Predictable signals of seasonal precipitation in the Yangtze–Huaihe River Valley

Authors


ABSTRACT

An effort is made to identify the ‘more predictable’ signals in seasonal precipitation in the Eastern Yangtze–Huaihe River Valley of China based on the observations recorded for the period of 1951–2004 at a network of 23 stations in this region. A recently developed methodology for decomposing the interannual variance of seasonal mean climate fields is applied to precipitation time series at the stations. This allows the total interannual variance to be separated into the variance of a slow component, or predictable signal, and the variance of a more noisier component of rainfall associated with intraseasonal variability. The potential predictability (signal-to-total ratio) is generally moderate in this region and is lower during winter from January to March (JFM) and higher in summer from May to July (MJJ) and autumn from November to January (NDJ). Empirical orthogonal function (EOF) analysis is then applied to the predictable-, intraseasonal- and total-covariance matrices, respectively. Leading EOF modes of the total component more often resemble the EOF modes of the intraseasonal component when the potential predictability is lower and vice versa. Temporal variation of the leading modes in the predictable components is more closely linked to the interannual variability of Eastern Pacific sea surface temperature (SST) in early summer (MJJ), North Atlantic SST in late summer (JAS) and El Niño/Southern oscillation in late autumn (NDJ). The rainfall SST connections found in these seasons persist throughout the entire 54-year period. The predictable rainfall modes also have apparent linkages to the intensity and position of the Western Pacific subtropical high during the early summer and late autumn with more rainfall occurring in the region when the height pattern is more intensive and extends further west. During the late summer, the predictable rainfall variability is accompanied by a large variation of tropospheric advection from north of the region, indicating the impact of an anomalous atmospheric circulation associated with the variation of summer North Atlantic Oscillation.

1. Introduction

The valley area spanning the regions adjacent to the Yangtze River (or Changjiang) in the south to the regions adjacent to the Huaihe River (or Huaihe) in the north (roughly 27°N–35°N, 110°E–125°E) is a geographically important region containing one of the major grain producing areas in China. As a result, the climate variability and predictability of this region have been the central concern of numerous previous monsoon studies, especially by Chinese scientists (Ding, 1994). In particular, the variability of precipitation in this region has been recognized as one of the prominent features associated with the behaviour of the East Asia summer monsoon (EASM). The variation of precipitation in this region reflects activities of the Mei-Yu front during the developing and retreating stages of the EASM (Tao and Chen, 1987; Ding and Chan, 2005).

As a fundamental aspect of climate predictability, variabilities of monthly to seasonal means of precipitation in the Yangtze–Huaihe River Valley and their relationships to the conditions of sea surface temperature (SST) in the global oceans have been studied extensively. A wide range of SST conditions has been suggested in relation to the interannual variability of the precipitation in this area. For example, some instantaneous and/or lead–lag relationships have been shown to exist between the precipitation in the Yangtze and Huaihe River Valley and the SSTs in the tropical Pacific Ocean (Liu and Ding, 1995; Chang et al., 2000; Gong and Ho, 2002; Wu et al., 2003; Feng and Li, 2011), the Northwest Pacific (Zhang et al., 2007), the tropical Indian Ocean and South China Sea (Zhang et al., 2003; Ding et al., 2010), and North Atlantic Oscillation (NAO) (Xin et al., 2008; Linderholm et al., 2011). And the relationships between the precipitation and SSTs have a dependence on the phases of the monsoon and/or SST evolutions. These precipitation–SST relationships also showed interdecadal changes (Chang et al., 2000; Gong and Ho, 2002; Xin et al., 2008). The wide range of the suggested precipitation–SST relationships indicates the complexity of the mechanisms that determine the interannual variability of the precipitation in this region.

Impact of low frequency intraseasonal variability is another important factor to be considered in understanding the predictability of seasonal mean precipitation in the Yangtze–Huaihe River Valley. For example, intraseasonal variability has played an important role in contributing to the severe floods in the Yangtze River during the summer of 1991, 1998, 2003 and 2007 (Lu and Ding, 1996; Mao and Wu, 2005; Chen et al., 2005, 2009; Xia et al., 2008). Compared to the slower evolution and relatively more persistence impact of SSTs, the impact of intraseasonal variability is considered to be noisy in the prediction of seasonal means as the intraseasonal variability is more short-lived and has more randomness in its origins over seasonal or longer timescales. For this reason, it is necessary to estimate the contribution of intraseasonal variability to the interannual variability of seasonal mean to better estimate the signal (e.g. the impact of SST) and thereby improve the prediction of seasonal means.

