East Asian Meiyu variability reflected in precipitation oxygen isotopes via western Pacific subtropical high

Abstract

It remains uncertain whether precipitation oxygen isotopes (δ¹⁸O) reliably capture East Asian Meiyu monsoon variability. Analyzing daily δ¹⁸O across the Yangtze-Huai River Basin from 28-34°N, we reveal a distinct spatial dichotomy. In the middle and northern Meiyu regions, δ¹⁸O robustly tracks Meiyu precipitation. Conversely, the southern Meiyu margin is decoupled from Meiyu variability, primarily reflecting upstream convection processes further south. We identify the western Pacific subtropical high (WPSH) as the central driver, creating a dynamic dipole: its northwestward extension enhances moisture transport and deep convection along its northwestern flank (driving isotopic depletion in the northern Meiyu region), while imposing subsidence and convective inhibition under its body (suppressing isotopic depletion in the southern Meiyu region). Importantly, these mechanisms persist on interannual timescales. Consequently, while northern δ¹⁸O records effectively capture Meiyu variability, southern records reflect distinct vertical constraints, necessitating spatially differentiated paleoclimate interpretations.

Introduction

The East Asian Summer Monsoon (EASM) is a key component of the Asian Summer Monsoon system, significantly influencing precipitation variability and extremes in East Asia. Fluctuations in precipitation during this season exert substantial effects on agriculture and socio-economic stability in the densely populated countries within the region^1,2,3. The onset of the East Asian Meiyu season marks the commencement of the EASM in subtropical areas, characterized by a distinct monsoon rainy period that typifies the summer monsoon progression^4,5,6. In China, this rainy system is primarily situated within the Yangtze-Huai River Basin and extends northeastward to South Korea (Changma), and Japan (Baiu). Reconstructing rainfall or monsoon variability in East Asia has relied heavily on oxygen isotope (δ¹⁸O) records derived from geological archives, particularly Chinese speleothems, which are believed to inherit precipitation δ¹⁸O signals^7,8,9. Despite recent advancements in understanding the climatic significance of East Asian speleothem and precipitation δ¹⁸O, a fundamental question remains: does δ¹⁸O signal effectively capture the specific variability of the Meiyu, or is it masked by larger-scale processes?

The “amount effect”¹⁰ serves as a foundational concept for interpreting speleothem and precipitation δ¹⁸O in the context of East Asian monsoon variability. The classical amount effect describes a negative correlation between monthly precipitation amounts and δ¹⁸O at local scales in tropical or monsoon regions¹⁰. However, strictly local correlations have proven unstable; observations at many sites within these regions reveal weak or insignificant relationships between local precipitation amount and δ¹⁸O over various timescales^11,12,13. In contrast, recent research has highlighted the influence of rainfall and convection processes along upstream moisture transport pathways, which isotopically deplete transported vapor and subsequent precipitation^14,15,16,17. This phenomenon, termed the “regional amount effect”¹⁸, refers to the integrated effects of convection over larger regions on vapor isotopes through rainout, rain-vapor exchange, and vertical redistribution of moisture^{19,20,21,22,23}. Consequently, precipitation within and downstream of these convective regions are isotopically depleted. Consistent with these modern precipitation isotope findings, recent studies on speleothem δ¹⁸O records have similarly underscored the importance of large-scale processes over local variables^7,8,24.

However, the EASM is a complex climate system characterized by spatially heterogeneous precipitation and circulation patterns^6,25,26, leading to considerable debate regarding the specific spatial footprint that Chinese speleothem δ¹⁸O records reflect. The definition of the “large-scale upstream integration” region remains elusive. For instance, Pausata, et al.²⁴ suggested that higher speleothem δ¹⁸O values in eastern China during the last glacial were linked to reduced precipitation in the Indian monsoon region rather than East Asia itself. Conversely, differences in speleothem δ¹⁸O between East Asia and India, as well as within East Asia, suggest that Chinese speleothem δ¹⁸O may capture site- or region-specific climate signals^27,28,29. Similar discrepancies are evident in precipitation δ¹⁸O studies^15,30,31,32. For example, early research at Nanjing in East China indicated that precipitation δ¹⁸O was influenced by rainout in remote ocean areas such as the Bay of Bengal³², whereas subsequent studies at the same site argued that rainout within southeastern China played a more significant role¹⁵. Additionally, studies in South China¹⁴, North China³³, and Japan¹⁶ have identified rainfall and convection processes occurring over spatial scales ranging from several hundred to over a thousand kilometers as dominant factors. These contradictions indicate that the spatial scale of upstream controls should not be overgeneralized.

Critically, while the western Pacific subtropical high (WPSH) is the primary dynamic driver of Meiyu variability^4,5, its influence on isotopic variability remains poorly understood. The relationship between atmospheric circulation and δ¹⁸O is complicated by the debate over “remote” versus “regional” controls^13,34,35. Specifically, it is unknown whether the WPSH exerts a uniform control on isotopes across the Meiyu region or if it induces spatially nonlinear responses.

In this study, we address these uncertainties by examining the relationship between East Asian Meiyu variability and precipitation δ¹⁸O at the process level. We analyze atmospheric controls on daily δ¹⁸O variations during the Meiyu season across four sites spanning the northern and southern sectors of the Meiyu region (Fig. 1). Building on the concept of the regional amount effect^18,36, we explicitly test whether precipitation variability in the Meiyu region influences precipitation δ¹⁸O and how the location and extent of the WPSH modulate this relationship. Our analysis reveals distinct atmospheric circulation anomalies linked to δ¹⁸O variations, highlighting a mechanism where the WPSH imposes contrasting dynamic controls between the northern and southern Meiyu regions. Further verification suggests that these findings based on daily data hold true at the interannual timescale. These findings hold practical implications for improving the interpretation of paleoclimate records in this transitional monsoon zone.

Fig. 1: Locations and climatological settings of the study sites.

Climatological (1998–2024) distribution of a total precipitation amount (mm) during the Meiyu season and b the fraction of precipitation amount during the Meiyu season relative to the summer (June-August) total (%). The red lines indicate the Meiyu region boundaries. The red dots denote the study sites (NJ Nanjing, HMQ Hemuqiao, WC Wucheng, CS Changsha). Maps created using Cartopy. Coastlines are from the Natural Earth public domain dataset. The Meiyu boundaries were processed in QGIS based on the Meiyu Monitoring Dataset.

