First, we compared the variation trends of the original δ¹⁸O sequences from all 30 caves over eastern China. As presented in Figure S1, although there were differences in record duration, dating error, and temporal resolution, the variation trends in stalagmite δ¹⁸O in various regions of eastern China exhibited obvious spatial differences during the past 2000 years. Specifically, the stalagmite δ¹⁸O records of most caves exhibited an increasing trend, and a number of caves located in northern, northeastern, and southwestern China exhibited a decreasing trend. This spatial difference is consistent with the results of a recent study that demonstrated that the ‘2-ka shift’ is evident in northeast Asia over the past two millennia (Zhao et al., 2021).

To quantitatively explore the variation trend in stalagmite δ¹⁸O in eastern China, we performed M-K trend test analysis on the 30 stalagmite records (Table 2 and Figure 1). Given the current availability of cave data, we have selected data ranging from 2 to 0 ka BP (before present [1950CE]) for trend analysis in our study. The results revealed that the stalagmite δ¹⁸O record in eastern China did not exhibit a consistent variation trend over the past 2000 years. Instead, 14 caves exhibited an increasing trend (at the 95% confidence level), eight caves exhibited a decreasing trend (at the 95% confidence level), and there was a non-significant trend in the other eight caves (Table 2). Furthermore, we conducted synthesized analysis for the stalagmite δ¹⁸O record of different variation types. As presented in Figure 2, the linear regression coefficient of all increasing trend stalagmite δ¹⁸O records was high (r²=0.61, p<0.01). The linear regression coefficient of all decreasing trend stalagmite δ¹⁸O records was also high (r²=0.47, p<0.01), while that of all trendless stalagmite δ¹⁸O records was very low (r²=0.003, p=0.38). Hence, the aforementioned findings indicate that stalagmite records display a variety of patterns of trend variations over the past 2000 years.

Based on the above-mentioned data, we further inferred that speleothem δ¹⁸O records during the past 2000 years suggested spatial differences in variation trends for different subregions over eastern China. Specifically, stalagmite δ¹⁸O records in North China (~35°N to the north, NC type) and South China (~26°N to the south, SC type) exhibited an overall decreasing trend, while those from Central-East China and the Yangtze River Valley (26°N-35°N, CEC type) demonstrated an obviously increasing trend. Overall, variation trends in stalagmite δ¹⁸O in eastern China likely are characterized by a spatially ‘-+-’ pattern during the past 2000 years. It should be noted that the above latitudinal division of the meridional zone differences is currently just an educated guess and is not very strict. For example, Hongyan Cave and Poya Cave in eastern Guizhou were located around 28°N, but their stalagmite δ¹⁸O exhibited a significant decreasing trend. It is conceivable that these caves are located at the boundary between the two climatic regimes and could have experienced gradual shifts in latitude over extended periods of time. To demonstrate the prevailing spatial pattern, we subsequently performed a synthesis of stalagmite δ¹⁸O records in three distinct regions of eastern China. As presented in Figure 3, stalagmite δ¹⁸O integrated sequences exhibited a significant decreasing trend during the past 2000 years from NC (r²=0.09, p<0.01) and SC (r²=0.09, p<0.01) and exhibited a significant increasing trend at CEC (r²=0.56, p<0.01). In fact, the long-term variation trends in cave δ¹⁸O records are primarily determined by the centennial-scale fluctuations over the last 2000 years. For example, it is suggested that during the two centennial-scale periods of ~0.4-0.8 ka and ~1.5-1.9 ka, the variations in cave δ¹⁸O were dominated by tripole patterns (‘-+-’ and ‘+-+’, respectively). Furthermore, the spatial pattern identified above is corroborated by the results of EOF analysis (EOF1, accounting for 22.63% of the variance) for 20 cave records out of a total of 30 based on their comparable time spans (0.26-1.88 ka) (see Figures S3 and S4). It is important to note that the current dataset derived from stalagmites remains relatively limited, thus potentially constraining the reliability of the EOF analysis findings to a certain extent. Generally, our finding is in agreement with those of previous publications, including both observations and climate modeling (e.g., Shi et al., 2017; Wang et al., 2022; Ge et al., 2023). Therefore, the findings mentioned above indicate that the stalagmite records from the three regions display distinct patterns of variation trends, all of which exceed the confidence threshold. As a result, this may indicate a common signal.

