Rainwater isotopes in central Vietnam controlled by two oceanic moisture sources and rainout effects

The interpretation of palaeoclimate archives based on oxygen isotopes depends critically on a detailed understanding of processes controlling the isotopic composition of precipitation. In the summer monsoonal realm, like Southeast Asia, seasonally and interannually depleted oxygen isotope ratios in precipitation have been linked to the summer monsoon strength. However, in some regions, such as central Vietnam, the majority of precipitation falls outside the summer monsoon period. We investigate processes controlling stable isotopes in precipitation from central Vietnam by combining moisture uptake calculations with monthly stable isotope data observed over five years. We find that the isotopic seasonal cycle in this region is driven by a shift in moisture source from the Indian Ocean to the South China Sea. This shift is reflected in oxygen isotope ratios with low values (− 8 to − 10‰) during summer and high values during spring/winter (0 to − 3‰), while 70% of the annual rainfall occurs during autumn. Interannual changes in precipitation isotopes in central Vietnam are governed by the timing of the seasonal onset and withdrawal of the Intertropical Convergence Zone, which controls the amount of vapour contributed from each source.

it remains unclear what drives the seasonal cycle in δ 18 O p throughout the entire year, especially for sites with a rainy season independent of the summer monsoon. The focus on these two seasons is problematic for mainland Southeast Asia, where the timing of peak rainfall varies strongly across the peninsula and is not restricted to strong summer monsoon rainfalls 36 (Fig. 1). For example, annual rainfall in central Vietnam is strongly affected by tropical cyclones that peak during autumn (September to November) and rainfall associated with northeasterly winds 37,38 (Fig. 1). This makes central Vietnam ideal to investigate the impact of non-summer monsoonal rainfall on the seasonal cycle in rainwater isotopes.
Previous modelling studies have identified the topography of mainland Southeast Asia as a crucial factor in modulating the timing of its rainfall season 39,40 . Autumn rainfall in central Vietnam is a direct consequence of local topography interacting with the annual migration of the Intertropical Convergence Zone (ITCZ), creating zones of intense rainfall and rain shadows depending on the season 40 . An inverse relationship between topography and the isotopic composition of rainfall has long been observed and is described as the 'altitude effect' 41 . However, the altitude effect is less pronounced in the tropics 42 and its relative importance is disputed 43 . Mainland Southeast Asia has two main mountain ranges located between Myanmar and Thailand to the west (the Tenasserim Hills), and between Laos and Vietnam to the east (the Truong Son Mountains or Annamite Mountains). Therefore, central Vietnam is ideal to investigate the importance of these effects on stable isotopes in precipitation there.
In this study, we combine modelling of monthly moisture uptake locations for precipitation arriving in central Vietnam and monthly stable isotope ratios collected over five years. We aim to identify controls on the seasonal cycle in precipitation isotopes from central Vietnam to understand processes at work and identify the primary controls on seasonal variability independently from the summer monsoon season. Further, we investigate the impact of rainout effects using a spatial array of precipitation isotope data from five sites in mainland Southeast Asia. Then we explore interannual variability of precipitation isotopes in order to estimate the potential of stable isotopes from central Vietnam for palaeoclimate reconstructions.  44 in East and Southeast Asia. Data for Ho Chi Minh City from Le Duy et al. 30 and Munksgaard et al. 31 and for Nakorn Phanom from Noipow 32 . Central Vietnam's boreal autumn rainfall regime is unique in the region and is represented by the Dong Hoi station. The map was generated using the Matplotlib Basemap Toolkit version 1.1.0 (https ://matpl otlib .org/basem ap/index .html).

Moisture source location and large-scale circulation.
There are three main air mass sources for mainland Southeast Asia: the Indian Ocean, the Pacific Ocean, and continental Asia 30,33,34 . Calculating the air mass trajectories using the Hybrid Single-Particle Lagrangian Integrated Trajectories (HYSPLIT) model helps to understand the general air movement; however, it is of limited use for identifying the location of moisture uptake 49 . Primary evaporation can occur anywhere along the trajectory path, leaving the trajectory analysis with a given uncertainty. We therefore calculate the monthly moisture uptake based on the specific humidity to estimate the uptake location for moisture arriving in central Vietnam using PySplit 50 . The analysis shows that from June to August, most moisture is taken up from the southern Bay of Bengal and during the rest of the year from the northern South China Sea (Fig. 3). During May and September rainfall originates from both sources, suggesting the switch from one source to the other occurs during these months.