Results

Relationship between Meiyu precipitation and its oxygen isotope composition

To explore the relationship between Meiyu precipitation and its δ¹⁸O values, we first examined the temporal variation of daily δ¹⁸O relative to the precipitation life cycle (Fig. 2). Data for the ten days before and after each Meiyu season were shown only for contextual comparison purposes, and they were not used in the following analysis. Overall, δ¹⁸O values exhibit a decreasing trend during the Meiyu onset and remain generally lower during the active phase compared to pre-onset levels. Consistent with this transition, at Nanjing, five of the ten highest daily δ¹⁸O values during the Meiyu season occurred at the onset, with an additional high value near the onset date (Fig. 2). This suggests that the establishment of the large-scale monsoon circulation, rather than isolated local storms, drives the primary isotopic depletion. Additionally, δ¹⁸O shows a lower-frequency decreasing trend from the early to late Meiyu season (Fig. 2). If equally divide the Meiyu season into three stages (Fig. S1), the decrease in δ¹⁸O values from the first to the second stage is -1.3‰ (p = 0.06) and from the second to the third stage is -1.1‰ (p = 0.09). However, these intraseasonal changes are much smaller than the synoptic fluctuations (>10‰), which align with the northward propagation of precipitation systems. This observation implies that δ¹⁸O is probably governed by the synoptic evolution of the rainband itself, originating from the south. Similar qualitative patterns are observed at the other three sites (Figs. S2–S4). Notably, Changsha, located at the southwestern edge of the Meiyu region (Fig. 1), shows δ¹⁸O fluctuations aligning with precipitation systems originating south of the Meiyu region (Fig. S2), hinting at a decoupling between the southern Meiyu margin and the primary Meiyu belt.

Fig. 2: Temporal variations of daily precipitation δ¹⁸O at Nanjing relative to regional rainband evolution.

a Time series of daily precipitation δ¹⁸O at Nanjing from ten days before the onset to ten days after the retreat of each Meiyu season from 2012 to 2016. b Latitude-time plot of daily precipitation amount (mm day^-1) averaged over the longitude band of 110–122°E during the same periods. c and d Same as (a) and (b) but for 2017–2021. Red circles highlight the ten highest δ¹⁸O observations, and blue circles highlight the ten lowest δ¹⁸O observations within Meiyu seasons from 2012 to 2021. Dashed vertical lines with dates indicate the onset and retreat of each Meiyu season. Dotted horizontal lines depict the latitude of Nanjing.

To quantitatively capture the “regional amount effect”—where isotopes integrate convection history along moisture transport pathways^14,18,19,33—we calculated correlations between Meiyu season daily δ¹⁸O and accumulated precipitation over optimal preceding windows (P_{acc_td}; where t in the subscript denotes number of days). We tested t values ranging from 1 to 15 days to identify the timescale yielding the most negative correlation across all grid points in an extended Meiyu region (25–35°N, 110-125°E) (Fig. 3). Results for Nanjing and Hemuqiao show optimal integration windows of t = 2 and t = 5 days, respectively. For Wucheng, a local correlation minimum occurs at t = 6 days; while correlations become slightly more negative beyond t = 9, we selected t = 6 to remain consistent with typical synoptic timescales reported in previous studies^14,19,33. In contrast, Changsha shows the most negative correlation at t = 1 day. In the framework of the regional amount effect^14,18,19,33, these optimal t values primarily reflect the isotopic memory of the air mass, representing the characteristic timescale over which upstream convective processes cumulatively imprint the isotopic signal on transported vapor. The variation in t across sites likely reflects differences in moisture transport pathways and the residence time of water vapor within the dominant convective systems. Notably, slightly adjusting the optimal t values does not substantially alter the spatial correlation patterns or the conclusions. While determining the precise controls on these site-specific timescales warrants future investigation, these windows were used to accumulate (precipitation) or average (other variables) meteorological fields in subsequent analyses.

Fig. 3: Determination of the optimal integration window (t) for the regional amount effect.

a Time series of the most negative correlation coefficients between Meiyu season daily precipitation δ¹⁸O at Nanjing and precipitation amount accumulated at each grid point within the extended Meiyu region (25–35°N, 110–125°E) over varying days prior to δ¹⁸O observations. b–d Same as (a) but for Hemuqiao, Wucheng, and Changsha, respectively. N indicates the sample size (number of daily observations). Dashed and dotted horizontal lines indicate significant levels of 90% and 95%, respectively.

Figure 4 illustrates the spatial distribution of correlations between Meiyu season daily precipitation δ¹⁸O and accumulated precipitation amount over the determined optimal t days. For Nanjing, significant negative correlations with P_{acc_2d} are evident in the middle and northern Meiyu region, extending in a southwest-northeast band toward South Korea. This negative correlation pattern does not imply a simple “local amount effect.” Rather, it reflects the spatial footprint of the organized Meiyu front. It aligns with the spatial orientation of the Meiyu rainband (Fig. 1), confirming that isotopic depletion is driven by the organized rainband. The absence of significant negative correlations further south (20–28°N) for Nanjing is physically consistent with the fact that the primary monsoon rainband shifts northward to the Yangtze River Basin during the Meiyu season⁶. Consequently, the region to the south acts primarily as a moisture conduit dominated by the Low-Level Jet³⁷, where moisture transport is efficient and less affected by the intense frontal rainfall and convection that characterizes the Meiyu convergence zone itself. Crucially, a distinct dipole emerges: while correlations are negative along the northwestern flank of the WPSH (the Meiyu rainband), significant positive correlations appear in the areas stretching from the northern South China Sea (SCS) to the western Pacific (Fig. 4a). Physically, this dipole pattern delineates the contrasting regimes of the WPSH: the negative correlation zone corresponds to the moisture-converging northwestern flank, while the positive correlation zone lies within the subsiding main body of the high. This positive correlation contrasts with previous studies in South China at Guangzhou and Hong Kong, where stronger convection in these areas would lead to lower δ¹⁸O values^14,38. This divergence highlights a fundamental spatial shift: as the monsoon marches northward, the Meiyu region enters a regime where isotopic variability is modulated by the interplay between the WPSH flank (isotopic depletion) and the WPSH body (isotopic enrichment), rather than a simple uniform convection effect.

