Effects of Low-Frequency Oscillation at Different Latitudes on Summer Precipitation in Flood and Drought Years in Southern China

Abstract

Based on the daily precipitation data from 753 meteorological stations provided by the National Meteorological Information Center (China) and the daily reanalysis data from NCEP/NCAR and ERA5 during the period from 1980 to 2020, the low-frequency (LF) precipitation characteristics of the typical summer flood and drought years in southern China and their relation to the LF atmospheric circulation at different latitudes are compared and analyzed, and extended-range forecasting signals are given. The results show that: (a) In both flood and drought years, summer precipitation in southern China is controlled by 10–20 day oscillation (quasi-biweekly oscillation, QBWO); (b) LF convection is active in southern China in both flood and drought years, but the convective center is southward in flood years, and the vertical meridional circulation is stronger. The key circulation systems of 500 hPa LF height field in flood and drought years include LF “two ridges and one trough” and LF “+”, “−”, “+” East Asia Pacific (EAP) teleconnection wave train in mid-high latitudes of Eurasia. However, the “two ridges and one trough” in flood years are more westward and meridional than in drought years, and the LF Subtropical High is stronger and more extensive, with more significant westward extension; (c) In flood (drought) years, there is northerly and then westerly (central westerly) dry-cold, northeasterly wet-cold, southwesterly (none), and southeasterly (including southerly across the equator) wet-warm water vapor channels. The sources of dry and wet cold air in flood (drought) years are located near Novaya Zemlya (the eastern West Siberian Plain), the Yellow Sea, and the Bohai Sea (Sea of Japan). Additionally, the sources of wet-warm water vapor include the Arabian Sea, the Bay of Bengal, the western Pacific Ocean, and the sea area of northeastern Australia (the western Pacific Ocean and the northern sea area of Australia); and (d) The LF predictive signals of outgoing longwave radiation (OLR) appear on −11 days, while the signals of the 500 hPa height field are on −9 days. There are both westward and eastward propagation predictive signals in flood years, and only westward spread signals in drought years.

1. Introduction

Atmospheric low-frequency oscillation (LFO) is prevalent in the global atmosphere and has a significant impact on global climate [1,2]. It mainly includes quasi-biweekly oscillation (QBWO) of 10–20 days and intraseasonal oscillation (ISO) of 30–60 days [3]. It is an important component of the large-scale circulation anomaly [4] and is closely related to the precipitation anomaly [5]. As China is located in the East Asian monsoon region, the effect of atmospheric LFO on precipitation over southern China, including the middle and lower reaches of the Yangtze River, South China, and Southwest China, is significant. Summer is one of the periods when precipitation is most concentrated in southern China. Therefore, flood disasters occur frequently, causing particularly prominent economic losses and serious social impacts [6,7]. Since floods in southern China are largely associated with LFO, many scholars have conducted in-depth studies on the effects of LFO on precipitation in various regions of southern China.

The LFO of summer precipitation in China is regulated by low-frequency (LF) signals from the tropical and mid-high latitude atmospheric circulation [8,9,10,11,12]. Lau et al. [13] pointed out the existence of significant LFO of 15–20 days in summer precipitation in eastern China, and many scholars have confirmed this conclusion through case studies [14,15,16]. In eastern China, the northward propagation of the LF signals at the tropical latitudes cooperates with the southward propagation of the LF wave trains at the mid-high latitudes in summer. Therefore, the above-normal convection activity, anomalous water vapor convergence, upward motion and anomalous high-level divergence are anchored over eastern China, contributing to the positive phase of ISO precipitation locally [17,18,19,20]. Summer precipitation over the middle and lower reaches of the Yangtze River has an obvious LFO period of 20–50 days [21,22]. The tropical western Pacific convective anomaly corresponds significantly to the Rossby wave trains (i.e., East Asia Pacific teleconnection wave train, EAP), which are distributed meridionally from the tropics to middle latitudes, leading to an increase in LF precipitation of 20–50 days in the Yangtze-Huaihe River Valley and the region in the south of the Yangtze River, which enhances the possibility of the occurrence of persistent extreme precipitation events. [23,24,25,26]. The summer precipitation in South China shows a significant LFO of 10–20 days [27,28]. The LF circulation at the mid-high latitude can influence the LF circulation over South China by the dispersion of the Rossby wave along the LF wave train. The westerly winds from high latitudes move southward, bringing dry-cold air and working together with the warm-wet air currents at low latitudes to produce precipitation [28,29,30]. The persistent extreme precipitation in Southwest China is concentrated in summer and characterized by LFO of 15–60 days. During precipitation, LF systems in each height layer and each latitude cooperate with each other in three-dimensional space to form a LF circulation, which is favorable for precipitation [31,32].

Some scholars have also focused on the similarities and differences of LF circulation systems affecting LF precipitation in southern flood seasons during drought and flood years. For example, in the middle and lower Yangtze River, the northward propagation of summer intraseasonal oscillation is stronger in drought years than in flood years, reaching near 50° N. In flood years, both the northward propagation from the low latitudes and the weak southward propagation from the mid-high latitudes are obvious, converging in the Yangtze River basin to form a strong center of oscillation [33]. Tong et al. [34] researched the LFO characteristics of precipitation and convection in drought, flood and drought-flood-coexistence years over the Yangtze-Huaihe basin. As regards the propagation of LF convection, it is found that the 8–16 day (16–32 day) LF convection mostly propagates in the southward (westward) direction in typical drought (flood) years. In the eastern region of southwest China, the meridional and zonal propagation of outgoing longwave radiation (OLR) 40-day LF convection in the summer of typical flood years is strengthened, and the LF convection is stronger than normal, which causes more precipitation, while in the summer of typical dry years, the meridional and zonal propagation of OLR 40-day LF convection is weakened, and the LF convection is weaker than normal, which causes less precipitation [35]. In South China, there are differences in the sources of cold and warm LF vapor affecting 10–20-day LF precipitation in flood and drought years of the pre-flood period. For example, during both flood and drought years, the primary LF water vapor sources include the south side of Lake Baikal, the northern Sea of Japan and the Yellow Sea, the South China Sea and the western Pacific Ocean, and the southern Sea of Japan is another water vapor source in flood years. In addition, there are two southwest warm water vapor flows and one southeast cross-equatorial water vapor flows in flood years. While in drought years, there is only one relatively weak warm water vapor that flows in the southwest direction [36].

