An Objective Method for Defining Meiyu Onset in Lower Reaches of the Yangtze River Basin

A unique feature in boreal summer is northward progression of monsoon rainband in East Asia from tropics to subtropics (beyond 40°N). A climatological Meiyu front often appears along the Yangtze River in June, persisting there for a few weeks. Accompanied to the subtropical front is persistent rainfall with heavy rainstorms over the Yangtze River Valley basin. The front is characterized by pronounced southerlies to its south that carry a large amount of warm and humid air northward and marked northerlies to its north that transport dry and cold air southward. As a result, strong meridional temperature gradient forms along the Meiyu front.

There is an open issue as how to define the onset of the Meiyu. In the current regional operational meteorological centers, the Meiyu onset is defined primarily based on local rainfall and persisting period by “Meiyu monitoring indices” or local criteria subjectively somehow. As a result, the defined onset dates sometimes differ greatly among neighboring cities due to localized precipitation characteristics. Given that the northward progression of the Meiyu rainband in East Asia is driven fundamentally by slowly evolving solar radiative forcing, the atmospheric circulation associated with the Meiyu onset must be of large scale. Thus, there is a need to systematically examine large-scale atmospheric dynamic and thermodynamic conditions prior to and during the Meiyu onset.

A number of studies are devoted to understand the interannual variability of the Meiyu onset. For instance, during the early Meiyu onset years, the South Asian high (Zhu et al., 2008; Zhao et al., 2018; Niu et al., 2021) and the subtropical westerly jet stream (Sampe and Xie, 2010; Li and Zhang, 2014; Hong et al., 2018) in the upper troposphere established earlier and shifted northward. Li et al. (2019) found that favorable conditions for the early Meiyu onset were a stronger than normal South Asian high in April and a weakened subtropical jet stream between 10° and 30°N. Blockings in the middle troposphere and the western Pacific subtropical high (WPSH) were also important in modulating the Meiyu onset (Ding et al., 2007). Liang et al. (2010) defined the Meiyu onset date based on 5-day running mean precipitation and the ridgeline of the WPSH. The intensity of Meiyu rainband and its northward moving speed were closely linked to Meiyu onset time (Liang and Ding, 2012). There is a significant positive correlation between the onset dates of Meiyu and summer monsoon over the South China Sea (Ding et al., 2020).

It has been shown that the interannual variability of the Meiyu onset date exhibited two dominant periods (Ding et al., 2013; Liang et al., 2018). One is at 2–4 yr and the other is at 5–8 yr. The interannual variability of the Meiyu onset date is greatly modulated by El Niño–Southern Oscillation (ENSO) and the Tropospheric Biennial Oscillation (TBO) over the Asian monsoon region (Li et al., 2006). When the El Niño (La Niña) phase prevailed in the equatorial Pacific, the Meiyu onset was likely to be late (early) (Wu et al., 2017; Pan et al., 2021). TBO is another important precursory signal influencing the monsoon onset in East Asia (Ding et al., 2013). Liang and Ding (2012) found that the key regions of sea surface temperature (SST) connected to the interannual variation of Meiyu were located in the tropical Pacific and Indian Ocean.

The objective of the current study is threefold. First, we intend to develop an objective large-scale Meiyu onset index (MOI) in East China that integrate the onset dates collected from regional climate centers. Second, we examine the interannual variability of the MOI, with a special focus on the atmospheric and oceanic conditions associated with the early and late onset groups. Third, we construct a statistical model suitable for real-time Meiyu onset prediction and conduct an independent forecast beyond the training period. The remaining part of this study is organized as the following. Sections 2 describes the data and analysis method. Large-scale dynamic and thermodynamic conditions associated with climatological Meiyu onset are examined and an integrated MOI is proposed in Section 3. In Section 4, we reveal the atmospheric and oceanic signals associated with the early and late onset groups. A statistical model is further constructed based on the precursory atmospheric and ocean signals in the preceding months and the forecast skill is assessed for an independent period. Finally, the conclusions are given in the last section.

