 2.5° resolution are employed (Kalnay et al. 1996). The sea surface temperature (SST) data used in this paper are from the Kaplan SST, version 2 (V2), data provided by the National Oceanic and Atmospheric Administration/Office of Oceanic and Atmospheric Research/Earth System Research Laboratory/Physical Sciences Division (NOAA/OAR/ESRL/PSD) in Boulder, Colorado (http://www.esrl.noaa.gov/psd/), with a resolution of 5° × 5°. The mei-yu data in the MLYRV are provided by the National Climate Center in China, including the observed rainfall amount, onset/ending date, and intensity of mei-yu, covering the period from 1885 to the present. The MOD in the MLYRV is defined by the daily rainfall data from five stations (Shanghai, Nanjing, Wuhu, Jiujiang, and Hankou) in the MLYRV and by the daily averaged latitude of the WPSH axis at 500 hPa. If the total rainfall amount of the five stations is more than or equal to 10 mm, and the rainfall amount at two or more stations is equal to or more than 0.1 mm, this day is defined as a rainy day. The date defined as the MOD should satisfy the following criteria: (i) the rainy days in any continuous period during the first 10 days after the mei-yu onset should be equal to or more than 5 days; (ii) the rainy days in any 10 days during the mei-yu period should be equal to or more than 4 days and the nonrainy days should be equal to or less than 4 days; (iii) the mei-yu period may be composed of one or more than one rainy period, where during each rainy period the rainy days should be equal to or more than 6 days and the total daily rainfall amount of the five stations should be more than 25 mm; and (iv) the ridge of the WPSH should remain between 20° and 25°N during the mei-yu period and the number of days in which the WPSH ridge is out of this range should be less than 5. In this study, summer refers to June–August (JJA) and spring refers to March–May (MAM).

The steps followed to predict the MOD are as follow. First, the year-to-year increments (YIs) of the MOD, SST, and atmospheric variables are calculated. According to Wang et al. (2000), Fan et al. (2008), and Fan and Wang (2009, 2010a,b), the YI is defined as the difference of a variable between the current year and the preceding year. Second, the correlation coefficients between the YI of MOD (MODYI) and the YIs of global atmospheric variables and SST in the preceding winter and spring during 1966–97 are calculated to find the physical predictors that have close associations with the MOD. Last, a physics-based statistical forecast model for MOD is established using the multiple linear regression method. The predictors are the YIs of variables closely related to MODYI based on the second step. The output of this model is MODYI. The MOD itself can be calculated from the MODYI. We chose the period from 1966 to 1997 to carry out the correlation analysis and to construct the prediction model for two main reasons. First, the period during which the climate is similar to that in the latest 10–15 yr should be chosen. Choosing the historical data in these 32 yr to train the statistical model is sufficient enough. Second, we should use the hindcast of MOD in the latest 10–15 yr to evaluate the performance of the prediction model. Thus, the dataset from 1966 to 1997 is used to train the statistical model and the MOD in 1998–2012 is hindcast to evaluate the performance of the prediction model.

3. The predictive signals of MOD in the preceding winter and spring

Figure 1 shows the correlation maps of the YIs of SST in the preceding winter and spring with reference to the MODYI during 1966–97. Figure 1a indicates that the significant negative correlations are located over the subtropical eastern and western Pacific. However, the correlation coefficient over the eastern equatorial Pacific is weakly positive, which does not pass the significance test at the 95% confidence interval. This indicates that the ENSO signal in the preceding winter is not strong enough to affect the MOD. This result is consistent with the conclusion in Tomita et al. (2011), whereas it is opposite to that of Chen and Zhao (2000). Note that the correlation between MODYI and YI of SST in spring over the eastern equatorial Pacific is significantly positive (Fig. 1b), indicating that the spring ENSO signal contributes to the prediction of MOD. If the spring SST over the eastern equatorial Pacific in the current year is warmer than that in the previous year, the MOD in the current year is later than that in the previous year. This result is consistent with Wang et al. (2009). It is well known that the ENSO cycle has a strong phase lock to the seasons, the SSTA over the eastern equatorial Pacific usually reaches its maximum in winter (Rasmusson and Carpenter 1982; Tziperman et al. 1997; Liu 2002). The ENSO signal in winter is an important precursory signal associated with the summer rainfall amount in the MLYRV, and is often used as a predictor in statistical forecast models to predict the summer rainfall amount in the MLYRV (Tong et al. 2006; Fan et al. 2008). However, the correlation of the ENSO signal in winter and MOD is not so significant that it may be incorrect to predict MOD with the ENSO signal in winter.

