Moisture sources and dynamics over the Southeast Tibetan Plateau reflected in dual water vapor isotopes

The Southeast Tibetan Plateau (SETP) has experienced a significant drying trend in recent decades, which is likely linked to shifts in moisture sources. To investigate the roles of ocean surface evaporation, continental air mass intrusion, and precipitation-vapor interactions, we present a three-year daily time series of near-surface vapor δ¹⁸O and d-excess data from the SETP station. Our analysis reveals that apparent negative correlations between d-excess and relative humidity over the Indian Ocean are primarily due to anticorrelated seasonal cycles, which become insignificant or marginal when examined seasonally. This result underscores the need for caution in interpreting d-excess as a conservative tracer of ocean surface evaporation. Instead, we identify local and upstream specific humidity as the primary determinants of non-monsoon season d-excess variability, which is influenced by the intrusion of cold and dry air from upper levels. During the summer monsoon season, both d-excess and δ¹⁸O reflect the effect of raindrop evaporation during transport, which decreases δ¹⁸O but increases d-excess. These findings offer new insights into the use of water isotopes to track moisture sources and circulation changes across the SETP, especially under varying seasonal circulation systems. In particular, the findings for d-excess will contribute to our understanding of shifts in moisture sources and provide a framework for interpreting d-excess in various hydroclimatic applications, including ice core studies.

1 Introduction

The Tibetan Plateau (TP) and its surrounding regions, known as the Third Pole and the Asian Water Tower, are critical sources of freshwater for billions of people (Immerzeel et al., 2020; Yao et al., 2022). Recent climate change has induced significant hydrological shifts, marked by drying trends in the Southeast TP (SETP) and wetting in the north (Yao et al., 2022; Jiang et al., 2023; Zhang et al., 2023). Atmospheric water vapor is the primary input to the hydrological system, making it essential to understand its sources and dynamics to diagnose regional water imbalances. Using a Lagrangian vapor tracking method, Zhang et al. (2023) suggested that the drying trend is associated with meteorological droughts propagating from a shift in moisture source regions. However, their conclusions and methodology are subjects of ongoing debate (Zhang et al., 2025; Zhao et al., 2025), underscoring the need for alternative approaches. As natural tracers of the water cycle, water stable isotopes offer valuable insights into moisture sources and dynamics (Galewsky et al., 2016; Bowen et al., 2019). However, the interpretation of water isotopes on the TP remains challenging because of complex fractionation processes and shifting circulation systems between summer monsoon and westerlies (Yao et al., 2013; Bershaw, 2018; Thompson et al., 2024; Li et al., 2025).

Recent studies have confirmed that monsoon convection at upstream locations along moisture transport pathways, rather than the local precipitation amount, controls summer monsoon season precipitation δ¹⁸O over the southern TP (He et al., 2015; Cai et al., 2017). This is related to the “amount effect” (Dansgaard, 1964), where higher precipitation leads to lower δ¹⁸O values due to continuous rainout associated with stronger convection following the Rayleigh distillation (Kurita et al., 2015; Ruan et al., 2019; Cai et al., 2025). Additionally, interactions between precipitation and water vapor play a significant role in depleting lower tropospheric vapor isotopes (Lee and Fung, 2008; Risi et al., 2008a; Kurita et al., 2011; Cai et al., 2018). Although the regional amount effect prevails during the monsoon season, this relationship weakens or reverses in the non-monsoon season when it is dominated by westerlies. This variability suggests the presence of additional controls, such as moisture source variability, kinetic fractionation, or shifts in atmospheric circulation patterns (Breitenbach et al., 2010; Yao et al., 2013; Cai and Tian, 2020; Guo et al., 2024).

Observations of vapor isotopes could help disentangle the different processes involved in the amount effect, particularly through examining the secondary parameter deuterium excess (d-excess). The d-excess, defined by Dansgaard (1964) as δ²H − 8δ¹⁸O, primarily reflects the effects of kinetic fractionation. During the rainout process, equilibrium fractionation is the dominant mechanism, whereas raindrop evaporation is associated with kinetic fractionation. Furthermore, limited precipitation during non-monsoon seasons makes it challenging to study a full seasonal cycle of the atmospheric water cycle, which can be compensated for by continuous monitoring of vapor isotopes. Although a few stations on the TP have monitored isotopic compositions in the vapor phase (Yu et al., 2015, 2016; Tian et al., 2020; Dai et al., 2021; Chen et al., 2024), there is limited knowledge about vapor d-excess.

Both theoretical predictions and observations over ocean surfaces indicate that d-excess reflects ocean surface evaporation conditions, such as sea surface temperature (SST) and relative humidity normalized to SST (RH_SST) (Craig and Gordon, 1965; Merlivat and Jouzel, 1979; Liu et al., 2014; Bonne et al., 2019). These relationships are frequently used to interpret d-excess values over the TP (Zhao et al., 2012; Shao et al., 2021; Chen et al., 2024; Liu et al., 2024). For example, Shao et al. (2021) reported significant correlations between an ice core d-excess record from the central TP and RH_SST over the northern Bay of Bengal (BOB) and Arabian Sea (AS). However, the correlation coefficient was only −0.44, with a steep slope of −0.99 ‰ %⁻¹, which differs from the typical range observed in oceanic regions (−0.3 ‰ %⁻¹ to −0.6 ‰ %⁻¹) (Uemura et al., 2008; Benetti et al., 2014; Liu et al., 2014; Bonne et al., 2019). This discrepancy suggests additional complexities, such as continental recycling and raindrop evaporation. Furthermore, many studies at other terrestrial sites have also questioned whether d-excess accurately preserves evaporation conditions from oceanic source regions (Welp et al., 2012; Aemisegger et al., 2014; Samuels-Crow et al., 2014; Fiorella et al., 2018; Wei and Lee, 2019).

In addition, ice core d-excess values at high altitudes are generally higher than those observed in precipitation at lower altitudes on the TP (Thompson et al., 2000; Tian et al., 2001; Zhao et al., 2012, 2017; Joswiak et al., 2013; Shao et al., 2021). High vapor d-excess values at high elevations have been observed elsewhere, such as on the Andes (Samuels-Crow et al., 2014). Such elevated d-excess values have been attributed to the mixing with subsiding air (Samuels-Crow et al., 2014; Sodemann et al., 2017). However, this mechanism remains unconfirmed on the TP.

