1 Introduction

The Tibetan Plateau (TP) is a unique area where the topography plays a key role in the regional atmospheric circulation and the hydrological cycle, which encompasses the largest number of glaciers outside of polar regions (Yao et al. 2012). Short instrumental records from inhabited areas depict a recent large warming trend (Liu and Chen 2000; Wang et al. 2008; Qin et al. 2009). Natural archives allow us to place recent changes in the broader context of natural climate variability. Among them, ice cores provide high-resolution proxy records of climate variability, archived through e.g. changes in water stable isotopes (usually expressed in a delta notation, such as δ18O used hereafter). However, their climatic interpretation requires us to decipher the climatic controls on high-altitude precipitation δ18O. While northern TP δ18O is mostly controlled by temperature-driven distillation, as in mid and high latitudes, the situation is more complex in the monsoon sector of the southern TP (Yao et al. 2013).

A number of studies have attributed recent increasing trends in south TP ice core δ18O to regional warming (Thompson et al. 2000; Yao et al. 2006a; Zhao et al. 2012), while process-based studies have stressed the importance of changes in precipitation amount and Indian monsoon intensity (Vuille et al. 2005; Kaspari et al. 2007; Joswiak et al. 2010). Some authors have also stressed that spatial patterns are related to changes in air mass trajectories and moisture sources (Tian et al. 2003; Johnson and Ingram 2004; Breitenbach et al. 2010).

The southern TP climate is characterized by strong seasonal shifts in atmospheric circulation, as winter air masses are advected by the westerlies while summer atmospheric circulation is dominated by the Indian monsoon (Fig. 1a). Large-scale circulation indeed plays a key role in the spatial and seasonal distribution of southern TP precipitation δ18O at intra-annual scales (Tian et al. 2001; Gao et al. 2011; Yao et al. 2013). Several studies have therefore suggested that TP ice cores record past variations of the regional atmospheric circulation (Qin et al. 2000; Kang et al. 2002, 2007b) which transports moisture from different sources (Tian et al. 2006), as also proposed for other locations (Meeker and Mayewski 2002; Kreutz et al. 2000; Hoffmann et al. 2003b; Xiao et al. 2004). So far, the exact processes relating atmospheric circulation and southern TP δ18O have not been precisely identified and quantified.

Fig. 1
figure 1

a General patterns of moisture transport towards the TP (purple, summer Indian monsoon; black, winter westerlies). The blue circle indicates the location of the ice core drilling site Noijingangsang. The position of the nearest meteorological stations (Lhasa and Nagarze) is shown by the red filled circles. b Mean seasonal cycle (1961–2006) of surface air temperature (T, °C, circles), precipitation amount (P, mm per month, vertical bars) at Lhasa (LS) and Nagarze (LKZ) meteorological stations, and monthly weighted precipitation δ18O (‰, 1994–2006, triangles) at Lhasa. Error bars show the inter-annual standard deviation of air temperature and precipitation amount

Full size image

Here, we present a new high-resolution isotopic record from an ice core recently drilled on the Noijinkansang (NK) glacier, southwest of Lhasa, which covers 2.7 km2 in the southern TP (Fig. 1a) and extends from 5540 to 6674 m a.s.l.. During May 2007, our ice core was drilled to bedrock (55.1 m depth) at 5950 m a.s.l. in a flat basin of the accumulation zone of the NK glacier (29°2′8.68″N, 90°11′54.10″E, Fig. 1a). The nearest meteorological stations, Nagarze (29.96°N, 90.33°E, 4430 m a.s.l.) and Lhasa (29.70°N, 91.13°E, 3650 m a.s.l.), show that more than 80 % of annual precipitation occurs during the summer Indian monsoon season (June–September, hereafter JJAS, Fig. 1b), with a large inter-annual variability (Gao et al. 2009; Tian et al. 2001) (Fig. 1b). Monthly precipitation δ18O from these stations displays a clear seasonal cycle, typical of southern TP, with maximum values in spring and minimum values in summer (Fig. 1b, Gao et al. 2009, 2013; Yao et al. 2013). This seasonal pattern is identified in the NK ice core and provides the basis for its age scale (Sect. 2).