The data used in the current study include (1) Meiyu onset dates collected from Shanghai Climate Center and Jiangsu Climate Center, (2) the National Centers for Environmental Prediction/Department of Energy (NCEP/DOE) Reanalysis II (Kanamitsu et al., 2002), (3) National Oceanic and Atmospheric Administration (NOAA) Climate Prediction Center (CPC) Merged Analysis of Precipitation (CMAP) data, and (4) NOAA Extended Reconstructed Sea Surface Temperature (ERSST) v5 dataset. While the reanalysis and CMAP products have a horizontal resolution of 2.5° × 2.5°, the ERSST has a horizontal resolution of 2° × 2°. The current analysis focuses on the period of 1981–2020.

The first goal of the current study is to obtain a large-scale Meiyu onset index over the lower reaches of the Yangtze River basin (LYRB). This index is defined based on the daily low-level wind, temperature, relative humidity, or precipitation fields. Integrated Shanghai and Jiangsu onset dates serve as a standard for deriving such an objective, large-scale index.

A composite analysis is carried out to reveal large-scale atmospheric circulation and SST features associated with early and late onset groups. Prior to the composite analysis, an interdecadal and long-term trend component is removed from the anomaly fields. This component is defined as a 5-yr running mean of the daily anomaly fields that are derived from the raw data by subtracting daily climatological fields.

In addition to regular meteorological fields, moist static energy (MSE) is also used to measure atmospheric condition prior to and during the Meiyu onset. This quantity is often used to measure the atmospheric convective instability (Hsu and Li, 2012; Wang L. et al., 2017; Wang L. C. et al., 2021). MSE considers both the temperature and specific humidity fields, and can be expressed as

where T denotes temperature, q denotes specific humidity, z denotes height, c_p = 1004 J kg⁻¹ K⁻¹ is the specific heat at constant pressure, g = 9.8 m s⁻² is the gravitational acceleration, and L_v = 2.5 $ \times $ 10⁶ J kg⁻¹ is the latent heat of vaporization. MSE represents how hot, how wet, and how high an air parcel is.

Currently, Meiyu onset at each city along the LYRB is determined by regional meteorological centers. The forecasters at each center make consultations based on local weather conditions. Such a method is somehow subjective, relying heavily on forecasters’ experience, and as a result, the onset dates among neighboring cities sometimes differ greatly. To remove possible discontinuation in space and to reflect large-scale atmospheric conditions, it is desirable to integrate these regional onset dates to a single date each year for the LYRB region (containing Shanghai and Jiangsu Province). The integration is weighted based on area. The area proportion of Jiangsu and Shanghai in the key analysis region (shown in a purple box in Fig. 2) is approximately 8 : 1.

Figure 1 shows the time evolution of the integrated Meiyu onset date in the LYRB during 1981–2020. The average Meiyu onset date is 19 June. Note that the integrated onset date time series exhibits a strong interannual fluctuation. The earliest onset date is 3 June 1996 and the latest onset is 9 July 1982. The standard deviation of the integration onset index is 8.8 days.

Figure 1.  Time evolution of the integrated Meiyu onset date index (solid) during 1981–2020, estimated based on area-weighted, subjectively-determined onset dates at cities over the LYRB. The dashed curve denotes the interannual component of the integration time series, and blue lines denote a half of the standard deviation of the time series.

The dashed curve in Fig. 1 is the interannual component of the integrated onset index, derived by subtracting a 5-yr running mean from the original time series. By doing so, we intentionally remove the interdecadal and long-term trend components so that we can focus on the interannual variability.

As mentioned earlier, it is critical and practically useful to construct an objective Meiyu onset index. To define such an index, one may first examine the large-scale circulation feature associated with the climatological Meiyu onset. Figure 2 illustrates the temporally evolving patterns of climatological mean precipitation, relative humidity and temperature at 1000 hPa at a 10-day interval from early to late June. To clearly illustrate the pattern change, climatological monthly mean values have been subtracted from these fields. Note that as time progresses from early to late June, near surface air temperature and moisture over the LYRB (i.e., purple box region) becomes warmer and moister. An obvious transition from dry to wet happens in the rainfall field in late June, suggesting that a climatological mean onset date must occur in late June, which is in good agreement with the mean onset date derived from the integration index.