Mountain valleys in the SETP are considered as significant pathways for transporting water vapor into the TP (Araguás-Araguás et al., 1998; Tian et al., 2007; Yao et al., 2013). To investigate these processes, we initiated a water vapor sampling campaign at the South-East Tibetan Plateau Station for integrated observation and research of alpine environment (SETP station) in June 2015. The primary objectives were to explore moisture sources and dynamics and their influence on vapor isotope compositions across different seasons. To achieve these goals, we analyzed the relationships between vapor isotopes and oceanic evaporation conditions, continental air mass intrusions, and precipitation-vapor interactions during different seasons. Finally, we discuss how our findings contribute to the interpretation of ice core records.

2 Data and methods

2.1 Atmospheric water vapor sampling

Vapor samples were collected at the SETP station (29°46^′ N, 94°44^′ E, 3326 m above sea level, and Fig. S1 in the Supplement) using a cryogenic trapping method. The sampling system included an air pump, a linked-ball-shaped glass cold trap, and an electric-powered system that creates and maintains a cold environment filled with 95 % ethanol as cold as below −80 °C. Ambient air was pumped from an inlet positioned approximately 8 m above ground level through a Teflon tube to a glass trap maintained at an operating temperature of −70 °C. The airflow rate was set to approximately 5 L min⁻¹, allowing the collection of 10–20 mL of water samples during each sampling session. The sampling durations were adjusted seasonally: 24 h in summer and extended to 48 h in winter when necessary to ensure adequate sample volume. The samples were collected at 20:00 Beijing Standard Time (12:00 UTC). The efficiency of the trapping method was verified by connecting an additional cold trap to the system, which showed no visible condensation in the additional cold trap (Yu et al., 2015). Further validation was achieved through comparisons with direct measurements using a cavity ring-down spectroscopy (CRDS, model Picarro L2130-i) at Lhasa, southern TP, confirming the reliability of this method for atmospheric water vapor sampling (Tian et al., 2020).

The sampling campaign ran from 25 June 2015 to 14 June 2018, yielding a total of 742 samples. These samples were stored frozen until analysis. Those collected before 28 June 2016 were measured at the Key Laboratory of Tibetan Plateau Earth System, Environment and Resources, Institute of Tibetan Plateau Research, Chinese Academy of Sciences, using a Picarro L2130-i analyzer. Samples collected after 28 June 2016 were measured at the Institute of International River and Eco-security, Yunnan University, using a Picarro L2140-i analyzer. The isotopic values were calibrated using three standard water samples, with the detailed calibration procedures described by Liu et al. (2024). The measurements are expressed relative to Vienna Standard Mean Ocean Water 2 (VSMOW2), with precisions of 0.1 ‰ for δ¹⁸O, 0.4 ‰ for δ²H, and 1.2 ‰ for d-excess.

2.2 Meteorological data

Daily local meteorological data prior to 2018, including the precipitation amount, air temperature, air pressure, and relative humidity at the SETP station, were obtained from the National Tibetan Plateau/Third Pole Environment Data Center (Luo, 2018). The specific humidity (q) at the SETP station was calculated using air temperature, air pressure, and relative humidity data following the established equations outlined in Huang (2018).

We further obtained meteorological variables such as 2 m air temperature, 2 m dew point temperature, SST, and others at 0.25°×0.25° and hourly resolution from the European Centre for Medium-Range Weather Forecasts fifth generation reanalysis (ERA5) (Hersbach et al., 2019). The RH_SST is estimated using ERA5 data: ${\mathrm{RH}}_{\mathrm{SST}}=e_{\mathrm{air}}/e_{\mathrm{sat}}$, where e_air is the vapor pressure of air and where e_sat is the saturation vapor pressure with respect to the SST. Additionally, precipitation data at 0.1°×0.1° and half-hourly resolutions were obtained from the Integrated Multi-satellitE Retrievals for GPM (V07) dataset (Huffman et al., 2023). Moreover, we used ERA5 data and meteorological data at 1°×1° and 3 h resolutions from the Global Data Assimilation System (GDAS) to calculate backward trajectories (see Sect. 2.4 for details).

Statistical analyses primarily involved linear correlations and regressions, with the coefficient of determination (R²) used to quantify the variance explained by each variable. In addition, we used composite analysis to reveal relationships between variables. For example, to identify general patterns in backward trajectories associated with d-excess exceeding 30 ‰, all the days with such high d-excess values were compiled into a collection. A composite map of trajectories from this collection was then constructed to reveal typical pathways under these conditions.

2.3 Theoretical framework for understanding isotope compositions and humidity

In addition to complex atmospheric circulation models, the evolution of vapor isotope compositions during different moistening and dehydration processes can be predicted through a compilation of atmospheric processes such as condensation, mixing, and raindrop evaporation (Worden et al., 2007; Noone, 2012; Galewsky et al., 2016). These processes shape distinct pathways of isotopic evolution in relation to atmospheric humidity.

The Rayleigh distillation model describes the progressive condensation of water vapor (Dansgaard, 1964). The isotopic composition of the remaining vapor, denoted as δ, can be expressed as $\mathit{\delta }=\left(\mathrm{1}+{\mathit{\delta }}_{\mathrm{0}}\right){\left(q/q_{\mathrm{0}}\right)}^{\mathit{\alpha }-\mathrm{1}}-\mathrm{1}$, where q is the specific humidity, and α is the fractionation factor. A subscript of 0 refers to the initial condition of the air mass. Raindrop evaporation introduces further complexity. As raindrops form at higher altitudes where vapor is depleted in heavy isotopes, their partial evaporation affects the surrounding vapor, leading to isotope values lower than those predicted by Rayleigh models (Worden et al., 2007; Risi et al., 2008a). This effect gives rise to “super-Rayleigh” trajectories, characterized by an inflated effective fractionation factor (α_e), defined as ${\mathit{\alpha }}_{\mathrm{e}}=\left(\mathrm{1}+\mathit{\varphi }\right)\mathit{\alpha }$, where ϕ quantifies deviations from equilibrium. Notably, Worden et al. (2007) and Noone (2012) have given different equations for such deviations, and this study aligns with the formulations by Noone (2012).

Air mass mixing also influences humidity and isotopic compositions. When a dry air mass mixes with a moist mass, the specific humidity of the mixed air can be described as $q=f_{\mathrm{dry}}{q}_{\mathrm{dry}}+f_{\mathrm{moist}}{q}_{\mathrm{moist}}$, where f represents the fraction of each air mass, with $f_{\mathrm{dry}}+f_{\mathrm{moist}}=\mathrm{1}$. Isotopic compositions are similarly derived by solving mass balance equations for the light and heavy isotopes, resulting in a hyperbolic relationship between δ and q. In other words, δ×q and q should have a linear relationship in the mixing process (Fiorella et al., 2018). The intercept of the regression between δ and $\mathrm{1}/q$ or the slope between δ×q and q provides an estimate of the moist end member's isotopic composition (Keeling, 1958).