This paper focuses on the climatic interpretation of the NK ice core δ18O record (hereafter NK δ18O). In Sect. 3, we explore the statistical relationships between inter-annual δ18O variations and local or regional temperature and precipitation data. Large-scale atmospheric teleconnections with NK δ18O are also investigated. Atmospheric reanalyses, climate indices and an atmospheric general circulation model (LMDZiso) are used to identify and quantify the role of changes in large-scale atmospheric circulation on NK δ18O. Section 4 reports our conclusions and perspectives.

2 Data

The NK ice core was cut into 2206 samples with intervals of 2 or 3-cm for analyses of stable isotopes, major soluble ions and carbonaceous aerosols, and 55 samples (~1 m resolution) for total beta radioactivity (hereafter β-activity) measurements. All the samples were transported to laboratory in frozen condition and melted at room temperature prior to analyses. Oxygen isotopes were measured using a MAT-253 mass spectrometer (with an analytical precision of 0.05 ‰) at the Key Laboratory of CAS for Tibetan Environment and Land Surface Processes, China. The β-activity was measured at the Analytical Laboratory of the Beijing Research Institute of Uranium Geology, China. The ice core is drilled on a vast and flat platform in the accumulation zone of the glacier, close to the ice divide. Field observations report no sign of surface wind mixing. Zhao et al. (2012) describes in details the ice core drilling and analytical procedures.

Below 16 m depth, the ice core density gradually increases to 0.9 g/cm3, with no indication of melt or stratigraphic perturbation. The chronology is based on the identification of seasonal cycles in δ18O and major soluble ions and verified using β-activity. The seasonal cycle of precipitation δ18O is used to identify annual layers from one maximum of ice core δ18O to the next one, therefore from spring (April–May) of 1 year to spring of the next year. The mean accumulation rate of the NK ice core of 470 mm w.e. is larger than the climatological precipitation amount at Lhasa and Nagarze (441.3 and 359.3 mm, respectively). Peaks in β-activity are detected at ~13.73 m (1987 Chernobyl event) and ~23.59 m (1963 nuclear bomb testing). The 55.1 m ice core is estimated to span 142 years (Fig. 2c). From repeated annual layer counting using sub-annually resolved δ18O and major soluble ions from the 2206 samples, we estimate the age scale uncertainty to be ±1 year from 2006 to 1963 and less than ±2 years prior to 1963 (Zhao et al. 2012).

Fig. 2
figure 2

(a) Inter-annual variation of simulated JJAS vapor δ18O anomaly from LMDZiso at the NK ice core site since 1958. (b) NK δ18O from LMDZiso simulated in the corresponding region from 1958 to 2006. Inter-annual variations of (c) Noijinkansang ice core δ18O from 1864 to 2006 compared with (d) JJAS temperature anomaly at Nagarze and Lhasa from 1961 to 2006, (e) annual temperature anomaly at Nagarze and Lhasa from 1961 to 2006, (f) simulated annual temperature anomaly from LMDZiso at the NK ice core site since 1958, (g) JJAS precipitation anomaly at Nagarze and Lhasa from 1961 to 2006, (h) simulated annual precipitation amount anomaly from LMDZiso at the NK ice core site since 1958. The blue lines show data at Nagarze station, the red lines show data at Lhasa station, and the gray lines show data from LMDZiso. The vertical yellow shades indicate the extreme years based on lowest/highest NK δ18O. On all panels, R (larger than 0.1 are shown only) and slopes show the correlations between NK δ18O and each climatic parameter

Full size image

From low-resolution measurements of δ18O and δD (220 merged samples), Zhao et al. (2012) observed an increasing trend of deuterium excess. They attributed this signal to a decrease of Indian monsoon precipitation on the southern TP through the reduced fraction of annual mean accumulation provided by summer precipitation (associated with a low deuterium excess, Zhao et al. 2012). Here, we expand the investigation of the NK water stable isotope record through our new high-resolution δ18O data.