Despite the large amount of research in this area, we are not aware that anyone has tried to estimate, or quantify, how much of the interannual variability in the seasonal mean rainfall is related to processes that are potentially predictable and how much is related to intraseasonal variability, which on seasonal and interannual time scales can be considered as noise. It is possible, e.g. that for some climate fields the contribution to the interannual variability from intraseasonal variability can be as large, or larger, than that from the predictable component (Frederiksen and Zheng, 2004). It is possible in that case for some of the leading modes of interannual variability to reflect the influence of intraseasonal variability more than the predictable component. We show below that this is indeed the case for rainfall in Yangtze–Huaihe River Valley. So in order to better understand the predictability of the rainfall, it is important to effectively remove this noise component from the analysis.

In this article, we will explore the potential predictability of the seasonal mean precipitation for the Yangtze–Huaihe River Valley by applying a methodology proposed by Zheng and Frederiksen (2004), i.e. capable of estimating the contribution of intraseasonal variability to the interannual variance of seasonal means using monthly mean values. Frederiksen and Zheng (2007) provide an overview of the history and details of the general methodology. The primary question we ask is what are the most predictable signals in the seasonal mean precipitation of the Yangtze–Huaihe River Valley and how may the potential predictability be estimated by removing the ‘intraseasonal noise’ from the interannual variability of seasonal means? Another objective of this study is to examine the role of SSTs over the global oceans as possible predictors that can generate skillful predictions for the most predictable signals in the seasonal precipitation over the Yangtze–Huaihe River Valley. We will also examine the circulation pattern associated with the variation of the most predictable seasonal mean signals in the Yangtze–Huaihe River Valley precipitations. A knowledge of all these factors will help us to better understand how one might improve the prediction skill of rainfall forecasts in the region.

The rest of this article is arranged as follows: The methodology and data used are described in Section 'Data and methodology'. The annual cycle of the potential predictability is discussed in Section 'Annual cycle of the potential predictability'. The predictable seasonal rainfall signals with improved predictability and their possible SST predictors are identified in Section 'Predictable modes and their possible predictors'. The associated circulation patterns with the most predictable seasonal mean signal are examined in Section 'Circulation patterns associated with predictable modes'. Summary and discussions are given in Section 'Summary and discussion'.

2. Data and methodology

2.1. Data

The monthly precipitation dataset used in this study are from a network of 23 weather stations operated by the China Meteorological Administration in the region of the Yangtze–Huaihe River Valley (Figure 1), as defined by Zhao and Qian (2010). Monthly means of precipitation recorded at these stations for the period of January 1951 to December 2004 are available for this study. For each station, and for each 3-month season, the monthly mean rainfall is divided by its climatological seasonal mean to produce a more uniform distribution. Monthly mean anomalies are then obtained with regard to the 1951–2004 climatological mean annual cycle, and the seasonal mean anomaly is then calculated as the 3-month mean of the monthly anomalies.

Figure 1.

Geographical location of the 23 rain gauges with 54 years record in Yangtze–Huaihe River Valley.

The monthly mean sea level pressure (MSLP), 500-hPa geopotential height field, the zonal and meridional velocities and specific humility (at 300, 400, 500, 600, 700, 850, 925 and 1000 hPa) datasets (on a 2.5° latitude × 2.5° longitude grid) used in this study are from the National Centers for Environmental Prediction-National Center for Atmospheric Research (NCEP-NCAR) Reanalysis (Kalnay et al., 1996). Our calculation covers the period from January 1951 to December 2004, being consistent with the data length of the precipitation data used in this study.

The monthly mean SSTs (on a 1.0° latitude × 1.0° longitude grid) used in this study is taken from the UK Met Office Hadley Centre SST dataset (HADISST1.1; Rayner et al., 2003) for the same 54-year period. The oceanic area used to produce the SST one-point correlation maps in Section 'Predictable modes and their possible predictors' is between 40°S and 60°N and between 40°E and 10°W.

2.2. Variance decomposition method and potential predictability

Zheng and Frederiksen (2004; hereafter ZF2004) proposed a methodology for extracting, from monthly mean data, spatial patterns of interannual (supra-annual) variability in seasonal mean fields that can be related to variability of slow and intraseasonal components. Frederiksen and Zheng (2007) provide a review of the underlying ideas and applications of the general methodology. Readers are referred to the papers for a detailed description of the methodology. Here, we present only a brief summary of the method.