Assuming a surface temperature of 25 °C and a relative humidity of 85 %, we utilize the evaporation model of Craig and Gordon (1965) to determine the isotopic composition of ocean evaporation. This results in δ¹⁸O $=-\mathrm{11.5}\mathrm{‰}$, δ²H = −81.4 ‰, and d-excess =10.6 ‰. These values serve as the wet end member for modeling the moistening process through mixing with ocean evaporation. For the dry end member, we consider a dehydrated air mass from the Rayleigh curve at q=0.5 g kg⁻¹, δ¹⁸O $=-\mathrm{60.3}\mathrm{‰}$, and δ²H = −418.0 ‰ (Fig. S2). The dehydration process via condensation is initiated at a relative humidity of 80% on the mixing line. Similarly, “super-Rayleigh” distillation involving partial rain evaporation also begins from this starting point. We explore two “super-Rayleigh” scenarios: Rain_evap_A assumes 2 % rain evaporation, whereas Rain_evap_B assumes 5 % rain evaporation, which is based on equations from Noone (2012). Additionally, we consider the influence of evapotranspiration over South Asia and the TP on atmospheric humidity and vapor isotope compositions over the SETP. Quantifying the isotopic compositions of land surface evapotranspiration is challenging. Given that precipitation δ¹⁸O over South Asia generally ranges from −1.0 ‰ to −5.0 ‰ (Bowen and Wilkinson, 2002; Terzer-Wassmuth et al., 2021) and that transpiration constitutes two-thirds or more of evapotranspiration (Good et al., 2015; Cao et al., 2022; Han et al., 2022), we assume a δ¹⁸O value of −5.0 ‰ as an upper bound for land surface evapotranspiration. Similarly, we assume a d-excess of 15.0 ‰ for this wet end member.

2.4 Backward trajectory and moisture source diagnostic

To investigate air mass transport and diagnose moisture sources and pathways toward SETP, we calculated backward trajectories using the Hybrid Single-Particle Lagrangian Integrated Trajectory model (HYSPLIT) (Stein et al., 2015). Trajectory calculations were driven by nested ERA5 (within the domain of 0–50° N and 40–120° E) and GDAS (globally but outside of the ERA5 domain) data to achieve higher resolution in the major potential source regions. In addition, the vertical motion was driven by the model vertical velocity. Air parcels were released from 5 locations: the study site and points displaced 0.2° in each cardinal direction. These releases occurred at 7 different vertical levels: 10, 50, 100, 200, 300, 400, and 500 m above ground level. For each day during the sampling campaign, trajectories were initiated every 3 h to calculate 10 d backward trajectories, resulting in 280 trajectories per day. Geographical and meteorological variables, including location, pressure, temperature, specific humidity, rainfall amount, boundary layer height, and terrain height along the trajectories, were stored at hourly intervals.

To quantify moisture contributions along trajectories to SETP's humidity, we applied the Lagrangian moisture source diagnostic method developed by Sodemann et al. (2008). This method uses mass balance principles along trajectories, interpreting increases in specific humidity (forward in time) as moisture uptake and decreases as moisture loss due to precipitation occurs. It also accounts for the reduced contribution of earlier moisture uptake due to precipitation en route. We previously adapted this method to identify moisture sources for precipitation in subregions of South Asia and East Asia (Cai et al., 2018; Cai and Tian, 2020). A more detailed description of the moisture source diagnostic method can be found in Sodemann (2025).

In this framework, the moisture source can be attributed into four categories: contributions within an extended boundary layer over (1) land and (2) ocean, (3) contributions from above the extended boundary layer, and (4) remaining unattributed sources. Following Sodemann et al. (2008), the extended boundary layer was parameterized as 1.5 times the boundary layer height. The diagnostic results indicated that approximately 7.0 % of the moisture arriving at the SETP remained unattributed, confirming that 10 d trajectories are sufficient to diagnose most moisture sources. Overall, the fractions of within-boundary-layer contributions are 60.2 % over land and 5.0 % over ocean, with an additional 27.8 % originating from above the extended boundary layer. To investigate the relationship between vapor d-excess and relative humidity (RH) at moisture sources, we diagnosed a set of RH-related variables. The RH^source is the weighted-mean RH at locations where within-boundary-layer moisture uptakes occur by weighting the individual within-boundary-layer moisture contributions. The RH^source_land and RH^source_ocean are similar to the RH^source but only apply to within-boundary-layer moisture uptakes over land and ocean, respectively. The ${\mathrm{RH}}_{\mathrm{SST}}^{\mathrm{source}\mathrm{\_}\mathrm{ocean}}$ is the weighted-mean RH_SST at locations where within-boundary-layer moisture uptakes occur over ocean by weighting the individual within-boundary-layer moisture contributions.

Additionally, this study emphasizes the contributions of air parcels themselves to the humidity of the SETP. This variable captures the history of the moisture and indicates how much moisture within each air parcel finally reaches the SETP. The moisture contribution of an air parcel to the SETP's humidity is a measure of the importance of upstream air. We calculated weighted-mean upstream variables using the moisture contribution of the air parcel along trajectories as the weight. We also applied K-means clustering to group trajectories, helping to identify major transport pathways. When calculating the mean trajectory for each cluster and the meteorological variables along each mean trajectory, the moisture contribution of the air parcel is also considered as the weight to calculate the weighted means.

Figure 1 Time series of observed vapor δ¹⁸O, d-excess, and daily local meteorological variables from 2015–2018: (a) δ¹⁸O, (b) d-excess, (c) air temperature, (d) relative humidity (RH), and (e) precipitation amount. The light blue shading highlights the summer monsoon season, whereas the light steel blue shading indicates the non-monsoon season.

Figure 2 Relationship between vapor δ²H and δ¹⁸O. The data are presented for different seasons: non-monsoon (November–April) as red dots, summer monsoon (JJAS) as navy diamonds, May as olive squares, and October as cyan triangles. The solid line indicates the global meteoric water line (GMWL). The dashed line indicates the local vapor line (LVL) estimated from all the δ²H and δ¹⁸O data points.

