Sanming (25°30′–27°07′N, 116°22′–118°39′E) is located in the mid-west of Fujian Province in coastal SEC in the frontal area of the EASM region (Figure 1). The study area has a typically humid subtropical monsoon climate. According to historical meteorological data from 1980 to 2000, the mean annual air temperature and precipitation in Sanming are 19.5°C and 1,560 mm, respectively. As shown in Figure 1, the uneven precipitation in this area is mainly concentrated from March to June, accounting for more than 57% of the annual precipitation. Due to the absence of long-term precipitation observations in Sanming, monthly precipitation data from 1951 to 2012 at the nearest meteorological station at Yong’an, 27 km from Sanming, were used to study the regional representativeness of the annual variation characteristics of precipitation at the sampling sites. We analyzed the correlation between the monthly precipitation data in Yong’an and in other regions, which suggests that precipitation over Sanming is highly representative of the southeastern region in China (Figure 1).

Precipitation samples were collected from a bare area in the Sanming Forest Ecosystem and Global Change Observation and Research Station (26°9′N, 117°28′E, altitude: 134 m a.s.l), from January 2019 to December 2020. Each precipitation event was sampled for isotope analysis using a ZJC-V Intelligent rainfall sampler (Supplementary Figure S1) which senses rainfall and opens a lid for collection. Rainwater flows directly into a refrigerated bottle through a water pipe, effectively preventing the evaporation of samples. After collection, each sample was poured into 20 ml and 2 ml polypropylene colorless plastic bottles with no space. These bottles were sealed with Parafilm sealing film to prevent evaporation and chilled at 4°C to prevent isotope fractionation. A total of 209 samples were collected on rainy days from 2019 to 2020.

Stable isotope composition (δ²H and δ¹⁸O) analyses were performed at the Stable Isotope Laboratory, School of Geographical Sciences, Fujian Normal University. The samples were analyzed using Picarro-L2140i cavity ring-down spectroscopy (CRDS). Three isotope standards and seven samples were examined as a group in this study. Seven samples were measured after the three standards, and each of the samples or standards was measured separately with seven injections. Taking into account the memory effect and stability of the instrument, only the final three injections were averaged and calibrated. The test findings were based on the samples’ calibrated values. Measurement calibration utilized three internal standards (δ¹⁸O: 19.13‰, −8.61‰, and −0.15‰; δ²H: 144‰, −63.4‰, and −1.7‰). The results were presented as δ-values relative to the standard Vienna Standard Mean Ocean Water (V-SMOW). The measurement accuracies were typically better than ±0.1‰ for δ¹⁸O and ±0.5‰ for δ²H.

IsoGSM is a general circulation model based on water isotope (Yoshimura et al., 2008), which is used to fill in missing data (June to July 2020) and explore the control of atmospheric processes on δ¹⁸O variability (Supplementary Table S1). IsoGSM has a time resolution of 6 h, and a horizontal resolution of 200 km, 28 vertical levels. Detailed descriptions of the model setup are available in previous studies (Yoshimura et al., 2008; Yoshimura et al., 2014; Yang et al., 2016). The monthly variations in precipitation and water vapor isotope compositions linked to synoptic weather cycles may be reasonably replicated by the IsoGSM. We first cross-compared the data from IsoGSM with observed data from Sanming station from 2019 to 2020 to confirm the reliability of the data from IsoGSM (Supplementary Figure S2). Result showed that the observed δ¹⁸O_p data are consistent with the simulated δ¹⁸O_p data from IsoGSM, which is a sign of good replication.

HYSPLIT was applied to locate and distinguish potential air mass source regions during the sampled rain episodes (Draxler and Hess, 1998). The model is obtained from the National Oceanic and Atmospheric Administration (NOAA) at the Air Resources Laboratory (https://ready.arl.noaa.gov/HYSPLIT.php). We set the end height for the backward trajectory model at 1500 m a.g.l. At the sampling site. The height is generally considered to be the largest water vapor transport volume and the precipitation is most likely to start (Aggarwal et al., 2004; Breitenbach et al., 2010; Wei et al., 2018). In addition, we computed the trajectories for 120 h back in time to reconstruct the original transport trajectory because it is sufficient to identify the air mass backward trajectories at the regional scale (Tao et al., 2021).