This new record will be compared in the next section with local and regional meteorological data. For this purpose, surface air temperature data are available from Lhasa and Nagarze meteorological stations (National Meteorological Information Center of the China Meteorological Administration). We also use 2.5° × 2.5° regional monthly precipitation amount data spanning 1979–2006 from the GPCP (Global Precipitation Climatology Project) (http://www.esrl.noaa.gov/psd/data/gridded/data.gpcp.html). Finally, we also use monthly regional surface air temperature and vertical winds data for the period of interest from NCEP reanalysis data (http://www.esrl.noaa.gov/psd/data/reanalysis/reanalysis.shtml).

To explore the impacts of El Nino/Southern Oscillation (ENSO) on NK δ18O, we use the Nino 3.4 sea surface temperature (SST) index derived from the HadISST1 data set (Rayner et al. 2003) for the period 1870–2006. The relationships between our NK δ18O and Pacific Decadal Oscillation for the period of 1900–2006 (PDO, http://jisao.washington.edu/pdo/PDO.latest, Zhang et al. 1997; Mantua et al. 1997) as well as Zonal index (defined by the normalized difference in zonal-averaged sea level pressure anomalies between 35 N and 65 N, representing the overall intensity of the surface westerly anomalies over middle latitudes) for the period 1864–2006 (http://ljp.lasg.ac.cn/dct/page/65592, Li and Wang 2003) are also analyzed.

LMDZiso is an atmospheric general circulation model which explicitly includes water stable isotopes (Risi et al. 2010). It captures the seasonal and intra-seasonal variability of TP precipitation δ18O (Gao et al. 2011; Yao et al. 2012; Gao et al. 2013), but with a cold and dry bias in the southern TP (Gao et al. 2011). Here, we use the inter-annual precipitation and vapor δ18O outputs from a simulation performed at 2.5° × 3.75° resolution, run from 1958 to 2006 in a “nudged” mode. The Atmospheric Model Inter-comparison Project (AMIP) protocol (Gates 1992) is used to prescribe monthly sea surface boundary conditions. The European Centre for Medium-Range Weather Forecasts (ECMWF) reanalysis results (Uppala et al. 2005) are used to nudge the horizontal wind fields with a relaxation time of 1 h. This warrants that the large-scale LMDZiso simulated wind fields are consistent with the observations, and allows to use of the LMDZiso framework to understand and quantify the role of changes in atmospheric circulation in the observed δ18O variability. A zoomed simulation using the zoom functionality of the LMDZ model, stretching the grid to the horizontal resolution of 50–60 km between 0° and 55°N and 60° and 130°E, are also used here to detect the inter-annual variability.

3 Results and discussion

3.1 Ice core δ18O variation

The NK δ18O record shows large seasonal variations, from −7.0 ‰ at 24.19 m depth to −24.7 ‰ at 16.27 m depth (not shown), with an annual mean value of −16.9 ± 1.7 ‰ from 1864 to 2006 (Fig. 2c). This record contains remarkable multi-decadal variations. Utilizing the regime shift detection algorithm of Rodionov (2004), six different regimes are identified with the regime shift years of 1911, 1921, 1957, 1973 and 1980. High and stable δ18O values are observed in the nineteenth century, above present-day levels. Two subsequent periods are marked by high δ18O, from the 1910 to 1920 and from 1958 to 1972, with lower values in-between. Since the 1980, NK δ18O has increased at a mean pace of 0.05 ‰ per year. A sharp decrease occurs from the 1973 to 1978, with the strongest annual δ18O minimum recorded in 1978. We have investigated whether similar anomalous low δ18O values are recorded in other Tibetan ice cores. A small minimum is recorded in 1978 in the Malan, Puruogangri and Dasuopu ice cores, and in 1976–1977 in the Guliya, Geladaidong and Tanggula ice cores (not shown). Because the dating uncertainties of these ice cores are usually around ±1 year (Hou et al. 2003; Yao et al. 2006a, b; Joswiak et al. 2010; Kang et al. 2007a; Yao et al. 1997; Wang et al. 2006), we conclude that there is a widespread negative δ18O anomaly in late 1970s in the Tibetan ice cores, most strongly expressed in NK and Tanggula ice cores.

