Stable isotope compositions of precipitation over Central Asia

Local meteoric water lines (MWL)

The local meteoric water lines (MWLs) of precipitation, δ²H, and δ¹⁸O, provide important information on the water cycle, water vapor sources, and water transport (Jonesl, Leng & Arrowsmith, 2007; Lutz, Thomas & Panorska, 2011). A Central Asia MWL (CAMWL) was established as δ²H = 7.30δ¹⁸O + 3.12 (R² = 0.95, n = 727, p < 0.01, R is the Pearson correlation coefficient) (Fig. 2) based on 727 precipitation groups. The slope of the CAMWL was slightly lower than that of the global MWL (GMWL), which was eight (Craig, 1961) and the Chinese MWL of 7.48 (Liu et al., 2014). Differences in the local MWLs slope often occur because of deviations in humidity at the source of moisture or evaporation. Central Asia is located in the Eurasian hinterland, and has a significant variation in the annual cycle and alternating dry or wet seasons. A low LMWL slope is associated with non-equilibrium conditions that affect falling raindrops during dry conditions (Liu et al., 2014), leading to the potential for significant sub-cloud evaporation.

The precipitation δ¹⁸O over Central Asia ranged from +2‰ to −25.4‰ with a mean of −8.7‰, and δ²H ranged from −4.2 ‰ to −191.4‰ with a mean of −67.1 ‰.

Seasonal variations in precipitation δ¹⁸O

The arid Central Asian region is affected by monsoons and westerlies, resulting in annual differences in climatic variables and precipitation δ¹⁸O. The variation in the annual air temperature represents a continental climate. The maximum and minimum temperatures occur in July and January, respectively. In this study, the temperature variation was unimodal, and the precipitation amount represented three distribution types (Fig. 3). The three precipitation patterns observed in different regions are shown in Fig. 3. (1) Maximum precipitation in occurred in the summer, with summer precipitation accounting for 41.4% of the total precipitation. These stations (including Hetian, Zhangye, Barabinsk, and Urumqi stations) are mainly distributed in northern Central Asia and are primarily influenced by the intensity and location of the westerly circulation. (2) Maximum precipitation values in winter and spring, with winter and spring precipitation accounting for 39.6% and 37.8% of the total precipitation, respectively, while the summer precipitation accounted for only 6.4% of the total precipitation. These stations (including the Kabul and Teheran stations) are mainly located in southern Central Asia and are primarily influenced by the Indian monsoon. (3) A well-distributed seasonal precipitation pattern was observed at Astrakhan and Saratov stations. These are present in northwestern Central Asia and are primarily influenced by airflow from the Arctic. This result also highlighted the complexity of the spatial–temporal variations in precipitation in Central Asia.

The annual temperature and precipitation influenced the annual precipitation δ¹⁸O cycle. The maximum δ¹⁸O occurred from June to August (JJA). Maximum precipitation δ¹⁸O at the Urumqi, Hetian, Barabinsk, and Teheran stations were −5.4‰, 0‰, 2‰, and −0.3‰ in August, respectively. The maximum values at the Zhangye and Kabul stations occurred in July. The maximum values occurred in June and September, respectively, at the Saratov and Astrakhan stations. All minimum values occurred from December to February (DJF). Therefore, the δ¹⁸O content in precipitation was higher in summer than in winter, and the seasonal pattern of the δ¹⁸O in precipitation reflected the climate regime during the annual cycle. In winter, air masses were relatively cold and dry and the amount of precipitation was very low, while in summer, the opposite was true.

Spatial characteristics of precipitation δ¹⁸O

Figure 4 shows the spatial distribution of precipitation δ¹⁸O throughout the year, and for summer and winter seasons at each station in Central Asia. The maximum value of the annual mean δ¹⁸O (−1.47‰) was at Cele station, located in southern Xinjiang, and the minimum value (−18.6‰) was observed at Hami station. For the summer months, the maximum δ¹⁸O value (2.8 ‰) was observed at Cele station, and the minimum value (−13.4‰) was observed at Altay station, located in northeastern Central Asia. The maximum δ¹⁸O value (−7.7 ‰) during the winter months was observed at the Teheran station, located in southern Central Asia, and the minimum value (−35.3‰) was observed at the Dabancheng station.