The associated meteorological data (precipitation amount, air temperature, and relative humidity) were monitored using an automatic HOBO rain gauge and a wireless temperature and humidity data logger next to the rainfall sampler, from January 2019 to December 2020. The National Climate Center provided monthly instrumental data for the Yong’an meteorological station from 1951 to 2010 (https://ncc-cma.net/cn/). To calculate the water vapor flux and analyze the relationships between regional integrated convective activity and daily precipitation stable isotope composition during the observation period, we obtained 850 hPa zonal and meridional wind components, specific humidity, and outgoing longwave radiation (OLR) data from the NOAA Physical Sciences Laboratory (https://psl.noaa.gov). KNMI Climate Explorer (https://www.knmi.nl), a web-based tool for high-resolution paleoclimatology, was used to conduct a spatial correlation test for the regional representation of the Sanming area (Trouet and Oldenborgh, 2013).

Craig (1961) first reported the regression line δ²H = 8 * δ¹⁸O + 10 as the global meteoric water line (GMWL) based on the positive connection between δ²H and δ¹⁸O found in natural meteoric waters throughout the world. With the increased establishment of GNIP program stations, numerous LMWLs have been built (Merlivat and Jouzel, 1979; Araguás-Araguás et al., 1998; Zhang et al., 2016; Wu et al., 2022). Although there are considerable geographical differences in climatic characteristics and sources of water vapor, the slope of the LMWL and GMWL are still in agreement. The correlation between daily stable precipitation isotope data at the Sanming station is defined here as the LMWL: δ²H = 8.29 * δ¹⁸O + 13.93 (R ² = 0.95, p < 0.0001, n = 209). Compared to the GMWL and the meteoric water line of China (CMWL) (δ²H = 7.9 * δ¹⁸O + 8.2, R ² = 0.98) (Zheng et al., 1983), the slope and intercept of Sanming LMWL were both higher. This is because GMWL and CMWL combine data from global or national studies that include several places with radically differing climates. The Sanming station, on the other hand, is located in a monsoon region and has a humid environment that results in a large slope and intercept. The correlation for the yearly precipitation isotopes in 2019 and 2020 was defined as δ²H = 8.44 * δ¹⁸O + 13.93 (R ² = 0.95, p < 0.0001, n = 117) and δ²H = 8.14 * δ¹⁸O + 13.94 (R ² = 0.95, p < 0.0001, n = 92), respectively (Figure 2A). The intercepts of the LMWL in 2019 and 2020 were almost consistent; however, the slope in 2020 was lower than that in 2019 which may have been due to the relative dryness and high evaporation in 2019.

The δ¹⁸O_p and δ²H data for the SM and NSM periods are shown in Figure 2B to demonstrate how the EASM affected the LMWL. The scatters are located above or near the GMWL in the SM (May-September). The NSM (October to April of the following year) scatters distribute below the GMWL. The slopes in both periods were smaller than those of the GMWL and LMWL. The slope of the LMWL in the NSM period was lower than the slope of 7.70 in the SM period, but its intercept of 14.98 was considerably higher than the intercept of 7.53 in the SM period, showing seasonal differences in the precipitation water vapor sources in this regional. Typically, the intercept and slope of the meteoric water line are positively related (Yao et al., 2018). However, in the monsoon region, because the average stable isotope value of the NSM period is more positive than that of the SM period, the scatter in the NSM waterline chart deviates from the upper right relative to the SM, resulting in a larger intercept of the LMWL during the NSM period than that during the SM period.