In summary, there have been many studies on the LF characteristics of summer precipitation in various regions of southern China and the influence of atmospheric LFO at the tropics and mid-high latitudes on them, but most research focused on the flood events. There are many studies focusing on the effects of atmospheric LFO at different latitudes on LF summer precipitation, while the comparative analysis of the differences of these effects in flood years and drought years is insufficient. Thus, this paper takes southern China as the whole research area, contrasts and analyzes the LF characteristics of the summer precipitation in flood and drought years, further reveals the LF characteristics of atmospheric circulation that affect LF precipitation at different latitudes, searches for the extended range forecasting signals, and provides meaningful reference information for the prediction of summer precipitation in southern China.

2. Data and Methods

2.1. Data

The data used in this study include (a) daily precipitation data recorded at 753 stations provided by the National Meteorological Information Center of China (http://data.cma.cn/ (accessed on 16 May 2022)), which contains daily observation data of precipitation from China’s basic, base, and general meteorological stations (this dataset, the Daily meteorological dataset of basic meteorological elements of China National Surface Weather Station V3.0, has undergone a more stringent quality control, and suspicious and erroneous data have been manually verified and corrected by the National Meteorological Information Center of China. The integrity and quality of the data are great.), (b) global daily reanalysis data of wind field, geopotential height field, etc., provided by the United States National Centers for Environmental Prediction/National Center for Atmospheric Research (NCEP/NCAR), which is using a state-of-the-art analysis/forecast system to perform data assimilation using past data from 1948 to the present, [37] (https://psl.noaa.gov/data/gridded/data.ncep.reanalysis.html (accessed on 16 May 2022)), (c) ERA5 daily reanalysis data of column-integrated water vapor flux and the corresponding divergence field, provided by the European Centre for Medium-Range Weather Forecasts (ECMWF), which combines model data with observations from across the world into a globally complete and consistent dataset using the laws of physics [38] (https://cds.climate.copernicus.eu/cdsapp#!/search?type=dataset (accessed on 16 May 2022)), and (d) OLR data from the National Oceanic and Atmospheric Administration (NOAA), with gaps then filled with temporal and spatial interpolation [39] (https://www.psl.noaa.gov/data/gridded/data.olrcdr.interp.html (accessed on 16 May 2022)). The data are all for 1980–2020. Due to the fact that the Chinese observation stations were built at different times, for the study purpose of this paper and to ensure the completeness and continuity of the data, we excluded the stations with less than 41 years of the length of the daily precipitation data from the 753 observation stations in China. Referring to Zhi et al. [40] and Fu and Bueh [41], southern China is defined as the region of 105–125° E, 20–35° N, within which there are 206 observation stations (Figure 1). Except for the daily precipitation data recorded at 753 stations (China), the horizontal resolution of the above data is 2.5° × 2.5°.

In addition, there was a significant interdecadal turn in the coupled atmosphere-ocean system in the middle and late 1970s [42,43]. This paper focuses on the interannual variation of LFO rather than the interdecadal variation. In order to exclude the possible influence of the interdecadal background on the interannual anomalies of atmospheric LFO in the middle and late 1970s, this paper mainly investigates the LF characteristics of summer precipitation in southern China from 1980–2020.

2.2. Methods

2.2.1. The Morlet Wavelet Analysis Method

The Morlet wavelet analysis method [44] is performed on the summer precipitation series in flood and drought years to reveal their significant LF periods, respectively. This technique is suitable for non-stationary processes and can yield localized time–frequency information that is not available from the traditional Fourier transform. The wavelet mother function is defined as

$$\phi \left(t\right)=e^{i2\pi t}{e}^{-\frac{t^2}{2}}.$$

(1)

When $x\left(t\right)$ is the square-integrable function (i.e., $x\left(t\right)\in L^2$), which is similar to the Fourier transform), the continuous wavelet transform is defined as

$$W\left(a,b\right)=\left(x,{\phi }_{a,b}\right)=a^{-\frac{1}{2}}{{\displaystyle \int }}_{-\infty }^{\infty }x\left(t\right){\phi }^{\ast }\left(\frac{t-b}{a}\right)dt,$$

(2)

where $a$ and $b$ correspond to the scale and the time shift parameters of the analyzing wavelet, respectively. $\ast$ denotes the complex conjugate.

Referring to the EMC (Empirical Monte Carlo) method used by Wang et al. [45], a significance test at α = 0.05 level is performed on the wavelet power spectrum.

2.2.2. The Butterworth Band-Pass Filter

The Butterworth Band-pass Filter [46] is used to extract the LF components for 10–20 days in the precipitation series. It is suitable for short data series. Its advantage is that the band-pass frequency can be freely selected, and the data will not be lost on both sides after the data are filtered. The original data series is $x\left(t\right)$, and the filtered one is $y\left(t\right)$. The filter formula is

$$y\left(t\right)=a\left(x\left(t\right)-x\left(t-2\right)\right)-b_{1}y\left(t-1\right)-b_{2}y\left(t-2\right).$$

(3)

The frequency response function is defined as

$$W\left(z\right)=\frac{a\left(1-z^2\right)}{1+b_{1}z+b_2{z}^2},z=e^{-2\pi if∆},$$

(4)

The coefficients $a,b_1\mathrm{and}{b}_2$ are calculated from the center frequency and the frequency values on both sides of the frequency band.

2.2.3. The Composite Analysis Method

The filtered LF precipitation series is divided into eight phases [47,48], and the corresponding atmospheric circulation elements are phase composited [49] (see the first paragraph of Section 3.2 for details). The t-test method is used to test for significance, and the formula of the t-test is

$$t=\frac{\overline{x}-\overline{y}}{\sqrt{\frac{S_1^2}{n_1}+\frac{S_2^2}{n_2}}},$$

(5)

where $\overline{x},s_1\mathrm{and}{n}_1$ represent the mean, standard deviation, and sample size of LF atmospheric circulation elements in certain phases, respectively; $\overline{y},s_2\mathrm{and}{n}_2$ represent the mean, standard deviation, and sample size of LF atmospheric circulation elements in summer for multiple flood (drought) years, respectively.

