A daily gridded precipitation dataset for 1951–2007 was obtained from APHRODITE (Asian Precipitation – Highly-Resolved Observational Data Integration Towards Evaluation of Water Resources, APHRO_MA_V1101R2 product, (21)) (Yatagai et al., 2009). The regional mean SPR (March–April) and EASM (May–September) precipitation amounts in China from 1951 to 2007 are shown in Fig. 1, which was exported based on this dataset using the free software Ferret (https://ferret.pmel.noaa.gov/Ferret, last access: 3 November 2019).

Monthly precipitation datasets of 160 meteorological stations in China for the period 1951–2014, obtained from the National Climate Center (https://www.ncc-cma.net/, last access: 3 November 2019), were used to characterize the percentage of spring (March–April) and EASM (May–September) precipitation amount relative to the annual precipitation amount in China.

Monthly mean δ18Op and precipitation amount data from meteorological stations across the MRC were obtained from the Global Network for Isotopes in Precipitation (GNIP) (http://www.iaea.org/, last access: 3 November 2019) (Table 1 and Fig. 1a). The monthly mean δ18Op data are used to compare the seasonal to interannual variation of δ18Op in the MRC. The stations near the coast from the southeastern region of the MRC (Fuzhou, Haikou, Hong Kong, Guangzhou) were excluded, because their precipitation amount and δ18Op are significantly influenced by typhoons in summer and autumn. Changsha station is the only GNIP station in the SPR region.

IsoGSM is a water isotope-permitting general circulation model (Yoshimura et al., 2008). We use the product of IsoGSM nudged toward the NCEP/NCAR Reanalysis 2 (Kanamitsu et al., 2002) atmosphere and forced with observed sea-surface temperatures (SST) and sea ice data (Yoshimura et al., 2008). A detailed description of the model setup can be found in Yoshimura et al. (2008) and Yang et al. (2016). IsoGSM can reproduce reasonably well monthly variabilities of precipitation and water vapor isotopic compositions associated with synoptic weather cycles. In order to verify the reliability of the simulated data from IsoGSM, we first cross-compare the data from GNIP Changsha station with those from IsoGSM during 1988–1992. The good replication indicates that both the precipitation amount and the δ18O data from the IsoGSM simulation are consistent (Supplement Fig. S1).

ENSO plays an important role in governing the climatic variation in the MRC (e.g., Feng and Hu, 2004; Xue and Liu, 2008; Zhou and Chan, 2007). We used the Southern Oscillation Index (SOI) and the Multivariate ENSO Index (MEI) to calculate the correlations between the phases of ENSO, the δ18Op, and the seasonal precipitation amount. The SOI is defined as the normalized pressure difference between Tahiti and Darwin. Negative and positive values of SOI represent El Niño and La Niña events, respectively. The data were obtained from the Australian Government Bureau of Meteorology (http://www.bom.gov.au/climate/current, last access: 3 November 2019). The MEI is based on six ocean–atmosphere variables (sea-level pressure, zonal and meridional components of the surface wind, SST, and total cloudiness fraction of the sky) over the tropical Pacific, and it is used to examine the role of ENSO in influencing the rainfall over the MRC. The MEI is defined as the first principal component of the abovementioned six variable fields. Therefore, it provides a more complete description of the ENSO phenomenon than a single variable ENSO index such as the SOI or Niño 3.4 SST (Wolter and Timlin, 2011). Positive and negative values of MEI represent El Niño and La Niña events, respectively. The data were obtained from the website of the Earth System Research Laboratory, National Oceanic and Atmospheric Administration (NOAA) (http://www.cdc.noaa.gov/people/klaus.wolter/MEI, last access: 3 November 2019). Tropical Pacific SST show a La Niña phase during the period from May 1988 to May 1989 and an El Niño phase during the period from May 1991 to June 1992. Therefore, we define 1988–1989 as La Niña years (1988 is the developing year and 1989 is the decaying year of the La Niña event) and 1991–1992 as El Niño years (1991 is the developing year and 1992 is the decaying year of the El Niño event) in this paper.