3 Results

3.1 General characteristics of vapor δ¹⁸O, d-excess, and local meteorological variables

Consistent with Yao et al. (2013), we defined June–September (JJAS) as the summer monsoon season. In contrast, November–April (November–April) was designated as the non-monsoon season, with May and October considered transition periods between the two seasons. In general, the δ¹⁸O values are relatively low during the summer monsoon season and relatively high during the non-monsoon season (Fig. 1a). The mean δ¹⁸O values are −18.4 ‰ for the non-monsoon season, −23.3 ‰ for the summer monsoon season, −16.9 ‰ for May, and −22.8 ‰ for October. δ¹⁸O shows a dramatic decrease at the onset of the summer monsoon. Conversely, from the end of the summer monsoon season to spring and early summer, the δ¹⁸O values gradually increase. Although the amount effect significantly influences this region, the seasonal variation in δ¹⁸O does not strictly align with local precipitation patterns. For example, although local precipitation clearly ceases after the summer monsoon (Fig. 1e), δ¹⁸O remains at relatively low levels. This behavior is consistent with precipitation δ¹⁸O values in the SETP, northeast India, and Bangladesh (Yao et al., 2013; Yang et al., 2017; Cai and Tian, 2020).

Although d-excess values are also lower during the summer monsoon season and higher during non-monsoon periods, the timing of seasonal transitions differs from that of δ¹⁸O (Fig. 1b). The mean d-excess values are 18.3 ‰ for the non-monsoon season, 11.9 ‰ for the summer monsoon season, 13.7 ‰ for May, and 14.9 ‰ for October. The highest d-excess values generally occur during the winter months, when the air temperature and relative humidity (RH) are at their lowest levels (Fig. 1c and d). Furthermore, d-excess starts to decrease in spring, earlier than the sharp drop in δ¹⁸O at the onset of the summer monsoon.

The linear relationship between paired δ¹⁸O and δ²H values (Fig. 2), along with their position relative to the global meteoric water line (GMWL, δ²H =8δ¹⁸O +10) (Craig, 1961), provides additional insights into isotopic fractionation processes (Putman et al., 2019). The local vapor line (LVL), estimated from all the δ²H and δ¹⁸O data points, is δ²H =7.96δ¹⁸O +14.04 (R²=0.98). This LVL plots above but approximately parallel to the GMWL. This relatively high intercept of the LVL reflects the continental location of the site and additional kinetic fractionation after ocean evaporation. The δ²H−δ¹⁸O relationship also varies seasonally. During the non-monsoon season, the LVL is δ²H =7.58δ¹⁸O +10.61 (R²=0.96), whereas during the summer monsoon season, it shifts to δ²H =7.53δ¹⁸O +0.91 (R²=0.99). Non-monsoon data primarily plot above both the GMWL and the overall LVL. Conversely, most monsoon season isotope data fall below the overall LVL, although the lowest δ value points during this period are positioned above the overall LVL, suggesting additional kinetic fractionation, such as rain evaporation (He et al., 2024). The vapor isotopes for May resemble those of the non-monsoon season but align more closely with both the GMWL and the LVL, whereas the data for October exhibit behaviors similar to those of the monsoon season observations.

Figure 3 Relationships between vapor isotopes (δ¹⁸O and d-excess) and specific humidity (q) from 2015–2017. (a) Scatter plot of δ¹⁸O against q. (b) Scatter plot of d-excess against q. Each data point is color-coded by month. The reference lines correspond to those in Fig. S2; their interpretations are detailed in Fig. S2 and Sect. 2.3. Note that only data before 2018 are shown (see text for details).

The relationships between δ¹⁸O and specific humidity (q) further indicate distinct seasonal patterns in moisture dynamics (Fig. 3a). Owing to the lack of avilability of local meteorological data for 2018, our analyses focused on data collected before 2018. During the non-monsoon season, particularly in the winter months, most data points are positioned above the Rayleigh distillation line but below a mixing line that represents an upper bound of hypothetical evapotranspiration over South Asia. This suggests a mixture of dry end members and moist end members. In contrast, during the summer monsoon season, data predominantly fall below the Rayleigh line, which is influenced by “super-Rayleigh” processes linked to rain evaporation.

Further insights come from examining δ×q versus q relationships, which highlight seasonal contrasts in moisture source signatures (Fig. S3). For the non-monsoon season, a simple estimation through the linear regression between δ×q and q suggests a moist end member with a δ¹⁸O of $-\mathrm{13.9}\mathrm{‰}\pm \mathrm{0.6}\mathrm{‰}$. The weighted annual mean precipitation δ¹⁸O at our site was approximately −14.5 ‰ (Yao et al., 2013). However, during the monsoon season, the overall estimation of δ¹⁸O for the moist end member through the linear regression between δ×q and q is significantly lower at $-\mathrm{30.9}\mathrm{‰}\pm \mathrm{1.8}\mathrm{‰}$, indicating that an additional moisture source from rain evaporation is more depleted in heavy isotopes. These results align with the distribution of δ¹⁸O-q data below the Rayleigh line during the summer monsoon season (Fig. 3a), underscoring the influence of different moisture sources and processes across seasons.

The relationships between d-excess and q also reflect seasonal contrasts in moisture dynamics (Fig. 3b). During the non-monsoon season months, a negative correlation is observed, where a lower q corresponds to a higher d-excess value (Figs. 1 and 3b). This relationship is particularly pronounced under dry and cold conditions. In contrast, during the summer monsoon season, no clear relationship between d-excess and q is apparent, with d-excess showing considerable variability of approximately 20 ‰ at any given q. These findings suggest that d-excess is less predictable than δ¹⁸O when using q, except at low humidity levels.

Figure 4 Moisture sources and transport pathways during different seasons from 2015–2017. (a) Spatial distribution of relative contributions of moisture from all air parcels over each 1°×1° box (shading) to humidity at the SETP station, along with specific humidity (q) along mean trajectories (weighted by moisture contributions) for the non-monsoon season of November–April (November–April). (b–d) Same as (a), but for the monsoon season of June–September (JJAS, b), May (c), and October (d), respectively. The dotted yellow and dashed green contours indicate the moisture contributions at 0.1 % and 1 %, respectively. The yellow crosses indicate the location of the SETP station. The black solid lines denote the Tibetan Plateau with an altitude contour at 3000 m.

3.2 Seasonal variability in moisture sources and transport pathways

To understand the drivers behind the seasonal variations in moisture dynamics, we analyzed the moisture sources and transport pathways during different seasons (Fig. 4). Our focus was on the contribution of moisture from historical air masses (last 10 d) to humidity at SETP.

During the non-monsoon season (Fig. 4a), moisture is mainly transported via two main pathways: one originating from the west of SETP, carried by the westerlies (clusters Nov–Apr2 and Nov–Apr3), and the other from the south, such as the BOB (cluster Nov–Apr1). The quantitative contributions are 84.8 % from the southern pathway and 15.2 % from the western branches combined. Interestingly, when considering only trajectories without accounting for moisture contributions, all three clusters appear to originate from the west or southwest of the SETP (Fig. S4). This discrepancy highlights the importance of distinguishing between pure air mass transport and actual moisture sources when trajectory data are interpreted.

