The Water Accounting Model-2layers (WAM-2layers) is an Eulerian posterior moisture tracking model which can track tagged moisture both forward and backward in time to determine the spatial and temporal distributions of moisture sources (Van der Ent et al., 2010; Van der Ent, 2014). In comparison with the commonly used Lagrangian models (e.g., the FLEXible PARTicle (FLEXPART) dispersion model and the Hybrid Single-Particle Lagrangian Integrated Trajectory (HYSPLIT) model) that concern the movement of “air particles” in the atmosphere and identify precipitation and evaporation events mainly based on the dynamic humidity information of the tracked particles, Eulerian models principally focus on moisture transport among fixed grids. In general, Lagrangian models are more accurate and run faster than Eulerian ones for short-term moisture tracking of single grid cells, while Eulerian models are more efficient for long-term moisture tracking over large target regions (Tuinenburg and Staal, 2020). More importantly, the selection of WAM-2layers enables us to consider the moisture budget from precipitation and evaporation separately on Eulerian grids (Van der Ent et al., 2013, 2014).

The basic principle of the forward moisture tracking for the tagged moisture (subscript g) in WAM-2layers in the lower layer is the atmospheric water balance (Findell et al., 2019): 1∂Sg,lower∂t=-∂Sg,loweru∂x-∂(Sg,lowerv)∂y+Eg-Pg±Fv,g, where Sg is the moisture storage in the atmospheric column; t is time; u and v are wind speeds in the zonal (x) and meridional (y) directions, respectively; Eg and Pg are evaporation entering and precipitation leaving the layer, respectively; and Fv is the vertical moisture exchange between the lower and upper layers. Note that Eg only applies to the lower layer. The “well-mixed” assumption is adopted in this model, which means that precipitation is assumed to be immediately removed from the atmosphere in the tracking process (i.e., Pg/P=Sg/S, where P and S are total precipitation and total column atmospheric moisture storage, respectively). The two vertical layers in the model are set to deal with the wind shear in the upper air. To better capture the vertical exchanges due to convection, turbulence and re-evaporation and to minimize the water balance losses between the two layers, the gross vertical flow is set to 4 times the vertical flow in the net flow direction and 3 times the vertical flow in the opposite direction. Although this is a simplification of the turbulent moisture exchange, physically reasonable results have been obtained in previous studies, and the general tracking has been validated against the online 3D tracking models (Van der Ent et al., 2013, 2014; Findell et al., 2019).

In the tracking process, the spatial resolution of the Eulerian grids is reduced to 1∘×1∘, and the time step is set as 0.25 h to maintain precision and numerical stability. We have tested the sensitivity of the moisture tracking results to the selection of different time steps. Figure S1 shows the comparison of two simulations using 15 min (0.25 h) and 10 min time steps, suggesting the stability of using different time steps for moisture tracking in the study area. In addition, the vertical separation between the two layers is prescribed as ∼812 hPa at the normal atmospheric pressure (Van der Ent et al., 2013). Note that the atmospheric pressure of the vertical separation varies with different surface pressure (the “half-level” pressure in different reanalysis products is defined as Pk-1/2=Ak-1/2+Bk-1/2Ps, where Ps is surface pressure, k represents different model levels and the values of Ak-1/2 and Bk-1/2 are defined independently for different reanalysis datasets). Based on our preliminary experiments, we select a tracking domain that covers nearly all the potential ocean and land source regions of the TP precipitation (30∘ S–80∘ N and 40∘ W–140∘ E).

In an Eulerian grid cell (x,y) at time t, the relative contribution ( %) of oceanic evaporation to precipitation is defined as 2ρo(t,x,y)=Mo(t,x,y)Mo(t,x,y)+Ml(t,x,y), where Mo is the absolute contribution of oceanic moisture and Ml is the absolute contribution of terrestrial moisture (including local recycling). The ocean and land distributions were defined according to the 1∘×1∘ gridded land–sea mask from the ERA-Interim. We further remove inland large lakes (considered as “sea” in the ERA-Interim, e.g., the Caspian Sea and the Black Sea in the Eurasian continent) from the mask. The final land–sea mask with 1∘×1∘ spatial resolution used in this study is shown in Fig. S2.