Figure 2.  Temporally evolving horizontal patterns of (upper) precipitation (shaded; mm day⁻¹) fields and (below) 1000-hPa relative humidity (shaded; %) and 1000-hPa temperature (contour; °C) fields at a 10-day interval in June 1981–2020 climatology. June monthly mean fields have been removed to clearly show the anomaly evolution patterns. The purple box (30°–32.5°N, 115°–122.5°E) denotes a key analysis domain.

Black curves in Fig. 3 illustrate the time evolutions of the box-averaged climatological daily precipitation, relative humidity, temperature, and MSE. To remove high-frequency fluctuations, a 5-day running mean has been applied. A significant jump in precipitation occurred on the day of Meiyu onset (19 June) and held for 23 days before falling back. However, the relative humidity, temperature, and MSE in the lower and middle levels also showed a significant jump on 19 June and continued to increase.

Figure 3.  Time evolutions of (a) the box-averaged climatological (black) daily precipitation (mm day⁻¹), (b) 1000- and 850-hPa relative humidity (%), (c) 1000- and 850-hPa temperature (°C), and (d) 1000- and 850-hPa MSE ($ \times $10⁴ J kg⁻¹). Vertically oriented dashed line denotes the climatological onset date 19 June. Green and brown curves at each panel represent the corresponding evolutions for an early and a late onset group.

From the climatological perspective, one may obtain the Meiyu onset criteria for the area-averaged temperature, relative humidity, precipitation, and MSE. All these variables exhibit a rising trend as the Meiyu arrives. Specifically, the climatological Meiyu onsets when the box-averaged rainfall reaches 7 mm day⁻¹, 1000-hPa (850-hPa) relative humidity reaches 82% (71%), 1000-hPa (850-hPa) temperature reaches 25°C (16°C), and 1000-hPa (850-hPa) MSE reaches 34 × 10⁴ J kg⁻¹ (33.4 ×10⁴ J kg⁻¹), respectively.

Can one apply the aforementioned climatological criteria to individual years? To address the question, we constructed a number of objective Meiyu onset date indices based on either a single variable or a combination of two variables, using the climatological thresholds shown in Fig. 3. Tables 1–2 list calculated correlation coefficients with the integration index (shown in Fig. 1) when either a single variable criterion or a combined two-variable criterion is used.

Based on the result above, we select an objective Meiyu onset index (MOI) based on the area-averaged 1000-hPa relative humidity and temperature for each year, because the combined two-variable criterion leads to the highest correlation coefficient (0.55) with the integration index. With this criterion, Meiyu at each year is started when the area-averaged 1000-hPa relative humidity exceeds 82% and the area-averaged 1000-hPa temperature exceeds 25°C.

Figure 4 shows the time series of the aforementioned objectively determined MOI. The mean Meiyu onset date is 20 June. The early Meiyu onset groups are including 1982, 1984, 1988, 1990, 1993, 1996, 1999, 2011, 2013, 2015, and 2020; and the late groups are 1983, 1987, 1989, 1991, 1992, 1997, 2000, 2005, 2010, 2014, 2018, and 2019. As this time series is significantly correlated with the integration index, the proposed objective method in general agrees with the traditional subjective method, from a large-scale point of view. The benefit of this new index is simple and objective, rather than subjective, and it well reflects large-scale atmospheric conditions in the LYRB region.

Figure 4.  Time series of an objectively determined MOI (solid curve) during 1981−2020. For comparison, the time series of the integration index (dotted curve) is also shown.

For comparison with Fig. 2, we examine the temporal evolution patterns of precipitation and 1000-hPa relative humidity and temperature anomalies the early and late onset groups. As seen from Fig. 5, there is a clear anomalous rain band along the LYRB throughout June for the early onset group. The LYRB appears warmer and moister since the middle June. For the late onset group, on the other hand, dry weather persists in the region throughout June, and surface temperature and relative humidity are lower than the climatology. The daily evolutions of these variables averaged over the key analysis domain for the early and late onset groups are shown in Fig. 3, for comparison.

Figure 5.  Temporally evolving horizontal patterns of (upper) precipitation (shaded; mm day⁻¹) and (below) 1000-hPa relative humidity (shaded; %) and temperature (contour;°C) anomaly fields for the (a) early and (b) late Meiyu onset group. The mean climatology at each 10-day interval has been removed.