2.2.4. The Inverse Algorithm

The atmospheric apparent heat source $Q_1$ is estimated by an “inverse algorithm” from Yanai et al. [50]

$$Q_1=C_p\left[\frac{\partial T}{\partial t}+V·\nabla T+{\left(\frac{p}{p_0}\right)}^k\omega \frac{\partial \theta }{\partial p}\right].$$

(6)

The whole vertical integration is $\langle Q_1\rangle$:

$$\langle Q_1\rangle =\frac{1}{g}{{\displaystyle \int }}_{p_t}^{p_s}{Q}_{1}dp,$$

(7)

where $T$ is temperature; $V$ denotes the horizontal wind vector; $\omega$ denotes the vertical component of velocity in $p$ coordinates; $\theta$ is potential temperature; $k=\frac{R}{C_p}$ ($R$ and $C_p$ represent the gas constant of dry air and specific heat at constant pressure, respectively); $p_s$ denotes the surface pressure, and $p_t$ denotes 100 hPa.

2.2.5. The Lead–Lag Correlation Method

The lead–lag correlation method [51] is used to obtain the early predictive signal of LF precipitation.

$$r_{xy}\left(j\right)=\frac{\frac{1}{n-j}{{\displaystyle \sum }}_{t=1}^{n-j}\left(x_t-\overline{x}\right)\left(y_{t+j}-\overline{y}\right)}{{\mathrm{s}}_x{\mathrm{s}}_y},$$

(8)

where $\overline{x}$ and ${\mathrm{s}}_x$ represent the mean and standard deviation of LF atmospheric circulation elements (sizes = $n$), respectively; $\overline{y}$ and ${\mathrm{s}}_y$ represent the mean and standard deviation of LF precipitation (sizes = $n$), respectively; $r_{xy}\left(j\right)$ is the lead–lag correlation coefficient with a time interval length of $j$.

2.2.6. The Calculation Method for the Water Vapor Budget

The water vapor income at a certain boundary is

$$WVI\left(k\right)={\left(-1\right)}^{k+1}{Q}_k{l}_k.$$

(9)

The net water vapor budget is

$$NWVB={{\displaystyle \sum }}_{k=1}^4{\left(-1\right)}^{k+1}{Q}_k{l}_k,$$

(10)

where $k$ = 1, 2, 3, and 4 represent the western, eastern, southern, and northern boundaries, respectively. $WVI\left(k\right)$ and $NWVB$ represent the water vapor income at the kth boundary and the net water vapor budget at the four boundaries, respectively. $Q_k$ is the average column-integrated water vapor flux at the kth boundary, and $l_k$ is the length of the kth boundary.

2.2.7. The Effective Degrees of Freedom

Due to the increased auto-correlation of the filtered data, the effective degrees of freedom are recalculated in the significance t-test of their lead–lag correlation coefficients [52,53,54].

$$EDoF=\frac{N}{{{\displaystyle \sum }}_{i=-\infty }^{\infty }{R}_{XX}\left(i\right)R_{YY}\left(i\right)},$$

(11)

where $EDoF$ is the effective degree of freedom of the sample; $N$ is the sample size; $R_{XX}$ and $R_{YY}$ are the auto-correlation coefficients of LF atmospheric circulation elements and LF precipitation, respectively.

3. Results

3.1. Comparison of the LF Characteristics of Summer Precipitation in Flood and Drought Years in Southern China

3.1.1. Selection of the Typical Flood and Drought Years

Figure 2 shows the normalized time series of the 206-station regional mean total summer precipitation in southern China from 1980 to 2020. It can be seen that there are obvious interannual and interdecadal variations in summer precipitation in southern China, among which the precipitation is below average from 1981 to 1992 and above normal from 1993 to 2002, and the interdecadal change is weakened after 2003, which is similar to the conclusions of Wu et al. [55] and Xu et al. [56]. Taking ±1 as the standard, ten flood years (1993, 1994, 1995, 1996, 1998, 1999, 2002, 2008, 2017, and 2020) and nine drought years (1981, 1985, 1988, 1989, 1990, 1992, 2003, 2004, and 2013) are selected as typical years. Accordingly, the differences in precipitation characteristics, such as significant LF periods, between flood and drought years in southern China are compared and analyzed.

3.1.2. Selection of Representative Station

In order to find representative stations characterizing LF precipitation, the spatial deviation distribution of the annual mean total summer precipitation over 10 (9) flood (drought) years in southern China is given (Figure 3). The spatial deviation of precipitation is defined as the difference between the total summer precipitation at a single station and the 206-station regional mean total summer precipitation. It can be seen that the the spatial deviation of the total summer precipitation in flood years is greater than that in drought years. The maximum value centers of the precipitation deviation in flood years are located at the border of Guangdong and Guangxi, the junctions of Jiangxi, Anhui, and adjacent provinces, respectively. (Figure 3a). The locations of the maximum value centers of drought years are similar to those of flood years, but the precipitation deviation is also large at the border of Zhejiang and Fujian, the border of Fujian and Guangdong, and northeast Sichuan (Figure 3b). In total, 47 (deviation values ≥ 100 mm) and 50 (deviation values ≥ 40 mm) stations are selected as representative stations in flood and drought years, respectively, to further compare the LF characteristics of precipitation in flood and drought years.

3.1.3. Comparison of LF Significant Periods of Precipitation in Flood and Drought Years

The Morlet wavelet analysis is performed on the representative-station regional mean daily precipitation series in each year of flood and drought years, respectively, and the significant LF periods are counted in

Table 1. The significant low-frequency (LF) period (significant at α = 0.05 level) and the calendar dates of phase 3 in each year of flood and drought years.