The Arctic Oscillation (AO) can also influence the climate and precipitation over the MRC (Gong et al., 2001, 2011; He et al., 2017; Li et al., 2014). It was suggested that a warmer winter in East Asia (a positive winter AO value) is associated with increased winter rainfall in southern parts of East Asia, and a positive spring AO is followed by increased rainfall in southern China but decreased rainfall in the lower valley of the Yangtze river (He et al., 2017). We also calculated the correlation between the AO index and the seasonal rainfall amount in our study area. The data were downloaded from the website of NOAA (https://www.cpc.ncep.noaa.gov/products/precip/CWlink/daily_ao_index/ao.shtml#forecast, last access: 3 November 2019).

The HYbrid Single-Particle Lagrangian Integrated Trajectory model (HYSPLIT) (Stein et al., 2015) was used to perform air mass back-trajectory calculations for the GNIP Changsha station during the period 1988–1992. In order to qualitatively assess the moisture source regions and transport paths for rainy season precipitation, only air mass back trajectories for precipitation-producing days were used. Trajectories were initiated four times daily (at 00:00, 00:60, 12:00, and 18:00 UTC) during precipitating days (> 1 mm precipitation per day) and their air parcel was released at 1500 m above ground level and moved backward by winds for 120 h (5 d). To identify the moisture uptake locations along the back trajectories during 1988–1992, we followed the method described in Sodemann et al. (2008) and Krklec et al. (2018). Two criteria (i.e., a more conservative threshold of positive gradient in specific humidity (0.2 g kg-1 within 6 h) and initial relative humidity of more than 80 %) were used to identify moisture uptake locations along the back trajectories. Following the methodology of Krklec et al. (2018), we calculated the contributions of moisture uptake locations en route to the precipitation in GNIP Changsha station and provided a map showing the percentage of moisture uptake contributing to Changsha precipitation during La Niña (1988–1989) and El Niño (1991–1992) years. A grid of 0.5∘×0.5∘ was used for the computation of the moisture uptake locations.

We calculated the mean ratios of spring (March–April) to annual precipitation (denoted as spring / annual) and EASM (May-to-September) to annual precipitation (denoted as EASM / annual) ratios for the period 1951–2014. Figure 2a shows that the mean percentage of spring / annual in southeastern China (about 20 to 33∘ N and 107 to 122∘ E, red rectangle in Fig. 2a), which range from 10 % to 25 % and from 0 % to 10 % in northern (about 33 to 53∘ N and 100 to 134∘ E, pink polygon in Fig. 2a) and southwestern regions of the MRC (about 20 to 33∘ N, 9 to 107∘ E, green polygon in Fig. 2a), respectively. The Jiangxi and the eastern Hunan provinces, the core regions of the SPR, show the highest mean percentage of spring / annual within the MRC (20 %–25 %), which is consistent with the results from the previous studies (Tian and Yasunari, 1998; Wan and Wu, 2009). Fig. 2b shows that the mean percentage of EASM / annual is 40 %–70 % in southeastern China and 70 %–95 % in other regions of the MRC. Conversely, the mean percentage of nonsummer monsoon precipitation to annual precipitation (NSM / annual) is 30 %–60 % in southeastern China and 5 %–30 % in other regions of the MRC, and it reaches the maximum in the SPR region (45 %–60 %). This indicates that the proportion of EASM precipitation (40 %–55 %) is nearly equivalent to the proportion of NSM precipitation (45 %–60 %) in the SPR region.

Usually, the SPR period lasts from March to mid-May (Wan and Wu, 2009) and the EASM period lasts from mid-May to September (Wang and Lin, 2002); however, the onset and/or retreat time of SPR and EASM and their intensities vary in different years (Zhou and Chan, 2007). The EASM starts late (late May to early June) and tends to be weaker during El Niño years (Huang et al., 2012), and EASM precipitation amount over southeastern China is reduced when the SPR starts later and lasts longer (until late May) (Wan et al., 2008a). Therefore, if we define the March-to-May precipitation as SPR and the June-to-September precipitation as EASM in El Niño years, the mean percentage of SPR / annual in the SPR region is 35 %–45 % (Fig. 2c), the mean percentage of EASM / annual is only 30 %–40 % (Fig. 2d), and the mean percentage of NSM / annual is 60 %–70 %. However, in other regions of the MRC, the mean percentage of EASM / annual (65 %–90 %) is still much higher than the mean percentage of NSM / annual (10 %–35 %) (Fig. 2c and d). Conversely, during La Niña years, the March–April and May-to-September precipitation should be defined as SPR and EASM precipitation, respectively (Fig. 2a and b). Therefore, the distribution of EASM vs. NSM precipitation amount in the SPR region is distinctly different from that in other regions of the MRC, and the ratio of EASM / NSM precipitation amount in the SPR region might be influenced by ENSO. We discuss this in detail in the Sect. 4.2.