Fig. 1.

Correlation maps of the YI of SST during the (a) preceding winter and (b) spring with reference to MODYI during 1966–97. The dark (light) shading represents statistically significant positive (negative) correlation at the 95% confidence level.

Fig. 1.

Correlation maps of the YI of SST during the (a) preceding winter and (b) spring with reference to MODYI during 1966–97. The dark (light) shading represents statistically significant positive (negative) correlation at the 95% confidence level.

Comparing Fig. 1a with Fig. 1b, it can be seen that the distribution patterns of the correlations of MODYI and YI of SST in winter are somewhat similar to those in spring, but the correlation values become larger than those in winter. Significantly negative correlations in spring are found over the western Pacific, which exhibits a horseshoe pattern. The significantly negative correlation over the North Atlantic is very similar to the result of Xu et al. (2001), indicating that the SSTA over the North Atlantic has an impact on the MOD. If the spring SST over the North Atlantic in the current year is warmer than that in the previous year, the MOD in the current year is earlier than that in the previous year, and vice versa.

Figure 2 shows the correlation maps of YI of geopotential height at 500 hPa H500, surface temperature TS, zonal wind at 200 hPa U200, sea level pressure SLP, and meridional wind at 850 hPa V850 with respect to the MODYI. The distribution of the correlation coefficients between YI of H500 and MODYI in the high to middle latitudes of the Northern Hemisphere exhibits a systematic wave pattern (Fig. 2a), characterized by three negative areas over the Ural Mountains, the eastern North Pacific, and the North Atlantic, as well as by two positive areas over eastern Asia and North America. This wave pattern indicates that when the East Asian trough and its upstream Ural high ridge develop, the strong cold air can easily invade China, leading to a colder winter in China. In this case, the MOD would be earlier, and vice versa.

Fig. 2.

As in Fig. 1, but for (a) H500, (b) TS, (c) U200, (d) SLP, and (e) V850 in winter. The boxes denote the areas from which we chose predictors of MOD. The dark (light) shading represents statistically significant positive (negative) correlations at the 95% confidence level.

Fig. 2.

As in Fig. 1, but for (a) H500, (b) TS, (c) U200, (d) SLP, and (e) V850 in winter. The boxes denote the areas from which we chose predictors of MOD. The dark (light) shading represents statistically significant positive (negative) correlations at the 95% confidence level.

Figure 2b shows that the MODYI is positively correlated with the YI of surface temperature in the previous winter over Eurasia and negatively correlated with the over the warm pool in the western Pacific. Note that the distribution of the correlation coefficients between MODYI and winter over the ocean is similar to that between MODYI and SST in winter (Fig. 1a). For example, there is an obvious negative correlation area with a horseshoe pattern. This is mainly due to the heating of the ocean to the surface atmosphere and the effective turbulence mixing in the atmospheric boundary layer. The atmospheric thermal conditions in its boundary layer can represent the thermal conditions of the sea surface. In winter, the Eurasian continent is relatively much colder than the Pacific Ocean. As such, the contrast of the MODYI– correlation coefficients between the Eurasian Continent and western Pacific represents the impact of the land–ocean thermal contrast on the MOD. When the surface temperature difference between land and ocean is large (small), the MOD is early (late).