3.2 Relationships with regional meteorological data

These results motivate further exploration of the climatic controls on NK δ18O, with a focus on the period from 1960 to 2006, where the age scale is well established. We first investigate statistical relationship of NK δ18O with regional meteorological data. Since more than 80 % of precipitation occurs in JJAS, we focus on the relationships between JJAS temperature and precipitation amount and NK δ18O. Since the 1960s, NK δ18O shows decadal variations which parallel those recorded in Nagarze and Lhasa JJAS temperature. Without detrending, NK δ18O inter-annual variations are significantly correlated with Nagarze JJAS temperature (R = 0.39, p < 0.05, Fig. 2), but not with precipitation. With detrending, NK δ18O inter-annual variations are still significantly correlated with Nagarze JJAS temperature (R = 0.47, p < 0.01), but neither with Nagarze precipitation nor with Lhasa temperature and precipitation. Using NCEP gridded JJAS temperature data, we detect significant positive correlations between NK δ18O and temperature in the central-northern TP and central India, with correlation coefficients <0.6 (Fig. 3a). Using GPCP gridded precipitation data, we identify a weak but significant anti-correlation between NK δ18O and JJAS precipitation in South East TP (95°–100°E and 25°–30°N, R > −0.4, p < 0.01), and a positive correlation with JJAS precipitation in the Bay of Bengal (65°–95°E and 10°–20°N, R < 0.6, p < 0.01, Fig. 3b).

Fig. 3
figure 3

Spatial distribution of linear correlation coefficients between NK ice core δ18O and regional air temperature and precipitation from observations and LMDZiso simulations, as well as winds and SST. a NCEP gridded JJAS surface air temperature data (1948–2006), b GPCP gridded JJAS precipitation data (1979–2006), c JJAS precipitation amount simulated from LMDZiso (1979–2006). d Correlation coefficients between NK ice core δ18O and LMDZiso JJAS vapor δ18O at 425 hPa hight (1979–2006). e Correlation coefficients between NK ice core δ18O and NCEP gridded JJAS zonal wind (U wind) at 500 hPa height (1979–2006). f Correlation coefficients between NK ice core δ18O and NCEP gridded JJAS meridional wind (V wind) at 500 hPa height (1979–2006). g Correlation coefficients between NK ice core δ18O and NCEP gridded DJF zonal wind at 500 hPa height (1979–2006). h Correlation coefficients between NK ice core δ18O and NCEP gridded DJF meridional wind at 500 hPa height (1979–2006). i Correlation coefficients between NK ice core δ18O and NCEP gridded JJAS SST (1982–2006). j Correlation coefficients between NK ice core δ18O and NCEP gridded DJF SST (1982–2006). On all maps, only correlations significant at the 95 % confidence level are shown

Full size image

3.3 Comparison with LMDZiso simulations and large scale climate fields

We now compare these data with LMDZiso outputs. The simulated temperature is in better agreement with the observed temperature at Nagarze station than that at Lhasa (Fig. 2e, f), especially for the cool periods (1968, 1977–1979, 1992–1997). However, the model overestimates the variability of precipitation amount, compared with observations at Lhasa and Nagarze stations (Fig. 2g, h), especially before 1980s. We observe a better model-data agreement between NK δ18O and the simulated water vapor δ18O than with the simulated precipitation δ18O for the period of 1958–2006 (Fig. 2a–c); this could be related to the intermittency of precipitation, or to post-deposition processes (Steen-Larsen et al. 2014).

Some inter-annual changes in the simulated annual temperature, precipitation amount and water vapor δ18O coincide with changes recorded in NK δ18O (Fig. 2a–c, f, h). Within LMDZiso, the percentage of the simulated precipitation δ18O which can be explained by a linear correlation with local simulated temperature and precipitation is weak and can only explain less than 10 % of precipitation δ18O variance (Fig. 2b, f, h), even when considering a multiple regression that includes the temperature and precipitation amount together. The lowest NK δ18O is observed and simulated in 1978, and it is associated with the most negative anomalies of simulated annual temperature (Fig. 2e), precipitation amount (Fig. 2h) and vapor δ18O (Fig. 2a).