Fig. 4: Spatial fingerprints of the regional amount effect at the four sites.

a Spatial distribution of correlation coefficients between Meiyu season daily precipitation δ¹⁸O at Nanjing and accumulated precipitation over the optimal 2-day window (P_{acc_2d}). The corresponding mean location and extent of the western Pacific subtropical high (WPSH) are depicted by the 5860 and 5880 gpm contours (thick dashed and solid contours, respectively). b–d Same as (a) but for Hemuqiao (P_{acc_5d}), Wucheng (P_{acc_6d}), and Changsha (P_{acc_1d}), respectively_. Only values passing the 90% significance level are shown. N indicates the sample size. Black dots denote the site. Yellow lines depict the Meiyu region boundaries. Maps created using Cartopy. Coastlines are from the Natural Earth public domain dataset. The Meiyu boundaries were processed in QGIS based on the Meiyu Monitoring Dataset.

At Hemuqiao (Fig. 4b), situated in the central Meiyu region, significant negative correlations with P_{acc_5d} are observed within the middle sectors of the Meiyu region. However, significant negative correlations are almost exclusively absent in the Meiyu region but shifted southward to the southern periphery or south of the Meiyu region for the southern sites at Wucheng and Changsha (Fig. 4c, d). Quantitively, significant negative correlations are found in 6.2% and 6.1% of grid points in the proximal Meiyu sector (defined here as 28–33°N, 110–122°E) for Wucheng and Changsha, respectively. This confirms that for the southern Meiyu margin, moisture transport from farther south plays the dominant role. In summary, Meiyu variability is effectively captured by δ¹⁸O in the northern and middle sectors (where the site lies downstream of the rainband), but the southern sector is dominated by variability from further south rather than the Meiyu front itself. It is important to clarify that “upstream” is relative: for Nanjing, the upstream region is the Meiyu belt itself; for Changsha, the upstream region lies further south in southern China.

To further characterize the spatial precipitation patterns driving isotopic extremes, we performed a composite analysis (Fig. 5 and S5-S7). High δ¹⁸O observations were selected as the top ten daily δ¹⁸O values during the Meiyu seasons for Nanjing, Hemuqiao, and Wucheng, respectively, while low δ¹⁸O comprised the bottom ten. For Changsha, where data is limited, the top five and bottom five daily δ¹⁸O values during the Meiyu season were used. High δ¹⁸O cases at Nanjing correspond to relatively dry conditions across East Asia (Fig. 5a); although the Meiyu region remains wetter than its surroundings (reflecting seasonal climatology), the frontal intensity is clearly suppressed. In sharp contrast, low δ¹⁸O events are associated with an intense, organized rainband extending from the southwest corner of the Meiyu region to South Korea, with maximum rainfall (>60 mm day^-1) located several hundred kilometers upstream (southwest) of Nanjing. This wetting is juxtaposed with heightened aridity south of the rainband during low δ¹⁸O episodes. This specific spatial configuration reinforces the correlation results (Fig. 4a), demonstrating that rigorous isotopic depletion at Nanjing requires an organized, intense Meiyu front rather than isolated local convection. While some high δ¹⁸O events tend to cluster near the Meiyu onset (Fig. 2), we find it is unlikely to bias these composites for two reasons. First, the background intraseasonal changes in δ¹⁸O (~1‰ per stage; Fig. S1) are an order of magnitude smaller than the synoptic difference between the high and low δ¹⁸O composite groups (13.2‰). Consequently, the composites primarily capture distinct circulation regimes of active and suppressed phases rather than a seasonal baseline. Second, the widespread dry anomalies observed in the high δ¹⁸O precipitation composite (Fig. 5a) physically resemble the break phases of the Meiyu system^6,39, regardless of their specific timing. Similarly, composite analyses for Hemuqiao, Wucheng, and Changsha (Figs. S5–S7) also support the findings from Fig. 4. Overall, these correlation and composite analyses indicate that variations in Meiyu precipitation are more effectively captured by δ¹⁸O signals at middle and northern sites, while δ¹⁸O at southern locations correlates more significantly with precipitation patterns south of the Meiyu region.

Fig. 5: Precipitation anomalies associated with isotopic extremes at Nanjing.

Composite of accumulated precipitation over 2 days prior to δ¹⁸O observations (P_{acc_2d}; mm day^-1) for a the ten highest and b the ten lowest Meiyu season δ¹⁸O cases at Nanjing. c The difference of P_{acc_2d} between low and high (low minus high) δ¹⁸O cases. Dotted areas in (c) indicate differences that do not pass the 90% significance level. The black star denotes the location of Nanjing. Black lines depict the Meiyu region boundaries. Maps created using Cartopy. Coastlines are from the Natural Earth public domain dataset. The Meiyu boundaries were processed in QGIS based on the Meiyu Monitoring Dataset.

Atmospheric circulation anomalies associated with Meiyu season precipitation δ¹⁸O variability

As demonstrated above, East Asian Meiyu precipitation variability is effectively captured by δ¹⁸O in precipitation within the middle and northern parts of the Meiyu region, specifically at Hemuqiao and Nanjing. This section examines the atmospheric circulation and moisture transport anomalies linked to these δ¹⁸O variabilities, with a particular emphasis on Nanjing due to its extensive observational records.