Flood Years			Drought Years
Year	LF Period (d)	The Dates of Phase 3 (Number: 35)	Year	LF Period (d)	The Dates of Phase 3 (Number: 36)
1993	10–20	18 June, 30 June, 20 July	1981	10–20, 30–60	11 June, 29 June, 11 July, 24 July, 11 August
1994	10–20, 30–60	9 June, 25 June, 5 July, 6 August	1985	10–20	25 June, 5 July, 12 August
1995	10–20	24 July, 3 August	1988	10–20, 30–60	28 June, 12 July, 28 July, 8 August
1996	10–20, 20–30	2 July, 15 July, 2 August, 17 August	1989	10–20	17 June, 30 June, 11 July, 26 July
1998	10–20	23 June, 3 July, 23 July	1990	10–20, 20–30	17 June, 1 July, 14 July, 1 August, 21 August
1999	10–20	8 June, 24 June, 16 July, 13 August	1992	10–20	5 July, 22 July
2002	10–20, 20–30	11 June, 21 July, 7 August	2003	10–20, 20–30	10 June, 25 June, 9 July, 24 July
2008	10–20, 20–30	13 June, 27 June, 8 July, 21 July, 15 August	2004	10–20, 60–90	24 June, 6 July, 19 July, 31 July, 14 August
2017	10–20	14 June, 18 July, 1 August, 3 August	2013	10–20	8 June, 25 June, 19 July, 2 August
2020	10–20, 60–90	8 July, 19 July, 5 August

. It is found that all flood and drought years have a significant LF period of 10–20 days. Figure 4 shows the wavelet analysis results of the representative-station regional mean of the daily precipitation series from January to December in 1995 (flood year) and 1992 (drought year) in southern China. In both 1995 and 1992, there is a significant 10–20-day LF period of precipitation series (Figure 4a,c). The LF components of 10–20 days (Figure 4b,d) reflect the intensity change of precipitation in each process. The precipitation is more (weaker) in the positive (negative) phase of 10–20-day LF components, which is similar to the findings of Li et al. [27].

In summary, in both flood and drought years, summer precipitation is most significant in the LF period of 10–20 days in southern China. Therefore, the following discussion focuses on the characteristics of atmospheric LFO of 10–20 days in flood and drought years and explores the correlation between the LF precipitation of 10–20 days and the LFO of 10–20 days.

3.2. Comparison of LF Characteristics of Atmospheric Circulation Affecting LF Precipitation in Flood and Drought Years

The LFO of precipitation is deeply affected by the LF signal of atmospheric circulation in low latitudes and mid-high latitudes [10,11,57]. How do the LF characteristics of atmospheric circulation in low latitudes and mid-high latitudes relate to summer LF precipitation in southern China differ in flood and drought years? Focusing on this problem, the following will start with the atmospheric element fields (such as OLR, wind field, height field, water vapor transport, etc.) that can reflect the LF signals at different latitudes and select the complete LF precipitation processes with peak and valley absolute values greater than 0.8 standard deviations according to the summer daily representative-station regional mean LF precipitation series of 10–20 days in each year of 10 (9) typical flood (drought) years, respectively. The selected precipitation processes are divided into phases 1–8, in which phase 3 (7) corresponds to the peak (valley) of the wet (dry) phase, and phase 1 (5) represents the transition of the oscillation from the dry (wet) phase to the wet (dry) phase, and phases 2 and 4 (6 and 8) represent half of the peak (valley) values. For example, it can be seen from Figure 4b,d that there are two complete processes of 10–20 days in 1995 and 1992, respectively. Phase 3 in 1995 (1992) is on 24 July and 3 August (5 July and 22 July). The dates of phase 3 in each year of flood and drought are summarized in Table 1. There are 35 and 36 phase 3 in flood and drought years, respectively. The corresponding LF element composited field (i.e., average field in a certain phase of 35 or 36 processes in flood and drought years, respectively) in eight phases can reflect the evolution of LF circulation characteristics during the development of LF precipitation.

The development and maintenance of the East Asian monsoon largely depend on the distribution of heat sources and heat sinks in Asia and nearby regions. Therefore, the LF precipitation in southern China affected by the monsoon is closely related to the distribution of atmospheric apparent heat sources [58,59]. According to the inverse algorithm in Section 2.2.4, the whole-layer integrated atmospheric apparent heat source <Q₁> is calculated, and then according to the composite analysis method in Section 2.2.3, the OLR, wind field, height field, water vapor transport, <Q₁>, etc. of each phase in flood and drought years is composited. Due to the limitation of space, this paper will focus on the atmospheric LFO characteristics of LF precipitation at the peak of phase 3 and compare their similarities and differences in flood and drought years.

3.2.1. OLR

OLR is the outgoing longwave radiation emitted by the earth-atmosphere system, which can accurately describe tropical weather systems that are rarely observed on the earth and sea surfaces. It can also reflect information such as the strength of convective development and large-scale vertical motion [35]. In this subsection, combined with the distribution of the divergent wind fields in the upper (150 hPa) and lower (850 hPa) troposphere and <Q₁>, the similarities and differences in the characteristics of the LF OLR fields of 10–20 days (Figure 5) at the LF precipitation peak of phase 3 are compared and discussed in flood and drought years.

As seen in Figure 5, in phase 3, no matter if it is in flood years (Figure 5a,c) or drought years (Figure 5b,d), (a) there is strong convective development in southern China, accompanied by convergence (divergence) in the lower (upper) troposphere, which is conducive to the increase in LF precipitation, but the convection active center in flood years is further southward than that in drought years; (b) in the south of the convection active region, there is an east–west (from the Bay of Bengal to the sea area east of the Philippine Islands) and a northeast–southwest (from the South China Sea to the tropical northwest Pacific Ocean) convection-inhibitive region in flood and drought years, respectively, but the convection-inhibitive region is further westward (further eastward) in flood (drought) years; (c) are the convection-active region in southern China and the convection-inhibitive region in the South China Sea belong to the same vertical circulation system? To address this problem, the latitude-height vertical profiles and LF vertical-meridional circulation of the LF divergence fields of 10–20 days along 110° E (Figure 5e) and 115° E (Figure 5f) are given for flood and drought years, respectively. It can be seen that in flood years, most of the water vapor in southern Chin heads south to the low latitudes, converges, and sinks after rising to the upper troposphere (Figure 5e). In drought years, while the water vapor in southern China converges and rises, a considerable component turns northward (Figure 5f), which means that the vertical meridional circulation is stronger in flood years than in drought years, and the configuration of the LF circulation system in the upper and lower troposphere is also more baroclinic. It is more conducive to the release of unstable energy [60] and the occurrence and maintenance of more LF precipitation.