There were remarkable spatial differences in precipitation δ¹⁸O over Central Asia. The spatial distribution of precipitation δ¹⁸O was mainly affected by the thermodynamics of water vapor condensation during the Rayleigh fractionation process and included meteorological elements, water vapor transport, and geographic factors (Dansgaard, 1964; Yurtsever & Gat, 1981; Rozanski, Araguas-Araguas & Gonfiantini, 1993).

Several studies have verified that altitude and latitude are the main geographic factors that affect changes in temperature and water vapor condensation, and can be referred to as geographic factor effects of precipitation δ¹⁸O (Liu, Tian & Yao, 2009; Yao et al., 2013). The correlation of precipitation δ¹⁸O with latitude, altitude, and longitude was investigated using partial correlation analysis, and the R were 0.53 (p < 0.05), 0.32 (p < 0.05), and 0.17 (p > 0.05), respectively, indicating that latitude is the most significant factor. In summer, latitude was significantly correlated with precipitation δ¹⁸O (R = 0.54, p < 0.05), while both latitude and longitude were significantly correlated with precipitation δ¹⁸O in winter (R = 0.31, p < 0.05; R = 0.67, p < 0.01, respectively). For the annual δ¹⁸O, the δ¹⁸O/LAT gradient was −0.42‰/°, implying that the precipitation δ¹⁸O was reduced by approximately 0.42‰ for every one-degree variation in latitude, which was larger than the gradient in China (−0.22‰/°). Similarly, the δ¹⁸O/ALT gradient was −0.001‰/m, which was lower than the global value of −0.0022‰/m and the value of −0.0016‰/m in China (Liu, Tian & Yao, 2009; Bowen & Wilkinson, 2002). In Central Asia, altitudes range from −18 to 4,200 m, with a complex topography and various climatic characteristics.

Several researchers have constructed models to explain the relationship between precipitation δ¹⁸O and geographic factors (Liu, Tian & Yao, 2009; Yao et al., 2013), including the Bowen and Wilkinson (BW) model (Bowen & Wilkinson, 2002). To confirm the effects of altitude and latitude on precipitation δ¹⁸O over Central Asia, we evaluated the relationship between precipitation δ¹⁸O and geographical factors using stepwise regression analysis.

In Central Asia, geographical controls on precipitation δ¹⁸O were best expressed by a stepwise linear regression, including altitude (ALT, m), latitude (LAT, °N) and longitude (LON, °E):

(1) $${\delta }^{18}\mathrm{O}=(-0.179)\times LAT+(-0.022)\times LON+0.00085\times ALT(R^2=0.51,p<0.05)$$

The best model accounting for JJA δ¹⁸O included both altitude (ALT, m) and latitude (LAT, °N):

(2) $${\delta }^{18}\mathrm{O}=(-0.632)\times LAT+(-0.002)\times ALT+23.851(R^2=0.64,p<0.05)$$

The DJF δ¹⁸O model was expressed by linear regression as:

(3) $$\begin{array}{l}{\delta }^{18}\text{O}=(-0.482)\times LAT+(-0.280)\times LON+(-0.001)\times ALT\\ +23.891(R^2=0.73,p<0.01)\end{array}$$

Our model comprehensively explained the effect of altitude, latitude, and longitude on precipitation δ¹⁸O over Central Asia, and revealed that the variation in precipitation δ¹⁸O depended on geographical factors at each station. These results showed that our models were suitable for use in Central Asia.

We used the aforementioned models to obtain the spatial distribution of precipitation δ¹⁸O over Central Asia (Fig. 5). The annual precipitation δ¹⁸O gradually decreased as the latitude increased (Fig. 5A). In the summer, higher δ¹⁸O values were observed in southwest Central Asia and lower δ¹⁸O values were observed in northern Central Asia and the Pan-Third Polar region (including the Tibetan Plateau, Iranian Plateau, and Tianshan Mountains) (Fig. 5B). This pattern shows the effects of latitude and altitude on precipitation δ¹⁸O. However, precipitation δ¹⁸O in Central Asia was higher than that in the Tibetan Plateau (Fig. 5B). This was mainly due to the transport of moist air originating from the Arabian Sea and the Bay of Bengal into Central Asia, which resulted in more summer precipitation (Zhao et al., 2014). A strong anti-cyclonic pattern was noted in the Arabian Sea, Indian subcontinent, and the Bay of Bengal, and was associated with a strong southerly flow controlling the Indian subcontinent and extending north up to the valley between the Iranian Plateau and Tibetan Plateau regions. Furthermore, the valley between the Iranian Plateau and the Tibetan Plateau is below 1,500 m, and the anomalous southerlies can transport moisture into Central Asia (Zhao et al., 2014; Zhao & Zhang, 2015). The cyclonic pattern observed in Central Asia was associated with this pattern, and an anomalous southwesterly wind may have transported moisture into Central Asia. Tian et al. (2001) indicated that there was a different moisture source between the northern and southern Tibetan Plateau, and that the northern limit of the summer monsoon was north of the Yarlung Zangbo River located in the middle of the Tibetan Plateau.