Figure 6 shows geopotential height at 500 hPa, wind fields at 850 hPa, and vertically integrated moisture flux for high and low δ¹⁸O cases at Nanjing, as well as their differences. The 5860 gpm contour remains relatively stable over land in both cases but exhibits a significant northward migration over the western Pacific during low δ¹⁸O cases (Fig. 6a, b). In contrast, the core of the WPSH, defined by the 5880 gpm contour, shows a pronounced westward extension and northward shift during low δ¹⁸O cases. This results in a significant intensification of geopotential height at 500 hPa over the western Pacific (anomalies > 20 gpm) between ~25–40°N, with a slight decrease in the southern regions (Fig. 6c). These changes in the WPSH modulate the atmospheric circulation and moisture transport patterns, particularly along its flanks. High δ¹⁸O cases at Nanjing are associated with southerly winds (~4 m s^-1) spanning from the northern SCS to Nanjing (Fig. 6d). In contrast, during low δ¹⁸O episodes, much stronger (>8 m s^-1) southwesterly winds dominate south of Nanjing, accompanied by a sharp dissipation of southerly winds north of Nanjing, indicating enhanced convergence along a robust horizontal wind shear zone (Fig. 6e). The atmospheric circulation anomalies between low and high δ¹⁸O cases are marked by significantly intensified southwesterly winds at 850 hPa along the 5860 gpm contour (Fig. 6f). This anomaly aligns with the northwestward extension of the WPSH center (5880 gpm). Such changes in the WPSH and wind patterns facilitate enhanced water vapor transport and the formation of organized rain-bearing systems.

Fig. 6: Atmospheric circulation anomalies driving isotopic variability at Nanjing.

a–c Composites of 500 hPa geopotential height (gpm) for a high δ¹⁸O cases, b low δ¹⁸O cases, and c their difference (low minus high), respectively. d–f Same as (a–c) but for 850 hPa wind (vector and shading; m s^-1). g–i Same as (a–c) but for vertically integrated moisture flux (vectors; kg m^-1 s^-1) and its divergence (shading; 10^-5kg m^-2 s^-1). Dotted areas in (c) indicate differences that do not pass the 90% significance level. Only differences passing the 90% significance level are shown in (f) and (i). Thick dashed and solid contours indicate the 5860 and 5880 gpm levels, respectively (black for high δ¹⁸O, brown for low δ¹⁸O). Maps created using Cartopy. Coastlines are from the Natural Earth public domain dataset. The Meiyu boundaries were processed in QGIS based on the Meiyu Monitoring Dataset.

Further analysis of the vertically integrated moisture flux and its divergence shows that the Meiyu region is already characterized by southwesterly moisture flux and convergence during high δ¹⁸O cases at Nanjing (Fig. 6g). In contrast, during low δ¹⁸O cases, while the overall moisture transport pattern from the northern SCS to the Meiyu region remains largely consistent directionally, its magnitude significantly amplifies over the Meiyu region (>500 versus ~250 kg m^-1 s^-1) (Fig. 6h). While changes in moisture flux and divergence are observable across a broader spatial scale, the significant differences between low and high δ¹⁸O cases are concentrated along the northwestern flank of the WPSH, including the Meiyu region itself (Fig. 6i). Furthermore, the strengthened moisture divergence (or weakened moisture convergence) over areas from the northern SCS to the western Pacific during low δ¹⁸O cases aligns with the reduced precipitation documented in these regions (Fig. 5). Following the concept of the regional amount effect^18,36, this marked intensification of moisture convergence over the Meiyu region (Fig. 6) favors decreased δ¹⁸O values in vapor and precipitation.

Beyond horizontal circulation, the northwestward extension of the WPSH fundamentally alters vertical motion and convective depth. Figure 7 illustrates the vertical velocity at 500 hPa and Outgoing Longwave Radiation (OLR; W m^-2) for high and low δ¹⁸O cases at Nanjing. During high δ¹⁸O cases, vertical ascent over the Meiyu region is present but weak (Fig. 7a), consistent with the weaker convection (OLR > 210 W m^-2) located upstream to the south and southwest (Fig. 7d). In contrast, during low δ¹⁸O cases, the WPSH extension significantly intensifies upward air motion along its northwestern flank while suppressing it over the northern SCS and the western Pacific (Fig. 6b, c). Vertical profiles reveal that this ascending motion over the Meiyu region deepens significantly, extending from ~950 to ~200 hPa, and is accompanied by significant lower-tropospheric moistening below ~600 hPa (Fig. S8). Crucially, this dynamic lifting couples with thermodynamic instability: in the middle and northern Meiyu regions, the advection of high equivalent potential temperature (${\theta }_{e}$) at 850 hPa provides the necessary fuel, while decreased convective inhibition (CIN) facilitates the release of this instability (Fig. S9). These conditions trigger deep convection, evidenced by OLR values dropping below 200 W m^-2 (Fig. 7e, f), leading to the rigorous isotopic depletion.

Fig. 7: Vertical motion and convection anomalies associated with isotopic extremes at Nanjing.

a–c Composites of 500 hPa vertical velocity ($\omega$; hPa s^-1; negative values indicate ascent) for a high δ¹⁸O cases, b low δ¹⁸O cases, and c their difference (low minus high), respectively. d–f Same as (a–c) but for Outgoing Longwave Radiation (OLR; W m^-2). Dotted areas in (c) and (f) indicate differences that do not pass the 90% significance level. Maps created using Cartopy. Coastlines are from the Natural Earth public domain dataset. The Meiyu boundaries were processed in QGIS based on the Meiyu Monitoring Dataset.

However, a distinct regime emerges concurrently in the south (roughly 18–28°N) in these composites, standing in sharp contrast to the favorable conditions observed over the Meiyu region itself. This region exhibits mid-tropospheric drying and subsidence, associated with the WPSH extension (Fig. 7 and S8). Despite the presence of high instability (high ${\theta }_{e}$ typically > 345 K) and abundant moisture, the increased convective inhibition creates a “suppression regime” where convection is capped (Fig. S9). This vertical constraint suggests a nonlinear isotopic response to the WPSH: unlike the linear coupling in the northern Meiyu region, isotopic depletion in the southern Meiyu region appears to be threshold-dependent, requiring strong dynamic forcing to overcome the inhibition. This hypothesis of spatially distinct driving mechanisms is quantitatively tested in the following section.