We compared the seasonal variations of precipitation amount and δ18Op in the SPR region with those in other regions of the MRC by using data from the GNIP stations. According to the spatial distribution of EASM precipitation as discussed in Sect. 3.1, we assigned Zhengzhou, Xi'an, Yantai, Shijiazhuang, and Tianjin GNIP stations to northern region of the MRC; Kunming, Guiyang, Zunyi, and Chengdu GNIP stations to southwestern region of the MRC; and Changsha, Guilin, Liuzhou, Nanjing, and Wuhan GNIP stations to southeastern China. Only the Changsha GNIP station is located in the SPR region (Table 1 and Fig. 1a).

The seasonal variation of δ18Op in the MRC is consistently related to the onset, advancement, and retreat of the EASM. The δ18Op values decrease in May as the summer monsoon starts (Fig. 3). The δ18Op values are relatively low during the monsoon season (June–August) (Fig. 3) because of the long-distance transport of water vapor from the distal Indian Ocean to the MRC. Along this pathway, progressive rainout associated with regional convection leads to more negative δ18Op values via Rayleigh distillation (Baker et al., 2015; He et al., 2018; Liu et al., 2010; Moerman et al., 2013; Tan, 2014). The δ18Op values become progressively higher as the EASM withdraws in September (Fig. 3). From October to next April, the δ18Op values are rather high (Fig. 3), resulting from the short-distance transport of water vapor from the western Pacific Ocean or local moisture recycling and local convection (He et al., 2018; Moerman et al., 2013; Tan et al., 2016; Wu et al., 2015). The low δ18Op values in winter in northern region of the MRC are caused by the temperature effect, but it is less important because of its small contribution to the amount-weighted mean annual precipitation δ18O (δ18Ow) (Cheng et al., 2012). Therefore, the seasonal δ18Op values over the MRC show a broadly consistent pattern reaching a maximum in March–April and a minimum in July–August in the MRC with the exception of low winter δ18Op values in northern region of the MRC.

Given that there are only a few years of data from those GNIP stations, we obtained the mean monthly precipitation amount from the nearest meteorological station to each GNIP station in the MRC for the period 1951–2014 (Fig. 3c). Both datasets show that the seasonal variation of precipitation amount in southeastern China, especially in the SPR region, is different from that in other regions of the MRC (Fig. 3b and c). The precipitation amount in March and April before the onset of EASM is high over southeastern China. It is even higher than the summer monsoon precipitation amount in June, July, and August in other regions of the MRC. In the SPR region, the summer monsoon precipitation amount in July–August is much smaller than the precipitation in March–April. However, in other regions of the MRC, the summer monsoon precipitation in July–August is the highest of the whole year.

We identified the moisture uptake locations along the back trajectories and calculated their contributions to the precipitation at the GNIP Changsha station during EASM and NSM seasons in a La Niña phase (1988–1989) with low δ18Op anomalies and in an El Niño phase (1991–1992) with high δ18Op anomalies (Fig. 4). The results show that the moisture uptake locations and contributions during the EASM season are similar between El Niño and La Niña phases as well as those during the NSM season. During the EASM season, the moisture sources are mainly from South China Sea–South China, the Bay of Bengal–Indochinese Peninsula, and the Indian Ocean, while the remaining ones are from North China–western Pacific (Fig. 4a and c). In previous studies researchers mainly focused on the variations in moisture source during the EASM season (Baker et al., 2015; Cai et al., 2017; Tan, 2014). In this study, however, we also analyzed the back trajectories during the NSM season because NSM precipitation contributes ∼50 % to the annual precipitation in the SPR region. It shows that the NSM moisture sources originate from the South China Sea and southern China; the remaining ones are driven from local evaporation. Compared to the moisture sources during the EASM season, very few moisture sources are indicated for the Bay of Bengal–Indochinese Peninsula and the Indian Ocean during the NSM season.