The change in moisture source location is linked to the seasonal migration of the ITCZ and the associated large-scale atmospheric circulation changes. The ITCZ in the West Pacific is located around 5 • S during December to February and north of 15 • N in June to August 51 (Fig. 4). During April, the ITCZ begins to move northwards from near the Equator crossing southern Vietnam in early May 52 . In mid-May, the ITCZ moves rapidly northwards within a few days 53 , explaining the southwesterly winds, which increase the contribution of moisture to mainland Southeast Asia from the Bay of Bengal. This effect is clearly visible in central Vietnam as a www.nature.com/scientificreports/ shift in the moisture uptake location towards the Indian Ocean during May (Fig. 3). The Indian Ocean remains the main moisture source location throughout summer (June to August) and mid-September (Fig. 4b). Beginning in September, the ITCZ slowly retreats southwards resulting in northeasterlies bringing intense rainfall into central Vietnam at this time 54,55 . Our moisture uptake analysis for central Vietnam shows this large-scale change in circulation pattern with a high contribution of moisture from the northern South China Sea from mid-September onwards (Fig. 3). As the ITCZ continues to move south during October to December, easterlies weaken and moisture is taken up more locally from mainland Southeast Asia and part of the South China Sea. The deuterium excess is a result of the different fractionation rates of oxygen and hydrogen isotopes during kinetic fractionation 60 . During kinetic fractionation, hydrogen fractionation is faster than oxygen fractionation and the rate of fractionation depends on relative humidity at the evaporation site 61 . Kinetic fractionation occurs mainly www.nature.com/scientificreports/ during evaporation, rather than condensation 61 . Therefore, the deuterium excess acts as a fingerprint of climatic conditions during primary evaporation, independent of rainout processes along the travel path 60 . Drier source regions with low relative humidity are associated with an increased deuterium excess and vice versa 62 . In our data, the deuterium excess in summer rainfall has lower values compared to winter and spring rainfall. Deuterium excess values of autumn rainfall clusters in between (Fig. 5). This confirms our previous results based on moisture source location and shows that the deuterium excess reflects two oceanic sources: the Indian Ocean during summer with most of the values around 5-10‰ and Pacific Ocean during spring, autumn and winter when most values range around 11-17‰. Further, the increased deuterium excess during winter reflects the lower relative humidity in the South China Sea during this time. The relative humidity in the Bay of Bengal is high during summer due to the strong convection related to the Indian Summer Monsoon, which is reflected by low deuterium excess values (Fig. 5).
We now consider in more detail the isotopic variability of rainfall during the transition periods between moisture sources in May and September (Fig. 3), when the deuterium excess shows a large variability (Fig. 5). The deuterium excess depends on the relative contribution from each source, which can vary from year to year. Rainfall from May 2017 is strongly depleted, which is caused by generally larger range of moisture source regions compared to 2014 ( Fig. 6 left column). Lower values in 2017 can be linked to an increased contribution of more distal moisture sources compared to 2014 and 2016. This is a first hint that interannual variability of oxygen isotopes in rainfall can be linked to source location.