From Fig. 2c, it can be seen that the significantly negative correlation between MODYI and the YI of U200 prevails around 30°N in East Asia, suggesting that the MOD is closely related to the upper-level westerly jet in the Northern Hemisphere. When the upper-level westerly jet is strong (weak), the MOD is early (late). Figure 2d shows that the MOD is negatively correlated with the YI of SLP over Eurasia, whereas it is positively correlated with the YI of SLP over the Indian Ocean and the northwestern Pacific. In boreal winter, due to the ocean–land thermal contrast, Eurasia is controlled by a massive cold high, but the ocean is under a warm low. Therefore, the MOD is related to the ocean–land pressure contrast. When the SLP difference between land and ocean is large (small), the MOD is early (late). It can be found from Fig. 2e that the MOD is positively correlated with the YI of V850 in East Asia. Comparing Fig. 2e with the climatological-mean V850 (not shown), we find that the positive correlation corresponds to the northerly winds in winter, indicating that the MOD is early (late) when the northerly wind in the lower troposphere is strong (weak). Through a comprehensive analysis of Figs. 2a–e, it can be found that in winter the Ural high and East Asian trough in high latitudes, the land–ocean temperature contrast, the upper-level westerly jet, the land–ocean pressure contrast, and the northerly wind in the lower troposphere all reflect the intensity of the East Asian winter monsoon (EAWM). If the EAWM in the current year is strong (weak), the MOD is early (late).

From the correlation maps of the YI of atmospheric variables in spring with respect to MODYI (not shown), we find that the distributions of correlation coefficients are similar to those in winter. However, due to the recession of the winter monsoon and the adjustment of the atmospheric circulation, both the location and scope of the significantly correlated region and the magnitude of the correlation coefficient are changed. The relationship between the MOD and East Asian trough becomes less significant. Due to the enhancement of the subtropical high from winter to spring, the relationship between MOD and H500 over the eastern Pacific becomes significant. The distribution of the correlation between MODYI and remains consistent with that in winter, indicating that the MOD is also related to the air–sea temperature contrast in spring. However, due to the warming of Eurasia and the retreat of the EAWM, the scope of the significantly positive region over Eurasia is reduced, and its location shifts northward in spring. The patterns of correlation coefficients between MODYI and YI of U200, SLP, and V850 are very similar to those in winter, suggesting that the MOD is still associated with the upper-level westerly jet, land–sea pressure difference, and lower-level northerly wind in spring. However, due to the regression of the EAWM, the upper westerly jet, the continental cold high, and the lower-level northerly wind are all weakened, with an obviously reduced scope for the significantly correlated region.

How does the EAWM affect the MOD? We propose a physical mechanism through a composite analysis. The years with a standard deviation of MOD greater than 1 are defined as late MOD years, whereas the years with a standard deviation of MOD less than −1 are defined as early MOD years. There are 7 yr with early MOD and 7 yr with late MOD after 1979 (Table 1). The climatological-mean MOD (average of 1971–2000) is 19 June. From Table 1, it can be found that the definitions of late and early MOD years are reasonable. The MOD in the years with early (late) MOD is ≥9 days earlier (later) than the climatological-mean MOD. Figure 3 shows the composite difference of SLP between the early and late MOD years. From Figs. 3a–c, we find that the SLP over East Asia in the early MOD years is higher than that in the late MOD years, indicating a stronger Siberian cold high in boreal winter in the early MOD years than that in the late MOD years. Besides, the winter and early spring TS (January–March) in the early MOD years is obviously lower than that in the late MOD years over East Asia (not shown). Thus, it can be concluded that the EAWM is stronger than normal in the early MOD years than it is in the late MOD years. This can also be demonstrated by the composite difference of atmospheric circulation at 500 hPa (Fig. 4). The H500 in high latitudes exhibits an alternative occurrence of positive–negative–positive deviations in the Northern Hemisphere in boreal winter and early spring (January–March), which shows a wave train pattern (Figs. 4a–c). The locations of the positive difference in northwestern Eurasia and the negative difference in northeastern Eurasia coincide with the Ural high ridge and the East Asian trough, respectively, implying that the Ural high ridge and East Asian trough are strong and that meridional circulation is prevalent in high latitudes in the early MOD years, which can constantly transport cold air masses from Siberia to East Asia. Thus, the EAWM is stronger in the early MOD years than in the late MOD years. This result is consistent with that from the correlation analysis (see Fig. 2).

Table 1.

The years with early and late MOD.

The years with early and late MOD.
The years with early and late MOD.
Fig. 3.

Composite difference of SLP (contours; hPa) between early and late MOD years. The shading indicates areas where the differences are statistically significant at the 90% confidence level according to a Student’s t test.

Fig. 3.

Composite difference of SLP (contours; hPa) between early and late MOD years. The shading indicates areas where the differences are statistically significant at the 90% confidence level according to a Student’s t test.