In principle, the amount-weighted mean annual precipitation δ18O (δ18Ow) can be calculated from the sum of monthly weighted isotopic values divided by the total amount of precipitation as 1δ18Ow=(PJan×δ18OJan+PFeb×δ18OFeb+…PDec×δ18ODec)/(PJan+PFeb+…PDec). Based on the characteristics of the precipitation amount and δ18O during EASM and NSM seasons in the MRC, Eq. (1) can be written in the following mode: 2δ18Ow≈(PEASM-mean×δ18OEASM-mean+PNSM-mean×δ18ONSM-mean)/(PEASM-mean+PNSM-mean)=EASM%×δ18OEASM-mean+NSM%×δ18ONSM-mean, where PEASM-mean and PNSM-mean are the mean precipitation amounts of EASM and NSM, respectively; δ18OEASM-mean and δ18ONSM-mean are the mean values of EASM and NSM precipitation, respectively; and EASM % and NSM % are the mean percentages of the EASM and NSM precipitation amounts, respectively.

Therefore, we can consider that the δ18Ow is controlled by both precipitation amount and δ18Op during the EASM and NSM seasons in the MRC. Given the relationship between monthly precipitation amount and δ18Op in the MRC (Fig. 3), we find that (1) in northern and southwestern regions of the MRC δ18Ow values are mainly controlled by the amount and δ18O of EASM precipitation, because the precipitation amount of the EASM with rather low δ18Op values accounts for 70 % of the annual precipitation and the NSM precipitation is only a small contribution to the δ18Ow (less than 30 %). (2) In southeastern China, especially in the SPR region, the precipitation amount of the NSM with rather high δ18Op values even exceeds that of the EASM with rather low δ18Op values, and it also has an important effect on δ18Ow. Hence, δ18Ow in the SPR region is affected by both EASM and NSM precipitation. In addition, except for the effect of the seasonal distribution of precipitation amount, the seasonal δ18O itself also attributes to the δ18Ow, which is related, among others, to the variations in integrated regional convection and moisture source and transport distance (Cai et al., 2018; Baker et al., 2015; Huang et al., 2017; Tan et al., 2016).

In order to separate the influences of precipitation seasonality and monthly δ18Op, we used the decomposition method used by Liu and Battisti (2015) and Cai and Tian (2016) to evaluate the role of changes in precipitation seasonality (δ18Ops; assuming that the monthly precipitation δ18Op in El Niño years 1988–1989 is the same as that in La Niña years 1991–1992). We then calculated the difference between precipitation δ18Ow in El Niño years and La Niña years and the change in precipitation δ18O (δ18Oiso; method is similar to that for calculating δ18Ops but assuming that the monthly precipitation amount is the same). The results for the Changsha station indicate that the difference in precipitation δ18Ow between El Niño years (1988–1989) and La Niña years (1991–1992) (i.e., El Niño minus La Niña) is 2.7 ‰, δ18Ops is 1.3 ‰, and δ18Oiso is 1.3 ‰. These results imply that the difference in δ18Ow between El Niño and La Niña conditions reflects the differences of both the δ18Op and the precipitation seasonality.

Tan (2014) suggested that positive (negative) δ18Ow anomalies during El Niño (La Niña) phases reflect more (less) water vapor originating from the nearby South China Sea and the western Pacific Ocean (characterized by rather high δ18Op values) relative to the remote Indian Ocean (showing comparable low δ18Op values). By using the HYSPLIT model, however, Cai et al. (2017) demonstrated that the moisture sources vary little between years with relatively high and low δ18O values (corresponding to El Niño and La Niña years) in the EASM region; hence, EASM precipitation is primarily derived from the Indian Ocean, while the Pacific Ocean moisture is a minor contributor. This is consistent with our results (Fig. 4). In addition, by using a Lagrangian precipitation moisture source diagnostic, Baker et al. (2015) suggested that the moisture uptake area in the Pacific Ocean does not differ significantly between summer and winter and is thus a minor contribution to monsoonal precipitation; changes in moisture transport, however, may impact the δ18O variation of EASM precipitation. Dayem et al. (2010) also proposed that several processes (e.g., source regions, transport distance and types of precipitation) contribute to the δ18Op variation. We found that the moisture sources in the Bay of Bengal–Indochinese Peninsula and the Indian Ocean were less important during the NSM season compared to the EASM season (Fig. 4). The moisture uptake area in the EASM season does not differ significantly between El Niño and La Niña years nor in the NSM season. Their contributions to the whole precipitation in El Niño and La Niña years, however, are different (Fig. 4). The variation in moisture source during the EASM period, to some extent, might contribute to changes in δ18Ow, but it is not the main factor. We emphasize the effect of NSM precipitation amount on δ18Ow in the SPR region, and we made an attempt to analyze the relationship between the seasonal precipitation amount and δ18Ow with ENSO phase on interannual timescale in the next section.