Different roles of moisture transport and convective triggering

The distinct atmospheric anomalies observed between the northern and southern regions suggest that precipitation δ¹⁸O is governed by spatially varying mechanisms. To clarify this, we first examined the Lagrangian moisture sources (Fig. 8), which reveal a sharp contrast in moisture origins. For the northern site (Nanjing), moisture is derived largely from the Meiyu region itself or transported through it from the southwest (Fig. 8a). To quantify these contributions, we integrated moisture uptake over representative geographic sectors. Specifically, 26.1% of moisture originates directly from the proximal Meiyu sector (28–33°N, 110–122°E), with substantial contributions (59.4%) from the broader middle and southern China domain (20–35°N, 100–122°E). Regarding oceanic contributions, the SCS sector (0–20°N, 100–122°E) contributes 15.5%, whereas the western Pacific sector (0-30°N, 122–160°E) contributes only 3.2%. This dominance of the southwestern pathway confirms that moisture is primarily transported via the monsoon flow rather than direct easterly trade winds.

Fig. 8: Lagrangian moisture sources for Meiyu season precipitation.

a Moisture source fraction (shading; 10^-3) for Meiyu season precipitation at Nanjing between 2012 and 2021. b–d Same as (a) but for Hemuqiao (2018–2019), Wucheng (2016–2019), and Changsha (2010). Moisture source contributions are calculated as the ratio of moisture uptake from each 1° × 1° grid box to the precipitation at the site. Vectors show the vertically integrated moisture flux (kg m^-1 s^-1) averaged during the corresponding period. Maps created using Cartopy. Coastlines are from the Natural Earth public domain dataset. The Meiyu boundaries were processed in QGIS based on the Meiyu Monitoring Dataset.

The spatial shift in moisture sources is not abrupt but gradual. The intermediate sites (Hemuqiao and Wucheng; Fig. 8b, c) exhibit a progressive southward and westward shift in moisture origins, bridging the distinct regimes of the northern and southern Meiyu regions. For the southern site (Changsha), the Meiyu region lies almost exclusively downstream; consequently, moisture contributions from regions north of Changsha account for only 5.5%. This source distribution explains why δ¹⁸O at Changsha is decoupled from precipitation variability within the Meiyu region (Fig. 4d). Furthermore, this spatial shift illuminates the complex relationship with the WPSH: while δ¹⁸O at Changsha correlates negatively with precipitation in the WPSH main body (Fig. 4d), Nanjing exhibits a positive correlation (Fig. 4a). These opposing signals suggest that the balance between moisture supply (thermodynamics) and vertical lift (dynamics) shifts fundamentally across the region.

To robustly distinguish these competing controls, we integrated statistical partial correlation analysis with physical precipitation decomposition (Figs. 9 and 10). Both methods yield consistent results revealing spatially distinct controls between the northern (“coupled regime”) and southern Meiyu regions (“dynamic-limited regime”). For the northern site at Nanjing, precipitation δ¹⁸O retains significant correlations with both moisture flux (Fig. 9a) and vertical velocity (Fig. 9b) when the other variable is controlled. This finding is independently corroborated by the decomposition analysis, where δ¹⁸O at Nanjing correlates significantly with both the thermodynamic ($\Delta {P}_{{thermodynamic}}$) and dynamic ($\Delta {P}_{{dynamic}}$) components. Mechanistically speaking, these results indicate a moisture-convection coupling: isotopic depletion in the active Meiyu region is co-limited by the supply of water vapor and the strength of convective triggering. Notably, the WPSH suppresses convection within its main body, despite enhancing convection at its northwestern flank (Fig. 9).

Fig. 9: Disentangling the roles of moisture transport and vertical instability at Nanjing.

a Spatial distribution of partial correlation coefficients between Meiyu season daily precipitation δ¹⁸O (2012–2021) and vertically integrated moisture flux, controlling for 500 hPa vertical velocity ($\omega$). b Partial correlation between δ¹⁸O and $\omega$, controlling for vertically integrated moisture flux. c Correlation between δ¹⁸O and the thermodynamic component of precipitation anomalies. d Correlation between δ¹⁸O and the dynamic component. The corresponding mean location and extent of the western Pacific subtropical high (WPSH) are depicted by the 5860 and 5880 gpm contours (thick dashed and solid contours, respectively). Only values passing the 90% significance level are shown. Maps created using Cartopy. Coastlines are from the Natural Earth public domain dataset. The Meiyu boundaries were processed in QGIS based on the Meiyu Monitoring Dataset.

Fig. 10: Disentangling the roles of moisture transport and vertical instability at Changsha.

a Spatial distribution of partial correlation coefficients between Meiyu season daily precipitation δ¹⁸O (2010) and vertically integrated moisture flux, controlling for 500 hPa vertical velocity ($\omega$). b Partial correlation between δ¹⁸O and $\omega$, controlling for vertically integrated moisture flux. c Correlation between δ¹⁸O and the thermodynamic component of precipitation anomalies. d Correlation between δ¹⁸O and the dynamic component. The corresponding mean location and extent of the western Pacific subtropical high (WPSH) are depicted by the 5860 and 5880 gpm contours (thick dashed and solid contours, respectively). Only values passing the 90% significance level are shown. Maps created using Cartopy. Coastlines are from the Natural Earth public domain dataset. The Meiyu boundaries were processed in QGIS based on the Meiyu Monitoring Dataset.

The southern Meiyu margin (Changsha) exhibits a decoupling of these factors. Here, precipitation δ¹⁸O shows no significant correlation with moisture flux in proximal areas, but significant correlations are found further upstream in the southwest (Fig. 10a). In contrast, precipitation δ¹⁸O shows significant negative correlations with vertical velocity at 500 hPa within a 2-3° zonal band surrounding Changsha (Fig. 10b). The decomposition analysis further reveals that δ¹⁸O at Changsha tracks the dynamic component almost exclusively, with negligible correlation to the thermodynamic component (Fig. 10c, d). For example, in the key upstream region (25–28.5°N, 110–115°E; identified in Fig. 8d), significant correlations with the dynamic component are found in 64.4% of grid points, compared to only 12.7% for the thermodynamic component (Fig. 10c, d). These results provide strong evidence for the “vertical constraint” mechanism proposed earlier. In the south, the northwestward extension of the WPSH imposes high convective inhibition and mid-level drying (as shown in Fig. 7 and S8). This creates a nonlinear threshold where high moisture availability alone is insufficient to drive isotopic depletion. Consequently, isotopic variability at the southern Meiyu margin is dictated almost entirely by the dynamic triggering required to overcome this WPSH-induced suppression, resulting in the observed high δ¹⁸O values during WPSH extension events (Fig. S10)—opposite to the δ¹⁸O-WPSH relationship at Nanjing.