In contrast, during the summer monsoon season (Fig. 4b), moisture transport is predominantly from the south of the SETP, which is driven by the summer monsoon. The pathways observed in May (Fig. 4c) represent a transition from the non-monsoon season (Fig. 4a) toward the dominant southerly transport observed during the summer monsoon (Fig. 4b). In comparison, the moisture sources and transport pathways exhibit a slight eastward shift during October compared with those during the summer monsoon season (Figs. 4d and S4d).

Another notable aspect of the moisture source distributions is the dominant contribution from proximal terrestrial regions, particularly those to the south of the SETP (Fig. 4). For example, the 1 % contour representing moisture contributions from air parcels over each 1°×1° grid box does not extend or barely extends into oceanic regions during any of the four seasons. This finding indicates that surface evaporation from oceanic regions such as the BOB and AS contributes minimally. Quantitatively, the within-boundary-layer contributions from oceanic regions are determined to be 2.5 %, 9.1 %, 4.6 %, and 2.0 % for non-monsoon, summer monsoon, May, and October, respectively. Most of the moisture originating over these oceanic regions is lost through precipitation before reaching the SETP, and what remains is replenished by evapotranspiration during transport over land. This finding raises an important question: do the vapor isotopes measured at the SETP still reflect the meteorological conditions at their oceanic sources?

Figure 5 Relationships between vapor d-excess and relative humidity scaled to sea surface temperature (RH_SST). (a) Regression of d-excess against RH_SST (shading and only values significant at the 95 % significance level are shown) and correlation coefficients between them (contours at an interval of 0.1 and only negative correlations are shown) for all the data from 2015–2017. (b) and (c) are the same as (a) but only for the data within the summer monsoon season (JJAS) or the non-monsoon season (November–April), respectively. The black dots indicate the location of the SETP station. The black solid lines denote the Tibetan Plateau with an altitude contour at 3000 m.

3.3 Role of ocean surface evaporation conditions at seasonal and intraseasonal time scales

Relationships between d-excess and ocean surface evaporation conditions, such as RH_SST and SST, were examined using data from 2015–2017 (Figs. 5a and S5a). The results indeed show negative correlations between d-excess and RH_SST over the northern Indian Ocean, particularly in the northern parts of the AS and BOB (Fig. 5a). Specifically, the regression slopes for this relationship across the northern Indian Ocean vary from higher than −0.1 ‰ %⁻¹ to values below −0.6 ‰ %⁻¹. Focusing on specific regions, the northern BOB (10–22° N and 80–99° E) and the eastern AS (7–20° N and 65–78° E; Fig. 5a) presented regression slopes within the range of −0.3 ‰ %⁻¹ to −0.6 ‰ %⁻¹ previously reported (Uemura et al., 2008; Benetti et al., 2014; Liu et al., 2014; Bonne et al., 2019). For example, the regional average RH_SST in the eastern AS shows an overall regression slope of −0.49 ‰ %⁻¹ ($r=-\mathrm{0.52}$ and p<0.01) (Fig. 6a), whereas the northern BOB has a slope of −0.52 ‰ %⁻¹ ($r=-\mathrm{0.55}$ and p<0.01) (Fig. 6b). However, the clustering of data points by season (Fig. 6) suggests that the apparent negative correlations might primarily stem from opposing seasonal trends. Similarly, apparent negative correlations between d-excess and SST also emerge over the northern Indian Ocean (Fig. S5a). However, both theoretical predictions (Merlivat and Jouzel, 1979) and in situ observations above the ocean surface (Liu et al., 2014; Bonne et al., 2019) reveal a positive correlation between d-excess and SST. These discrepancies lead us to speculate that the overall correlations between the SETP vapor d-excess and surface evaporation conditions over the northern Indian Ocean are likely driven by seasonal variability.

Figure 6 Relationships between SETP vapor d-excess and relative humidity normalized to sea surface temperature (RH_SST) averaged over (a) the eastern Arabian Sea (7–20° N and 65–78° E) and (b) the Bay of Bengal (10–22° N and 80–99° E) from 2015–2017. Each data point is color-coded by month. The solid black lines indicate the linear regression between all the data points. The dashed orange lines indicate linear regression for data during the non-monsoon season (November–April), and the dashed dark blue lines indicate data during the summer monsoon (JJAS). The slopes (‰ %⁻¹), r values, and p values for the three data groups are also shown.

The relationship between d-excess and RH_SST was further analyzed by distinguishing between the summer monsoon and non-monsoon seasons. During the summer monsoon season, the negative correlation diminishes significantly, with correlation coefficients dropping below 0.3 (Fig. 5b). In contrast, significant correlations are present during the non-monsoon season (Fig. 5c), potentially due to intraseasonal variations where d-excess peaks in winter and decreases at the beginning and end of the non-monsoon season (Fig. 1b), possibly accompanied by opposing RH_SST trends. Although the correlation is significant during the non-monsoon season, the explained variance in d-excess remains low, at a maximum of 10 %–16 % over the northern BOB. Similarly, correlations with SST over the northern Indian Ocean also become negligible when seasons are considered separately (Fig. S5). To account for transport time, we examined correlations betweend-excess and RH_SST from 1–11 d prior to the d-excess observation dates during the summer monsoon (Fig. S6) and no n-monsoon seasons (Fig. S7), respectively. The results are consistent with those shown in Fig. 5, indicating that considering these lagged timeframes does not enhance the correlation between d-excess and RH_SST.

Figure 7 Composite of moisture sources and transport pathways for high and low d-excess days during the non-monsoon season from November–April. (a) Spatial distribution of the relative contribution of moisture from all air parcels over each 1°×1° box (shading) to humidity at the SETP station, along with specific humidity (q) along mean trajectories (weighted by moisture contributions) for d-excess values higher than 30 ‰ during the non-monsoon season (n=10). (b) Same as (a) but for d-excess values lower than 10 ‰ (n=8). The yellow crosses indicate the location of the SETP station. The black solid lines denote the Tibetan Plateau with an altitude contour at 3000 m.

Table 1 Correlation coefficients between d-excess and the RH-related variables for different seasons from 2015–2017. Values with significance levels exceeding 99 %, between 99 % and 95 %, 95 % and 90 %, are in bold italics, bold, and italics, respectively.

Figure 8 Relationships among vapor d-excess, local specific humidity (q), weighted-mean upstream q, weighted-mean upstream air temperature (T), and weighted-mean upstream air altitude during the non-monsoon season from November–April. (a) Scatter plot of d-excess against local q. (b) Scatter plot of local q against upstream weighted-mean q. (c) Scatter plot of upstream q against upstream air T. (d) Scatter plot of upstream q against upstream air altitude. All the upstream variables are mean values along backward trajectories weighted by the moisture contribution of air parcels. The solid curves indicate the log (a, c, d) or linear (b) regression between the respective variables, with the correlation coefficients indicated by the numbers.