Figure 5.  (Continued).

In this section, we intend to reveal atmospheric circulation and SST conditions associated with the early and late onset group. A composite analysis was carried out. The early and late onset groups are defined based on a half of the standard division of the objectively determined MOI shown in Fig. 4.

Figure 6 illustrates the patterns of anomalous precipitation, 850-hPa wind, and SST fields associated with the early and late onset groups. In the early onset composite, a negative SSTA appears in the preceding winter, and accompanied with the cold SSTA and suppressed rainfall in central equatorial Pacific, and enhanced rainfall in the Maritime Continent. In the preceding spring, as the cold SSTA persists, pronounced easterly anomalies appear in the equatorial western and central Pacific. The anomalous easterlies turn northward, leading enhanced northward moisture transport and a positive rainfall anomaly near Philippines. In June, a Pacific–Japan (PJ)-like rainfall pattern (Nitta, 1987; Wang et al., 2000; Lin et al., 2018; Li et al., 2020) appears in East Asia, with enhanced (suppressed) convection near Philippine (Taiwan). An anticyclonic circulation anomaly is stimulated by the local suppressed convection in Taiwan. Southerly anomalies to the west of the anomalous anticyclone transport warm and moist air northward, leading to earlier onset of the Meiyu in the LYRB. A nearly mirror image of the circulation patterns appears in the late onset composite.

Figure 6.  Patterns of anomalous precipitation (shaded; mm day⁻¹), 850-hPa wind (vector; m s⁻¹), and SST (contour;°C) fields from DJF to June for the early Meiyu onset group (left) and the late Meiyu onset group (right). The green (red) dots indicated the precipitation (SST) exceeding 90% confidence level.

The low-level MSE, specific humidity, high-level wind, and geopotential height (GPH) evolution patterns associated with the early and late onset groups are shown in Fig. 7. Consistent with the SSTA feature, a negative MSE and specific humidity center appear in the preceding winter for the early onset group. A marked signal of a negative MSE center appears in the preceding spring just south of the LYRB, elongated from South China to South Japan. As June approaches, a strong moist zone is set up along the LYRB, extending eastward. The establishment of the large positive MSE center signifies the early onset of the Meiyu in the region. Again, a mirror image of the moisture and MSE patterns can be seen clearly in the late onset composite.

Figure 7.  Evolutions of 1000-hPa MSE (shaded; J kg⁻¹), specific humidity (green and brown contour; g kg⁻¹), 200-hPa wind (vector; m s⁻¹), and 200-hPa geopotential height (red and blue contour; gpm) anomaly fields in the early (left) and late (right) Meiyu onset groups based on the objective MOI. The red, green, and purple dots indicate the area exceeding 90% confidence level for the MSE, specific humidity, and GPH fields, respectively.

An interesting feature is the westward migration of a positive specific humidity anomaly from subtropical Pacific in the preceding winter to East Asian coast in June in the early onset groups (left panel of Fig. 7). Accompanied with this westward propagation is the westward movement of southerly wind anomaly in subtropical Pacific induced by La Niña (left panel of Fig. 6). It is likely that such a westward moisture migration may be partially responsible for the increase of local moisture over the LYRB. Another interesting feature is the contrast of a negative and a positive geopotential height center in high-latitudes (beyond 60°N) in the upper troposphere between the early and late onset groups (Fig. 7). This is accompanied with a weakened (strengthened) Aleutian low and a strengthened (weakened) westerly jet along 60°N in the early (late) onset group. The southern branch of the circulation pattern is a part of the Pacific–North America teleconnection caused by tropical forcing associated with La Niña (El Niño). The early signals in preceding winter may be used as predictors for constructing a statistical model for the Meiyu onset forecast.

In this section, we intend to construct a statistical model to predict the Meiyu onset. A lagged regression analysis to the time series of the current objective MOI was performed for many meteorological variables in the preceding months from December to May. Considering the independence among predictors, we finally come up with three predictors, as shown in Fig. 8. The first one is 200-hPa geopotential height anomaly averaged in JFM over a high-latitude region (i.e., the red box in Fig. 8a). This reflects the high-latitude circulation feature. The second is 1000-hPa relative humidity averaged in NDJ in the subtropical North Pacific (i.e., the blue box in Fig. 8b). This reflects subtropical moisture feature. The third is time tendency of 1000-hPa temperature anomaly between April–May and March–April averaged over the two red boxes shown in Fig. 8c, which reflects the ENSO development during northern spring.