Additionally, the link between source location and interannual variability is evident during September to November rainfall (Fig. 6). Considering the averaged values for this time, two clear clusters for 2014/15 and 2016/17 are evident (Fig. 6). This relation is reflected in the source location ( Fig. 6 middle columns). In 2016 and 2017, moisture arrives from a more distal location, such as near Sri Lanka and the Gulf of Thailand (Fig. 6). This is not the case in 2014/15 and 2018, when moisture arrives from the proximal eastern Bay of Bengal and off the coast from northern and central Vietnam (Fig. 6). This interannual variability in moisture source is related to the timing in the withdrawal of the ITCZ from Vietnam 64 . An early withdrawal (2014/15) causes the formation of an anti-cyclonic circulation in the South China Sea, partly blocking the south-westerly flow from bringing moisture from the Bay of Bengal to central Vietnam. This leads to more positive deuterium excess. A late withdrawal enables more moisture contribution from the southern Bay of Bengal (south of Sri Lanka) and the Gulf of Thailand (Fig. 6) 40,65 , leading to more negative deuterium excess values. Several studies highlighted that tropical cyclones affect interannual autumn rainfall variability in central Vietnam 30,37,66 , thus it is of importance to consider tropical cyclone activity as a potential factor in modulating δ 18    Several studies linked the seasonal cycle in rainwater isotopes in Asia to a shift from oceanic to continental sources including Southeast Asia 24,33 , Taiwan 68 , Thailand 34 and southern China 69 . Our findings suggest that in central Vietnam δ 18 O p is controlled by a shift between two oceanic sources, rather than a shift between a continental and oceanic source (Fig. 3). Establishing the Local Meteoric Water Line (LMWL) is a useful tool to investigate the contribution of continental moisture or secondary evaporation along the travel path. Plant transpiration returns the water vapour with nearly no change in deuterium excess, whereas open water or soil evaporation leads to an increased deuterium excess and sub-cloud evaporation to a decreased deuterium excess 70 . This means that an increased deuterium excess represents an increased continental moisture contribution and a decreased deuterium excess indicates secondary evaporation. The LMWL for central Vietnam is: δ 2 H = (8.6 ± 0.3‰) × δ 18 O + (15 ± 2‰) with an r 2 of 0.98 (n = 54) calculated following Sachs et al. 71 . We calculate the LMWL for each season and find that during autumn, spring and winter, the LMWL slope is increased by 0.5‰ and the intercept by 5.6‰ compared to the GMWL (Fig. 5). This deviation suggests a limited contribution of continental moisture (lake, river or soil moisture) from mainland Southeast Asia during this time of the year 70 . During summer the LMWL has a slope of 6.7, which is less than 8 of the GMWL and an intercept of − 2.5 ± 5.4 ‰ (Fig. 5). This decrease again suggests a limited influence of continental evaporation. The depletion in the deuterium excess could be a result of re-evaporation of raindrops after condensation (sub-cloud evaporation) 2,60,70 .

Spatial variability of δ 18
O p across mainland Southeast Asia. Thus far, we have focused on explaining the mechanisms controlling the temporal variability of rainwater isotopes in central Vietnam. However, we have not explained the spatial evolution, nor fully considered processes occurring during condensation, rather than evaporation. Investigating how rainwater isotopes vary along the moisture travel path, is useful to understand the importance of rainout processes. These effects are strongest when moisture travels via landmasses and for central Vietnam this is the case during summer (June to August). To explore the spatial variability of δ 18 O p in mainland Southeast Asia during this time, we investigate the effects of local topography and rainout. Rayleigh fractionation due to rainout occurs not only while moisture travels landwards, but also due to orographic uplift 41 . During summer, air masses reaching central Vietnam have a western source and cross two mountain ranges before reaching our study site 72 , thus making this time ideal to understand these processes. We perform two simulations: one with mainland Southeast Asia being represented as a flat land surface (Simulation 1) and one considering local topography to understand the effects of rainout and orographic uplift (Simulation 2). Simulation 1 (Fig. 7) models the rainout in water isotopes from the Bay of Bengal to Hainan (Haikou) and predicts a rainout effect of − 0.68‰ per 1000 m. Considering the GNIP δ 18 O p data across the transect, it is clear that the rainout effect alone cannot correctly explain this spatial variability. Simulation 2 shows that the spatial development of δ 18 O p along the transect is better simulated when topography is included in the model (Fig. 7). We simulate the effects of the Tenasserim Hills (700 m elevation in the simulation), positioned between Myanmar and Thailand and of the Truong Son Mountains (1500 m elevation in the simulation), located between This simulation is in good agreement with observed data (Fig. 7) and shows that the altitude effect alone can explain 70% of the spatial variability in rainwater isotopes during summer. During the summer months, the vertical temperature gradient is greater compared to winter, amplifying the altitude effect during this time 68 . The altitude effect has also been shown to be larger during the rainy season, related to stronger rainout 42 . While the dry season is during summer in central Vietnam, Fig. 4 shows that this is not the case for the rest of the Peninsula, which undergoes the rainy season during summer. Sturm et al. 42 simulated the altitude effects in tropical South America and found that, during the rainy season, more than 60% of the δ 18 O p variance is related to altitude according to observed data, which is similar to our results. Our simulations suggest that the rainout and altitude effects combined can explain the spatial variability of oxygen isotopes in mainland Southeast Asia during summer. However, the relation between δ 18 O p and altitude in mainland Southeast Asia needs more detailed investigation, ideally using isotope-enabled general circulation models. Our results are preliminary and hint towards the importance of orography and local advection for the spatial evolution of water stable isotopes.