To further explore the relationships between d-excess and RH over moisture source regions, we analyzed the correlations between d-excess and the RH-related variables for all observations from 2015–2017 and during the four seasons (Table 1). Overall, d-excess is significantly negatively correlated with these RH-related variables at moisture sources at the annual scale, whereas correlations during specific seasons weaken to very low levels. The strongest correlation is between d-excess and RH^source from 2015–2017 ($r=-\mathrm{0.48}$, p<0.01). Notably, correlations with ${\mathrm{RH}}_{\mathrm{SST}}^{\mathrm{source}\mathrm{\_}\mathrm{ocean}}$ are weak overall (Table 1). In summary, vapor d-excess at the SETP is less likely a conservative tracer of surface evaporation conditions (neither RH_SST nor SST) over the northern Indian Ocean. Therefore, interpreting d-excess in meteoric water or paleo archives from the TP as a proxy for Indian Ocean evaporation conditions should be approached with caution.

3.4 Role of dry and cold air intrusion during the non-monsoon season

Both theoretical predictions from the Rayleigh model and observations during the non-monsoon season suggest that d-excess increases as q decreases when q reaches extremely low values (Fig. 3b). In addition, the results for both air mass transport and moisture transport reveal the significant role of the westerlies (Figs. S4a and 4a). Based on this evidence, we propose that during the non-monsoon season, vapor isotopes are influenced by the mixing of cold and dry air transported by westerlies from higher altitudes with surface vapor. Furthermore, surface vapor influenced by recycled moisture from terrestrial evapotranspiration would further elevates d-excess at a given q (Fig. 3b).

We performed a composite analysis of moisture sources and transport pathways for the highest (higher than 30 ‰, and n=10) and lowest (lower than 10 ‰, and n=8) d-excess observations during the non-monsoon season (Fig. 7). High d-excess values are associated primarily with moisture transported by westerlies from regions west or southwest of the SETP, such as over the TP and northern India. In addition, backward trajectories for these cases show air masses characterized by extremely low q values, reaching below 2 g kg⁻¹ along the mean trajectories (weighted by moisture contribution) across the TP (Fig. 7a). Conversely, for low d-excess cases, the moisture transport pathways shift toward more humid regions south of the SETP, including northeast India, Bangladesh, and the BOB (Fig. 7b). This contrasting moisture transport pattern between high and low d-excess cases aligns with our hypothesis that high d-excess values are associated with dry and cold air transported by westerlies.

The influence of cold and dry air intrusions was further investigated through an analysis of relationships involving d-excess, local q, weighted-mean upstream q, weighted-mean upstream air temperature, and weighted-mean upstream air altitude (Fig. 8). The upstream variables represent weighted averages along the 10 d backward trajectory, where the weights correspond to the moisture contribution at each time step (Sect. 2.4). The non-monsoon season d-excess shows robust negative correlations with both local q ($r=-\mathrm{0.65}$, p<0.01; Fig. 8a) and upstream q ($r=-\mathrm{0.48}$, p<0.01). Furthermore, local q is strongly linked with upstream q (r=0.83, p<0.01; Fig. 8b), which is associated with air masses characterized by low temperatures and high altitudes (Fig. 8c and d). Additionally, the properties of the upstream air could also impact δ¹⁸O. Indeed, the δ¹⁸O values during high d-excess cases are lower than those during low d-excess cases (at a significance level of 95.3 %). The overall correlation coefficient between δ¹⁸O and d-excess during the non-monsoon season is −0.29 (p<0.01). Notably, the correlations between δ¹⁸O and q are weaker than those observed for d-excess, with local q showing r=0.42 (p<0.01) and upstream q showing r=0.41 (p<0.01). The relationship between non-monsoon season δ¹⁸O and humidity is expressed mainly as the relationship between δ×q and q (r=0.82 for local q and r=0.80 for upstream q). Spatial correlations between vapor isotopes (δ¹⁸O and d-excess) and 2 m air temperature as well as humidity measured by the 2 m dew point temperature also support these findings (Fig. S8). Significant negative correlations between d-excess and dew point temperatures exist over the Southeast TP, northeast India, and northern Bangladesh. In contrast, δ¹⁸O shows significant positive correlations with air temperature over the Indian subcontinent and northwestern Southeast Asia.

As shown in Fig. 3b, extremely high d-excess values are predicted at very low q levels. Previous studies have shown that as q approaches zero, vapor d-excess can approach 7000 ‰ following the Rayleigh distillation trajectory (Bony et al., 2008), a behavior inherent to the definition of d-excess (Dütsch et al., 2017). High d-excess values have also been observed in low-humidity environments, such as polar regions (Bonne et al., 2014; Steen-Larsen et al., 2017) and high altitudes (Webster and Heymsfield, 2003; Samuels-Crow et al., 2014; Sodemann et al., 2017). Therefore, we infer that the increasing trend of d-excess with decreasing local q, upstream q, and regional dew point temperature is due to enhanced mixing with dry and cold subsiding air transported by westerlies from high altitudes. The relationships between upstream q and upstream air temperature as well as altitude further support this inference, indicating that low-humidity conditions are associated with the presence of subsiding dry and cold air from high altitudes (Fig. 8c and d). Therefore, vapor d-excess during the non-monsoon season not only provides insights into specific humidity levels but also reveals the source of humidity.

3.5 Role of precipitation-vapor interaction during the summer monsoon season

In contrast to the significant dependence of d-excess on q during the non-monsoon season, no correlation is detected (r=0.04, p=0.51) during the summer monsoon season. The behavior of δ¹⁸O also differs between the two seasons (Fig. 3). During the summer monsoon season, δ¹⁸O-q plots below the Rayleigh curve, indicating that the vapor has experienced precipitation-vapor interactions through rain evaporation (Fig. 3a). Partial rain evaporation in an unsaturated atmospheric environment leads to kinetic fractionation, which decreases d-excess values in raindrops while increasing d-excess values in the surrounding vapor (Risi et al., 2008b). This effect of precipitation-vapor interactions on vapor isotopes has been suggested as a primary mechanism driving the amount effect in tropical regions (Risi et al., 2008a; Kurita et al., 2011; Galewsky et al., 2016; Bowen et al., 2019). Therefore, we hypothesize that vapor isotopes during the summer monsoon season at the SETP station are influenced by the extent of precipitation-vapor interactions.