Fig. 4.

Composite difference of H500 (contours; gpm) between early and late MOD years. The shading indicates areas where the differences are statistically significant at the 90% confidence level according to a Student’s t test.

Fig. 4.

Composite difference of H500 (contours; gpm) between early and late MOD years. The shading indicates areas where the differences are statistically significant at the 90% confidence level according to a Student’s t test.

The EAWM can affect the MOD through its impact on the EASM. Many studies (Chen et al. 2000, 2013; Huang et al. 2004) have pointed out that the EAWM interacts with the EASM and that the ocean is an important linkage between the two. Li (1998) indicated that frequent activity during strong winter monsoons can trigger ENSO-related SSTA, which can last from the winter to the following summer and have impacts on the summer monsoon. From the composite difference of the SST between the early and late MOD years (Fig. 5), one can see that the negative value prevails over the central-eastern equatorial Pacific, exhibiting a La Niña pattern, which is consistent with the studies of Chen et al. (2000, 2013) and Wang et al. (2009). Note that the negative SST differences over the central-eastern Pacific are not the largest in January or February. They increase and expand gradually from January to June and are larger in spring than in winter. This is why the ENSO signal associated with the MOD is stronger in spring than in winter.

Fig. 5.

Composite difference of winds at 850 hPa (arrows) and SST (contours; °C) between early and late MOD years. The shading indicates areas where the SST differences are statistically significant at the 90% confidence level according to a Student’s t test. The thick arrows indicate wind differences that are statistically significant at the 90% confidence level according to a Student’s t test.

Fig. 5.

Composite difference of winds at 850 hPa (arrows) and SST (contours; °C) between early and late MOD years. The shading indicates areas where the SST differences are statistically significant at the 90% confidence level according to a Student’s t test. The thick arrows indicate wind differences that are statistically significant at the 90% confidence level according to a Student’s t test.

According to the theory of Matsuno (1966) and Gill (1980), the Rossby wave response to negative SSTA is excited, namely, a pair of anomalous anticyclone and associated equatorial easterly anomalies, in the early MOD years. In addition, the westerly wind anomalies can be found over the equatorial Indian Ocean. The easterly wind anomalies over the equatorial Pacific and the westerly wind anomalies over the equatorial Indian Ocean imply an enhanced Walker circulation in the early MOD years. Ji et al. (1997) and Zeng et al. (2011) demonstrated that a strong EAWM generally corresponds to an enhanced Walker circulation. The easterly and westerly anomalies converge over the western Pacific near Indonesia and the Philippines, producing intensified ascending motion and convection there. Note that the easterly anomaly over the equatorial Pacific is the strongest in February among the 6 months, as shown in Fig. 5b, indicating that the Walker circulation is the strongest in February, too.

After February, though, the EAWM decays and the easterly anomaly over the equatorial Pacific weakens, while the negative SSTA persists and its scope is enlarged due to the ocean memory (Figs. 5c–f). Therefore, the anomalous easterly wind in the lower troposphere also is maintained from the winter to the following spring and early summer, which converges around Indonesia and the Philippines with the anomalous westerly wind from the equatorial Indian Ocean, making stronger than normal ascending motion and convection there in the early MOD years (Figs. 5c–f). In addition, the negative differences of outgoing longwave radiation (OLR) are more obvious around Indonesia and the Philippines in the early MOD years than in the late MOD years (Fig. 6). This enhanced ascending movement and convection can persist from winter to early summer in the early MOD years, leading to a stronger than normal ascending branch of the Walker circulation over the western Pacific, which intensifies the descending motion in midlatitudes through the Hadley cell over the western Pacific. In this case, the subtropical high over the western North Pacific becomes stronger than normal and shifts westward and northward in the early MOD years, which can be manifested by the positive differences of SLP (Fig. 3f), H500 (Fig. 4f), and OLR (Fig. 6f) over East Asia and the western North Pacific. As a result, the MLYRV is controlled by an anomalous anticyclone in the early MOD years, which can be seen from Fig. 5f. The southwesterly wind to the west of the anticyclone can induce earlier summer monsoon arrival in the MLYRV than normal, leading to the early MOD. Because the weak winter monsoon tends to coincide with the El Niño pattern of SSTA (Chen and and Zhao 2000; Chen et al. 2013; Wang et al. 2009), the late MOD years correspond to the reverse condition of the early MOD years.