NK δ18O is weakly positively correlated with JJAS simulated precipitation amount in the tropical Indian Ocean (R < 0.6, p < 0.1) and, as in GPCP, anti-correlated with LMDZiso precipitation in the southeast TP (R > −0.3, p < 0.1) (Fig. 3c). NK δ18O is most strongly correlated with vapor δ18O at 425 hPa for the period of 1979–2006 (pressure level closest to NK ice core drilling site elevation) in the tropical Indian Ocean (10°S–10°N, 85°E–130°E) and the west Pacific Ocean (Fig. 3d). In addition, NK δ18O is significantly positive correlated with JJAS zonal wind (U wind) at 500 hPa in 32°N–40°N and 50°E–80°E from 1979 to 2006 (Fig. 3e). Meanwhile, NK δ18O is significantly negatively correlated with JJAS U wind at 500 hPa over the tropical Indo-Pacific Ocean during the same period (Fig. 3e). After 1979, the significant positive correlation between NK δ18O and meridional wind (V wind) exists in northwest TP (Fig. 3f). NK δ18O is also significantly positively correlated with the observed SST since 1982 in the tropical Indo-Pacific Ocean (between 5°N and 20°S, Fig. 3i). In winter, the significantly positive correlation only exists in 0°–10°N and 95°E–120°E with U wind, and the Persian Gulf as well as a small part of Arabian Sea with V wind (Fig. 3f, h). NK δ18O show significantly positive correlation with winter SST in the western and northern Arabian Sea and the Philippines (Fig. 3j).

These statistical analyses suggest large-scale atmospheric circulation drivers of NK δ18O inter-annual variability. The mechanisms relating large-scale atmospheric circulation and tropical precipitation isotopic composition are complex, and have been explored in atmospheric models. Following earlier works of Hoffmann (2003) and Vuille et al. (2003) for the Andes, Brown et al. (2006) identified strong imprints of monsoons, meridional migration of the ITCZ and ENSO in tropical precipitation isotopic composition, through processes affecting rainout intensity and convection height. Other ice cores in the central TP, Arctic, Antarctic, and subpolar regions were also interpreted to depict rapid changes in atmospheric circulation after the mid-1970s (Kang et al. 2002; Kaspari et al. 2007; Schneider et al. 2012; Kelsey et al. 2012).

The statistical analyses show that local temperature or precipitation account for less than 50 % of NK δ18O variability in both observations and LMDZiso, and point to processes associated with moisture transport and large-scale atmospheric circulation, at the inter-annual scale. At the intra-summer scale, different controls were identified for Lhasa precipitation δ18O variability (Gao et al. 2013). Combining daily precipitation isotopic composition measurements and satellite information on convection and water vapor isotopic composition, He et al. (2015) showed a dominant control through convective activity above North India: enhanced convection (and increased precipitation above N. India) depletes surface water δ18O, and this depletion is further amplified along the orographic uplift of moister air, and finally by enhanced subsidence of depleted vapor towards Lhasa. Our study suggests that controls at the intra-seasonal time scale cannot be extrapolated at the inter-annual scale. Indeed, the statistical analyses show that, at the inter-annual scale, enhanced JJAS precipitation above the southern Bay of Bengal and enriched vapor δ18O in the tropical ocean and west Pacific Ocean at 425 hPa level are consistent with an δ18O enrichment at NK high elevation site, responding to a common large-scale mode of variability.

We now discuss the mechanisms which may affect the recent NK δ18O trend, since 1980. While warming is expected to induce enrichment in the NK δ18O, one process is expected to compensate for temperature effects and weaken the NK δ18O increase after 1980. Enhanced westerlies (Yuan et al. 2014), accounting for a larger proportion of annual precipitation, also provide depleted moisture though long distance transport.