Given above-mentioned spatial difference, it is likely that a spatially ‘-+-’ mode has dominated stalagmite δ¹⁸O records of eastern China over the past 2000 years. To further visually illustrate the trends observed in the stalagmites, we conducted spatial interpolation of the trends in cave δ¹⁸O by using the inverse distance weighting (IDW) method (Figure 4a). The result indicates that δ¹⁸O records from NC (5 caves) and SC (8 caves) exhibited a decreasing trend, while those from CEC (17 caves) suggested an increasing trend. Although there was uncertainty in the spatial pattern of stalagmite δ¹⁸O variations due to the currently limited dataset, the overall ‘-+-’ mode observed above was generally robust.

This spatial pattern of cave δ¹⁸O reminds us that we should compare it to the dominant tripole mode of interdecadal-scale summer rainfall variability over eastern China that has been substantially investigated over the past two decades (Hsu and Liu, 2003; Hsu and Lin, 2007; Ding et al., 2008; He et al., 2016; Hua et al., 2021; Duan et al., 2023c). For example, Hsu and Lin (2007) suggested that the tripole precipitation pattern in East Asia during summer is a zonally elongated and meridionally banded structure (alternatively varying from 20°N to 50°N), and the increasing (decreasing) phase of the pattern is characterized by more (less) rainfall in central and eastern China, Japan, and South Korea, and less (more) rainfall in northern and southern China. Using two sets of model outputs from 20 models, Duan et al. (2023c) investigated the driving mechanisms of the tripole-leading rainfall mode over eastern China and observed that it was related to a circumglobal zonal wave train propagating along a westerly jet stream. As depicted in Figure 4b, the dominant mode (EOF1, explaining 18.0% of the variance) of summer (July-August) precipitation during the period of 1961-2018 clearly exhibited a tripole pattern over eastern China, and this is consistent with findings from previous research (Chiang et al., 2017; Zhang et al., 2018b). Hence, the spatial variation in cave δ¹⁸O and its comparison to the tripole mode of summer rainfall across eastern China both contribute to advance our understanding of the spatial patterns of paleomonsoon circulation and its dynamic mechanisms across diverse temporal scales.

In fact, the spatial characteristics of paleomonsoon precipitation in East Asia have also received widespread attention, and various methods and approaches such as geological carriers and climate numerical simulations have been used to perform analyses from orbital to centennial-interdecadal scales (Liu et al., 2015; Rao et al., 2016a, 2016b; Zhu et al., 2017; Chen et al., 2018b; Zhang et al., 2018b; Dai et al., 2021; Cao et al., 2024). For example, at the orbital timescale, a potential tripole precipitation pattern occurred in monsoonal China over the past 425 ka as simulated by the Norwegian Earth System Model (NorESM-L) that illustrated that more (less) precipitation in northern and southern China and less (more) precipitation in Central-East China during strong (weak) EASM periods that were associated with high (low) boreal summer insolation (Dai et al., 2021). At the suborbital/millennial timescale, climatic humidity (monsoon precipitation) over eastern China exhibited a spatially ‘-+-’ tripole mode during the early and late Holocene but a spatially ‘+-+’ tripole mode during the mid-Holocene as suggested by a comprehensive comparison of published records from lakes, peat, and cave stalagmites (Rao et al., 2016a, 2016b). At the centennial timescale, a north-south dipole mode and a ‘sandwich’ tripole mode of warm season precipitation in eastern China occurred in the second and third modes during the past half millennium, although the first leading mode exhibited coherent variations over most of China (Shi et al., 2017). Cave δ¹⁸O records have been widely recognized as a reliable proxy for paleo-precipitation isotopes and monsoon circulation across different temporal scales, as isotopic composition of precipitation and sources of water vapor inherited by the stalagmites δ¹⁸O are influenced by climatic factors such as Pacific Decadal Oscillation (PDO), Western Pacific subtropical high (WPSH), and the migration of monsoonal rain belts in eastern China (Tan, 2016; Wang et al., 2021, 2022; Lin et al., 2024). However, in numerous previous studies, researchers have predominantly focused on the spatial consistency of the stalagmite record. In this study, we are the first to observe the spatial differences in variation trends of speleothem δ¹⁸O records over the past 2000 years as partly supported by recent studies (Wang et al., 2022; Hu et al., 2024). In conclusion, the tripole mode would have dominated monsoon precipitation on various timescales, and this study could provide novel information regarding the tripole mode of monsoon circulation during the past 2000 years.