implications for palaeoclimate research. The seasonal variability of rainwater isotopes in central Vietnam is controlled by a seasonal shift between two moisture source locations: the Bay of Bengal and the South China Sea. This shift in source location is regulated by the position of the ITCZ (Fig. 8) [74][75][76][77] . Further, variability in the ITCZ position is tightly linked to broader tropical ocean dynamics including ENSO and the IOD 78 . This link is mainly established via variability in the local Hadley circulation, which is modulated by changing sea surface temperatures (SSTs) in the western Pacific and thus ENSO 79   These similarities suggest a link between interannual variability in δ 18 O p , and the ITCZ position and ENSO events across mainland Southeast Asia. This means that palaeoclimate records from central Vietnam based on δ 18 O p can potentially be used to reconstruct regional climate dynamics related to ocean SST anomalies and the evolution of ITCZ through time.
In addition to δ 18 O p variability across mainland Southeast Asia, studies also investigated the effects of ENSO and the IOD on local rainfall amount. They found a strong spatial variability in the relation of SST anomalies and rainfall anomalies across the Peninsula 36 and throughout the seasons 84 . This emphasises that rainfall amount and δ 18 O p do not necessarily correspond, since the impact of ENSO on δ 18 O p appears to be consistent across the Peninsula, whereas ENSO induced rainfall variability is spatially heterogeneous 85 . Therefore, δ 18 O p from mainland Southeast Asia can potentially be used to trace large-scale circulation patterns, like shifts in the ITCZ and ENSO, while proxies for local rainfall can be used to investigate how these climate drivers control the local hydroclimate. However, our dataset is limited in time and restricts any long-term observations.

conclusion
The timing of the rainy season in central Vietnam differs strongly from the rest of mainland Southeast Asia, with peak rainfall from September to November. Therefore, central Vietnam offers the opportunity to investigate the seasonal evolution of stable isotopes in precipitation unconfounded from the summer monsoon strength. Our data show, that there is a strong seasonal cycle in the isotopic composition, with low values during summer and high values during spring/winter, while the seasonal cycle in precipitation amount peaks in autumn. We find that the seasonal isotopic variability in δ 18 O and δ 2 H is controlled by a shift between two oceanic moisture sources. During summer, the Bay of Bengal is the main source location and during the remaining part of the year, the South China Sea contributes most moisture. Thus, precipitating moisture arriving in central Vietnam crosses mainland Southeast Asia only during summer. By simulating the spatial variability of δ 18 O in summer precipitation we find that rainwater isotopes across mainland Southeast Asia are mainly related to the altitude and rainout effect. Thus, both effects amplify the seasonal difference in the isotopic composition between summer and spring/autumn/winter rainfall even further.
By investigating the transition period from one source to another (once in May and again in September to November), we find that during May the relative distance of moisture uptake within the Bay of Bengal differs from year to year and thus drives interannual variability in the isotopic composition of precipitation. A similar process can be observed in September to November, where lower values in δ 18 O and δ 2 H are related to a longer travel distance of moisture within the Bay of Bengal, but also to an additional source of moisture contributed from the southern South China Sea. During years with higher mean values for September to November, this second source within the southern South China Sea is absent and moisture from the Bay of Bengal arrives from a more proximal location, close to the coast of Myanmar and Thailand. These small changes within each moisture source are governed by the timing in ITCZ migration varying per year. This means that the relative travel distance and contribution of moisture from each source is regulated by the duration in which the ITCZ remains north of central Vietnam.