The ENSO is a coupled ocean–atmosphere phenomenon controlling the interannual variation in precipitation amount and δ18O over southeastern China (e.g., Feng and Hu, 2004; He et al., 2018; Huang et al., 2017; Moerman et al., 2013; Tan et al., 2014; Xue and Liu, 2008; Yang et al., 2016). Our analysis of the 1988–1992 data from the Changsha GNIP station suggests that the mean value of δ18Ow (-6.73 ‰) in La Niña years (1988–1989) is significantly more negative than during El Niño years (1991–1992; -4.11 ‰). However, there is no significant variation in the annual precipitation amount between La Niña and El Niño years (Fig. 5a). The difference of δ18Ow between La Niña and El Niño phases cannot be explained by variations in annual precipitation amount. This is consistent with the analyses based on instrumental meteorological data (Huang et al., 2017; Tan, 2014) and climate simulations (Yang et al., 2016). Previous studies showed that during El Niño years, the EASM is generally weak and the integrated regional convection decreases in the EASM region, thereby leading to higher δ18Op values, while the effect of La Niña is opposite (Cai et al., 2018; Gao et al., 2013; Zwart et al., 2016). Continental moisture recycling or local convection during the NSM season has limited impact on δ18Op relative to the integrated regional convective activities during the EASM season. As we discussed in Sect. 4.1, however, the difference in δ18Ow between El Niño and La Niña years is influenced by both the δ18Op and the precipitation seasonality. Indeed, there is more summer monsoon precipitation in June to September during La Niña years (1988–1989) but more SPR in March–April during El Niño years (1991–1992), though the annual precipitation amounts are similar (Fig. 5b). We find that the δ18Ow variability is broadly consistent with the variation in the ratio of EASM / NSM precipitation amount during 1988–1992 (Fig. 5). Unfortunately, the data series of the Changsha GNIP station is too short (5 years) to evaluate the relationship between the EASM / NSM ratio and δ18Op in the SPR region. Therefore, we used the average precipitation data from 11 meteorological stations (1951–2014) in Jiangxi Province and the eastern Hunan Province (Fig. 2, i.e., from the core area of the SPR) as well as the δ18O data obtained from the IsoGSM simulation (1979–2009) to examine the relationship between ENSO, AO, δ18Ow, and precipitation amount in the SPR region on interannual timescales.

We calculated correlation coefficients between the simulated δ18Ow; the SOI; the MEI; the EASM / NSM ratio; and the annual, EASM, and NSM precipitation amounts for 1979–2009 (Table 2 and Fig. 6). The results show that the time series of the simulated δ18Ow data significantly correlates with the SOI (r=-0.52, p < 0.01) and the MEI (r=0.51, p < 0.01), consistent with the positive relationship between the ENSO index and δ18Ow observed in modern precipitation (Huang et al., 2017; Tan, 2014; Yang et al., 2016) as well as in the δ18O records of speleothems and tree-ring cellulose (Tan, 2016; Xu et al., 2013, 2016a, b; Zhang et al., 2018). Furthermore, the same relationship holds for the Changsha GNIP station (Fig. 5a). This indicates that the precipitation δ18Ow is higher (lower) during the El Niño (La Niña) phase in the SPR region. There is, however, no significant correlation between the δ18Ow with the annual, EASM, or NSM precipitation amounts. This indicates that on interannual timescales, the δ18Ow is not controlled by the annual or EASM precipitation amount in southeastern China, consistent with the result based on instrumental data from the Changsha station (Fig. 5a) and other studies (Tan et al., 2014; Yang et al., 2016). The time series of the simulated δ18Ow data correlates with the EASM / NSM ratio (r=-0.36, p < 0.05) (Fig. 6 and Table 2), suggesting that the precipitation δ18Ow may be influenced by the precipitation seasonality (i.e., EASM / NSM ratio) modulated by ENSO on interannual timescales.