3.4 Causes of the late 1970s NK δ18O shift: statistical analysis

In order to further explore the processes controlling inter-annual variations of NK δ18O, we focus on the sharp decrease observed in the late 1970s (Fig. 2c). Meteorological data show cold conditions in 1978 at both stations (Fig. 2e) as well as in the central TP (Kang et al. 2007b), with the coldest summer encountered in 1976 at Nagarze and Lhasa (Fig. 2d).

During the mid-1970s, shifts in North Pacific climate regime were associated with a deepening of the Aleutian Low and a strengthening of the mid-latitude westerlies (Miller et al. 1994; Nakamura et al. 1997; Mantua et al. 1997; Zhang et al. 1997; Wu et al. 2005), which affected Indian monsoon strength (Trenberth and Hurrell 1994; Wang and Ding 2006). This regime shift also involved changes in the PDO and ENSO (Guilderson and Schrag 1998; Mantua and Hare 2002; Alexander et al. 2002). The ENSO influence on TP moisture and ice core δ18O are also found at Dasuopu (Brown et al. 2006) and other tropical ice cores (Vuille et al. 2003; Hoffmann 2003a; Bradley et al. 2003). This motivates an investigation of the relationships between NK δ18O and Pacific modes of variability.

The spectral analysis of NK δ18O, performed with a wavelet and Multi-Taper method (not shown) reveals a significant periodicity at 2.7 years, also suggesting an ENSO influence on NK δ18O. During several intervals, NK δ18O is significantly correlated with Nino 3.4 SST index for different seasons. During the cold PDO phase of 1943–1976 (Fig. 4e), NK δ18O is positively correlated with December-February (DJF) Nino 3.4 SST anomalies (R = 0.32, p = 0.06 and up to 0.44, p = 0.01 when excluding 1973, Fig. 4a, b). Since 1980, during the warm phase of PDO, NK δ18O is anti-correlated with May–September (MJJAS) Nino 3.4 SST anomalies (R = −0.32, p = 0.11 and up to R = −0.45, p = 0.02 excluding 1984) (Fig. 4a, c, e). The NK δ18O minimum at the end of the 1970s coincides with a PDO phase shift (from negative phase to positive phase, Grigholm et al. 2009), an inflexion point of the zonal index (onset of a fast increase, Fig. 4a, d) as well as ENSO (increase in frequency and intensity). This again indicates that the most remarkable NK δ18O negative anomaly is associated with changes of the large-scale atmospheric circulations, although a partial role of temperature cannot be excluded.

Fig. 4
figure 4

Comparison of (a) the NK δ18O anomalies, (b) December–February Nino 3.4 SST anomalies from 1870 to 2006, (c) May–September Nino 3.4 SST anomalies from 1870 to 2006 and (d) Zonal index from 1864 to 2006 as well as (e) December–February PDO from 1900 to 2006. Thin lines show annual data and thick lines show 10-year smoothing. Horizontal gray lines depict zero levels

Full size image

The mechanism by which the PDO/ENSO could affect the southern TP ice core δ18O was hypothesized through the Indian monsoon and westerlies changes. An “atmospheric bridge” connects the ENSO/PDO-driven large-scale atmospheric circulation changes with the SST, wind, and humidity from the North Pacific to Indian Oceans (Alexander et al. 2002), which then alter the strength and intensity of the Indian monsoon. In the decade prior to the climate shift, SST changed in the western subtropical Pacific, leading to a shift of atmospheric circulation and changes of SST in other parts of Pacific, then driving ENSO variability by modulating the trade winds (Barnett et al. 1999; Vimont et al. 2001, 2003; Alexander et al. 2002). After 1976, the cooling of SST of central Pacific and subtropics associated with central Pacific El Niño forced changes in the extra-tropical atmospheric circulation (Lorenzo et al. 2010), and surface wind anomalies switched from easterly to westerly. In return, tropical SST changes associated with the recurrence of El Niño from the mid 1970s to the late 1980s reinforced north Pacific changes and westerly wind anomalies (Wu et al. 2005). The warm phase of PDO during the boreal winter season affects subtropical and tropical SST persisting into the next monsoon season (Krishnamurthy and Krishnamurthy 2013). This pattern explains the anti-correlation between NK δ18O and May–September (MJJAS) Nino 3.4 SST anomalies since 1980. They drive changes in equatorial trade winds that reinforce the equatorial Walker circulation and lead to enhanced ascending motion in the central Pacific and the equatorial Indian Ocean, weakening the Indian monsoon (Krishnamurthy and Krishnamurthy 2013).