In the winter, precipitation δ¹⁸O gradually decreased with increasing longitude (Fig. 5C). The δ¹⁸O values in Central Asia was lower than that in the eastern Asia and Mongolian Plateau. The moisture transported from the Eurasian continent and the Mediterranean Sea into northern Central Asia generated more winter precipitation and westerly wind was needed to transport moist air into inland regions.

Correlation between precipitation δ¹⁸O and meteorological variables

We investigated correlations between monthly precipitation δ¹⁸O, monthly air temperature (Fig. 6), and precipitation at eight stations in Central Asia (Fig. 7). As shown in the figure, monthly precipitation δ¹⁸O at each station had a significant positive correlation with temperature (R = 0.51–0.96, p < 0.01). Moreover, the latitude effect with δ¹⁸O gradually increased as the temperature increased. The effect of temperature was the strongest at the Barabinsk station, followed by those at the Urumqi, Hotan, Zhangye, Saratov, and Kabul stations, while it was the weakest at the Astrakhan and Teheran stations. The gradients between δ¹⁸O and air temperature ranged from 0.28‰/°C at the Saratov and Teheran stations to 0.68‰/°C at the Hetian station. These results were similar to global mid- and high-latitudes with gradients of 0.55‰/°C (Rozanski, Araguas-Araguas & Gonfiantini, 1993) and elsewhere in China with gradients of 0.36‰/°C (Gu, 2011).

Significant positive correlations between precipitation δ¹⁸O and precipitation were observed at four stations (Zhangye, Barabinsk, Hetian, and Urumqi), and only one station had a significant negative correlation (Teheran station). Correlations for other stations were not significant. These results reveal an indeterminant dependence of precipitation δ¹⁸O on precipitation, which is in accordance with the insignificant effect of precipitation in inland regions (Gu, 2012). The δ¹⁸O–precipitation gradients ranged from −0.08‰/mm (Teheran station) to 0.20 ‰/mm (Zhangye station). In addition, the Zhangye and Teheran stations had a significant precipitation effect as a result of their location on the edge of the East Asian monsoon and Indian monsoon regions, respectively.

The Urumqi station has the longest and most systematic GNIP observations in Central Asia, with an observation period spanning from 1986 to 2003. Based on a subset of selected predictors, we established a model using stepwise regression analysis and the corresponding monthly temperature and precipitation data. The δ¹⁸O model in Urumqi station was expressed by linear regression as:

(4) $${\delta }^{18}\mathrm{O}=(-0.009)\times P+0.417\times T-15.53(R^2=0.69,p<0.01)$$

We reconstructed and estimated the monthly precipitation δ¹⁸O time series from 1985 to 2013 at Urumqi station (Fig. 8A). The reconstructed precipitation δ¹⁸O was high-positively correlated with the observed values at Urumqi stations (R = 0.83, p < 0.01), and the results clearly depicted the seasonal cycle of precipitation δ¹⁸O, except for some values during extreme cold or hot months. Furthermore, the reconstructed precipitation δ¹⁸O values for most spring and autumn seasons are similar to the observations. This indicates that basic fractionation mechanisms can be reflected using the reconstructed model. The reconstructed model indicates that the temperature effect of precipitation δ¹⁸O values was reflected in Urumqi Station. Several previous studies confirmed the temperature effect of precipitation isotopes existed in arid Central Asia (Yao et al., 1999, 2013; Wang et al., 2016). Thus, the selected control factors and established regressions can be used to reconstruct the long-term variation in precipitation δ¹⁸O, and also act as a proxy of an historic environment. In High Asia, the climatic controls on precipitation isotopes was used as a base for ice core sampling (Tian & Yao, 2016). In Central Asia, the δ¹⁸O of ice cores were in good agreement with temperatures at nearby stations, which indicates its value in paleoclimate reconstruction using ice cores (Tian et al., 2006; Zhang & Wang, 2018).