Relationships on the interannual timescale

The preceding analysis establishes a clear relationship between Meiyu variability and precipitation δ¹⁸O on synoptic timescales. To determine if these mechanisms persist on the interannual timescale, we conducted a correlation analysis between the amount-weighted mean δ¹⁸O of Meiyu season precipitation at Nanjing and the Meiyu season mean P_{acc_2d} from 2012 to 2021 (Fig. 11a). Broadly, the negative correlation within the Meiyu region and positive correlation in regions from South China to the western Pacific, as observed in the daily analysis (Fig. 4a), persist on the interannual timescale.

Fig. 11: Interannual atmospheric controls on Meiyu season δ¹⁸O at Nanjing.

a Correlation between amount-weighted mean Meiyu season δ¹⁸O and Meiyu season mean P_{acc_2d} (2012-2021, derived from IMERG). b Correlation between amount-weighted mean June-July δ¹⁸O and precipitation amount (1987–1992 and 2012–2021, derived from CMFD). c Correlation between amount-weighted mean June-July δ¹⁸O and 500 hPa geopotential height (1987–1992 and 2012–2021). Dotted areas indicate where the correlations do not pass the 90% significance level. In (c), thick black contours show the western Pacific subtropical high (WPSH; 5860/5880 gpm) in 2019 (highest June-July δ¹⁸O), and brown contours show 2016 (lowest June-July δ¹⁸O). The black stars denote the location of Nanjing. The black lines depict the Meiyu region boundaries. Maps created using Cartopy. Coastlines are from the Natural Earth public domain dataset. The Meiyu boundaries were processed in QGIS based on the Meiyu Monitoring Dataset.

However, the short observation period (N = 10 years) may limit the statistical robustness of this interannual analysis. To extend the record, we utilized data from Nanjing’s operation as a GNIP station during 1987-1992, which provided monthly precipitation isotope monitoring. We first examined the representativeness of Meiyu season δ¹⁸O relative to seasonal and annual δ¹⁸O values (Fig. S11). Overall, the Meiyu season δ¹⁸O is lower than the June-July δ¹⁸O (when most Meiyu activity occurs), the May-October δ¹⁸O (the rainy half-year), and the annual weighted mean δ¹⁸O values (Fig. S11). Crucially, however, the variability is highly coherent: the amount-weighted mean Meiyu season δ¹⁸O correlates strongly with June-July δ¹⁸O (R² = 0.81), May-October δ¹⁸O (R² = 0.82), and annual weighted mean δ¹⁸O (R² = 0.76). These findings underscore the robustness of using Meiyu season δ¹⁸O as a reliable indicator for broader seasonal and annual variations.

Given that the Meiyu season typically spans from mid-June to mid-July, we used the June-July weighted-mean δ¹⁸O to further discuss interannual variations over the combined period (1987-1992 and 2012-2021; N = 16 years). As shown in Fig. 11b, June-July δ¹⁸O is significantly negatively correlated with precipitation amount in the Meiyu region. This confirms that the “Meiyu footprint” observed in daily data scales up to control interannual variability. In addition, June-July δ¹⁸O exhibits significant negative correlations in North China, while positive correlations are observed south of the Meiyu region. This overall dipole correlation pattern closely resembles the leading mode of interannual East Asian summer precipitation variations in recent decades, characterized by the “Southern Flood–Northern Drought” (or vice versa) pattern⁴⁰.

Regarding atmospheric circulation, the interannual variation of June-July δ¹⁸O at Nanjing shows significant negative correlations with geopotential height at 500 hPa over the western Pacific (Fig. 11c), mirroring the daily-scale findings. This suggests that lower June-July δ¹⁸O values are associated with a northwestward extension of the WPSH on interannual timescales. This relationship is further supported by examining specific years. The lowest June-July δ¹⁸O in 2016 coincided with a northwestward extension of the WPSH, whereas the highest δ¹⁸O in 2019 was characterized by a retreated WPSH (Fig. 11c). While previous studies have linked low-latitude East Asian precipitation isotopes to El Niño Southern Oscillation (ENSO)^12,41, we find no significant direct correlation between Nanjing June-July δ¹⁸O and the Oceanic Niño Index (ONI) from the Climate Prediction Center in the current dataset (r = 0.34, p = 0.2). This suggests that precipitation δ¹⁸O in the Meiyu region serves as a direct footprint of the regional atmospheric circulation (WPSH) rather than a linear proxy for remote oceanic forcing.

Discussion

Although Chinese speleothem δ¹⁸O records have provided invaluable insights into past hydroclimate variations, their climatic interpretation remains a subject of considerable debate^8,29,42. Central to this controversy is the question of what specific spatial footprint of the monsoon system these δ¹⁸O records reflect, a complexity compounded by the intricate nature of the Asian Summer Monsoon itself. Conflicting interpretations range from local precipitation amount^43,44 to precipitation changes in remote regions, such as the Indian monsoon²⁴. Bridging these perspectives, our findings on precipitation δ¹⁸O across the Meiyu region corroborate the “regional amount effect” framework^18,36, demonstrating that rainfall and convection along upstream moisture transport pathways play the dominant role.

Crucially, our analysis refines the definition of “upstream” for this transitional climate zone. We find that atmospheric processes linked to rainfall and convection over a specific integration window (several days) prior to arrival are the primary drivers. Consequently, precipitation δ¹⁸O at sites in the middle and northern Meiyu region (e.g., Nanjing) is closely tied to the Meiyu rainband itself, as moisture traverses the active convection zone before arriving. In contrast, at the southern Meiyu margin, precipitation δ¹⁸O shows no direct connection to Meiyu precipitation, as the moisture is transported from regions farther south where the imprint of the Meiyu front is absent. These results align with recent findings from other parts of East Asia, suggesting that precipitation δ¹⁸O integrates rainfall and convection processes within a finite spatial scale—typically several hundred to over a thousand kilometers upstream^{14,15,16,17,33,35}. This conclusion is further supported by recent tree ring δ¹⁸O studies in South and East China, which similarly indicate that precipitation signals captured by tree ring δ¹⁸O records are also limited to a relatively confined geographic area^45,46,47. As correlations with remote regions may be inflated by spatial autocorrelation of atmospheric variables³⁸, our findings argue for a shift in focus toward a reevaluation of the mechanisms linking δ¹⁸O variations to atmospheric variables in distant regions, highlighting the importance of carefully considering the specific regional integration footprints of individual proxy sites.