Figure 3 shows the temporal characteristics of daily δ¹⁸O_p and d-excess (d = δ ²H − 8 × δ ¹⁸O, a parameter associated with the meteorological conditions of the source region (Dansgaard, 1964; Breitenbach et al., 2010) and corresponding meteorological factors, i.e., surface air temperature, relative humidity, local precipitation amount) in Sanming from 2019 to 2020. As shown in Figure 3, the daily mean temperature exhibited cyclical variations ranging from 1.3°C to 29.4°C, with an average air temperature of 19.5°C and the maximum temperature occurring in July. Relative humidity in Sanming ranged from 52.6% to 94.5%, with an average relative humidity of 75%. The daily precipitation in Sanming ranged from 0 to 157.4 mm, with the maximum and minimum monthly precipitation occurring in May or June, and November, respectively. Notably, the region receives the most abundant precipitation from March to May, accounting for 25%–50% of the annual precipitation. During the observation period, δ²H, δ¹⁸O_p, and d-excess values ranged from −99.07‰ to 47.39‰, −12.94‰–4.37‰, and −4.05‰–30.38‰, with quantity-weighted means of −33.96‰, −5.72‰ (δ¹⁸O_w), and 11.77‰ (d _w), respectively. We performed a correlation analysis between environmental parameters and δ¹⁸O_p to explore the impact of environmental factors on precipitation isotopes. Poor correlations were found between the daily scale precipitation δ¹⁸O_p and precipitation (R ² = 0.085, p < 0.0001, n = 209), temperature (R ² = 0.09, p < 0.0001, n = 209), and relative humidity (R ² = 0.02, p = 0.07, n = 209) (Supplementary Figure S3). The findings showed that on a daily scale, local environmental parameters had less impact on the seasonal variations in precipitation isotopes.

The Sanming stable precipitation isotope data showed a clear seasonal variation (Figure 4A), similar to other reports globally (e.g., Xie et al., 2011; Zhou et al., 2019; Bedaso and Wu, 2020). To determine the seasonal variations in δ¹⁸O_p, the year was subdivided into SM and NSM periods. The SPR period was also included in the comparison owing to its high precipitation. Specifically, the δ¹⁸O_p values ranged from −12.94‰ to −0.15‰ with an average value of −7.41‰ in the SM period. In the NSM period, δ¹⁸O_p values ranged from −6.91‰ to 4.37‰ with an average value of −3.50‰. In the SPR period, δ¹⁸O_p values ranged from −5.72‰ to 4.37‰ with an average value of −2.08‰. In general, the δ¹⁸O_p values in the SM period were lower than those in the NSM period and fluctuated widely. For both the SM and NSM periods, the δ¹⁸O_p values in 2020 were lower than those in 2019, and the degree of fluctuation was relatively consistent (Figure 4A). The data distribution for the SPR period was similar to that of the NSM period, but the average value was slightly higher. Although the SPR period had higher precipitation, its δ¹⁸O_p average value was still higher than that of the SM and NSM periods, which is different from the “amount effect” (Dansgaard, 1964).

The d-excess of precipitation had also been used to explore the seasonal variations in precipitation caused by the shift of water vapor sources in different seasons. For the d-excess value, the variation ranged from −4.05‰ to 21.01‰ with a d _w value of 9.70‰ in the SM and from 0.85‰ to 30.38‰ with a d _w value of 14.47‰ in the NSM. The d-excess values in the SM season were more negative than those in the NSM season, which can be attributed to the regional climatic feature. During the SM period, it is dominated by the EASM, resulting in light precipitation and negative d-excess values. While in NSM, the regional climate is controlled by westerlies and inshore air mass, the d-excess values are more positive. This finding is consistent with the results of Zhang et al. (2009), who observed that in southwest China, the water vapor generating precipitation from low-latitude oceans with high humidity has small d-excess values, whereas that from westerly or inland recirculation with low humidity has large d-values. During the SPR period, the d-excess values varied from 0.85‰ to 26.99‰ with a d _w value of 8.41‰ (Figure 4B), similar to the NSM period. This similarity suggests that the source of water vapor that caused more precipitation during the SPR period was similar to the source of water vapor during the NSM period.