Applying a 2-year smoothing, the time series of the simulated δ18Ow data significantly correlates with the annual precipitation (r=-0.89, p < 0.01), the EASM precipitation (r=-0.91, p < 0.01), and the EASM / NSM ratio (r=-0.81, p < 0.01) (Table 2 and Fig. 6). This indicates that on interannual to decadal timescales the precipitation δ18Ow might reflect changes in EASM precipitation amount and also the annual precipitation amount and the EASM / NSM ratio, because the EASM / NSM ratio and annual precipitation amount are significantly dominated by the EASM precipitation amount (Table 2).

To explore the relationship between ocean-atmospheric circulation (e.g., ENSO, AO) and the seasonal precipitation amount, we calculated correlation coefficients between the SOI; MEI; AO index; the EASM / NSM ratio; and the annual, EASM, and NSM precipitation amounts for 1951–2010 (Table 3). The mean value of October to next June SOI correlates with the EASM precipitation amount (r=0.26, p < 0.05), NSM precipitation amount (r=-0.51, p < 0.01), and the EASM / NSM ratio (r=0.52, p < 0.01) (Table 3). The mean value of October to next June MEI correlates with the EASM precipitation amount (r=-0.29, p < 0.05), NSM precipitation amount (r=0.54, p < 0.01), and the EASM / NSM ratio (r=-0.55, p < 0.01) (Table 3). This indicates, on interannual timescales, decreased EASM precipitation during the developing stage of El Niño and increased NSM precipitation during the mature stage of El Niño, resulting in lower EASM / NSM ratios during El Niño phases and vice versa. There is, however, no significant correlation between the SOI, MEI, and the annual precipitation amount. In addition, the EASM / NSM ratio significantly correlates with the EASM (r=0.64, p < 0.01) and the NSM (r=-0.70, p < 0.01) (Table 3).

Previous studies found that decreased summer rainfall in the south of the Yangtze river occurs during the developing stage of El Niño, resulting from a southward shift of the subtropical high associated with colder SST in the western tropical Pacific and weak convective activities in the South China Sea and the Philippines (Huang and Wu, 1989; Zhang et al., 1999). Kong and Tu (2003) found that there is less EASM rainfall in May–September in the lower reaches of the Yangtze river valley during 14 El Niño events since the 1950s. The same relationship is observed in the May–October rainfall reconstruction based on tree-ring cellulose δ18O (Xu et al., 2016a) during El Niño phases. Cooler summer SST in the western Pacific led to a weakened western Pacific subtropical high resulting in less rainfall during May–October in the middle-to-lower reaches of the Yangtze river (Liu and Li, 2011; Xu et al., 2016a) and vice versa. Increased rainfall in autumn, winter, and spring (i.e., NSM) occurred in southern China during the mature stage of El Niño (Wan et al., 2008b; Wang et al., 2000; Zhang et al., 1999, 2015; Zhou, 2011; Zhou and Wu, 2010). During these phases, lower-level southwesterly anomalies over the South China Sea transport more moisture into southeastern China, leading to increased NSM precipitation (Wang et al., 2000; Zhang et al., 1999; Zhou, 2011; Zhou and Wu, 2010). These conclusions are consistent with our findings in the SPR region. It is notable that there is no significant variation in EASM precipitation amount in our study area during the decaying stage of El Niño, although increased summer rainfall was observed in southern China (Huang and Wu, 1989).

We also find that the May AO index significantly and negatively correlates with the annual (r=-0.42, p < 0.01) and EASM (r=-0.39, p < 0.01) precipitation amounts in the SPR region (Table 3). This is consistent with previous observations that the positive May AO index is followed by decreased summer precipitation amount in the lower Yangtze river valley (Gong and Ho, 2002; He et al., 2017). It was suggested that a stronger May AO is associated with a northwards movement of the summer jet stream, leading to drier conditions in the lower Yangtze river. The positive spring AO gives rise to warmer equatorial SSTs between 150–180∘ E and weakens summer subtropical high in the western North Pacific. Consequently, decreased summer precipitation occurs in the lower Yangtze river (Gong et al., 2011). There is, however, no significant correlation between the AO index and the NSM precipitation amount and the EASM / NSM ratio in the SPR region (Table 3). This might be because the influence of AO on the winter climate varied spatially and temporally, resulting from the unstable relationship between the AO index and the East Asian winter monsoon (He et al., 2017; Li et al., 2014). This indicates that the AO mainly influences the changes in EASM and annual precipitation amount but not the precipitation seasonality (i.e., EASM / NSM ratio) in the SPR region. The February AO index positively correlates with precipitation amount in February (r=0.28, p < 0.05).

