Introduction

Interactions between the atmosphere and terrestrial environments have increasingly become the focus of research, especially since recent, human-mediated climate change has been acknowledged (Pielke et al. 1998; Williamson et al. 2008; Pachauri et al. 2014). In semi-arid and arid mountain regions, interactions among climate, glacial dynamics, water capacity of mountain lakes, and their influence on river-water discharge, need to be understood. The high mountain ranges of Central Asia are of particular interest within this context because of the complexity of regional circulation mechanisms (Ramaswamy 1962; Aizen et al. 2001; Chen et al. 2010), the steep environmental gradients originating from pronounced relief, and their importance as freshwater sources for people who live in the region. Therefore, it is necessary to understand linkages between the atmosphere and terrestrial environments, and studies should not be based on modern observation alone, but rather should explore long-term processes, through paleoenvironmental inference.

Lake sediments can provide valuable paleoclimate information, but most paleolimnological studies in Central Asia have been carried out either on the Tibetan Plateau, in lower to mid-elevation mountain areas of western Central Asia, or in bordering basins (Van Campo and Gasse 1993; Shen et al. 2005; Zhao et al. 2009; Zhu et al. 2015). A literature search showed that studies from high mountain regions, like the Pamir, are rare (Huang et al. 2014; Lauterbach et al. 2014). Such studies focused on glacial activity (Zech et al. 2005; Abramowski et al. 2006; Liu et al. 2014), fluvial system development, and climatic-tectonic interactions (Fuchs et al. 2015). It remains unclear whether large and deep lakes in mountainous environments react rapidly to climate change or with some delay, or whether they are mainly internally driven and thus relatively unresponsive to climate change (Williamson et al. 2008).

Here we present a sediment record from Lake Karakul that dates back to ca. 29 cal ka BP. We used sediment geochemical data as proxies for conditions external to the lake and for exploring internal lake processes. Our first objective was to assess the reliability of these proxy variables. Second, we aimed to infer past climate changes, and by comparison with other records, we attempted to infer trends in past moisture availability. Third, we investigated whether internal lake processes were driven by or were independent from regional climate changes.

Study area

Lake Karakul (Tajik for ‘black lake’) is located at an altitude of 3915 m above sea level (asl) in the north-east of the autonomous Gorno-Badakhshan region of Tajikistan, in the eastern Pamir Mountains (Fig. 1), which are adjoined by the Hindukush and Karakorum ranges in the south and the Tian-Shan ranges in the north. The modern lake is brackish, with a pH of 9.2 (Mischke et al. 2010). The endorheic basin is supplied with freshwater that mainly originates from surrounding glaciers, snow fields and frozen ground, and enters the lake largely through two main streams in the north and south of the western sub-basin, and five smaller streams from the north, north-east and south in the eastern sub-basin (Mischke et al. 2010). Direct discharge from precipitation in the catchment is of minor importance (Ni et al. 2004) because precipitation is generally low, with an annual mean of only 82 mm, but characterised by high variability. The Westerlies, which are today the main moisture source (Böhner 2006), are blocked by the mountain ranges to the west. Air temperatures recorded by the meteorological station at Lake Karakul reveal a mean annual temperature of −3.9 °C, mean TJanuary −18.1 °C and mean TJuly 8.5 °C (measurement period 1934–2007, excluding 1995–2003). Low winter temperatures cause ice cover from November until May, with up to one metre of ice thickness. A maximum Secchi depth of 19 m was recorded during the open-water season (Ergashev 1979), whereas Secchi depths of 10.7–11.7 m were recorded in June 2008 (Mischke et al. 2010) and 3.4 m in August 2013. Potamogeton pamiricus and Characeae colonize the littoral zone (Hammer 1986). The terrestrial vegetation in the catchment is dominated by alpine steppe taxa such as Stipa, Oxytropis and Artemisia. Direct human impact on the lake and its catchment is low.

Fig. 1
figure 1

Lake Karakul in the Pamir Mountains, Tajikistan, in arid Central Asia (Landsat-8 (OLI) image, 22 July 2014) and retrieved core KK12-1 including the age-depth model (for details, see ESM)

Full size image

Materials and methods

Fieldwork

Core KK12-1 (composite length 10.87 m) was collected during a field campaign in spring 2012 using UWITEC coring equipment (“Niederreiter 60”) at a water depth of 12 m. Coring was done from the ice cover on Lake Karakul, 1.8 km from the eastern shoreline (core site coordinates: 39.0176°N, 73.5327°E; Fig. 1). Core KK12-1 was collected using a single ice hole. Small gaps occur between each 2-m core segment. Recovered sediment between 568 and 525 cm depth was disturbed during the previous core drive and is not in situ. Nine additional surface sediment samples were collected from different depositional settings (e.g. slackwater, pond, dry river channel, sand dune; Electronic Supplementary Material [ESM] Table S2) in the eastern to northern vicinity of the lake and used as reference material.