Sanming has a typical subtropical monsoon climate, with obvious seasonal variation in precipitation. The precipitation isotope results also showed marked seasonal variation (Figure 4). We calculated the d-excess values based on isotope data which reflected meteorological indicators in the moisture source area are less influenced by the local environment and can be used to identify the precipitation moisture sources (Dansgaard, 1964; Breitenbach et al., 2010). The consistent interannual fluctuation of the δ¹⁸O_p and d-excess values during the 2-year monitoring period reveals that they are likely to be governed by transitions in water vapor sources and atmospheric circulation (Figure 3). Therefore, we used the HYSPLIT backward trajectory model to track water vapor during the SM and NSM periods. To exclude the possible influence of moisture transport on the results obtained on non-precipitation days, we excluded the trajectory and only selected water vapor on precipitation days during the study period for the simulation analysis. In addition, we calculated the water vapor flux during the observation period to compensate for the fact that HYSPLIT can only reflect the frequency of water vapor sources.

The cluster analysis results based on all backward trajectories for the SM/NSM seasons and the vertically integrated mean water vapor transport during the observation period are presented in Figures 5, 6. On a seasonal scale, water vapor flux dispersion was generally low and precipitation amounts were small during the NSM season. Moisture is usually transported by westerly winds and most of it is inland water vapor, with positive δ¹⁸O_p and d-excess values. During the SM season, the amount of water vapor transported from the ocean increased, and the overall water vapor flux dispersion was more marked than that during the NSM season. The SM brings a large amount of low-latitude tropical oceanic water vapor (73.38% in total) to the Sanming region, mainly transported by the southwest water vapor channel, the SCS channel, and the northwest Pacific channel (Figure 5B). Consistent with the results in Figure 3, moisture derived from oceanic sources is relatively depleted, while the moisture associated with continental or local sources is more enriched. The δ¹⁸O_w (−3.50‰) and d _w (14.47‰) values for NSM were positive, reflecting relatively short transport pathways and possible sub-cloud evaporation effects during the precipitation process (local sources) or long-distance transport associated with less pronounced rainout process (continental sources) (Figures 5A, 6A). SM has negative δ¹⁸O_w (−7.41‰) and d _w (14.47‰) values, indicating that progressive rainout along long-distance transport pathways of air masses should be considered (Figures 5B, 6B). Previous studies have also found that the negative d-excess values during the summer season are associated with weak kinetic isotope fractionation over the oceanic source with warm and humid conditions, whereas the cold and dry air masses from both continental and local sources cause high d-excess values in autumn and winter (Liu et al., 2008; Li et al., 2020). Our results (Figures 3, 5, 6) are consistent with these findings and suggest that moisture sources play a significant role in explaining the observed seasonal variation in δ¹⁸O_p. Notably, there is no single source of water vapor on a seasonal scale, with various water vapor mixing and dominating by turns occurring at different periods (Li et al., 2020; Guo et al., 2021). Therefore, due to the seasonal variation in precipitation isotopes in the region, the paleoclimate δ¹⁸O records of this region may be able to reflect the strength of EASM in the region.

SPR is the “early summer rainy season” that occurs from SEC (south of 30 N and east of 110 E) to southern Japan from 13 to 27 pentad (early March to mid-May) (Ding, 1992; Tian and Yasunari, 1998; Wan and Wu, 2008b). To further investigate the main controlling factors of precipitation stable isotopes during the SPR period, we show the backward trajectory analysis, water vapor flux field, and average OLR during the SPR period in Figure 7. In terms of backward trajectory analysis, water vapor sources are mainly divided into two categories: inland and offshore water vapor (Figures 7A, B). Although the precipitation amount was large during the SPR period, the δ¹⁸O_p value was still relatively positive and different from that in the SM season (Figure 4). Evidently, the shift in moisture sources is not the dominant factor in the precipitation isotope composition change during the SPR period. As shown in Figure 7C, the relatively high average OLR values in the surrounding ocean indicate weaker convection activity and shorter water vapor transport pathways during this period. Suppressed convective activity at the water vapor source and transport trajectory weaken the rain rainout effect of water vapor (Cai and Tian, 2016), which leads to positive δ¹⁸O_p values at the sampling site. Whether the SPR period represents the onset of the EASM has long been debated. Previous studies have found that SPR is a portion of the summer monsoon rainfall, which is a signal of the establishment of the East Asian subtropical monsoon (i.e., a part of the EASM) (Ding et al., 1994; He et al., 2008). Other studies have argued that SPR is an extension of winter atmospheric circulation caused by increased southwest wind speed on the southern side of the Tibetan Plateau (Tian and Yasunari, 1998; Wan et al., 2008a). Combined with the previous results, we suggest that SPR should be considered as part of the NSM season based on the positive value of δ¹⁸O_p.