The first evidence supporting this hypothesis is the significant correlation between δ¹⁸O and d-excess during the summer monsoon season ($r=-\mathrm{0.55}$, p<0.01; Fig. 9a). In addition, δ¹⁸O and d-excess tend toward weak correlations at high δ¹⁸O values but stronger correlations when δ¹⁸O values are low (Figs. 9a and S9). To explore this further, we categorized days with daily precipitation of at least 2 mm as “rainy days” and those with daily precipitation of less than 2 mm as “non-rainy days”. This distinction is based on the premise that precipitation-vapor interactions cannot occur in the absence of precipitation. The analysis reveals that the δ¹⁸O values during rainy days are significantly lower than those during non-rainy days, whereas the d-excess values show the opposite trend (p<0.01 for both δ¹⁸O and d-excess; Fig. 9b and c). Furthermore, the correlation between δ¹⁸O and d-excess becomes stronger on rainy days ($r=-\mathrm{0.69}$, p<0.01), although a weaker negative correlation persists even on non-rainy days ($r=-\mathrm{0.40}$, p<0.01). Even when a stricter threshold of 0 mm is applied for non-rainy days, the negative correlation between δ¹⁸O and d-excess remains significant ($r=-\mathrm{0.37}$, p<0.01). Moreover, correlations with local precipitation amounts are weak for both δ¹⁸O ($r=-\mathrm{0.31}$, p<0.01) and d-excess (r=0.26, p<0.01). These findings lead us to infer that vapor isotopes during the summer monsoon season at the SETP are influenced by not only local precipitation-vapor interactions but also the history of precipitation-vapor interactions that occurred before the vapor reached the region.

Figure 9 Relationships between SETP vapor d-excess and δ¹⁸O during the summer monsoon season. (a) Scatter plot of d-excess against δ¹⁸O and linear regression lines between them. (b) Distribution of δ¹⁸O values, with the dashed lines indicating values at the lower and upper quartiles and the solid lines indicating the mean values. (c) Same as (b) but for d-excess. Orange colors indicate data observed when the daily precipitation amount is less than 2 mm, and blue colors indicate data observed during days when the precipitation amount is not less than 2 mm. The r values for both lines are indicated in (a), and both are significant at the 0.01 level.

To further investigate the role of precipitation-vapor interactions, we use the total precipitation amount (P_acc) as an indicator of precipitation-vapor interactions, considering the cumulative effect over several days preceding sampling. Our analysis examined correlations between vapor isotopes (δ¹⁸O and d-excess) and P_acc over periods ranging from 1–10 d prior to sampling (Figs. S10 and S11). Vapor d-excess reaches an optimal correlation with P_acc when it is considered 3 d before sampling (P_{acc_3d}). Vapor δ¹⁸O shows a slightly longer memory and reaches an optimal correlation approximately 5–6 d before sampling. Figure 10 shows the spatial distribution of these correlations, where d-excess positively correlates with P_{acc_3d} across an approximately 5°×5° region surrounding the SETP and extending southwestward to the Himalayas (Fig. 10a). In contrast, δ¹⁸O values show significant negative correlations in similar regions (Fig. 10b). Interestingly, even on non-rainy days, significant regional-scale correlations persist, albeit weaker and with a smaller spatial extent (Fig. S12).

These findings provide further insights into understanding the mechanisms driving the amount effect. A negative correlation between δ¹⁸O and P_acc has also been observed in precipitation and can be attributed to either continuous rainout (Kurita et al., 2015; Cai and Tian, 2016; Ruan et al., 2019) or precipitation-vapor interactions (Lawrence et al., 2004; Worden et al., 2007; Risi et al., 2008a; Kurita et al., 2011). Although continuous rainout, explained by the Rayleigh distillation model, accounts for the decreasing trend of δ¹⁸O with increased rainfall, d-excess remains relatively stable unless the specific humidity decreases to very low levels (approximately 4 g kg⁻¹ in Fig. 3b, for example). The positive correlation between vapor d-excess and P_{acc_3d} provides an additional constraint, suggesting that the amount effect is not solely a result of rainout but also involves precipitation-vapor interactions, which significantly influence vapor isotopes in the lower troposphere.

Figure 10 Relationships between vapor isotopes for rainy days (local daily precipitation amount not less than 2 mm) and total precipitation amount at the regional scale during the summer monsoon season. (a) Spatial distribution of correlation coefficients between d-excess and total precipitation amount during the 3 d prior to sampling (P_{acc_3d}). (b) Same as (a) but for δ¹⁸O. Only values significant at the 95 % significance level are shown. The black dots indicate the location of the SETP station. The black solid lines denote the Tibetan Plateau with an altitude contour at 3000 m.

4 Implications for interpreting TP ice core isotope data

Interpreting d-excess in meteoric water and ice cores on the TP is complicated by evaporation conditions over the northern Indian Ocean (RH_SST and SST) and continental recycling (Pang et al., 2012; Zhao et al., 2012; Joswiak et al., 2013; An et al., 2017; Shao et al., 2021). Attempts have been made to establish relationships between vapor d-excess and RH_SST (Chen et al., 2024; Liu et al., 2024), as well as between ice core d-excess and RH_SST (Shao et al., 2021) or SST (Zhao et al., 2012). However, based on our results (Sect. 3.3), the apparent relationships are primarily a result of similarities in the seasonality of these variables. The preservation of oceanic source region conditions by vapor d-excess has also been questioned at other continental sites (Welp et al., 2012; Aemisegger et al., 2014; Samuels-Crow et al., 2014; Fiorella et al., 2018; Wei and Lee, 2019). Instead, these studies emphasized the roles of other processes, such as continental recycling and mixing with subsiding air masses. For example, Aemisegger et al. (2014) suggested that high transpiration contributions could lead to a lack of correlations between d-excess and RH_SST as transpiration is non-fractionating under the steady-state assumption (Yakir and Sternberg, 2000). In addition, uncertainties associated with trajectory calculations and moisture source diagnostics, especially the resolution of the driven data, could also have an impact on the relationship between d-excess and RH_SST. While our study employs nested ERA5 and GDAS data with a resolution of 0.25°×0.25° from ERA5 in key moisture source regions, using coarser GDAS data does not alter our conclusions (not shown). However, higher-resolution data or regional high-resolution models could enhance the accuracy of trajectory calculations and reduce uncertainties.