With the increase of SST and zonal wind in the tropical Indo-Pacific Ocean, the evaporation of surface ocean water strengthens, providing local vapor which is not depleted in δ18O. The increased local evaporation replenishes the monsoon flow and may explain the reduced depletion of NK δ18O. In the southern TP, these processes can be compensated by the fact that a weaker Indian monsoon also leads to an increase in moisture transported from western remote sources (e.g. Persian Gulf), which is characterized by low δ18O. El Niño events further reduce the strength of Indian monsoon through the descending Hadley branch, while the situation is more ambiguous during La Niña events (Mantua and Hare 2002; Krishnamurthy and Krishnamurthy 2013; Sano et al. 2013; Miyasaka et al. 2014). The transport of vapor over the tropical Indo-Pacific Ocean changes with winds corresponding to the PDO/ENSO phase. In El Nino winter, the westerlies decreases due to the warmer SST in tropical Indian Ocean, resulting in more water vapor transport to north; after 1980, during the warm phase of PDO, cyclones prevail in the west of Philippines in La Nina winter, helping stronger moisture transport from Indian Ocean to continent; before 1980, the condition is opposite (Yuan et al. 2014). The influences of El Nino/La Nina on anticyclone/cyclone close to the Philippines however are inconsistent and asymmetric (Wu et al. 2010). Meanwhile, the deepening of the Aleutian Low and strengthening of the winter mid-latitude westerlies are associated with colder conditions (Miller et al. 1994; Alexander et al. 2002; Mantua and Hare 2002), further depleting precipitation δ18O.

3.5 Causes of the late 1970s NK δ18O shift: analysis using LMDZiso

The role of large-scale circulation is now explored using LMDZiso JJAS precipitation δ18O. The model simulates too enriched precipitation δ18O, a feature common to other atmospheric models (Yao et al. 2013), but also underestimates the magnitude of inter-annual δ18O variations, and does not produce decadal variability. We note that the model caveats could be linked to the atmospheric model resolution, as it cannot resolve small-scale topographic features which may play a key role on the inter-annual variability of high elevation precipitation δ18O (Gao et al. 2011). The zoomed simulation obviously improves the intra-annual precipitation δ18O, but not its inter-annual variability. We also stress the fact that the ice core record can be associated with a noticeable noise level, due to post deposition processes such as wind scouring, and does include a signal from the non-monsoon season. However, LMDZiso does produce a sharp decrease in NK precipitation δ18O in the late 1970s, and minimum values in 1978. We therefore use the framework provided by the LMDZiso simulation to understand the processes controlling the JJAS δ18O depletion in 1978–1979, compared to 1982–1983 levels (Fig. 5). The associated precipitation δ18O anomaly has a magnitude of ~1.5 ‰ (Fig. 5a), extends over all the TP and is also depicted in the vapor δ18O (~1.7 ‰, Fig. 5c).

Fig. 5
figure 5

Attribution of the precipitation δ18O anomaly from 1978–1979 to 1982–1983 using the LMDZiso model. a Decomposition of the precipitation δ18O anomaly into a sum of five contributions due to rain-vapor disequilibrium (deseq), fractionation coefficients (α), temperature (T), local relative humidity (h), and residual effects (residual) including convection and large-scale transport (see Appendix in Supplementary Material). b Map showing the 1982–1983 pattern of JJAS vapor δ18O at the lowest atmospheric model level (colors) and pattern of JJAS winds at the lowest atmospheric level (vectors). c Same as (b) but for the difference between 1982 and 1983 and 1978–1979. Vertical profiles of difference between 1983 and 1978 for zonal (d) and meridional (e) winds along 90°E longitude from 30°S to 60°N, from NCEP reanalyses, downloaded from http://www.esrl.noaa.gov/psd/data/gridded/data.ncep.reanalysis.derived.pressure.html. The grey shading depicts the TP topography. The blue square shows the location of NK ice core