Firstly, the annual cycle is removed from the data. The monthly time series of each climate anomaly is then conceptually decomposed into two components consisting of a seasonal ‘population’ mean and a residual departure from this mean. Thus, if xym represents sample monthly values, within a season, in month m (m = 1, 2, 3) in year y (y = 1, …, Y, where Y is the total number of years), we use the following decomposition (e.g. ZF2004),

display math(1)

Here, μy is the seasonal population mean in year y, and ϵym is a residual monthly departure of xym from μy and arises from intraseasonal variability. The vector {ϵy1,ϵy2,ϵy3} is assumed to comprise a stationary and independent annual random vector with respect to year. Equation (1) implies that month-to-month fluctuations, or intraseasonal variability, arise entirely from {ϵy1,ϵy2,ϵy3} (e.g. (xy1 − xy2 = ϵy1 − ϵy2)).

We represent an average taken over an independent variable (i.e., m or y) by replacing that variable subscript with ‘o’. For example, xyo indicates the seasonal average of xym in year y and xoo, the average of xym taken over all months and years. With this notation, a seasonal mean can be expressed as

display math(2)

where ϵyo is associated with intraseasonal variability and μy with the interannual variability of external forcing and slowly varying (interannual/supra-annual) internal dynamics. ZF2004 refer to these as the intraseasonal and ‘slow’ component, respectively, of the seasonal mean xyo, which is also referred to as the ‘total’ component. Since dynamic prediction has a skillful range of about 10 days, the interannual variability arising from intraseasonal variability is not predictable at seasonal time scales. Therefore, we shall also refer to the intraseasonal component as the unpredictable component. In addition, we shall also refer to the slow component as the potentially predictable component.

Following Madden (1976), we define the potential predictability as the ratio between the variance of the predictable component (V(μy)) and the variance of the total component (V(xyo)) (Zheng and Frederiksen, 1999), which can be expressed as

display math(3)

It represents the fraction remaining after the removal of the intraseasonal component from the total. The larger the potential predictability, the more likely the seasonal mean precipitation anomalies can be predicted.

Suppose now that we have two climate variables xym and xym that satisfy Equations (1) and (2). Then, ZF2004 derived the following estimate of the interannual covariance V(ϵyo,ϵyo) of the intraseasonal components,

display math(4)

where,

display math(5)
display math(6)
display math(7)
display math(8)

In order to reduce the estimation error, the estimated math formula has to be constrained. If both of xym and xym are pressure variables, math formula is constrained to lie within the interval [0, 0.1] (ZF2004). Otherwise (i.e. at least one of xym and xym is a rainfall variable), math formula is empirically constrained to lie within the interval [−0.1, 0.1] (Zheng and Frederiksen, 2006).

Since the covariance between two seasonal means can be estimated as

display math(9)

the covariance between two seasonal means can be decomposed, using Equation (4), as

display math(10)

where the first term on the right-hand side can be rewritten as

display math(11)

and will be referred to as the ‘residual’ covariance after removing the variability of the intraseasonal component. It is worth emphasizing that this covariance, in general, consists of not only the covariance between μy and μy (i.e. the slow components of the climate variables), but also their interaction terms with ϵyo and ϵyo. In the case where the intraseasonal and slow components are independent, the residual covariance reduces to the covariance of the slow component. When this is not the case, V(xyo,xyo) − V(ϵyo,ϵyo) may still be better related to the covariance between the two slow components than is V(xyo,xyo).

The covariance matrices for the components of the seasonal mean are adjusted to ensure that they are positive semi-definite using the method described by Grainger et al. (2008).

2.3. Identifying predictable signals

Once the intraseasonal and residual cross-covariance matrices have been estimated, an empirical orthogonal function (EOF) analysis is conducted to identify the leading predictable modes of each covariance matrix. For convenience, we shall refer to the EOFs of the covariance matrices defined by Equations (9), (4) and (11) as the total, intraseasonal and predictable modes of interannual variability in the seasonal mean rainfall. The corresponding principal component (PC) time series of each predictable mode is obtained by projecting the field of 3-month mean (seasonal) precipitation anomalies onto the corresponding EOF mode for each season and year in the time series. A more detailed description is given in ZF2004. The potential predictability of the PCs can also be calculated by Equation (3).