We conclude that palaeoclimate archives from central Vietnam, based on stable isotopes, have the potential to capture changes in the relative moisture contribution sourced from the Indian Ocean versus the South China Sea and thus track variability in the past position of the ITCZ. Our findings suggest that a long-term northwards displacement of the ITCZ would result in more negative values in the isotopic composition of precipitation in central Vietnam and a southwards displacement in more positive values.

Methods
Data collection. Meteorological and water isotopic data was obtained from the International Atomic Energy Agency 44 WISER database, which gives access to data from GNIP. The GNIP station at Dong Hoi represents central Vietnam in this study and includes five years of data from 2014 to 2018. The year of 2015 is incomplete, with March, April, May and August missing, and contains seven data points for stable isotopes. The deuterium excess was calculated following Craig 59 . Further, we use GNIP data from Yangon (Myanmar), Bangkok (Thailand), Nakorn Phanom (Thailand) and Haikou (Hainan) to compare observed data with the modeled spatial evolution of δ 18 O p .
Modelling moisture uptake with pySplit. The HYSPLIT model 86 was developed to track air parcels moving towards or from a set location. The model incorporates meteorological data and reconstructs the movement of air masses. Backward trajectories were calculated using frequency analysis with new trajectories every 6 hours for 30 days at a height of 1500 m above mean sea level, using the NCEP Global Data Assimilation System (GDAS) 87 model one-degree archive. The flow at this level is representative of the large-scale, near surface flow in the tropics and captures the moist surface in which convergence occurs 88,89 . Moisture uptake is calculated following Sodemann et al. 90 using PySplit 50 . Briefly, the change in humidity over 6 h periods of a HYSPLIT trajectory is calculated. When values are positive, greater than a threshold (0.2 g/kg), and the change in humidity occurs within the planetary boundary layer, it is assumed that the moisture required is taken up from surface evaporation.  94 , and a vapour transport model 95 . This approach allows us to predict the evolution of δ 18 O p by incorporating orographic precipitation and the regional water-energy budget 92 . Parameters used in this study are listed in Table 1. Initial vapour content and advective velocity are derived from the mean u and v wind fields of the ERA5 dataset 56 . Potential evapotranspiration from the CGIAR-CSI Global-Aridity and Global-PET Database 96,97 and dryness index was calculated using GNIP data. To understand the effects of topography on oxygen isotopes, we compare a flat surface simulation (simulation 1) with a simulation incorporating local topography (simulation 2). Two mountains are used in simulation 2, represented by simple Gaussian bell shapes with peak elevation of 700 m and a width of 130 km for the Tenasserim Hills and peak elevation of 1500 m and a width of 160 km for the Truong Son Mountains. Both simulations use the same climatic parameters (Table 1), which are representing typical conditions during the period of June to August. Moisture arriving in central Vietnam only crosses landmasses during this time of the year, thus allowing us to investigate continental rainout and altitude effects.

Data availability
All data used for this research are available in the methods and at: GNIP isotope data is accessible at http://www. iaea.org/water . GPCC Precipitation, Interpolated OLR and NCEP Global Data Assimilation System data were provided by the NOAA/OAR/ESRL PSD, Boulder, Colorado, USA, from their Web site at https ://psl.noaa.gov/. The ERA5 database can be accessed at https ://cds.clima te.coper nicus .eu/cdsap p#!/home. The annual tropical cyclone reports can be access at: https ://www.usno.navy.mil/JTWC/annua l-tropi cal-cyclo ne-repor ts.
Received: 13 June 2020; Accepted: 16 September 2020 Table 1. Model input used in the simulations 1 and 2 for the period of June to August. Initial vapour content and advective velocity are derived from the ERA5 dataset 56 . Potential evapotranspiration from the CGIAR-CSI Global-Aridity and Global-PET Database 96,97 and dryness index was calculated using GNIP data.