As shown in Figure 3, there is a dramatic and sustained depletion of δ¹⁸O_p from May to June, which coincides with the dates of DBR precipitation (i.e., a time that occurs during the dragon boat race from late May to mid-June). DBR is associated with the alternation of winter and summer monsoons (Wu et al., 2017) and is usually considered the beginning of the SCSSM, which always accompanies the end of the SPR period. Therefore, to explore the possible relationship between the SCSSM and δ¹⁸O_p in DBR precipitation in Sanming, a comparative analysis of the OLR [suggestive of strong convection (Yang et al., 2017)] was conducted in the SCS from mid-April to early June (Figure 8).

As shown in Figure 8, the bias of δ¹⁸O_p was not steep and had many fluctuations, and the OLR had multiple variations during the DBR period. According to Yang et al. (2017) observed that the summer monsoon is featured by intermittent activity on the synoptic scale. In addition, previous studies have reported a negative correlation between δ¹⁸O_p and monsoon intensity (Tian et al., 2001; Vuille et al., 2005). It is worth noting that the small decreases in each fluctuation correspond roughly to the occurrence of low OLR values, whereas the timing of the slight decrease does not coincide with the timing of the OLR lows (Figure 8). More specifically, there are four decreases in δ¹⁸O_p in the 26th, 30th, 33rd, and 38th pentads, while the OLR lows occurred on May 7 (26th pentad), 26 (30th pentad), June 8 (32nd pentad), and June 29 (36th pentad). This consistent lag further indicated that the δ¹⁸O_p values in the Sanming region reflect the convection situation in the SCS promptly and can provide a good indication of the timing and process of SM onset. This result also highlights the potential of the precipitation isotope composition to capture atmospheric circulation signals.

With the gradual onset of the SCSSM, the isotope values gradually become negative and maintain this level with an average δ¹⁸O_w value of −7.41‰ (Figure 3). During the SM period, water vapor from the ocean is the primary source of SM precipitation in the Sanming area, which causes negative precipitation δ¹⁸O_p values. Recently, many studies have emphasized the critical role of large-scale deep convective activity in the upstream regions. More specifically, the upstream rainout and convection activities along air mass trajectories can deplete the heavy isotopes in water vapor, which explains the isotope seasonal variability at downstream sites (Risi et al., 2008; Crawford et al., 2013; Aggarwal et al., 2016; Zwart et al., 2016; Ansari et al., 2020). Therefore, we analyzed the large-scale atmospheric convection intensity, as measured by the average OLR in the SM (Figure 9). The relatively low average OLR values around the water vapor source during the SM period indicated that the atmospheric convective intensity along the water vapor transport path was frequent and intense (Figure 9). Therefore, the strong convective activity of water vapor increases the depleting effect in the moisture source and along the transport pathway, resulting in negative δ¹⁸O_p values in the Sanming area (Lee and Fung, 2008; Risi et al., 2008; Cai et al., 2017; Ruan et al., 2019; Zhou et al., 2019), which is similar to the results shown in Figure 3. These observations further verify that, although the transfer of moisture sources has a non-negligible effect on the precipitation isotope composition in the SM and NSM seasons, the convective processes on the water vapor sources and transport pathways also play an essential role in the changes in precipitation isotope composition.