Laboratory analysis

Core sampling, lithology and age-depth model

The six sediment core segments were opened at the Alfred Wegener Institute in Potsdam. The “archive” halves were used for smear-slide analysis and non-destructive work such as initial core description and X-ray fluorescence (XRF) scanning. The “work” halves were subsampled for radiocarbon dating and sedimentological analyses (e.g. CNS (carbon, nitrogen, sulphur), total organic carbon (TOC), grain-size distribution and stable isotopes on carbonates).

The lithology from 1087 to 525 cm is characterised mainly by silt and sandy silt, with thin layers of sand (Fig. 1). Macrophyte remains are relatively rare or lacking in this section of the core. Sediments from 503 cm depth to the top of the core are characterised by silt with a few thin layers of sandy silt. Sediments above 465 cm depth contain very abundant macrophyte remains. The macrophyte remains are Potamogetonaceae leaves that are embedded horizontally in the sediments.

Fourteen samples, nine consisting of plant remains, four of bulk sediment and one of an in situ-collected living charophyte were dated at the Poznań Radiocarbon Laboratory, Poland (ESM Table S1). Radiocarbon ages were corrected for the ‘reservoir effect’ of Lake Karakul using the published value of 1.420 ka (Mischke et al. 2010) and the apparent age of the newly collected macro-algae. The lake reservoir effect calculation was based on the 14C concentration of the atmosphere in the year of the sampling (2012). Thus, the post-bomb 14C concentration of the atmosphere in 2012 was used for the calculation of the F14C (fraction modern) of 0.8914 (Reimer et al. 2004; Mischke et al. 2013). The obtained F14C value is equivalent to a lake reservoir effect of 1315 years. The average of the two determined lake reservoir effects, i.e. 1368 years, was applied to correct the radiocarbon age data from the core. No terrestrial macrofossils were found in the core, and we tentatively assume that the lake reservoir effect did not change markedly through time. The mean lake reservoir effect of 1368 years was used in the age-depth model (Fig. 1; ESM Fig. S1) created with the Bacon package (Blaauw and Christen 2011) in R (R Core Team 2012). Bacon parameter settings largely follow Blaauw and Christen (2011), but we set the mean accumulation rate to 20 year cm−1 prior to the model run and 20 for the memory strength to account for bioturbation effects, originating from the dense macrophyte cover. The obtained age-depth models were fairly robust despite variations in the parameter settings.

Geochemistry

Smear slides were prepared and inspected approximately every 40 cm. Total carbon (TC), total nitrogen (TN), total sulphur (TS) and TOC were analysed for 97 samples taken at 10-cm intervals. The freeze-dried and finely ground samples for element analyses were weighed into tin capsules for CNS determinations and measured using an Elementar vario EL III (CNS) analyser. An Elementar vario MAX C was used to quantify TOC. TN values in the lower part of the core were occasionally at, or close to the detection limit at 0.1%, however the data were still used to calculate TOC/TN ratios. Total inorganic carbon (TIC) was calculated by subtracting TOC from TC values.

Stable isotope analyses on authigenic carbonates were conducted on 98 samples, which were bleached for 24 h with 2.5% NaOCl to remove organic matter and sieved with a 36-µm mesh to clean the samples of detrital components and shells. The remaining authigenic carbonate fraction was freeze-dried, ground for homogenization and loaded into Labco vials. Measurements of δ13C Carb and δ18O Carb were undertaken with a ThermoFinnigan Gasbench II linked to a DELTAplusXL mass spectrometer and measured after digestion with 103% H3PO4 at 72 °C for 60 min at the German GeoForschungsZentrum (GFZ) in Potsdam. The stable isotope data are reported in per mil relative to VPDB and were calibrated using NBS19 with an analytic precision error of <0.08% for both δ13C and δ18O on reference material.

Grain-size analysis

Preparation of 97 samples for grain-size analysis included 24 h of treatment with CH3COOH (10%) to remove carbonate, sieving with a 2-mm sieve to remove abundant plant remains in most parts of the core, and shaking the samples on a platform shaker, initially in a 0.3% H2O2 solution, with an addition of 10 ml of H2O2 (35%) every second day, for up to six weeks to remove the remaining organic matter. Grain-size analysis was performed with a Coulter LS 200 Laser Diffraction Particle Analyser, which was equipped with a 1-mm sieve to protect the lens. At least two subsamples were measured for every sample.

XRF analysis

High-resolution non-destructive XRF measurements were performed at the Alfred Wegener Institute, Bremerhaven, with an Avaatech XRF Core Scanner. An additional 26 single samples were measured for calibration purposes (see the ESM for details).