Fig. 6.

Composite difference of outgoing longwave radiation (W m−2) between early and late MOD years. The shading indicates areas where the SST differences are statistically significant at the 90% confidence level according to a Student’s t test.

Fig. 6.

Composite difference of outgoing longwave radiation (W m−2) between early and late MOD years. The shading indicates areas where the SST differences are statistically significant at the 90% confidence level according to a Student’s t test.

Based on the above analysis, we conclude that the EAWM impacts the MOD through two related interactions. One is the tropical ocean–atmosphere interaction, which changes the Walker circulation. The other is the tropical–extratropical interaction, through which the Walker circulation anomaly can affect the Hadley cell over the western Pacific, and then change the location and intensity of the WPSH. The MOD is the result of these two interactions.

4. The physics-based statistical forecast model

The predictors used in this study are chosen based on the correlation analysis in section 3. The area-averaged variables with correlation coefficients significantly larger than 0.4 within the black boxes in Fig. 2 are selected as the predictors. Due to the exchange of heat between the surface atmosphere and ocean, the surface temperature distribution is consistent with SST, which can be confirmed by Figs. 1a and 2b. Therefore, we chose TS as one of the predictors, instead of SST, which leaves a total of 10 predictors. Because the number of predictors is still too large and the predictors may be related to each other, it is necessary to eliminate some of them to avoid redundancy. To establish the best prediction model, a method similar to the stepwise regression is used to choose the predictors. The historical data from 1966 to 1997 are used to establish the multiple linear regression equation, and a hindcast of the MOD during 1998–2010 is performed. The predictors are introduced into the regression equation step by step. If the introduction of one predictor can improve the prediction skill, it will be kept; otherwise, it will be deleted.

Through a large number of sensitive experiments, five predictors are ultimately chosen, including 1) averaged YI of H500 in winter around the Ural Mountains in western Eurasia (35°–80°N, 10°–80°E) with correlation coefficients less than −0.4 (Fig. 2a), which is denoted by H500_WUAI; 2) averaged YI of H500 in winter over eastern Eurasia (25°–55°N, 80°–170°E) with correlation coefficients larger than 0.4 (Fig. 2a), which is denoted by H500_EUAI; 3) averaged YI of TS in winter over Eurasia (25°–55°N, 60°–160°E) with correlation coefficients larger than 0.4 (Fig. 2b), which is denoted as TS_UAI; 4) averaged YI of TS in winter over the western Pacific (25°S–55°N, 100°E–140°W) with correlation coefficients less than −0.4 (Fig. 2b), which is denoted by TS_WPI; and 5) averaged YI of U200 in winter around 30°N (10°–50°N, 60°E–180°) with correlation coefficients less than −0.4 (Fig. 2c), which is denoted by WJI. Table 2 lists all the predictors and the correlation coefficients between them and MODYI. Using the above five predictors, we have the following linear regression equation:

 
formula
Table 2.

Correlation coefficients between MODYI and the predictors.

Correlation coefficients between MODYI and the predictors.
Correlation coefficients between MODYI and the predictors.

Figure 7 compares the simulated and observed MODYI and MOD during 1966–97. It can be seen from Fig. 7a that the MODYI shows an obvious interannual variation, which is captured by the forecast model. The correlation coefficient of the simulation and observations is 0.77. The original MOD can be calculated using MODYI, and it can also be reproduced well by the forecast model (Fig. 7b). The correlation coefficient of the observed and simulated MODs is 0.64, which is slightly lower than that of the MODYI.

Fig. 7.

Comparison of simulated (dashed line) and observed (solid line) (a) MODYI and (b) MOD during 1966–97. The horizontal solid line is the average MOD date, which is 17 Jun.

Fig. 7.

Comparison of simulated (dashed line) and observed (solid line) (a) MODYI and (b) MOD during 1966–97. The horizontal solid line is the average MOD date, which is 17 Jun.