Given the above correlation between cave δ¹⁸O and summer rainfall, we further explored the possible mechanisms responsible for the tripole mode of cave δ¹⁸O during the last 2000 years. Notably, the key issue is to interpret the climatic significance of Chinese cave δ¹⁸O, particularly when we observed spatial difference of δ¹⁸O records in contrast to previously recognized spatial similarity of δ¹⁸O records on various timescales (Cheng et al., 2019; Liu et al., 2020). In general, cave δ¹⁸O predominantly reflects the signal of precipitation δ¹⁸O (δ¹⁸O_p), and thus, factors that exert a significant influence on δ¹⁸O_p such as water vapor source and transport pathways are primarily considered for both modern times and the past 2000 years. Previous studies have indicated that changes in cave δ¹⁸O values are associated with moisture sources on various timescales (Dayem et al., 2010; Maher and Thompson, 2012; Tan, 2014; Tan, 2016; Lin et al., 2024), with negative δ¹⁸O values reflecting relatively long-distance water vapor transport and vice versa. Consequently, atmospheric-oceanic processes that directly or indirectly regulated water vapor sources are considered when exploring the driving mechanisms for the tripole mode of summer monsoon rainfall and δ¹⁸O_p, including tropical Pacific sea temperature, PDO, Southern Oscillation Index (SOI), Intertropical Convergence Zone (ITCZ), westerly jets, North Atlantic Oscillation (NAO), and solar radiation (Sung et al., 2006; Jing et al., 2014; Tan, 2016; Chiang et al., 2017; Yang et al., 2017; Wang et al., 2021; Xue et al., 2022; Zhang et al., 2022b; Zhao et al., 2023; Chen et al., 2024a).

Recently, Lin et al. (2024) used the Community Atmosphere Model version 3 (CAM3) with embedded water-tagging and stable water isotopes modules to illustrate that the variation of oceanic moisture from the Pacific Ocean and the North Indian Ocean played a vital role in the decadal variation of the δ¹⁸O in summer precipitation of northern China that was modulated by the atmospheric circulation anomalies influenced by PDO. Thus, we compared the synthesized cave δ¹⁸O timeseries to atmospheric circulation factors that may have regulated variations of modern δ¹⁸O_p and cave δ¹⁸O via direct or indirect effect on changes in water vapor transport. As presented in Figure 5, the reconstructed timeseries for PDO_sst, NAO, and SOI exhibit persistent long-term trends that are consistent with the decreasing trends of cave δ¹⁸O from NC and SC. For example, the PDO_sst is overall positively related to NC_syn (r=0.11, p=0.11) and SC_syn (r=0.40, p<0.01) (Table 3) that support the observed positive correlation between modeled δ¹⁸O_p in northern China and the PDO index during 1951-2020 (Lin et al., 2024). Additionally, the NAO was negatively correlated with NC_syn (r=-0.23, p<0.01) and positively correlated with CEC_syn (r=0.43, p<0.01) and SC_syn (r=0.44, p<0.01) (Table 3). We also note a significant negative correlation between SOI and NC_syn (r=-0.25, p<0.01), and this is consistent with the observed relationship between SOI and Heshang Cave δ¹⁸O from 1900 to 2000 (Tan, 2016). Given the good correlation between cave δ¹⁸O and NAO, PDO_sst, and SOI, particularly during 0.4-0.8 ka and 1.5-1.9 ka, we further explored the potential processes responsible for the tripole modes of monsoon precipitation. Recently, Wang et al. (2022) successfully separated monsoon (PC1) and precipitation signals (PCs 2 and 3) using principal components analysis on three individual stalagmite δ¹⁸O records spanning North, Central, and South China over the past millennium. Their results suggested that the tripole rainfall pattern was revealed by PC2 that was highly correlated with PDO index, and this correlation exceeded the 95% confidence level (r=0.21, n=113). This supports our finding that decreasing PDO_sst corresponds to a “+-+” tripole structure in monsoon precipitation during 0.4-0.8 ka (Figure 5), thus demonstrating that North Pacific dynamics dominate the tripole precipitation structure in East China. In fact, another cave record case from southwest China indicates that a warm/cold PDO period (positive/negative phase) would lead to the deficit rainfall over India, thus making the δ¹⁸O of water vapor transport to southwest China less/more negative due to the weaker Rayleigh distillation process during the last ~300 years (Wang et al., 2021). Many studies have suggested that the Western Pacific subtropical high (WPSH) acts as the bridge between the precipitation patterns in East China and the PDO. Generally, the position of the WPSH moves northward, and Central China is characterized by high temperatures and less rainfall (positive δ¹⁸O values). The northward shift of the WPSH increased the rainfall in South China and North China (negative δ¹⁸O values), and this is associated with the monsoon trough and Meiyu front (Wang et al., 2022). The situation is opposite during a weak monsoon period, as a positive PDO and negative AMO leads to a southward shift of the WPSH resulting in a “-+-” tripole mode in precipitation over eastern China.