Given the relationship between δ18Ow, SOI, MEI, and seasonal precipitation amount, we find that less EASM during the developing stages of El Niño and more NSM precipitation during the mature stages of El Niño lead to lower EASM / NSM ratios, resulting in higher δ18Ow values in the SPR region during El Niño phases and vice versa. We therefore suggest that over the SPR region the precipitation seasonality (i.e., the EASM / NSM ratio) modulated by ENSO primarily influences the interannual variability of δ18Ow. The AO mainly influences changes in EASM and annual precipitation amount but not the precipitation seasonality (i.e., EASM / NSM ratio) in the SPR region.

Although speleothem δ18O records have massively improved our understanding of the EASM variability on different timescales, the significance and quantification of these proxy records is still a subject of debate, because speleothem δ18O is influenced by several competing factors. We emphasize that the spatial differences in seasonal precipitation over the MRC are key to understanding the speleothem δ18O–climate relationship. Fig. 1 illustrates that (1) Wanxiang (Zhang et al., 2008), Dayu (Tan et al., 2009), Huangye (Tan et al., 2010), Wuya (Tan et al., 2014), Shihua (Li et al., 2017), and Xiaobailong (Tan et al., 2017) caves are located in the northern and southwestern part of the MRC, where δ18Ow is primarily controlled by the EASM and annual precipitation amount. Therefore, these records show a significant correlation with the instrumental precipitation and the regional drought/flood (D/F) index obtained from historical documents (e.g., Li et al., 2017; Liu et al., 2008; Tan et al., 2009, 2010, 2014, 2017; Zhang et al., 2008). (2) Dongge (Yuan et al., 2004), Heshang (Hu et al., 2008), Hulu (Wang et al., 2001), Yuhua (Jiang et al., 2012), and E'mei (Zhang et al., 2018) caves are located in southeastern China, where δ18Ow is not only affected by EASM precipitation but also by NSM precipitation. Hence, according to Wang et al. (2001), speleothem δ18O from Hulu cave reflects the ratio of summer-to-winter precipitation amount. Factors related to the NSM (e.g., moisture source, integrated regional convection, precipitation seasonality, winter temperature) have also been taken into consideration in the interpretation of speleothem δ18O in southeastern China (e.g., Baker et al., 2015; Cai et al., 2018; Cheng et al., 2016; Clemens et al., 2010; Dayem et al., 2010; Zhang et al., 2018). On the other hand, the high percentage of NSM precipitation with relatively high δ18Op values in southeastern China should be an important reason why δ18Ow and speleothem δ18O are much lower and their variability is much larger in southwestern China than in southeastern China (Li et al., 2016; Liu et al., 2010; Zhang et al., 2018), except for the influence of integrated regional convection and moisture source and transport distances during the EASM season.

We find that the precipitation seasonality modulated by ENSO mainly controls the δ18Ow values in the SPR region, with lower (higher) EASM / NSM ratios associated with El Niño (La Niña) phases resulting in higher (lower) δ18Ow values. Therefore, we suggest that the interannual variability of speleothem δ18O in the SPR region is primarily controlled by precipitation seasonality (i.e., the EASM / NSM ratio) modulated by ENSO. In addition, the ENSO index in the SPR region also significantly correlates with the EASM precipitation amount on interannual timescales and the precipitation δ18Ow negatively correlates with the EASM precipitation amount on interannual-to-decadal timescales, implying that additional studies are needed to disentangle the main driving factor(s) (e.g., EASM precipitation amount vs. EASM / NSM ratio) operating on different timescales. Few speleothem δ18O records have been published for the SPR region so far (Jiang et al., 2012; Zhang et al., 2018). Such long-term records, however, are critically needed to examine the climate–proxy relationship both on interannual and on decadal to millennial timescales.