Three atmospheric reanalysis products, namely the European Centre for Medium-Range Weather Forecasts (ECMWF) interim reanalysis dataset (ERA-Interim) (Dee et al., 2011), the National Aeronautics and Space Administration Modern-Era Retrospective Analysis for Research and Applications version 2 dataset (MERRA-2) (Gelaro et al., 2017) and the Japanese 55-year Reanalysis dataset (JRA-55) (Kobayashi et al., 2015), are used to drive the WAM-2layers. Variables used in our simulations include surface pressure (temporal resolution: 6 h), precipitation (3 h), evaporation (3 h), specific humidity (6 h), wind fields (6 h), total column water (6 h) and vertically integrated moisture fluxes (6 h). Note that for specific humidity and wind fields, 17 model layers are selected to represent the moisture distribution from the surface to the top of the atmosphere (see details in Table S2). To better describe the water transport in the atmosphere, the moisture considered in this study represents all possible phases of water in the atmosphere, including water vapor, cloud liquid water and cloud frozen water. Note that JRA-55 does not provide liquid and frozen water fluxes; thus, we only consider water vapor flux for this dataset. The time span of moisture tracking is 1979–2015 for ERA-Interim and JRA-55, and 1980–2015 for MERRA-2. All variables are temporally resampled to 0.25 h and spatially interpolated to 1∘×1∘ grids using bilinear interpolation for consistency.

We further retrieve the 1∘×1∘ gridded monthly averaged data from the ERA5 (the fifth generation of atmospheric reanalysis product produced by the ECMWF) (Hersbach et al., 2020) to examine the large-scale changes in evaporation, precipitation, horizontal wind fields and vertical velocity in the upper atmosphere during 1979–2015. The event-based precipitation δ18O data from 19 observation stations of the Tibetan Network for Isotopes in Precipitation (TNIP) are also used in this work (Yao et al., 2013).

The absolute contribution (mm per year, season or month) and relative contribution (the percentage of total sink precipitation) of global oceanic evaporation to the TP precipitation over the past 3 decades are tracked forward using an Eulerian moisture tracking model (WAM-2layers) and three reanalysis products (ERA-Interim, MERRA-2 and JRA-55). The spatial patterns of absolute contributions (Fig. 1a–e) agree well with the previous understanding that the ISM brings a large amount of Indian Ocean moisture to the southeastern TP and concentrates around the Brahmaputra Canyon region in summer (Xu et al., 2014; Wu et al., 2015). In the monsoon season (summer, Fig. 1c), the absolute contribution of oceanic moisture exhibits a sudden drop from more than 1000 mm along the southern TP to only about 100 mm in the central-northern TP, after traveling through the orographic barriers of the Gangdise and the southern slope of the Tanggula Mountains. This massive oceanic moisture also stretches westward along the southwestern slope of the Himalayas. In comparison, in spring and winter (Fig. 1b and e), a relatively weak westerly moisture sink appears around the Pamirs (∼100 mm season-1), as induced by the prevailing orographic precipitation (Curio and Scherer, 2016).

From the perspective of relative contribution, oceanic evaporation is responsible for 36 %–39 % of the total moisture condensation (precipitation) over the TP, and spatially, the relative contribution gradually decreases from more than 50 % along the southeastern edge of the TP to less than 20 % in the central-northern TP (Fig. 1f). Seasonally, the relative contributions of oceanic evaporation to the TP precipitation are 33 %–41 %, 36 %–39 %, 35 %–38 % and 51 %–54 % in spring, summer, autumn and winter, respectively. In spring and autumn (Fig. 1g and i), the relative contribution decreases gradually from the southern and southwestern TP to the central-northern TP. In summer (Fig. 1h), the relative contributions exhibit roughly a latitudinal distribution, with the 50 % isoline located between 25 and 30∘ N and the 30 % isoline between 30 and 35∘ N. This pattern is distinct from that of the absolute contribution featured by oceanic moisture concentrating around the southeastern TP. It is also notable that the westernmost TP (the Pamir region) shows the lowest relative contribution from oceans in summer in comparison with that in other seasons. This indicates a more active role that terrestrial evapotranspiration plays in the western TP during the monsoon season.

The largest contrast between the absolute and relative contributions occurs in winter when the absolute contribution reaches the lowest level while the relative contribution peaks (Fig. 1e and j). Based on the backward tracking of seasonal precipitation sources for the entire TP (Fig. S5), in winter, the moisture contribution from the westerlies-dominated oceans (the Mediterranean, the Red Sea, the Persian Gulf and even the Atlantic) is much higher than that from the cold and dry Eurasian continent (Fig. S5d–f). Although the Mediterranean, the Red Sea and the Persian Gulf are much smaller than the Atlantic, their moisture contribution to the TP precipitation is non-negligible. In fact, the total contributions of these three oceanic source regions can be greater than that of the Atlantic during both the monsoon and non-monsoon seasons (Table S3). In winter, the westerlies push the 50 % isoline of the relative oceanic contribution eastward to the mid-eastern TP, with the relative contribution well above 30 % for the entire TP. The spatial patterns on annual and seasonal scales observed here are also consistent among simulations driven by different reanalysis datasets (Figs. S3 and S4). Figure 2a shows a thorough comparison of monthly oceanic moisture contributions to precipitation over the entire TP among different simulations. The absolute and relative oceanic contributions are in general out of phase: high absolute (relative) contributions in summer (winter) and low absolute (relative) contributions in winter (summer), no matter which reanalysis dataset is used.