Conversely, some studies have suggested the important impact of the mid-latitude westerly jet and NAO toward precipitation changes in northern China (Wu et al., 2009; Yao et al., 2024), and this is also indicated by the correlation between abnormal negative δ¹⁸O values in NC and reconstructed NAO (Figures 5a and 5b). This is due to the observation that a negative phase NAO will enhance precipitation in NC due to the southward flow pattern of mid-latitude westerly jet (Chen et al., 2006; Sung et al., 2006; Herzschuh et al., 2019; Hu et al., 2024). Also, the observation data suggested that modern precipitation δ¹⁸O is derived from northern China (e.g., Changchun and Shijiazhuang), thus indicating relatively higher δ¹⁸O values during monsoon season and lower δ¹⁸O values during non-monsoon season that were obviously influenced by moisture originating from within the continent and the Western Pacific Ocean moisture channel (Zhang et al., 2021a). This supports the impact of NAO and mid-latitude westerly induced moisture transport on precipitation/stalagmite δ¹⁸O in NC. Additionally, the abnormal negative δ¹⁸O values in NC and SC during 0.4-0.8 ka is compared to the increasing SOI (Figures 5c-5e), and this represents a transition from El Niño to La Niña status during this period. The influence of La Niña-like status on tripole pattern of monsoon precipitation has been proposed during the mid-Holocene based on paleo-rainfall record comparisons from North, Central, and South China (Rao et al., 2016a, 2016b). Generally, the increasing SOI (Walker Circulation)/La Niña-like status corresponds to weakened WPSH, and this leads to decreased water vapor from West Pacific into the eastern China (Tan, 2016; Zhang et al., 2022a) and results in a decrease in the local water vapor transportation and subsequently reduces the values of the δ¹⁸O in precipitation and in stalagmites in NC and SC. Consequently, we suggest that PDO_sst, NAO, and SOI have played significant roles in modulating centennial-scale fluctuations in cave δ¹⁸O and monsoon precipitation in eastern China, ultimately resulting in the long-term trends observed in this study. Furthermore, the seasonality of precipitation and related differences in water vapor fluxes and paths for different subregions in eastern China should not be ignored (Lin et al., 2024). For example, spring rainfall (also known as non-monsoon rainfall) in Southeast China can exert a significant effect on the depletion of ¹⁸O in precipitation and stalagmites (Zhang et al., 2018a, 2020). In conclusion, the anomalies in cave δ¹⁸O from NC and SC during the last 2000 years are regulated by internal forces (atmospheric-oceanic processes) as discussed above.