Multivariate statistics

Different transport and deposition processes can be deciphered from grain-size distribution data using end-member modelling analyses (EMMA; Weltje 1997). The obtained grain-size data were analysed with the EMMAgeo package (Dietze and Dietze 2013) in R (R Core Team 2012). To achieve the best fit between modelled and measured data, we performed several model runs with different percentiles (P3/P97, P4/P96, P5/P95, P6/P94, P7.5/P92.5 and P10/P90; ESM Table S2). The minimum number of potential end-members was determined by the cumulative variance that explained 95% of the total variance of the original grain-size data, whereas the maximum number of potential end-members was determined by the mean coefficient of determination, which reflects the explained variance (mean r2; Dietze et al. 2012). To avoid overfitting, we used a model with P3/P97 and four robust and one residual EM, representing the unexplained variance.

Several statistical analyses were applied to portray the main structure in the data set, such as the relationship among single variables, and between variables and samples. To gain better insight into the system we differentiated between external and internal lake variables. A detailed description of the proxies used is included in the first part of the discussion. Lake external parameters are in general characterised as parameters of an externally forced nature, such as grain-size distribution, which is assumed to reflect input to the lake of minerogenic proxies for detrital matter. On the other hand, lake internal parameters are those related to internal processes and mainly form within the lake. Principal component analyses (PCA) were applied after the data were square-root transformed, centred and standardized in CANOCO version 5 (Ter Braak and Šmilauer 2012). A stratigraphically constrained incremental sum of squares analysis (CONISS) in Tilia version 1.7.16 was applied to the square-root transformed data to identify significant changes in the data and assign zones (Grimm 1987).

Results

Age-depth relationship in core KK12-1

A lake-reservoir effect of 1.368 ka was determined using dates on two in situ-collected living charophytes. Results of the lake reservoir-effect-corrected and calibrated samples (ESM Table S1) and the age-depth model (Fig. 1; ESM Fig. S1) show an age of ca. 29 cal ka BP at the base of core KK12-1. The mean sediment accumulation rate is 0.49 mm a−1 and ranges from 0.15 mm a−1 in the lower part of the core to 0.84 mm a−1 at the top, which is, in addition to changes in sediment input, mainly a consequence of changes in the abundance of macrophyte remains in the core.

TIC, TOC, TOC/TN, δ18O Carb and δ13C Carb

Core KK12-1 consists of a sequence of fine-grained, greyish sediment with a variable amount of plant remains. TIC ranges between 1.7 and 8.6% (median: 3.0%; Fig. 2b) and is generally characterised by high sample-to-sample variation. Lowest values occur between 982 and 692 cm (26.3 and 17 cal ka BP) and the highest values between 682 and 402 cm (16.5 and 5.3 cal ka BP). TOC ranges from 0.1 to 2% at 1082 and 452 cm (28.5 and 6.6 cal ka BP), followed by a sharp increase with values reaching 5.3% in the upper part of the core (median 1.6%).

Fig. 2
figure 2

Geochemical and sedimentological data for core KK12-1 including PCA Axis Scores. CONISS was used to identify zone boundaries. a.1 and a.2—for lake external parameters, b—for lake internal parameters

Full size image

TOC/TN ratios display low values. Results from below 692 cm (17 cal ka BP) are unreliable because TN was close to the detection limit. Highest values were determined between the two maxima at 332 and 172 cm (4.1 and 2.0 cal ka BP; Fig. 2b).

Values of δ13C Carb vary between 1.4 and 4.9‰ (median: 3.7‰). δ18O Carb displays values from −7.4 to −0.3‰ (median −3.1‰). Both stable isotope datasets show similar trends over the core, with high values from the core base to 1025 cm (28.6 to 26.5 cal ka BP), a rapid decrease and generally lower values until 652 cm (14.7 cal ka BP), a period of higher values until 292 cm (3.5 cal ka BP) and a decreasing trend to the core top (Fig. 2a.1).