To validate the performance of this physics-based statistical forecast model, a hindcast for 1998–2010 is carried out. Table 3 shows the observed and hindcast MODs. The hindcast MOD with a bias of less than 5 days is considered to be correct. Based on this criterion, 8 yr are predicted correctly: 1998, 1999, 2002, 2003, 2004, 2006, 2008, and 2010. The success rate is 61.5%, which shows a good level of skill for the forecast model in predicting MOD, especially for some years with extreme flooding, such as 1998, 1999, and 2003. Note that the years with large bias have a common problem, namely, the mei-yu intensity (mei-yu intensity = total amount of mei-yu rainfall per day during a mei-yu period) in these years is below the long-term-averaged value (63 mm day−1) and is even zero in some years (e.g., 2000, 2002, and 2009). The possible reason is as follows. In addition to the EAWM anomaly in the preceding winter, the MOD can also be affected by some other stochastic weather systems that occur in summer. For example, the movement and landfall of a typhoon can influence the WPSH, leading to the change of the MOD. For the years with lower than normal mei-yu intensity, the forcing of the atmosphere and ocean anomaly is weak, causing weaker than normal atmospheric systems for mei-yu. As a result, the MOD becomes more sensitive to the disturbance of some uncertain factors, and the predictability of MOD in the years with weak mei-yu intensity is lower than that with strong mei-yu intensity.

Table 3.

Observed and hindcast MOD and MODYI for 1998–2000.

Observed and hindcast MOD and MODYI for 1998–2000.
Observed and hindcast MOD and MODYI for 1998–2000.

5. Conclusions and discussion

To predict the onset date of mei-yu in the MLYRV quantitatively, a physics-based statistical forecast model is established in this study. The predictand of this model is the year-to-year increment of the MOD, and the predictors associated with the MOD are analyzed as the input of the forecast model.

By analyzing the atmosphere and ocean conditions in the preceding winter and spring before the mei-yu onset, we found that the MOD has strong predictability. The predictive signals appear in the preceding winter, and persist through spring until summer. These predictive signals contain the intensity of the Ural high and the East Asian trough in high latitudes, the intensity of upper-level westerly jet in middle latitudes, the contrast of land–sea temperature and pressure in the preceding winter and spring, etc. All of these predictive signals are related to the intensity of the EAWM, namely, the intensity of the EAWM is the crucial signal affecting the MOD. The MOD usually tends to be early when the EAWM is strong, whereas it tends to be late when the EAWM is weak.

The EAWM can influence the MOD by affecting the EASM through the tropical ocean–atmosphere interaction and the tropical–extratropical interaction. The ocean is a critical linkage between the EAWM and EASM. In the years with early MOD, the EAWM is strong, corresponding to a strong Walker circulation and a La Niña pattern of SSTA. The negative SSTA over the central-eastern equatorial Pacific can be maintained from winter to summer due to the ocean memory, leading to a Matsuno–Gill response, namely, the easterly anomalies generated over the central-eastern equatorial Pacific in the surface atmosphere. In addition, there are westerly anomalies over the Indian Ocean. The easterly and westerly anomalies converge around Indonesia and the Philippines over the western Pacific, leading to intensified ascending motion and convection, which further intensify the Hadley cell over the western Pacific. As a result, the descending branch of the Hadley cell is also enhanced, leading to a stronger than normal WPSH. The WPSH shifts northward and westward, and the MLYRV is controlled by an anticyclone anomaly. The southeasterly anomalies to the west of the anticyclone anomaly tend to enable the southwesterly monsoon arrival at the MLYRV earlier, resulting in an early MOD. The condition in the years with late MOD is the opposite.

Through many sensitivity experiments, five predictors are chosen to establish the linear regression equation. After choosing predictors, a hindcast of MOD in the 13 yr from 1998 to 2010 is performed using the statistical forecast model. The prediction of MOD is successful in 8 out of the 13 yr, which gives a success rate of 61.5%. The forecast model can predict the MOD in the years with strong mei-yu intensity more accurately, especially for extreme flooding cases. This characteristic of the forecast model is useful for the seasonal climate prediction in China, because we are more concerned about the MOD in flood years. This model can advise people and local governments on how best to prepare for a flooding event. In operational prediction, it is feasible to predict the rainfall amount and intensity of mei-yu first to determine the credibility of the predicted MOD from this forecast model. If the predicted mei-yu intensity is strong and there is a great probability of the occurrence of flooding during the mei-yu period in the MLYRV, the predicted MOD from this forecast model should be used for disaster prevention.