Furthermore, it is also noted that the high-value area of <Q₁> in southern China basically corresponds to those of convection and precipitation (Figure 3a,b) in both flood (Figure 5a,c) and drought (Figure 5b,d) years. This may be related to the small sensible heat and evaporation of the ground during a rainstorm, and the net heating of the atmosphere caused by the latent heat released by precipitation [61]. Gill [62] shows that the upper (lower) level on the northwest side of the anomalous heating source excites anticyclonic (cyclonic) circulation, which strengthens and sustains the South Asian High (SAH) at the upper level. Southern China is located in the ascending motion area in the south of the upper air jet stream entrance area in the north of the SAH, which is conducive to the occurrence and maintenance of heavy precipitation [63]. This is also confirmed by the previous analysis of lower tropospheric convergence and upper tropospheric divergence in southern China.

In short, in both flood and drought years, at the peak of phase 3 of LF precipitation, there is a closed LF vertical meridional circulation in southern China and its south, which means that the water vapor continues to converge and rise in southern China and is conducive to the maintenance of LF precipitation. However, in flood years, the convective center is southward, and the vertical meridional circulation is stronger, which means that the LF precipitation is more abundant.

3.2.2. 500 hPa Height and Wind Field

The continuous and stable high latitude circulation at 500 hPa provides the cold air required to ensure the continuous and stable maintenance of summer precipitation in southern China. In southern China, the warm-wet airflow in the west of the Western Pacific Subtropical High (WPSH) intersects with the cold air from the north, which results in precipitation [64]. Therefore, it is indicative to study the phase change of the 500 hPa LF height field with time during summer in southern China.

According to the 500 hPa height and wind field in flood and drought years composited by phase 3 of LF precipitation (Figure 6), on the one hand, no matter if it is in flood years or drought years, a LF “+”, “−”, and “+” East Asia-Pacific (EAP) teleconnection wave train anomaly is shown along the East Asian coast from south to north. Incidentally, the LF circulation in the mid-high latitudes of Eurasia is a “+”, “−”, and “+” anomalous wave train, forming a “two ridges and one trough” pattern. The positive anomaly in the north of the former wave train overlaps with the positive anomaly in the east of the latter wave train. On the other hand, there are obvious differences in the shape, position, and intensity of the circulation systems in the two wave trains in flood and drought years. For instance, the positive anomaly in the south of the EAP wave train is more extensive and extends westward than in drought years, causing the WPSH to become stronger and more obvious westward in flood years. It is noteworthy that in Figure 5a–d, the area near the Philippine Islands is an apparent heat sink region, which may also be one of the reasons why the WPSH is maintained at 20° N [63]. In flood years, the “two ridges and one trough” are generally more westward than in drought years, and the central through and eastern ridge have a greater meridional extent than in drought years.

The differences between the above LF circulation systems in flood and drought years lead to obvious differences in the intensity and source of LF water vapor and cold air. From one aspect, warm-wet airflow takes different paths. The WPSH (Figure 6a) in flood years is stronger than that in drought years (Figure 6b), with a larger range and more obvious westward extension, which can transport more warm-wet airflow (southwest and southeast warm-wet airflow) from the tropical north Pacific to the Bay of Bengal to the southern region. In drought years, the WPSH is mainly located in the tropical western Pacific, which transports a relatively small range of warm-wet airflow to southern China (southeast warm-wet airflow) from the South China Sea and the tropical northwest Pacific. From another, there are obvious differences in the paths of dry and wet cold air. In flood years (Figure 6a), the LF trough to the west of Lake Baikal and the LF trough to the east of the Caspian Sea are superimposed north and south, bringing the cold air from the vicinity and west of Novaya Zemlya to southern China via the east of the West Siberian Plain, the east of the Caspian Sea, the Iran plateau, the north of the Indian Peninsula, and the southwest of China (first northerly and then westerly cold air). This cold air also intersects with the cold-wet air (cold-wet airflow in the northeast direction) near the Yellow Sea and Bohai Sea brought by the LF southern cyclone. In drought years (Figure 6b), the LF trough above and to the west of Lake Baikal and the LF trough in the southern region merge into a northwest–southeast stepped trough so that the cold air flows southward to southern China via the east of the West Siberian Plain, the east of Lake Balkhash, and central and western China (cold air in the central west direction), and intersects with cold-wet water vapor from the southern Japan Sea brought by the LF southern cyclone (cold-wet air in the northeast direction).

In conclusion, the main LF circulation systems in the middle troposphere affecting the LF precipitation in flood and drought years are the “+”, “−”, and “+” EAP teleconnection wave train along the East Asian coast and the “two ridges and one trough” in the mid-high latitudes of Eurasia. The differences in the location and intensity of these LF circulation systems in flood and drought years lead to significant differences in the paths of cold and warm air affecting precipitation in flood and drought years. In flood years, there are northerly and then westerly dry-cold, northeasterly wet-cold, and southwesterly and southeasterly wet-warm airflow. However, in drought years, there are central and westerly dry-cold, northeasterly wet-cold, and southeasterly wet-warm airflow. In addition, dry-cold air and warm and cold wet air are both stronger in flood years than in drought years, and the convergence of cold and warm airflow creates stronger ascending motion (Figure 5e), causing LF precipitation to become more abundant in flood years than in drought years.

3.2.3. Column-Integrated Water Vapor Flux

In Section 3.2.2, the paths of dry and wet airflow affecting LF precipitation in flood and drought years are fundamentally expounded. As a physical quantity combining wind field and water vapor field, water vapor flux can not only reflect the characteristics of atmospheric circulation but also better characterize the distribution of water vapor [65]. The following will verify and supplement the conclusions in Section 3.2.2 by further comparing and analyzing the similarities and differences of LF water vapor transport characteristics in flood and drought years.