The backward tracking of precipitation for the entire TP shows that the moisture sources could extend far west to the Atlantic as driven by the mid-latitude westerlies and far south to the southern Indian Ocean as dominated by the monsoon system (Fig. S6a–c). Given the importance of the interactions between the westerlies and the ISM in determining the TP's climate and hydrologic cycle (Xu et al., 2008; Yao et al., 2013, 2018; Curio and Scherer, 2016), we divide the major oceanic source regions into the western oceans part (WO, including the Mediterranean, the Red Sea, the Persian Gulf and the eastern Atlantic) and the Indian Ocean part (IO), as shown in Fig. S6d.

Annually, the absolute contribution of moisture from the WO decreases from above 100 mm along the western and southern TP to around 20 mm in the central and northeastern TP (Fig. 3a). Seasonally (Fig. 3b–e), in addition to most parts of the western TP during the non-monsoon seasons, the southwestern edge of the TP, particularly in spring and summer, is also substantially influenced by the moisture transported from the WO. In comparison, the relative contribution gradually weakens from the northwest to the southeast of the TP on both annual and seasonal scales (Fig. 3f–j). This is consistent with the prevailing orographic precipitation dominated by westerly moisture transport and the zonal movement of the westerlies (Curio and Scherer, 2016). However, the relative contribution of the WO drops to below 10 % over the entire TP in summer, because the outbreaking of the ISM and the enhanced evapotranspiration from the wetting Eurasian continent dominate the available moisture over the TP (Fig. S5a–c). For the monthly variations in the WO moisture contribution to the TP precipitation (Fig. 2b), the absolute and relative contributions exhibit a phase shift of about three months, with the high absolute (relative) contribution in spring (winter) and the low absolute (relative) contribution in autumn (summer).

Considering both absolute and relative contributions to the regional precipitation, the moisture contribution of the IO is significantly higher than that of the WO in most parts of the TP, except for the northwesternmost TP. Along with the onset and retreat of the ISM, the absolute contribution from the IO exceeds 500 mm during summer in the southeasternmost TP and falls to below 100 mm during winter nearly over the entire TP (Fig. 4b–e). The relative contribution from the IO exhibits roughly a latitudinal gradient on an annual scale (Fig. 4f), with the 30 % and 10 % isolines located in the southern and northern TP, respectively. This zonal pattern lasts from spring to autumn (Fig. 4g–i), indicating a dynamic balance between the IO moisture contribution and the total moisture convergence during the period. This synchronism is largely broken in winter when the highest level of the relative contribution from the IO shifts to the southwesternmost TP (the south of the Gangdise Mountains) (Fig. 4j), with most IO moisture from the Arabian Sea (Fig. S5d–f). In fact, in the westerlies-dominated winter, the IO is still the major oceanic moisture source of precipitation in the southern TP, while the WO is the dominating oceanic moisture source for the western and northern TP. For the monthly variations of the IO moisture contribution (Fig. 2c), the absolute contribution reaches its maximum in summer owing to the significantly enhanced ISM, whereas the relative contribution peaks in both summer and winter. Additional simulations based on MERRA-2 and JRA-55 datasets both suggest similar spatial patterns of moisture contribution from different oceans (Figs. S7–10).

Figure 5 shows the annual trends of oceanic moisture contributions to the TP precipitation during 1979/1980–2015 estimated based on three simulations (driven by ERA-Interim, MERRA-2 and JRA-55). Despite some spatial discrepancies, all three simulations reveal a rough dipole pattern spanning from the southeastern TP to the central-northern TP. The most notable area is the Brahmaputra Canyon region, the most important moisture transport channel for the TP (Hren et al., 2009), which has gone through the most significant decrease in oceanic moisture contributions. Specifically, the moisture contributions from both the monsoon-dominated IO and the westerlies-dominated WO show decreasing trends in the Brahmaputra Canyon region (Fig. S11). Despite the enhanced evaporation from most oceanic source regions during 1979–2015 (Fig. 6a), substantial loss of oceanic moisture may occur before the moisture reaches the target region, as indicated by the increased precipitation sink along the moisture transport pathway (e.g., over the Indian subcontinent and the Bay of Bengal; Fig. 6b). In addition, we observe significantly weakened eastward and northward winds in the lower atmosphere (700 hPa) around the Brahmaputra Canyon (Fig. 7), suggesting that less moisture may transport through the region in the lower atmosphere. Meanwhile, we also observe significantly weakened upward motion in the region, which further verifies the weakened moisture convergence around the Brahmaputra Canyon region.