Relationship of precipitation δ¹⁸O with general atmospheric circulation

To reveal the relationship between precipitation δ¹⁸O and large-scale atmospheric circulation, we analyzed the correlation between precipitation δ¹⁸O at each station using long-term observation data and atmospheric circulation (e.g., IMI, SASMI, and WCI).

Our results showed that the precipitation δ¹⁸O at the Zhangye station was positively correlated with the EASMI (R = 0.62, p < 0.05), indicating that the East Asian summer monsoon has an important effect on precipitation δ¹⁸O in Zhangye. It is located on the eastern side of the arid northwest China and at the northeastern edge of the Tibetan Plateau as well as at the transition zone between the East Asian monsoon and westerlies. The East Asian monsoon can affect the precipitation δ¹⁸O in the westerly and monsoon transition regions.

The precipitation δ¹⁸O was negatively correlated with the IMI at both the Urumqi and Teheran stations (R = −0.57 and −0.41, respectively, p < 0.05), revealing an important effect of the Indian monsoon on precipitation δ¹⁸O. In general, the stronger the monsoon, the lower the δ¹⁸O value, and vice versa. The Teheran station is close to the Indian Ocean and is one of the main moisture paths for the Indian monsoon moving towards the north. Therefore, precipitation δ¹⁸O at the Teheran station was mainly influenced by the Indian monsoon. However, Urumqi is located in the Asian hinterland and is indirectly influenced by the Indian monsoon. At Urumqi, precipitation δ¹⁸O is affected by an anomalous moisture transport path through a multi-step process. A weakened Indian monsoon can cause an anomalous cyclone in the middle and upper troposphere in Central Asia, resulting in cooling, which is directly correlated with the increased precipitation in Xinjiang (Zhao et al., 2014). Furthermore, the Indian monsoon affects moisture transport from the Bay of Bengal and the Arabian Sea to Xinjiang through a two-step process (Zhao et al., 2014). The extreme summer precipitation in northern Xinjiang was controlled by moisture sources originating from the Indian Ocean, which was closely related to a stronger meridional circulation (Huang et al., 2017). The anomalous circulation can also transport some moisture from the Indian Ocean along the eastern periphery of the Tibetan Plateau to North Xinjiang (Huang et al., 2017).

The moisture in Central Asia is mainly transported by the westerly circulation from the North Atlantic Ocean and the Eurasian continent. The westerly circulation is one of the most important factors affecting precipitation δ¹⁸O over Central Asia. However, we found a weak positive correlation between precipitation δ¹⁸O and the WCI index. This may be related to the moisture transported by the westerly circulation, which originates from the Atlantic and Arctic oceans and dissipates along the way. The water vapor transported by the westerly is exhausted by the time it reaches Central Asia because of substantial evapotranspiration and evaporation below the clouds in arid regions.

The precipitation d-excess parameter can provide supplementary information to precipitation δ¹⁸O and δ²H values, and it is largely controlled by the moisture source and water transport paths. Thus, it may be the most important indicator to represent the regional moisture source and atmospheric circulation (Merlivat & Jouzel, 1979). In addition, the precipitation d-excess is closely related to the re-evaporation of raindrops during condensation and precipitation, and is affected by temperature and relative humidity during evaporation (Merlivat & Jouzel, 1979; Jouzel, Merlivat & Lorius, 1982; Jouzel, Froehlich & Schotterer, 1997; Wang, 2014). In Central Asia, the precipitation d-excess values ranged from +21.66 ‰ to −8.5‰, with an average of 8.5 ‰. The temporal variations in precipitation d-excess shows more positive values in spring and more negative values in summer and the opposite pattern for δ¹⁸O. This variation is in agreement with variations in the conditions under which the moisture source evaporates. There is a similar seasonal pattern in precipitation d-excess and δ¹⁸O of Central Asia and that of the westerly climate regime zone (Yao, 2009). However, there is a dissimilarity with the annual cycle in Lhasa, which is affected by southwest monsoon air mass (Wang, 2014). This demonstrates that the moisture transported to Central Asia was derived predominantly from the westerlies and the polar air masses. Moreover, the high precipitation d-excess values showed that recycled moisture derived from local sources makes a significant contribution to precipitation, especially in Central Asia. Yao et al. (2020) also suggested that the warming and increased moisture content of the atmosphere contributed to the local moisture cycle and increased precipitation recycling in eastern Central Asia.