Conversely, it appears that the increasing trend in cave δ¹⁸O from CEC corresponds to the overall southward ITCZ shift and the decreasing solar irradiance over the past 2000 years (Figure 5). The mean zonal locations of the ITCZ that are affected by temperature gradients between the northern and southern hemispheres resulting from changes in the Atlantic meridional overturning circulation (McGee et al., 2014; Chiang et al., 2015) have been regarded as a fundamental force driving the Asian summer monsoon changes on various timescales (Wang et al., 2004, 2005; Dykoski et al., 2005; Fleitmann et al., 2007; Wu et al., 2020). A significant positive correlation is indicated between the reconstructed ITCZ index and CEC_syn (r=0.15, p<0.05) (Table 3), and this supports that southward movement of ITCZ induced by the decreased NHSI would have weakened the convective activity over the tropical Indian Ocean, ultimately leading to significantly higher δ¹⁸O of ASM precipitation in the downstream area (Yang et al., 2016; Ruan et al., 2019; Xu et al., 2019). Similarly, the reconstructed proxy of total solar irradiance (TSI) is negatively correlated to three synthesized cave δ¹⁸O timeseries, of which the coefficient with SC_syn is high (r=-0.43) at the 99% significance level (Table 3). This further provides evidence for the inversely correlated relationship between solar irradiance and cave δ¹⁸O on a long-term timescale (Figure 5), as changes in the solar irradiance induce seasonal shifts in meridional convergence in concurrent trade winds from both hemispheres and local convection with their northernmost and southernmost positions reached during boreal summer and winter, respectively (Huang et al., 2013). Therefore, the observed long-term increasing trend in cave δ¹⁸O from CEC is jointly influenced by internal (ITCZ) and external (solar irradiance) factors, and this is supported by the majority of previously published stalagmite records (Wang et al., 2008; Cheng et al., 2016).

In previous studies, sea surface temperature (SST) anomalies vary along with PDO evolvement, and the SST cooling in mid-latitude North Pacific and SST warming in off-equator central-eastern Pacific are generally intensified from the developing stage to the decaying stage for the positive PDO phase and vice versa (Liu et al., 2023). Therefore, the intensified SST anomalies in tropical and midlatitude North Pacific both favor the meridional tripole pattern of summer precipitation anomalies in eastern China through tropical and extratropical dynamical processes during the decaying stage of positive PDO. Additionally, the volcanic double or clustered eruptions are speculated to have played a crucial role on the shift of the PDO phase (Chen et al., 2024c). Conversely, the persistently decreasing (positive to negative) trend of NAO from the Medieval Climate Anomaly (MCA) to the Little Ice Age (LIA) implies that changes in SSTs in the western equatorial Pacific (Mann et al., 2005; Emile-Geay et al., 2007) and the tropical Indian Ocean (Hurrell et al., 2004) induced decreased eastern and central tropical Pacific SSTs resulting in an initial strengthening of the NAO through tropospheric dynamics (Hoerling et al., 2001; Graham et al., 2007). Furthermore, the reconstructed SOI is significantly correlated with solar irradiance and mean Northern Hemisphere temperature (Yan et al., 2011), both of which exhibited a remarkable decreasing trend over the past 2000 years (Figure 5; Mann et al., 2008). The above-mentioned observations demonstrate that SSTs in the North Atlantic and Pacific as well as solar radiation and volcanic eruptions act as the main drivers responsible for the observed long-term trends in PDO_sst, NAO, and SOI during the past ~1000 and ~2000 years. However, it should be noted that there still remains controversy regarding the reconstruction of these factors (including PDO, NAO, SOI/ENSO) using different methods and datasets (Trouet et al., 2012; Tan, 2016). We suggest that more research, including proxy reconstructions and model simulations, should be performed in the future to improve our understanding of behaviors of the monsoon system and the underlying forcing factors at long-term timescales.

In conclusion, the spatial differences in Chinese cave δ¹⁸O records during the past 2000 years did not strictly adhere to the declining trend in solar radiation. Instead, it notably exhibited a spatial tripole mode on both centennial and long-term scales. We infer that such spatial patterns of precipitation isotopes are closely connected to precipitation seasonality and the corresponding changes in water vapor sources and transportation, and this provides a reference for the projection of the low-frequency variation mode of monsoon circulation with ongoing warming conditions.