According to the LF column-integrated water vapor flux and the corresponding water vapor flux divergence fields at the peak of phase 3 of LF precipitation of 10–20 days (Figure 7), it is clear that the precipitation in southern China is controlled by LF cyclonic water vapor circulation in both flood (Figure 7a) and drought years (Figure 7b), and cold and warm water vapor converge in southern China, which is favorable to the formation of LF precipitation. However, the LF water vapor circulation and water vapor channels that affect LF precipitation are different in flood and drought years. The main manifestations are as follows: (a) There are differences in the northwest cold air paths at high latitudes. The path of dry-cold air in flood years is consistent with what the 500 hPa LF circulation field reflects (Figure 6a), i.e., there is a dry-cold air path, first northerly and then westerly, going southward to southern China through the eastern West Siberian Plain, while it is long and contains less water vapor, and the characteristics reflected by the water vapor flux field are weaker. The cold air transport path from the eastern West Siberian Plain to southern China in drought years is also consistent with the cold air path of the central westerly reflected by the 500 hPa LF circulation field (Figure 6b); (b) there are differences in the cold water vapor flow in the northeast, whose path is consistent with the northeasterly cold-wet path reflected by the LF 500 hPa circulation field (Figure 6). The LF anticyclonic water vapor circulation in the northern part of the LF EAP water vapor wave train in flood years is located in northeastern China. It cooperates with the LF cyclonic water vapor circulation controlling southern China to transport the wet-cold water vapor from the Yellow Sea and the Bohai Sea to southern China. In drought years, the cold-wet water vapor from the Sea of Japan is brought into southern China mainly by the LF cyclonic water vapor circulation located in the area of southern China–Sea of Japan; (c) there are differences in the warm-wet water vapor flow in the southwest direction. In flood years, LF anticyclonic water vapor circulation is present in the Indian Peninsula and Indochina Peninsula, which forces the warm-wet water vapor from the Arabian Sea and the Bay of Bengal to enter southern China, while the circulation does not arise in drought years; (d) there are differences in the warm-wet water vapor flow in the southeast direction. In flood years, the LF anticyclonic water vapor circulation exists in the northeastern Philippine Islands, transporting warm-wet water vapor from the western Pacific Ocean to southern China via Indonesia and the South China Sea. In drought years, the LF anticyclonic circulation near the Philippine Islands is less extensive than that in flood years, transporting warm-wet water vapor from the western Pacific Ocean to southern China via the South China Sea; and (e) there are differences in the cross-equatorial warm-wet water vapor flow, and the cross-equatorial flow is further eastward in flood years. An anomalous LF anticyclonic water vapor circulation exists in the northeastern sea area of Australia during flood years, which brings the warm-wet water vapor from eastern Australia into the water vapor channel of the western Pacific. In drought years, there is an anomalous anticyclonic water vapor circulation in the northwest sea area of Australia, which causes the water vapor from the northern sea area of Australia and the water vapor from the western Pacific Ocean to converge above the Kalimantan Island and move towards southern China. By comparing Figure 6 and Figure 7, it is noted that the water vapor flux field and the 500 hPa LF circulation field reflect the same water vapor source and the water vapor channel.

As demonstrated in

Table 2. Composite of the LF water vapor income at each boundary (10⁶ kg/s), net water vapor budget (10⁶ kg/s), and water vapor channels of southern China based on the peak of phase 3 of LF precipitation of 10–20 days in flood and drought years.

	Flood Years	Drought Years
southern boundary	89.7	92.7
northern boundary	−14.9	33.5
western boundary	21.1	23.4
eastern boundary	−32.0	−87.5
net budget	63.9	62.1
water vapor channel (number)	the northerly and then westerly dry-cold, the northeasterly cold-wet, and the southwesterly and the southeasterly warm-wet channels (4)	the central westerly dry-cold, the northeasterly cold-wet, and the southeasterly warm-wet channels (3)

, the LF water vapor budget at each boundary of southern China (105–125° E, 20–35° N) at the peak of phase 3 of LF precipitation in both flood and drought years is consistent with that reflected by the LF column-integrated water vapor circulation field (Figure 7). The southern (eastern) boundary is the main input (output) boundary for LF water vapor in southern China. However, there are some differences in the LF water vapor budget of each boundary, and the net budget is greater in flood years than in drought years. In flood years, there are two input boundaries (southern and western boundaries) and two output boundaries (northern and eastern boundaries) in southern China, in which the water vapor income of the southern and western boundaries is 89.7 × 10⁶ and 21.1 × 10⁶ kg/s, respectively, and the water vapor expenditure of the northern and eastern boundaries is 14.9 × 10⁶ and 32.0 × 10⁶ kg/s, respectively, with a net water vapor budget of 63.9 × 10⁶ kg/s. In drought years, there are three input boundaries (southern, northern, and western boundaries) and one output boundary (eastern boundary), in which the water vapor income of the southern, northern, and western boundaries is 92.7 × 10⁶, 33.5 × 10⁶, and 23 × 10⁶ kg/s, respectively, and the water vapor expenditure of the eastern boundary is 87.5 × 10⁶ kg/s, with a net water vapor budget of 62.1 × 10⁶ kg/s. In addition, the water vapor channels in flood and drought years are also summarized in Table 2, respectively.

To summarize, under the cooperation of LF cyclonic and anticyclonic water vapor circulation at high and low latitudes, the northerly and then westerly (central westerly) dry-cold, the northeasterly cold-wet, and the southwesterly (none) and the southeasterly warm-wet water vapor channel come into being in flood (drought) years. Among them, the southeasterly warm-wet water vapor channel contains the southeasterly (southerly) cross-equatorial water vapor flow. The source of dry-cold air in flood (drought) years is located near Novaya Zemlya (the eastern West Siberian Plain), the source of wet-cold air lies in the Yellow Sea and Bohai Sea (the Sea of Japan), and the sources of warm-wet water vapor include the Arabian Sea, the Bay of Bengal, the western Pacific Ocean and the northeast sea area of Australia (the western Pacific Ocean and the northern sea area of Australia).

3.3. Predictive Signals of LF Precipitation in Flood and Drought Years

In order to understand the characteristics of LF atmospheric circulation in the early stage that is significantly related to LF precipitation in flood and drought years, and to find the meaningful LF predictive signals at different latitudes, the lead–lag correlation coefficients of LF OLR and the 500 hPa height field and LF precipitation are calculated, respectively. The expression of negative days, such as −11 days, indicates that the date of LF circulation elements is 11 days ahead of LF precipitation.

3.3.1. OLR

Firstly, the lead–lag correlation coefficients distribution of LF OLR 11~0 days ahead of LF precipitation in flood (Figure 8) and drought (Figure 9) years is analyzed. The negative (positive) correlation area corresponds to the LF negative (positive) anomaly of OLR, i.e., the convection-active (-inhibitive) region. Note that the data sample of flood (drought) years is 10 × 92 = 920 (9 × 92 = 828) in the correlation plots. Due to data persistence introduced by temporal filtering, the effective degree of freedom is around 470 out of the total sample size of 920 (828). With 470 independent samples, a correlation value of 0.076 is 90% significant (shaded areas in Figure 8 and Figure 9).