Beyond defining spatial scales, our study offers a novel dynamic mechanism for understanding the spatial heterogeneity of the East Asian Summer Monsoon (EASM). The EASM is not a monolith but exhibits marked variability across different latitude zones over a range of timescales^6,25,26. The capability of δ¹⁸O to capture this sub-regional heterogeneity is critical. Recent high-resolution speleothem δ¹⁸O records reveal notable differences across sub-regions of East Asia^29,42, a feature our modern process study explains dynamically.

Unlike previous studies at lower latitudes in East Asia that emphasize the migrating intertropical convergence zone (ITCZ) or the ENSO-driven Walker Circulation^12,14,17, our findings identify the western Pacific subtropical high (WPSH) as the dominant driver of isotopic variability in subtropical East Asia. Specifically, we establish a distinct “North-South Isotopic Dipole” driven by the WPSH, characterized by two contrasting regimes: (1) In the Northern Meiyu region (Coupled Regime): The northwestward extension of the WPSH enhances low-level moisture transport and vertical ascent along its northwestern flank. Here, isotopic depletion is driven by a coupling of thermodynamic fuel and dynamic lift, leading to deep convection and low δ¹⁸O. (2) In the Southern Meiyu region (Suppressed Regime): The WPSH extension imposes subsidence and high convective inhibition (CIN) under its main body. Here, isotopic variability is dynamically limited; despite high moisture availability, convection—and the associated isotopic depletion—is suppressed unless strong dynamic forcing overcomes the inhibition.

This mechanistic understanding explains the observed nonlinear δ¹⁸O-WPSH relationship between the northern and southern Meiyu regions. Importantly, these synoptic mechanisms scale up to interannual timescales, where amount-weighted mean δ¹⁸O at Nanjing exhibits significant correlations with both Meiyu region precipitation and the WPSH extension. These consistent findings confirm that East Asian Meiyu variability is effectively captured by δ¹⁸O in the middle and northern Meiyu regions. However, paleo-δ¹⁸O records from the southern Meiyu margin likely capture a different dynamic response to the same large-scale circulation, necessitating a spatially differentiated interpretation of East Asian proxy records. Given the complexity of these nonlinear responses, future coordinated monitoring of precipitation and water vapor isotopes across East Asia at high temporal resolutions is needed to fully map the boundaries of this north-south contrast.

Methods

Precipitation isotope data

Our work compiles precipitation δ¹⁸O data from multiple sources to ensure good spatial coverage across the Meiyu region in the Yangtze-Huai River Basin (Fig. 1 and Table S1). These data include sub-daily and daily precipitation δ¹⁸O observations from four stations: Nanjing¹⁵, Hemuqiao⁴⁸, Wucheng⁴⁹, and Changsha⁵⁰. The datasets are primarily daily during the Meiyu seasons. In the rare instance of sub-daily sampling (restricted to a single day: 17 June 2010 at Changsha), we converted the data to a daily value by calculating the precipitation-amount-weighted mean δ¹⁸O. This step is strictly based on mass balance conservation to represent the isotopic composition of the accumulated daily water mass and does not introduce statistical artifacts into the broader daily time series.

Among these sites (Table S1), Nanjing provides the longest continuous record of daily δ¹⁸O data, spanning from 2011 to 2021 and covering the Meiyu seasons from 2012 to 2021. To extend our analysis to interannual timescales, we also utilized monthly precipitation δ¹⁸O data for Nanjing (1987–1992) from the Global Network for Isotopes in Precipitation (GNIP). The reported analytical uncertainties for δ¹⁸O in both datasets are ~0.1‰^15,51. Although the daily data and GNIP datasets do not overlap in time, preventing a direct comparison, they exhibit consistent seasonal δ¹⁸O cycles (Fig. S12). This consistency confirms that there are no significant systematic biases between the sampling protocols. For the analysis of interannual variations, daily and monthly isotope data were aggregated to seasonal (Meiyu season or June-July) amount-weighted means. Similar to the daily conversion, this weighting is applied solely to satisfy isotopic mass conservation, ensuring that the seasonal values reflect the total precipitation input.

Meteorological data

To ensure comprehensive temporal coverage, precipitation data were sourced from two high-resolution datasets. The primary dataset consists of half-hourly precipitation data with a 0.1° × 0.1° resolution obtained from the Integrated Multi-satellitE Retrievals for Global Precipitation Measurement (GPM) mission (IMERG V07)⁵². These were accumulated to calculate daily total precipitation (mm day^-1). To extend the analysis to the historical period (1987-1992) for interannual discussion, we supplemented monthly precipitation data (0.1° × 0.1°) for 1987-1992 and 2012-2021 from the China Meteorological Forcing Data (CMFD) version 2.0⁵³. A comparative evaluation confirms that IMERG and CMFD exhibit consistent interannual variability for June-July precipitation, particularly within the Meiyu region (Fig. S13), justifying their combined application. Hourly meteorological variables—including geopotential height (gpm), wind (m s^-1), specific humidity (g kg^-1), temperature (K), convective inhabitation (CIN; J kg^-1), and vertically integrated water vapor flux (kg m^-1 s^-1) and its divergence (kg m^-2 s^-2)—were obtained from the European Centre for Medium-Range Weather Forecasts fifth generation reanalysis (ERA5)⁵⁴ at a resolution of 0.25° × 0.25°. The geopotential height at 500 hPa was used to characterize the western Pacific subtropical high (WPSH). Specifically, the location and extent of the 5860 and 5880 geopotential meter (gpm) contours at 500 hPa were used to describe the WPSH boundaries^55,56. In addition, daily Outgoing Longwave Radiation (OLR; W m^-2) data (1.0° × 1.0°) from the National Centers for Environmental Information (NCEI) were incorporated as a metric of convection depth and intensity. For the Lagrangian moisture source diagnostic, meteorological fields (1° × 1°, 3 h resolution) from the Global Data Assimilation System (GDAS) were used together with ERA5 data.