3. Annual cycle of the potential predictability

Our analysis starts with an initial estimation of the potential predictability (Equation (3)) of total observed variance of seasonal mean precipitation at the 23 stations. The averaged potential predictability of 23 stations is shown in Figure 2 for all seasons. Roughly speaking, the potential predictability of seasonal precipitation in the Yangtze–Huaihe River Valley is moderate with the predictable variance explaining at best about 16% of the total variance, which is to say that the precipitation in this area is mainly associated with intraseasonal variability. There is a distinct annual cycle in the potential predictability. The potential predictability is low during the winter seasons, with a minimum of less than 1% during the 3 months from January to March (JFM). Along with the development of the EASM, the potential predictability increases, and reaches a maximum of about 16% during the 3 months from May to July (MJJ). Then it decreases during the summer–autumn transition seasons (JAS and ASO), followed by an increase during the autumn seasons and reaches a second high during November to January (NDJ).

Figure 2.

Spatial mean of potential predictability in each station, expressed as a percentage of the total variance.

The method of Zheng and Frederiksen (2004) provide an efficient tool to identify coherent seasonal predictable signals with generally larger potential predictability by separating the variance due to the intraseasonal component (noise) from the variance in the slow component (signal) in the construction of the covariance matrix. ZF2004 showed that their method generally produced larger potential predictability in their EOF patterns than were possible from doing a normal EOF analysis using the total field without this separation. Table 1 provides a summary of the percentage variance explained and the potential predictability of the two leading slow and total EOFs for rain over our region of interest and for each of the twelve 3-month season. Also shown is a breakdown of the total interannual variance in the seasonal rainfall into an intraseasonal and slow component. In most seasons, the intraseasonal component has by far the largest variance compared with the slow component. So the intraseasonal component is largely responsible for the interannual variance of rainfall over this region. From Table 1, it can also be seen that the first slow EOF in each season explains by far the largest percentage of the variance in the slow component. Our discussion below will be focused on the seasons having better potential predictability.

Table 1. Variability of total, intraseasonal and potentially predictable component of the seasonal rainfall pattern (left three columns). The potential predictability of seasonal precipitation PCs with/without application of ZF2004 (right four columns). The percentage of the variance explained is shown in brackets
SeasonTotal VarIntra VarSlow Var(% explained var) Predictability
S-PC1S-PC2T-PC1T-PC2
MAM2.361.860.50(85%) 0.34(11%) 0.33(52%) 0.31(18%) 0
AMJ2.832.360.47(65%) 0.54(22%) 0.26(38%) 0.22(26%) 0.24
MJJ3.152.350.80(68%) 0.53(20%) 0.37(43%) 0.42(20%) 0.16
JJA3.722.760.96(57%) 0.51(23%) 0.41(38%) 0.47(19%) 0.21
JAS4.223.660.56(54%) 0.38(23%) 0.26(37%) 0.10(14%) 0.25
ASO4.273.960.31(43%) 0.20(27%) 0.41(35%) 0(16%) 0
SON5.234.580.65(56%) 0.42(24%) 0.29(39%) 0(21%) 0.41
OND5.564.800.76(83%) 0.46(13%) 0.26(54%) 0(19%) 0.09
NDJ5.224.240.98(94%) 0.51(4%) 0.50(63%) 0(16%) 0
DJF4.243.760.48(73%) 0.21(13%) 0.18(66%) 0.14(12%) 0
JFM2.772.720.05(49%) 0(35%) 0(59%) 0(14%) 0
FMA2.251.980.27(77%) 0.06(12%) 0.32(56%) 0(14%) 0

4. Predictable modes and their possible predictors

In this section, we examine the EOF modes for the total, predictable and intraseasonal components of the seasonal precipitation in the Yangtze–Huaihe River Valley and search for the associated possible predictors for the most predictable modes. In order to find the SST predictors, we calculated the correlation between the PC time series of the predictable modes and the one-season lead (in the previous 3-month period) and simultaneously seasonal mean SST anomalies. Possible predictors could then be selected as the area mean of SSTs in oceanic areas where the one-season lead correlation has maximum values with the PC time series.

4.1. Predictable rainfall modes related to Kuroshio current

From Table 1, the most potentially predictable PCs of the 3-month mean precipitation for MAM, AMJ, MJJ, JJA and JAS are their PC1s, for which the explained variance are shown in Table 1 in brackets. Their corresponding modes are presented in the right column of Figure 3. At the phase shown here, it is wetter everywhere. The structures have some similarities in AMJ, MJJ and JJA, with a maximum centred in the southwest and a minimum in the northeast of the Yangtze–Huaihe River Valley. However, the pattern structure in MAM and JAS differs from the others. The rainfall pattern in MAM has a south–north spatial structure while JAS has an east–west spatial structure.