Figure 8.  Correlation coefficients, with the objectively derived MOI, of (a) 200-hPa geopotential height anomaly field averaged in JFM, (b) 1000-hPa relative humidity anomaly field in NDJ, and (c) difference of 1000-hPa temperature anomaly between April–May and March–April. Dots indicate the correlation coefficient exceeding 90% confidence level.

Table 3 shows the correlation coefficients between the three predictors and the MOI, as well as correlation coefficients among the three predictors. As one can see, the maximum correlation occurs between the near-surface temperature tendency in the equatorial Pacific and MOI, followed by 1000-hPa relative humidity in preceding NDJ over the subtropical Pacific. The inter relations among the three predictors, on the other hand, are statistically insignificant, suggesting that they are relatively independent.

A linear regression model is constructed with the MOI as predictand. For the data training period of 1980–2010, the following linear regression model is derived:

where Y denotes normalized predictand, and X₁, X₂, and X₃ denote the normalized predictors listed in Table 3.

Figure 9 shows both the reconstruction time series and independent forecast result for the period of 2011–2020. For the training period (1980–2010), the correlation coefficient between the constructed time series (red curve) and the objective MOI (black curve) is 0.68, which is significant at a 99% confidence level. If one roughly separates the Meiyu onset onto three categories, early, normal, and late onset, then the hitting rate is 60%.

Figure 9.  Normalized Meiyu onset date time series of the MOI (black), reconstructed (before 2010), and forecasted (after 2010) MOI (red) during 1981–2020.

Using the model derived from the training period, we further conduct an independent forecast for the period of 2011–2020. The correlation between the objective MOI and the forecasted time series during the independent forecast period is 0.64, which exceeds the 95% confidence level. The hitting rate for the independent prediction period also reaches 60%.

In this study, an objective method is proposed to determine the Meiyu onset date over the LYRB. This method is derived based on the large-scale atmospheric conditions prior to and during the onset period. Two key variables are selected. They are 1000-hPa relative humidity and 1000-hPa temperature averaged over the LYRB. While the mean onset date is 20 June, the MOI shows a strong interannual fluctuation. The resultant MOI is in good agreement with an area-weighted integration index derived from local onset dates determined subjectively by local climate centers over the LYRB region.

A composite analysis was conducted to illustrate the large-scale atmospheric and SST conditions associated with the early and late onset. For the early onset, the most pronounced signal is the SSTA in the preceding winter and spring in the equatorial Pacific. The La Niña-like SSTA induces an anticyclone circulation with pronounced easterlies in the equatorial Pacific and southerlies in the western North Pacific. The southerlies advect higher mean moisture northward, leading to enhanced convection near Philippines. As summer arrives, a PJ-like pattern occurs in East Asian with a positive precipitation anomaly in Philippines and a negative precipitation anomaly and an anticyclonic flow near Taiwan. Southerly anomalies associated with the anticyclone transport warm and moist air northward into the LYRB, leading to the Meiyu onset earlier. An opposite circulation pattern appears in the late onset composite.

A linear regression model was further constructed to predict the MOI with three predictors in preceding months. In addition to the tendency of SSTA in the equatorial region in preceding spring, additional two predictors that are independent to the SSTA tendency are low-level relative humidity in the subtropical Pacific and upper-tropospheric geopotential height in the Arctic region. The linear regression model reproduces well the MOI during the training period (1981–2010) and performs reasonably well during a 10-yr independent forecast period (2011–2020).

The proposed new index can be easily implemented for real-time detection of the Meiyu onset over the LYRB region in national and regional operational forecast centers. The linear regression model developed in the current study can be readily implemented for real-time forecast. It is worth mentioning that a further improvement of the Meiyu onset index is possible. For example, one may consider more variables such as daily precipitation criterion or a larger area in defining the Meiyu onset. The linear regression model can be further improved by considering additional parameters such as land surface moisture and snow cover fields. It will be interesting to compare the statistical model forecast skill to that from a dynamic climate model.