Regional precipitation is fueled by moisture that is either transported directly from oceans or recycled from lands (Gimeno et al., 2020a). In most cases, oceanic moisture contributes to less than 50 % of the total precipitation in the TP (except for the southeasternmost TP in summer and the western TP in winter; Fig. 1). Compared with the relative contribution of oceanic moisture, the absolute contribution of oceanic moisture (especially that from the IO) exhibits a spatial pattern highly consistent with precipitation distributions over the TP on both annual and seasonal scales (Fig. S12). Previous studies have demonstrated that the precipitation seasonality over the TP is determined jointly by the zonal movement of the westerlies and the onset and retreat of the ISM, suggesting the potential connections between oceanic moisture contributions and precipitation dynamics (Xu et al., 2014; Curio et al., 2015; Yao et al., 2018). Indeed, the long-term (1979–2015) trends of precipitation and oceanic moisture contributions have similar dipole patterns (e.g., decreased precipitation in the Brahmaputra Canyon region but increased precipitation in the central-northern TP) (Figs. 5 and 6b). To examine whether the decreased precipitation around the Brahmaputra Canyon region is mainly due to the changes in oceanic moisture contributions, we carried out additional backward moisture tracking simulations for the southeastern TP region (SETP) (Fig. S13a). The SETP defined here roughly covers areas dominated by the decrease in both precipitation and absolute oceanic moisture contribution (the purple rectangle in Fig. 5). As shown in Fig. S13b, the moisture contributions of both the westerlies-dominated western sources and the ISM-dominated southern sources to the SETP decreased over time, and most source regions with substantial decreases are over land. Meanwhile, only a few areas in the southwestern slope of the Himalayas and the southwestern corner of the TP show enhanced moisture contribution to precipitation in the SETP. Actually, Jiang and Ting (2017) have suggested that the dipole precipitation pattern over the TP was likely attributable to the interactions between the ISM and the TP. Nevertheless, more thorough analyses of the land–atmosphere interactions with physics-based models at higher spatial resolution are still needed to better understand these interactions and their response to climate change.

Based on numerous precipitation isotopic observations and ice-core records since the 1990s, previous studies have identified three distinct climate regions in the TP, as governed by the westerlies (northern TP; hereafter westerlies domain), the ISM (southern TP; hereafter monsoon domain) and their interactions (hereafter transition domain) (Tian et al., 2007; Joswiak et al., 2013; Yao et al., 2013). Theoretically, moisture delivered by the ISM tends to have relatively low isotope values due to strong convection activities along its transport paths, whereas moisture delivered by the westerlies in general has relatively high isotope values (Bowen et al., 2019; Cai and Tian, 2020). Here we further investigate the relationships between the simulated oceanic moisture contributions and precipitation δ18O observations at 19 monitoring stations over the TP (Figs. 8 and S14–17).

Compared with other oceanic moisture contributions, the relative contribution of oceanic moisture from the IO is strongly correlated with precipitation δ18O (Fig. 8), and nearly all stations show negative correlations (Table S4). In the monsoon domain, δ18O is high in spring and low in summer (note that the δ18O axes are revised), and correspondingly, the relative contribution from the IO is low in spring and high in summer. In comparison, in the westerlies domain, high δ18O is associated with low relative contributions from the IO during the monsoon season, while low δ18O is associated with high relative contributions from the IO during the non-monsoon seasons. In the transition domain, the seasonal cycles of δ18O and the relative contribution from the IO are similar to those in both the monsoon domain and the westerlies domain. Note the mismatch between the summer peaks of the relative moisture contribution from the IO and the low δ18O values in autumn at Lulang, Nuxia and Bomi stations near the Brahmaputra Canyon, which is likely due to the impact of the moisture transport from Southeast Asia or the Pacific Ocean driven by the trough embedded in the south branch of the westerlies (Cai and Tian, 2020). Additional backward moisture tracking of the monthly moisture sources for the SETP (which covers Lulang, Nuxia and Bomi stations) also suggests that the moisture sources gradually extend to Southeast Asia and the western Pacific Ocean during September and October (Fig. S18).

Except for several sites in the southernmost TP, our results confirm the theory that a higher percentage of oceanic moisture contribution from the ISM-dominated IO is associated with a lower precipitation δ18O value over the TP. As for the relative contribution from the westerlies-dominated WO (Fig. S15), the absolute contribution from the WO (Fig. S16) and the absolute contribution from the IO (Fig. S17), their relationships with the water isotope ratios are much weaker. Based on the seasonality of water isotope ratios, previous studies have identified a northern boundary around 34–35∘ N for the impact of the ISM on the TP, where the Tanggula Mountains serve as a main orographic barrier (Tian et al., 2007; Joswiak et al., 2013; Yao et al., 2013). Quantitatively, this geographical barrier reflected in water isotope ratios in general aligns with the 10 %–20 % (∼20 %) isoline of the relative contribution from the IO (entire ocean sources) in summer (Fig. 4h).