Figure 3.

Total (left column), intraseasonal (middle column) and potentially predictable (right column) rainfall modes for (a–c) MAM-PC1; (d–f) AMJ-PC1; (g–i) MJJ-PC1; (j–l) JJA-PC1; (m–o) JAS-PC1, respectively. Contour interval is 0.05 and positive contours are shaded. Contour interval is 0.05 and positive contours are shaded.

The modes of the seasonal mean total rainfall (Figure 3, left column) and the intraseasonal modes of the rainfall (Figure 3, middle column) are also shown. Comparing the total precipitation patterns with the intraseasonal and predictable modes, we can see from Figure 3 that the intraseasonal modes are more similar to the total ones than the predictable modes in MAM, AMJ and JAS. Recall that these seasons have lower predictability (Figure 2) and so one would expect the intraseasonal component to dominate the rainfall variability. The spatial difference between the predictable, the total and the intraseasonal modes shows how important the decomposition is when focusing on the issue of predictability.

The predictable modes in these five seasons have similar correlation patterns with SST, as shown in Figure 4. A remarkable feature that appears consistently in both the simultaneous and one-season lead SST correlation pattern is the maximum values in the Northwest Pacific Ocean around the Kuroshio region. It indicates that these predictable rainfall patterns are associated with a significant warming in the Northwest Pacific Ocean around the Kuroshio region. Possible connections between the Kuroshio SST and the East Asian summer monsoon and the Western Pacific subtropical high have also been noticed in previous studies (Zhang et al., 2007). It is shown that the correlation between the Kuroshio SST and the predictable rainfall PC gets higher from MAM to JJA and become lower in JAS (Figure 4), which is consistent with the change in the raw potential predictability (Figure 2) in these seasons.

Figure 4.

One-point correlation maps of one-season lead SST (left column) and contemporary SST (right column) associated with potentially predictable rainfall modes for (a, b) MAM-PC1; (c, d) AMJ-PC1; (e, f) MJJPC1; (g, h) JJA-PC1; (i, j) JAS-PC1, respectively. Shading indicates correlation coefficients significant above 95% confidence level according to the Student t-test.

Previous studies have found a significant interdecadal variation in the Pacific around the end of 1970s that may account for the observed changes in the EASM–SST relationship (Chang et al., 2000; Han and Wang, 2007; Wang et al., 2008). To investigate if there are interdecadal shift in the relationship between the predictable precipitation of the Yangtze–Huaihe River Valley and the SST, we compare the correlation of spring and summer (MAM to JAS) SST associated with the slow rainfall PC1 between 1951 and 1977 and 1978 and 2004 (not shown). The significant positive correlations located in the Northwest Pacific Ocean and the Kuroshio region holds in the two shorter periods. However, the correlation shows remarkable differences in the eastern tropical Pacific Ocean between the two periods, especially in spring and early summer (MAM to MJJ). In 1978–2004, significant positive correlation appears in the eastern tropical Pacific Ocean for both the contemporary and one-season lead SST from MAM to MJJ, implying a close relationship with ENSO in the later period. In 1951–1977, significant negative correlation is located in the eastern tropical Pacific Ocean for contemporary SST in AMJ and MJJ, and the ENSO correlation disappears for contemporary SST in MAM and one-season lead SST from MAM to MJJ. This may be the reason that the ENSO SST is not significantly correlated with MJJ slow EOF1 for the entire 54-year period. In contract, the association between the Yangtze–Huaihe precipitation and the Kuroshio SST is rather stable throughout the entire 54-year period.

4.2. Predictable rainfall modes related to ENSO

The most potentially predictable PCs for ASO, SON, OND, NDJ and DJF are ASO-PC2, SON-PC1, OND-PC1, NDJ-PC1 and DJF-PC1 according to Table 1. Their variance explained can also be found in Table 1 (in bracket). The predictable rainfall patterns for these five 3-month periods (Figure 5, right column) have a similar south–north gradient in the loadings.

Figure 5.

Total (left column), intraseasonal (middle column) and potentially predictable (right column) rainfall modes for (a–c) ASO-PC2; (d–f) SON-PC1; (g–i) OND-PC1; (j–l) NDJ-PC1; (m–o) DJF-PC1, respectively. Contour interval is 0.05 and positive contours are shaded.