Full size image

We use a decomposition of the associated δ18O anomaly into several terms, which account for the effects of rain-vapor disequilibrium, fractionation coefficients, local temperature, temperature at the source, local relative humidity, and residual effects including convection and large-scale transport (See Appendix in Supplementary Material). This approach shows a key role of changes in the intensity of Rayleigh distillation due to changes in temperature. In 1978–1979, LMDZ simulates TP regional cooling, consistent with meteorological data, which explains a depletion of about 0.7 ‰. Residual effects including large-scale transport however account for an additional depletion of about 1 ‰.

The pattern of changes in atmospheric circulation (Fig. 5c) depicts a reduced northward monsoon flow, and a large North-Easterly anomaly of −0.5 m/s (Fig. 5b, c). Combined with the average JJAS vapor δ18O South West-North East gradient of about −2.4 × 10−3 ‰ per km (Fig. 5b) and a residence time of water vapor of about 10 days, this anomalous circulation pattern can account for the residual anomaly of low-level water vapor δ18O anomaly of ~−1 ‰. The precipitation δ18O composition then follows tightly that of the low-level water vapor, variations in precipitation-vapor differences accounting for less than 0.2 ‰ of precipitation variations.

The large-scale circulation effects are clearly visible in the latitudinal and vertical structure of JJAS zonal and meridional winds from NCEP reanalyses along a transect at 90°E for 1978 and 1983 (Fig. 5d, e). Regarding westerly wind (U, Fig. 5d), the data depict a decrease of low level winds (<500 hPa) at 0°–20°N, and an increase of high level winds (500 hPa) at 20°N–30°N in 1983 compared to 1978. The meridional profile displays a decrease of the low atmosphere southerly winds in 1983 compared to 1978 (Fig. 5e). This illustrates the decrease of Indian monsoon and increase of westerlies in 1983 compared to 1978, consistent with the statistical analysis and the tele-connections evidenced in Sect. 3.4.

In summary, the abrupt depletion of NK δ18O in 1978 is captured by the LMDZiso simulation. It is mostly attributed to a depletion of water vapor δ18O due to the change in temperature and in large-scale atmospheric circulation. The effect of large-scale atmospheric circulation on vapor δ18O is found to be larger than that of temperature.

4 Conclusions and perspectives

In this manuscript, we have reported the δ18O signal recorded by a new Southern Tibetan ice core. In contrast with earlier studies, our record does not show unprecedented warming during the last decades. It is dominated by decadal variations, including a remarkable isotopic depletion during the late 1970s. Statistical analyses stress a significant anti-correlation with precipitation amount (p < 0.05) in the surrounding areas, as well as zonal wind at 500 hPa in the tropical Indo-Pacific oceans, and a positive correlation with precipitation amount in the Bay of Bengal, as well as with surface temperature in TP, India and summer tropical SSTs. Significant relationships are detected with large-scale modes of variability (ENSO and PDO). The importance of large-scale atmospheric circulation is further supported by the ability of the atmospheric general circulation model LMDZiso to resolve the 1978 isotopic anomaly. The model shows that this anomaly extends over all of TP, and is positively related to an anomaly in water vapor δ18O in the tropical Indo-Pacific region. Within the model framework, we isolated the driver of this anomaly, which is mostly due to an abnormal atmospheric circulation, marked by enhanced westerlies, and a reduced monsoon flow, while air temperature also affects this anomaly through changes in Rayleigh distillation, but with a smaller effect than that of circulation alone.

This work demonstrates the ability of TP annual ice core δ18O to record abrupt shifts in atmospheric circulation and provides a methodology to evaluate the magnitude of local versus large-scale controls on ice core δ18O. This will motivate further integration of high-resolution precipitation δ18O records from multiple archives (ice cores, but also tree ring cellulose, speleothems and lake sediment records) in order to map the spatial and temporal variability of δ18O, and to characterize past changes in regional atmospheric circulation.