The East Asian Meiyu season typically commences around mid-June and retreats during mid-July⁴. In this study, we employed standardized onset and retreat dates based on the Meiyu monitoring index from the National Standardization Administration of China, as identified by Yao, et al.⁵⁷ for 1961–2020. For 2021, as the dataset was not yet updated, the onset and retreat dates were determined to be 9 June and 11 July, respectively, based on official announcements from meteorological services following the same monitoring standard. Geographically, the Meiyu region extends approximately from 28°N to 34°N and 110°E to 123°E (Fig. 1a), and is divided into three subregions from south to north following the boundaries defined by the National Standardization Administration of China, as digitized and provided by Yao, et al.⁵⁷. While the onset and retreat dates can exhibit slight latitudinal variations between the northern and southern sectors, we applied the unified dates defined for the entire Meiyu region. This approach ensures methodological consistency, anchoring our analysis to the macro-scale Meiyu circulation regime rather than local precipitation occurrences. Climatologically, the southern-central sector of the Meiyu region experiences the largest total Meiyu rainfall (Fig. 1a). Meiyu rainfall generally accounts for >60% of total summer (June-August) precipitation (Fig. 1b) and approximately 30% to 40% of the total annual precipitation in the region.

Statistical methods and precipitation decomposition analysis

To determine the relationships between precipitation δ¹⁸O relates and meteorological variables, we employed linear Pearson correlation analysis. We first calculated correlations at the grid-point scale to map the spatial distribution of significant associations between large-scale meteorological fields and local precipitation δ¹⁸O. In addition, composite analysis was utilized to characterize the atmospheric patterns driving isotopic variability. Specifically, we identified extreme cases (e.g., the ten highest daily δ¹⁸O values recorded at Nanjing during all Meiyu seasons) and constructed composite averages of precipitation and atmospheric circulation fields to isolate the distinct patterns associated with these anomalies.

To disentangle the potential coupled effects of horizonal moisture transport and convection constrained by vertical instability, we applied partial correlation analysis. We used the 500 hPa vertical velocity ($\omega$) as a proxy for vertical instability and the vertically integrated moisture flux to represent moisture supply. The partial correlation between δ¹⁸O and moisture flux, independent of vertical velocity, is denoted as $r\left({\delta }^{18}O,{Moisture\; flux}\right|\omega )$. Conversely, $r\left({\delta }^{18}O,\omega \right|{Moisture\; flux})$ denotes the partial correlation between δ¹⁸O and vertical velocity when vertically integrated moisture flux is controlled.

To mechanistically distinguish between the effects of moisture availability and vertical instability on precipitation and isotopic variability, we decomposed precipitation changes into thermodynamic and dynamic components using a moisture budget decomposition method^58,59. Specifically, the precipitation anomaly ($\Delta P$) during the Meiyu season is approximated based on the vertical moisture advection equation:

$$\begin{array}{l}\Delta P\approx \Delta {P}_{{dynamic}}+\Delta {P}_{{thermodynamic}}+{Residual}\end{array}$$

(1)

To avoid confusion with the isotopic notation (δ), we used the capital delta $(\Delta )$ to denote the anomaly from the mean, and the overbar ($\bar{X}$) denotes the mean state. The dynamic component is defined as:

$$\begin{array}{l}\Delta {P}_{{dynamic}}=-\frac{1}{g{\rho }_{w}}\int \left(\Delta \omega \cdot \frac{\partial \bar{q}}{\partial p}\right){dp}{\rm{\#}}\end{array}$$

(2)

This term represents precipitation changes driven by anomalies in vertical motion ($\Delta \omega$) acting on the mean background moisture gradient. It serves as a proxy for convective triggering and circulation strength (e.g., ascent or subsidence associated with the WPSH). The thermodynamic component is defined as:

$$\begin{array}{l}\Delta {P}_{{thermodynamic}}=-\frac{1}{g{\rho }_{w}}\int \left(\bar{\omega }\cdot \frac{\partial \Delta q}{\partial p}\right){dp}{\rm{\#}}\end{array}$$

(3)

This term represents precipitation changes driven by anomalies in specific humidity ($\Delta q$) acting on the mean vertical motion. It serves as a proxy for moisture transport and availability. Here, $g$ is the gravity, ${\rho }_{w}$ is the density of water, $\omega$ is the vertical pressure velocity, and $q$ is the specific humidity. The integration is performed from the surface (1000 hPa) to the upper troposphere (100 hPa) to capture the full atmospheric column moisture budget.

Finally, to identify the physical drivers of isotopic variability, we calculated the spatial and temporal correlations between the observed precipitation δ¹⁸O and these two independent physical components ($\Delta {P}_{{thermodynamic}}$ and $\Delta {P}_{{dynamic}}$). The statistical significance levels (p values) of all correlations and composite differences were assessed using the Student’s t-test.

Lagrangian moisture source diagnostic

We quantified moisture sources using backward air mass trajectories calculated via the Hybrid Single-Particle Lagrangian Integrated Trajectory model (HYSPLIT)⁶⁰. The model was driven by a nested meteorological dataset, combining high-resolution ERA5 data (within the target domain of 20–40°N and 100–125°E) with global GDAS data for the outer domain. Trajectories were released from a 5-point cluster (the target site plus 0.1° displacements in cardinal directions) at 25 vertical levels extending from the surface to ~300 hPa. For every precipitating day during the Meiyu season, trajectories were initiated at 3 h intervals and tracked backward for 10 days, yielding a density of 1000 trajectories per day.

Moisture uptake locations and their contributions were quantified following the “WaterSip” Lagrangian moisture source diagnostic method⁶¹. In principle, this method identifies moisture uptake events based on specific humidity increases along the trajectory, accounting for subsequent precipitative losses prior to arrival at the target site. The resulting moisture source contributions were aggregated into 1° × 1° grid boxes to map the spatial distribution of moisture sources. We have previously successfully applied this nested-data approach to Asian monsoon regions^23,33. Further methodological details can be found in Sodemann⁶¹.