Acknowledgments

The authors thank the three anonymous reviewers for their constructive suggestions and comments, which led to a significant improvement in the manuscript. This work is sponsored by the National Natural Science Foundation of China (Grants 41205075 and 41205044); the Natural Science Foundation of Jiangsu Province, China (BK2012062); and the National Public Welfare Research Found of China (GYHY201106017). This work was also supported by the Jiangsu Collaborative Innovation Center for Climate Change.

REFERENCES

REFERENCES
Chang
,
C. P.
, and
T.
Li
,
2000
:
A theory for the tropical tropospheric biennial oscillation
.
J. Atmos. Sci.
,
57
,
2209
2224
, doi:.
Chen
,
W.
,
H.-F.
Graf
, and
R.
Huang
,
2000
:
The interannual variability of East Asia monsoon and its relation to the summer monsoon
.
Adv. Atmos. Sci.
,
17
,
48
60
, doi:.
Chen
,
W.
,
J.
Feng
, and
R.
Wu
,
2013
:
Roles of ENSO and PDO in the link of the East Asian winter monsoon to the following summer monsoon
.
J. Climate
,
26
,
622
635
, doi:.
Chen
,
X. F.
, and
Z. G.
Zhao
,
2000
: Analysis and prediction of the rainy season. The Study of the Prediction of Rainfall during Flooding Season in China and Its Application (in Chinese), X. Chen, Ed., China Meteorological Press, 65–99.
Ding
,
Y. H.
,
1992
:
Summer monsoon rainfalls in China
.
J. Meteor. Soc. Japan
,
70
,
373
396
.
Ding
,
Y. H.
,
2004
: Seasonal march of the East-Asian summer monsoon. East Asian Monsoon, C.-P. Chang, Ed., World Scientific, 3–53.
Ding
,
Y. H.
,
J. J.
Liu
, and
Y.
Sun
,
2007
:
A study of the synoptic–climatology of the mei-yu system in East Asia (in Chinese)
.
Chin. J. Atmos. Sci.
,
31
,
1082
1101
.
Fan
,
K.
, and
H. J.
Wang
,
2009
:
A new approach to forecasting typhoon frequency over the western North Pacific
.
Wea. Forecasting
,
24
,
974
973
, doi:.
Fan
,
K.
, and
H. J.
Wang
,
2010a
:
A prediction model for Atlantic named storm frequency using a year-by-year increment approach
.
Wea. Forecasting
,
25
,
1842
1851
, doi:.
Fan
,
K.
, and
H. J.
Wang
,
2010b
:
Seasonal prediction of summer temperature over northeast China using a year-to-year incremental approach
.
Acta Meteor. Sin.
,
24
,
269
237
.
Fan
,
K.
,
H. J.
Wang
, and
Y. J.
Choi
,
2008
:
A physically-based statistical forecast model for the middle-lower reaches of the Yangtze River valley summer rainfall
.
Chin. Sci. Bull.
,
53
,
602
609
, doi:.
Gill
,
A. E.
,
1980
:
Some simple solutions for heat-induced tropical circulation
.
Quart. J. Roy. Meteor. Soc.
,
106
,
447
462
, doi:.
Huang
,
R.
,
W.
Chen
,
B.
Yang
, and
R.
Zhang
,
2004
:
Recent advances in studies of the interaction between the East Asian winter and summer monsoons and ENSO cycle
.
Adv. Atmos. Sci.
,
21
,
407
424
, doi:.
Ji
,
L.
,
S.
Sun
,
K.
Arpe
, and
L.
Bengtsson
,
1997
:
Model study on the interannual variability of Asian winter monsoon and its influence
.
Adv. Atmos. Sci.
,
14
,
1
22
, doi:.
Kalnay
,
E.
, and Coauthors
,
1996
:
The NCEP/NCAR 40-Year Reanalysis Project
.
Bull. Amer. Meteor. Soc.
,
77
,
437
471
, doi:.
Li
,
C. Y.
,
1998