From Figure 5, we can also see the total (left column) and intraseasonal (middle column) modes of rainfall. We see that there are more similarities between the total and intraseasonal patterns than between the total and predctable modes in these seasons, especially in ASO, SON and DJF. The predictable and intraseasonal patterns show very different structures. Note that the predictable mode is more similar to the total mode in NDJ when the potential predictability becomes highest compared to all of the other 3-month periods from ASO to FMA (Figure 2).

Both the simultaneous and one-season lead SST correlations show very similar patterns with high values in the eastern tropical Pacific Ocean from ASO to NDJ, which suggests the impact of El Niño/Southern Oscillation (ENSO) on rainfall variability during these seasons. We also notice that accompanied with the phase of El Niño, the contemporary SST in the South China Sea reaches an anomalously high value. As suggested by previous studies (Huang and Wu, 1989; Webster and Yang, 1992; Wang et al., 2000), ENSO–monsoon interaction also exist in the EASM with the Western Pacific subtropical high being a major factor controlling the interaction of the two dynamic system over this region. As suggested by other studies (Zhang et al., 1999; Wang et al., 2000; Wu et al., 2003), the influence of ENSO on the EASM rainfall is not limited to the summer and also exists in the autumn. The influence of the ENSO on the rainfall of Yangtze–Huaihe River Valley has an evolution starting in ASO and peaking in NDJ before retreating during DJF. The potential predictability follows this seasonal change in the ENSO influence.

5. Circulation patterns associated with predictable modes

In order to aid the physical interpretation of our slow rainfall modes, in this section, we will examine the atmospheric circulations associated with the potentially predictable modes. We do this by estimating the covariance between the PC time series of the slow EOFs and the slow component of the circulation field, using Equation (11). This will allow us to focus more on the predictive characteristics of the relationship. These slow covariance patterns are associated with very slowly varying (interannual to supra-annual) external forcings and internal dynamics (Frederiksen and Zheng, 2004) and correspond to the more potentially predictable signals in the seasonal mean precipitation and circulation variables.

As we have discussed in the above section, the predictable rainfall signal in the Yangtze–Huaihe River Valley is mainly associated with the variations in the Kuroshio SST in the northwest Pacific and ENSO SST in the eastern tropical Pacific. We choose MJJ and NDJ as two typical seasons representing the Kuroshio-related and ENSO-related evolutions, respectively, because the potential predictability reaches maximum values in these two 3-month periods (Figure 2). In addition, we will examine the circulation patterns in JAS to understand the relationship between the rainfall variability in the Yangtze–Huaihe River Valley and NAO.

5.1. Circulation patterns in MJJ

The 3-month period from May to July is a typical time during which the summer monsoon develops over China. In general, the Mei-Yu front moves progressively northward in the Yangtze–Huaihe River Valley area from the adjacent of Yangtze River in south to the adjacent of Huaihe River in north (Ding and Chan, 2005). The movement of the Mei-Yu front is accompanied by the intensification and northward progression of the subtropical high over the Pacific Ocean. There is an apparent difference between the location of the subtropical high during the positive and negative phases of the most predictable mode in MJJ (Figure 7(a) and (b)). In the composite of the positive phase, the subtropical high is more intense and extends more westward with the contour line of 5860 m (draw in boldface) in the 500-hPa geopotential height field reaching over to the coastline of South China at the west-most longitude of 110°E. As the subtropical high extends anomalously more westward, more moisture is brought in from the south into the Yangtze–Huaihe River This result is consistent with the findings by previous authors (Zhang et al., 2007). In contrast, in the composite of the negative phase (Figure 7(b)), the subtropical high is weaker with an absence of the 5880 m maximum centre in the 500-hPa geopotential height field and a much less westward extend of the 5860 m contour line that reaches only to the east side of Philippine with the west most longitude of 130°E.

Figure 6.

One-point correlation maps of 1-season lead SST (left column) and contemporary SST (right column) associated with potentially predictable rainfall modes for (a, b) ASO-PC2; (c, d) SON-PC1; (e, f) OND-PC1; (g, h) NDJ-PC1; (i, j) DJF-PC1, respectively. Shading indicates correlation coefficients significant above 95% confidence level according to the Student t-test.

Figure 7.

Covariance maps of the composite maps of 500 hPa geopotential height for (a) the positive phase, (b) the negative phase and (c) vertical integrated (at 300, 400, 500, 600, 700, 850, 925 and 1000 hPa) moisture flux and 925 hPa moisture convergence and (d) SLP associated with potentially predictable rainfall modes of MJJ-PC1.