The direct contribution of oceanic vapor to humidity at the SETP is very limited (Fig. 4), implying an even smaller contribution over the TP, as the SETP is at the forefront of moisture transport toward the TP (Fig. S1). The dominant terrestrial origin indicates significant continental recycling. Terrestrial processes such as transpiration and evaporation introduce isotopically enriched moisture with high d-excess signatures. Interestingly, vapor δ¹⁸O exhibits a noticeable positive correlation with the fraction of within-boundary-layer moisture contribution over land during the non-monsoon season (r=0.47, p<0.01), suggesting that enhanced continental recycling would increase the δ¹⁸O values (Fig. 3a). However, correlations between the fraction of terrestrial moisture sources and δ¹⁸O for other seasons or with d-excess are either insignificant or marginal (Table S1 in the Supplement). Further quantification of the effects of continental recycling on vapor isotopes requires detailed knowledge of the isotopic compositions of evapotranspiration fluxes. In this study, we utilized a simplified assumption regarding the isotopic composition of these fluxes to explore their influence on vapor isotopes. Therefore, future research should prioritize characterizing the isotopic signatures of both evaporation and transpiration fluxes, as well as determining the ratio between these two fluxes. This will provide deeper insights into how continental recycling shapes vapor isotope compositions.

Seasonal changes and long-term variations in precipitation and ice core isotopes have been interpreted as shifts in moisture sources between recycled terrestrial moisture and oceanic sources or their relative contributions (An et al., 2017; Yang and Yao, 2020). Oceanic moisture is typically associated with the summer monsoon, whereas westerlies bring moisture from continental recycling or even the Mediterranean Sea. Water isotopic signatures on the TP are thought to reflect this interplay between the summer monsoon and non-monsoon seasons (Tian et al., 2007; Pang et al., 2012; Joswiak et al., 2013). Despite seasonal shifts in moisture sources, continental recycling prevails throughout the year (Fig. 4). Our alternative perspective explains the high d-excess induced by westerlies as dry and cold air intrusions rather than surface evaporation or evapotranspiration. Although the interplay between the summer monsoon and westerlies remains valid, but we emphasize changes in air mass properties driven by different circulation systems.

The proposed alternative interpretation aligns with findings from the Andes (Samuels-Crow et al., 2014) and Corsica (Sodemann et al., 2017), potentially explaining the abnormally high d-excess in high-altitude ice cores, as mentioned in the Introduction. This is because the specific humidity at these ice core sites is extremely low, and prolonged interactions with cold and dry air may further modify snow isotope compositions (Wahl et al., 2022; Ma et al., 2024). In addition, intense precipitation-vapor interactions during the summer monsoon represent another potential source of elevated d-excess (Sect. 3.5). When this high d-excess vapor contributes to subsequent precipitation, its signal can be inherited in the resulting precipitation (Risi et al., 2008b). However, a clear relationship between TP precipitation d-excess and monsoon convection has yet to be established, partly because of the limited attention given to d-excess in previous studies (Yao et al., 2013). Local raindrop evaporation may counteract this effect by reducing raindrop d-excess values. The overall positive correlation between precipitation d-excess and altitude across Asia has sometimes been attributed to stronger evaporation at lower altitudes (Bershaw, 2018). For snowfall on glaciers, however, evaporation from falling snowflakes is less likely due to cold temperatures and the short distance between the cloud base and the glacier surface. Therefore, elevated vapor d-excess values caused by accumulated precipitation-vapor interactions upstream associated with monsoon convection could be another source of the high d-excess values in ice cores.

5 Conclusions

We present a three-year daily near-surface vapor isotope dataset collected at the SETP station, which is at the major channel for moisture entering the TP. The paired measurements of vapor isotopes and specific humidity reveal distinct moisture sources and dynamics between the non-monsoon and summer monsoon seasons, which is consistent with the findings from the Lagrangian moisture diagnostic method. Despite significant negative correlations between d-excess and normalized RH over the northern Indian Ocean when all seasons are considered, these correlations weaken or even disappear when analyzed within individual seasons.

During the non-monsoon season, vapor d-excess is influenced primarily by specific humidity at both local and upstream scales. Air that has undergone significant dehydration, situated at the lower end of the Rayleigh distillation, is expected to have extremely high d-excess values. Backward trajectory analyses and moisture source diagnostics reveal that the intrusion of cold and dry air driven by westerlies during the non-monsoon season leads to an increasing trend in the d-excess as the specific humidity decreases. This process also contributes to a weak negative correlation between d-excess and δ¹⁸O. Furthermore, compared with the summer monsoon season, δ¹⁸O primarily reflects mixing processes involving a relatively enriched moist end-member.

During the summer monsoon season, rain evaporation and “super-Rayleigh” processes emerge as the dominant processes shaping vapor isotope compositions. First, δ¹⁸O systematically shifts below the Rayleigh distillation curve, aligning with predictions of “super-Rayleigh” distillation caused by partial rain evaporation. Second, δ¹⁸O is inversely correlated with d-excess, indicating that kinetic fractionation is a source of depleted vapor, which cannot be attributed solely to rainout. Third, at the regional scale, δ¹⁸O is significantly negatively correlated with the total precipitation amount, whereas d-excess is positively correlated with the total precipitation amount.

These findings will aid in interpreting δ¹⁸O and d-excess records from Tibetan Plateau glaciers, offering refined insights into past hydroclimatic conditions and challenging assumptions linking ice core isotopes to oceanic evaporation alone. These new insights into vapor d-excess during the non-monsoon season provide an alternative framework for interpreting the high d-excess values in high-altitude TP ice cores. The introduction of high d-excess values by subsidence air from high altitudes could be a general phenomenon, as similar findings have been reported elsewhere (Samuels-Crow et al., 2014; Sodemann et al., 2017). Additionally, the findings on summer monsoon season moisture dynamics help disentangle the different effects of rainout and precipitation-vapor interactions in the context of the amount effect (Galewsky et al., 2016; Bowen et al., 2019). Although this study questions the earlier interpretation of TP d-excess as an indicator of oceanic evaporation conditions (Zhao et al., 2012; Shao et al., 2021; Chen et al., 2024; Liu et al., 2024), other studies have also raised doubts about the preservation of these signals inland (Welp et al., 2012; Aemisegger et al., 2014; Samuels-Crow et al., 2014; Fiorella et al., 2018; Wei and Lee, 2019). Further research is needed to determine how far inland oceanic evaporation signals can be preserved during the transport from coastal areas. Moreover, we acknowledge the use of simplistic assumptions regarding the isotopic compositions of evapotranspiration fluxes, highlighting the need for deeper investigations into the isotopic compositions of these fluxes to comprehend the effects of continental recycling. Furthermore, the focus on lower tropospheric vapor sources contrasts with that on precipitation sources at higher levels, which may differ and require additional exploration. Finally, the resolution of meteorological data may influence the accuracy of trajectory calculations and moisture tracking results. Future research should consider the use of higher-resolution meteorological data or the implementation of regional high-resolution models to increase the precision of these analyses.