: A further study on the triggering effect of strong East Asian winter monsoon on El Niño. Research on Climate Variability and Impact in China, C. Li, Ed., China Meteorological Press, 207–218.
Li
,
C. Y.
, and
M.
Yanai
,
1996
:
The onset and interannual variability of the Asian summer monsoon in relation to land–sea thermal contrast
.
J. Climate
,
9
,
358
375
, doi:.
Li
,
C. Y.
,
S. Q.
Sun
, and
M. Q.
Mu
,
2001
:
Origin of the TBO-interaction between anomalous East-Asian winter monsoon and ENSO cycle
.
Adv. Atmos. Sci.
,
18
,
554
566
, doi:.
Liu
,
Z.
,
2002
:
A simple model study of ENSO suppression by external periodic forcing
.
J. Climate
,
15
,
1088
1098
, doi:.
Matsuno
,
T.
,
1966
:
Quasi-geostrophic motion in the equatorial area
.
J. Meteor. Soc. Japan
,
44
,
25
43
.
Meehl
,
G. A.
,
1997
:
The South Asian monsoon and the tropospheric biennial oscillation
.
J. Climate
,
10
,
1921
1943
, doi:.
Ninomiya
,
K.
, and
T.
Murakami
,
1987
: The early summer rainy season (baiu) over Japan. Monsoon Meteorology, P.-C. Chang and T. N. Krishnamurti, Eds., Oxford University Press, 93–121.
Rasmusson
,
E. M.
, and
T. H.
Carpenter
,
1982
:
Variations in tropical sea surface temperature and surface wind fields associated with the Southern Oscillation/El Niño
.
Mon. Wea. Rev.
,
110
,
354
384
, doi:.
Sampe
,
T.
, and
S. P.
Xie
,
2010
:
Large-scale dynamics of the mei-yu–baiu rainband: Environmental forcing by the westerly jet
.
J. Climate
,
23
,
113
134
, doi:.
Tao
,
S. Y.
, and
L.
Chen
,
1987
: A review of recent research on the East Asia summer monsoon over China. Monsoon Meteorology, Oxford Monographs on Geology and Geophysics, Vol. 7, Oxford University Press, 50–92.
Tomita
,
T.
,
T.
Yamaura
, and
T.
Hashimoto
,
2011
:
Interannual variability of the baiu season near Japan evaluated from the equivalent potential temperature
.
J. Meteor. Soc. Japan
,
89
,
517
537
, doi:.
Tong
,
J.
,
Q.
Zang
,
D.
Zhu
, and
Y.
Wu
,
2006
:
Yangtze floods and droughts (China) and teleconnections with ENSO activities, 1470–2003
.
Quat. Int.
,
144
,
29
37
, doi:.
Tziperman
,
E.
,
S. E.
Zebiak
, and
M. A.
Cane
,
1997
:
Mechanism of seasonal–ENSO interaction
.
J. Atmos. Sci.
,
54
,
61
71
, doi:.
Wang
,
H. J.
,
G. Q.
Zhou
, and
Y.
Zhao
,
2000
:
An effective method for correcting the seasonal–interannual prediction of summer climate anomaly
.
Adv. Atmos. Sci.
,
17
,
234
240
, doi:.
Wang
,
J.
,
J. H.
He
,
X. F.
Liu
, and
B. G.
Wu
,
2009
:
Interannual variability of the mei-yu onset over Yangtze–Huaihe River valley and analysis of its previous strong influence signal
.
Chin. Sci. Bull.
,
54
,
85
92
.
Xu
,
H. M.
,
J. H.
He
, and
M.
Dong
,
2001
:
Interannual variability of the mei-yu onset and its association with North Atlantic Oscillation and SSTA over North Atlantic
.
Acta Meteor. Sin.
,
59
,
694
706
.
Ying
,
K.
,
X.
Zheng
,
X. W.
Quan
, and
C. S.
Frederiksen
,
2013
:
Predictable signals of seasonal precipitation in the Yangtze–Huaihe River valley
.
Int. J. Climatol.
,
33
,
3002
3015
, doi:.
Zeng
,
G.
,
W. C.
Wang
,
Z.
Sun
, and
Z.
Li
,
2011
:
Atmospheric circulation cells associated with anomalous East Asian winter monsoon
.
Adv. Atmos. Sci.
,
28
,
913
926
, doi:.