Figure 7(c) shows the slow covariance of the moisture flux vertically integrated through the entire layers and the 925-hPa moisture convergence associated with the predictable rainfall modes in MJJ. As can be seen in Figure 7(c), the large value of moisture convergence centre is situated over the Yangtze–Huaihe River Valley, responsible for the positive anomaly of the slow component of rainfall in this region. And there are two main moisture transport branches over the region of interest: one is the transport by the EASM, an anticyclone along the west margin of the Western Pacific high, fetching warm and moist air from South China Sea into the Yangtze–Huaihe river basin; another is the transport by the mid-latitude westerlies, a strong north westerly circulation, bringing moisture from the north China to central-eastern China.

Another notable feature in the atmospheric circulation associated with the MJJ most predictable mode is the anti-cyclonic anomalies in the sea level pressure (SLP) north to the Yangtze–Huaihe River Valley. The anti-cyclonic anomalies in SLP have two maximum centres – one is over the Ural Mountain and another over the Okhotsk Sea (Figure 7(d)). Previous studies have found that the dual blocking high condition is the most favourable for prolonged Mei-Yu heavy rainfall (Ding, 1990; Tao and Zhang, 1998; Wu and Wang, 2002).

5.2. Circulation patterns in NDJ

The impact of ENSO for the rainfall pattern in NDJ is confirmed by the covariance between the predictable precipitation PC1 and the SLP (Figure 8(d)), which shows an anomalous low in the eastern tropical Pacific Ocean and an anomalous high in the Western tropical Pacific Ocean.

Figure 8.

Covariance maps of the composite maps of 500 hPa geopotential height for (a) the positive phase, (b) the negative phase and (c) vertical integrated (at 300, 400, 500, 600, 700, 850, 925 and 1000 hPa) moisture flux and 925 hPa moisture convergence and (d) SLP associated with potentially predictable rainfall modes of NDJ-PC1.

During NDJ, the difference between the composite patterns of 500 hPa geopotential height for the positive and negative phases of the most predictable precipitation mode (Figure 8(a), (b)) is similar to that described in MJJ (Figure 7(a), (b)). The subtropical high is more westward extended in the positive phase than in its negative phase. As the subtropical high retreats to the south, the north edge of the 5860 m contour is around 20°N. The anomalous southwestlies bring moisture to the southern part of Yangtze–Huaihe River Valley and cause the anomalous wetter condition in that area while the seasonal precipitation is less than normal in the north, and vise versa when the subtropical high weaker and has less westward extension.

A notable feature of anomalous moisture convergence in the south of the Yangtze–Huaihe River Valley and divergence centre located in the north of this region can be seen in Figure 8(c), which may explain the north–south spatial structure of the slow rainfall pattern in NDJ. Similar to MJJ, the mid-latitude westerlies makes a dominant contribution to the moisture transport over the Yangtze–Huaihe River Valley while moisture flux from the South China Sea also plays an important role (Figure 8(c)).

5.3. Circulation patterns in JAS

Another remarkable feature in the SST evolution associated with the predictable mode of the seasonal precipitation in the Yangtze–Huaihe Rivers Valley is the Atlantic SST triple pole during summer. The triple structure appears in the 3-month season of MJJ and peaks in the 3-month season of JAS in both the one-season lead and simultaneous PC-SST correlations (Figure 4(e)–(j)). This result is consistent with the results recently obtained by other authors investigating the connection between the North Atlantic triple anomalous SST pattern and the EASM. The study by Wu et al. (2009) indicated that the summer triple anomalous SST pattern is a result of the anomalous NAO occurring in the previous spring. And the North Atlantic triple SST anomalies may impact the EASM by exciting downstream sub-polar teleconnections across the northern Eurasia.

The atmospheric circulation patterns associated with the most predictable precipitation mode of JAS are shown in Figure 9(a)–(d). Firstly, a centre of maximum covariance is seen over the North Atlantic (Figure 9d) and thus verifies the connection between the summer NAO and the summer precipitation in the Yangtze–Huaihe River Valley (Hasanean, 2004; Linderholm et al., 2011). Anomalous north-to-south moisture transport is a dominant feature over eastern China (Figure 9(c)), indicating the importance of anomalous moisture conditions in the north of the Yangtze–Huaihe River Valley during this time of year. The anomalous divergence located in the east of Yangtze–Huaihe River Valley as well as the anomalous convergence along central-south China and south Japan, are responsible for the east–west spatial pattern of the JAS predictable precipitation. On the other hand, the composite patterns of 500 hPa geopotential height for the most predictable mode in JAS (Figure 9(a), (b)) show that, the subtropical high extends westward in the positive phase rather than the negative phase. It brings more moisture to the west area of Yangtze–Huaihe River Valley and results in wetter than normal conditions there.