As demonstrated in Figure 8, from −11 to −6 days, the abnormally active LF convective signal in flood years first appears in Somalia and the equatorial Indian Ocean to the east (−11 days). Over time (−8 to −6 days), the convective signal develops and propagates from south of the Indian Peninsula to the Philippine Islands via the Indochina Peninsula along the southwest-northeast direction, and then enters southern China on −4 days. From −3 to 0 days, the LF convective signal moves northward to cover southern China and stays there continuously, leading to the occurrence of heavy LF precipitation.

As can be seen from Figure 9, the abnormally active LF convective signal in drought years appears in the south of the Indochina Peninsula, the South China Sea, and the east of the Philippine Islands on −11 days, intensifies, moves toward southern China from −8 to −6 days, expands into southern China on −4 days, and further intensifies and gradually occupies southern China from −2 to 0 days. The vigorous convective activity is conducive to the occurrence of LF precipitation.

In summary, the LF OLR negative anomaly appearing near Somalia (the south of the Indochina Peninsula) on −11 days in flood (drought) years is an important LF predictive signal reflecting the summer precipitation of southern China in flood (drought) years. Special attention should be paid to the LF negative OLR anomaly which moves to the rain area of southern China in the northeast (north) direction in flood (drought) years.

3.3.2. 500 hPa Height Field

As the 500 hPa LF height field can better reflect the information of the water vapor at mid-low latitudes and of the cold air at mid-high latitudes, the lead–lag correlation coefficients of the LF 500 hPa height field that is 9~0 days earlier than the LF precipitation in flood (Figure 10) and drought (Figure 11) years is calculated and analyzed below. The significant negative (positive) correlation area corresponds to the LF negative (positive) anomaly of the 500 hPa height field. The effective degree of freedom is about 370 at an α = 0.1 significance level, so the critical value of the correlation coefficient is about 0.083 (shaded areas in Figure 10 and Figure 11). It can also be seen from Section 3.2.2 that the key circulation systems of 500 hPa LF height field in flood and drought years include the LF “+”, “−”, and “+” EAP teleconnection wave train along the East Asian coast and the LF “two ridges and one trough” at mid-high latitudes of Eurasia (Figure 6a). Therefore, the predictive signals are searched by focusing on the LF anomalies related to the formation of the two wave trains.

Firstly, the predictive signals related to EAP are found. On −9 days, there is a LF negative anomaly in the north of East Siberia. From −7 to −4 days, the negative anomaly strengthens significantly and gradually splits into northeastern China and the Bering Sea in the southwest and southeast direction, respectively. After that, the two negative anomalies move westward and southward, respectively. On 0 days, the two anomalies form a large northeast-southwest trough (similar to Figure 6a) from the east of the Kamchatka Peninsula to southern China. From −9 to −5 days, the LF positive anomalies in central and eastern China, northeastern China, and the Bering Strait gradually extend southward and strengthen. On −4 days, the positive anomaly, located in the area from the east of the Caspian Sea to the eastern coast of China, meets with the positive anomaly that moves southward to the central North Pacific Ocean and extends westward. Additionally, then the positive anomaly continues moving southward. On 0 days, a large area from the Indian Peninsula to the east of the Philippine Islands is controlled by a LF positive anomaly. At this point, the “+”, “−”, and “+” EAP teleconnection wave train has formed from south to north along the East Asia coast in the South China Sea, southern China, and the area from the east of Lake Baikal to East Siberia, respectively, which is conducive to the deepening of the East Asian Trough and the strengthening of the westward extension of the WPSH.

Subsequently, the predictive signals of “two ridges and one trough” at the mid-high latitudes of Eurasia are searched. As shown in Figure 10, on −9 days, there is an east–west “+”, “−”, and “+” LF teleconnection wave train in the area from Western Europe to the Mediterranean, on the Eastern European Plain, and in the Central Siberia. The wave train keeps intensifying and extending eastward over time. From −4 to −2 days, the original positive anomaly from Western Europe to the Mediterranean Sea and the original positive anomaly of Central Siberia are weakened in the eastward movement. On 0 days, the two positive anomalous signals are strengthened again and form the “+”, “−”, and “+” LF teleconnection wave train in the Ural Mountains, Central Siberia, and the area from Lake Baikal to East Siberia, respectively, which is favorable to the maintenance of the circulation of two ridges and one trough at mid-high latitudes. Combined with the LF EAP teleconnection wave train along the East Asian coast, the cold air from the high latitudes and the wet-warm water vapor from the tropics converge in southern China, causing the LF heavy precipitation.

In conclusion, for the prediction of the EAP teleconnection wave train along the East Asian coast, special attention should be paid to the split and extension of the LF negative anomaly of East Siberia to the southwest and southeast on −9 days, the southward movement of the LF positive anomaly near northeastern and eastern China, as well as the southward and westward movement of the LF positive anomaly near the Bering Sea. For the prediction of the “two ridges and one trough” system, special attention should be paid to the strengthening and eastward movement of the “+”, “−”, and “+” LF teleconnection wave train on the early −9 days in the area from Western Europe to the Mediterranean, on the Eastern European Plain, and in Central Siberia, respectively.

For drought years (Figure 11), on −9 days, there are LF positive anomaly centers in northeastern New Guinea, the north of Central Siberia, and the area from Lake Baikal to eastern China, respectively. LF negative anomaly centers are located near the Sea of Okhotsk and the northwestern Pacific Ocean. These anomaly signals gradually move westward and strengthen over time. It is worth noting that on −3 days, the LF positive anomaly in northeastern New Guinea splits into a small LF positive anomaly center to the south of the Kamchatka Peninsula, and the center keeps moving toward the north of the Japanese archipelago. On 0 days, the “+”, “−”, and “+” EAP teleconnection wave train anomaly is presented in the South China Sea, southern China, and the Sea of Japan, respectively, on the East Asian coast from south to north. The LF teleconnection wave train of “+”, “−”, and “+” appears at the mid-high latitudes of Eurasia in the vicinity of the Ural Mountain, Lake Baikal, and the Sea of Japan, respectively, which is conducive to the maintenance of the circulation of “two ridges and one trough”. The interaction of the two wave trains causes the cold and warm airflow to converge in southern China, resulting in LF heavy precipitation in southern China.