Figure 9.

Covariance maps of the composite maps of 500 hPa geopotential height for (a) the positive phase, (b) the negative phase and (c) vertical integrated (at 300, 400, 500, 600, 700, 850, 925 and 1000 hPa) moisture flux and 925 hPa moisture convergence and (d) SLP associated with potentially predictable rainfall modes of JAS-PC1.

6. Summary and discussion

From the results presented in the above sections, the following major points emerge:

The predictable signals in the seasonal mean precipitation over the Yangtze–Huaihe Rivers Valley have been identified by an application of the methodology of ZF2004 that separates slow varying signals from the noise introduced by intraseasonal and synoptic variability. The predictable signals identified with the application of ZF2004 method have higher potential predictability because of the reduction of the noise from the variance of seasonal mean values.

Using the ratio of the variance of the predictable signal to the variance of total seasonal mean anomalies as a primary estimation of the potential predictability, the annual cycle of the change in the potential predictability for the seasonal precipitation over the Yangtze–Huaihe River Valley has been examined. The potential predictability is lowest during winter (JFM), reaches highest value in summer (MJJ), decreases during summer–autumn transition seasons (JAS and ASO) and increases in the autumn (NDJ).

Possible SST predictors for one-lead predictions of the seasonal precipitation in the Yangtze–Huaihe Rivers Valley may come from the seasonal mean SST anomalies in the area of the Kuroshio current for the prediction of rainfall during MAM to JAS, while the eastern tropical Pacific Ocean provides a possible predictor for autumn rainfall from ASO to DJF. So there are two dominant evolutions for the predictable rainfall patterns in Yangtze–Huaihe River Valley. One is the Kuroshio-related evolution from MAM to JAS; the other is the ENSO-related evolution from ASO to DJF. In the former case, it is not possible to say from our study whether this relationship is a signal of local SST forcing or if it is indicative of atmosphere–ocean-coupled variability. In addition, the relationship between predictable seasonal rainfall in Yangtze–Huaihe River Valley and the ENSO during spring and early summer (MAM to MJJ), however, has been found to be unstable in the two periods before and after late 1970s. The NAO has some influence on the predictable rainfall modes in JAS. There is a close association between the annual cycle of changes in the potential predictability and the seasonal change of the relation between predictable rainfall modes and SST. The potential predictability is higher/lower when the correlation between the most predictable rainfall PC and SST in Kuroshio or Western Pacific is higher/lower.

Teleconnections are found between the predictable rainfall modes and circulation. The high latitude blocking high, Southern Oscillation and the north Atlantic triple pole are significantly related to the slow component of rainfall in the Yangtze–Huaihe River Valley in MJJ, NDJ and JAS alternatively. The distribution of regional moisture sinks and sources around the Yangtze–Huaihe basin are consistent with the spatial patterns of the slow component of rainfall in MJJ, NDJ and JAS. Most of the water vapour is transported into the Yangtze–Huaihe River Valley through the southern as well as northern boundary in MJJ and NDJ, indicating the important role of the EASM circulation and the westerlies in supplying water vapour and supporting the regional rainfall. In JAS, the moisture inflow from the north dominates the moisture transport to the Yangtze–Huaihe region. The intensification and progression of subtropical high in Western Pacific Ocean has a close relationship with the predictable seasonal precipitation in MJJ, NDJ and JAS.

In the future, we plan to partition China into several regions and apply this methodology to predict seasonal precipitation in each region for all seasons. In this way, we hope, we are able to establish a seasonal precipitation prediction scheme for China.

Acknowledgements

We would like to thank Dr. Simon Grainger for useful comments. We gratefully acknowledge the anonymous reviewers for their constructive comments, which helped greatly in improving the quality of this manuscript. We also grateful to the editors for their hard work and suggestions on this manuscript. This work was supported by National Program on Key Basic Research Projects of China (grant nos. 2010CB951604 and 2010CB428402), and the National Natural Science Foundation of China General Program (grant no. 40875062).

Ancillary