To sum up, for the prediction of the EAP teleconnection wave train along the East Asian coast in drought years, special attention should be paid to the strengthening and westward movement of the LF negative (positive) anomaly in the Sea of Okhotsk and the northwestern Pacific Ocean (northeastern New Guinea) on −9 days. Besides, it should also be noted that the positive anomaly center splits off from the positive anomaly in the northeast of New Guinea and heads northeast on −3 days. As the two wave trains share the negative (positive) anomaly center in the EAP wave train located in the area from Lake Baikal to southern China (near the Sea of Japan), for the prediction of the “two ridges and one trough” system, attention should be paid not only to the anomaly center of the EAP teleconnection wave train mentioned above but also to the strengthening and westward movement of the LF positive anomaly centers in the north of Central Siberia and the area from the Lake Baikal to eastern China on −9 days.

4. Discussion and Conclusions

Based on the analysis of the LF characteristics of summer precipitation of southern China in typical flood and drought years, this paper further compares and investigates the similarities and differences of atmospheric LFO characteristics at different latitudes affecting the LF precipitation in flood and drought years and reveals the early predictive signals. The main conclusions are as follows.

(a) The precipitation is below average before 1992 and above normal from 1993 to 2002. After 2003, the interdecadal variability is weakened while the interannual variability increases. In both flood and drought years, summer precipitation is most significant in the LF period of 10–20 days in South China, followed by 30–60 days.

(b) The main characteristics of the atmospheric LFO at low and mid-high latitudes that affect LF precipitation in flood and drought years are compared and analyzed, and the main cold and warm vapor flow channels and sources in flood and drought years are given. It is found that at the peak of phase 3 of LF precipitation, LF convection is active in southern China in both flood and drought years and forms a closed LF vertical-meridional circulation with the convection inhibitive zone to its south. Additionally, the circulation is convergence (divergence) in the lower (upper) troposphere in southern China. The continuous convergence and rise of water vapor are conducive to the maintenance of LF precipitation. However, the convective center is southward, and the vertical-meridional circulation is stronger in flood years, resulting in more abundant LF precipitation than in drought years.

(c) The key 500 hPa height field LF circulation systems affecting LF precipitation in flood and drought years at mid-high latitudes include the LF “two ridges and a trough” at the mid-high latitudes of Eurasia and the LF “+”, “−”, and “+” EAP teleconnection wave train along the East Asian coast, which intersect in northeast Asia. However, the “two ridges and one trough” in flood years are generally further westward than that in drought years, and the central trough and the eastern ridge are more meridional than that in drought years. The LF WPSH in flood years is stronger than that in drought years, and its westward extension is more obvious.

(d) There are significant differences in the channels and sources of cold air at mid-high latitudes and wet-cold and wet-warm water vapor at mid-low latitudes in flood and drought years. In flood (drought) years there is a northerly and then westerly (central westerly) dry-cold, northeasterly wet-cold, southwesterly (none), and southeasterly warm-wet water vapor channel. Among them, the southeast warm-wet water vapor channel contains a southeasterly (southerly) cross-equatorial water vapor flow. The source of dry-cold air in flood (drought) years is located near Novaya Zemlya (the eastern West Siberian Plain), the source of wet-cold air is in the Yellow Sea and Bohai Sea (the Sea of Japan), and the sources of warm-wet water vapor include the Arabian Sea, the Bay of Bengal, the western Pacific Ocean and the sea area of northeastern Australia (the western Pacific Ocean and the sea area of northern Australia).

(e) The LF predictive signals at mid-low and high latitudes are found. For OLR, the LF OLR negative anomaly appearing near Somalia (the south of the Indochina Peninsula) on −11 days in flood (drought) year is an important LF predictive signal reflecting the summer precipitation of southern China in flood (drought) years. Special attention should be paid to the LF negative OLR anomaly which moves to southern China in the northeast (north) direction in flood (drought) years. For the 500 hPa height field, special attention is paid to the predictive signals of the EAP teleconnection wave train and “two ridges and one trough”. For instance, for the prediction of the EAP teleconnection wave train along the East Asian coast in flood years, special attention should be paid to the split and extension of the LF negative anomaly of East Siberia to the southwest and southeast on −9 days, the southward movement (southward and westward movement) of the LF positive anomaly near northeastern and eastern China (near the Bering Sea). In flood years, attention should be paid to the strengthening and eastward movement of the “+”, “−”, and “+” LF teleconnection wave train in the early −9 days in the area from Western Europe to the Mediterranean, on the Eastern European Plain, and in Central Siberia, respectively. In drought years, attention should be paid not only to the strengthening and westward movement of the LF predictive signals of the EAP teleconnection wave train but also to the predictive signal of the western ridge on −9 days, i.e., the strengthening and westward movement of the LF-positive anomaly centers in the north of Central Siberia and the area from Lake Baikal to eastern China (

Table 3. The source and moving direction of the predictive signals of the LF key systems that affect the LF precipitation at low and mid-high latitudes in flood and drought years.

		Flood Years	Drought Years
OLR	convective source (−11 d)	Somalia	the south of the Indochina Peninsula
	moving direction	northeast	north
WPSH	source (−9 d)	the northeastern and eastern China; the Bering Sea	the northeastern New Guinea
	moving direction	south; southwest	west
the East Asian Trough	source (−9 d)	the East Siberia	the Sea of Okhotsk and the northwestern Pacific
	moving direction	southwest	west

).

In addition, there is more than one source of LFO causing atmospheric circulation anomalies, and the superposition of multiple oscillation sources at different latitudes is more conducive to the occurrence of persistent heavy precipitation than a single one alone [20,23,25,29,30]. In this paper, the propagation characteristics of atmospheric LF signals and early forecast signals at different latitudes are analyzed, but the formation mechanism of these signals is not explored in detail. Note that the predictive signals in our paper are obtained by a statistical method, and the filtered data are used. By using a 10–20 day filter, the time series in each case would have a power concentrate on 15–16 days. Therefore, the correlation plot with 7–8 days’ difference shows about the same pattern with opposite signs. In the future, it is planned to further investigate and establish the relationship between the filtered and unfiltered predictive signals. Furthermore, the topic of how to improve the forecast accuracy of LF heavy precipitation processes by considering multiple pre-forecast factors is not discussed in depth. These are worth further study in the future.