Grain-size distribution and results of end-member modelling

Mean grain-size varies mainly between 7 and 31 µm (median 12 µm). A slight decreasing trend in mean grain-size and generally less variation were identified in the upper 500 cm (<8.1 cal ka BP). The final end-member model (P3/P97) results in four robust EMs and one residual EM (Fig. 3a). All four EMs show bimodal or polymodal grain-size frequency distribution curves, with the main peak at 2–4 µm for EM1, 10–20 µm for EM2, 20–50 µm for EM3 and 100–300 µm for EM4. EM1 accounts for 15.2% of the detrital sediments over the core, EM2 for 30.1%, EM3 for 18.5% and EM4 for 35.6%. EM1 and EM2 have relatively high abundances throughout the core, whereas EM3 and EM4 show larger fluctuations (Fig. 2a.2). The very fine-grained EM1 reaches its highest values, up to 67%, between 672 and 622 cm (15.8 and 13.6 cal ka BP). EM2 is most abundant from 982 to 942 cm (25.8 to 25.0 cal ka BP) with a maximum of 78%, and between 772 and 732 cm (21.0 to 19.5 cal ka BP), with a maximum of 76%. It is constantly recorded in medium to high proportions (>25%) from 492 cm upward (8 cal ka BP to present). EM3 shows peaks of 61% at 832 cm (22.7 cal ka BP), 60% at 452 cm (6.6 cal ka BP) and 65% at 72 cm (0.8 cal ka BP). Peaks for EM4, with maxima up to 73, 79 and 72%, are recorded from the core base until 992 cm (28.6 to 26.2 cal ka BP), from 822 to 782 cm (22.4 to 21.2 cal ka BP) and at 502 cm (8.6 cal ka BP).

Fig. 3
figure 3

End-member modelling for core KK12-1. a—Loading of grain-size-based end-member modelling of Lake Karakul, b—grain-size distribution of the nine reference samples in the vicinity of Lake Karakul

Full size image

XRF data

XRF scanning analyses revealed values from 1.7 to 70.3 for Sr/Rb ratios (median 11.0), 1.4 to 4.6 for Zr/Rb ratios (median 2.1), and 17.3 to 703.1 for Fe/Mn ratios (median 114.3). Br has been reported to be an indicator for organic matter in lake sediments (Biskaborn et al. 2013). Because of the good correlation between TOC and XRF Br, we quantified TOC Br from the peak area intensities of Br, to use it as a high-resolution proxy for the organic carbon content. TOC Br ranges from 0.8 to 5.7% (median 1.51%), showing a rapid increase at 449 cm (6.6 cal ka BP) and high variability in the upper part of the core. The calibrated TiO2 concentrations vary between 0.64 and 0.79% (median 0.69%). TiO2 shows relatively large fluctuations, with an overall decrease from the base to the middle of the core, and lower but constantly increasing values towards the top. Peaks in Sr/Rb and Zr/Rb ratios are mainly correlated with high EM4 portions.

Ordination results of sediment parameters

The first two PCA axes of the lake-external parameters explain 62.9% of the variance (Fig. 4a). TiO2, EM1 and EM2 plot on the positive end of the first axis. Opposite to EM1 and correlated with each other are the Zr/Rb ratio and the δ18O Carb values, which are also correlated with the δ13C Carb values, the Sr/Rb ratio and EM3 in the lower left of the PCA plot. EM4 and EMres plot in the upper left and are negatively correlated with EM2. The first two PCA axes of the lake-internal parameters explain 81.1% of the variance of the data. TOC, TOC Br and the TOC/TN ratio plot in the lower right quadrant, whereas TIC and the Fe/Mn ratio plot in the upper right part of the PCA plot (Fig. 4b).

Fig. 4
figure 4

Biplot for principal component analyses (PCA) of geochemical and sedimentological parameters for core KK12-1. a—for external parameters zone 1: 28.6–13.3 cal ka BP, zone 2: 13.3–6.6 cal ka BP, zone 3: 6.6 cal ka BP to present; b—for internal parameters zone 1: 28.6–18.4 cal ka BP, zone 2: 18.3–6.6 cal ka BP, zone 3: 6.6 cal ka BP to present

Full size image

The stratigraphically constrained cluster analysis suggests a division of the core into three zones for the external parameters. External parameter zone 1 (core base to 613 cm; 28.6 to 13.3 cal ka BP) shows negative PCA axis 1 scores up to 1012 cm depth (26.5 cal ka BP) and mostly positive values above, whereas PCA axis 2 scores are almost exclusively positive (Fig. 2a.2). Zone 2 (602 to 452 cm; 13 to 6.6 cal ka BP) plots with a decreasing trend from positive to negative axis 1 scores and variable, mainly positive axis 2 scores. The external parameter zone 3 (442–0 cm; 6.6 cal ka BP to present) is characterised by increasing trends for axis 1 and slightly increasing axis 2 sample scores (Fig. 2a.2).

The stratigraphically constrained cluster analysis for the internal parameters also results in three zones. Samples from internal parameter zone 1 (core base to 713 cm; 28.6–19.2 cal ka BP) present negative PCA axis 1 and positive and negative PCA axis 2 scores. PCA axis 1 sample scores for the internal parameter zone 2 (712–462 cm, 17.5–6.9 cal ka BP) show an increasing trend with mainly negative values, whereas axis 2 scores are positive throughout zone 2, after rapidly increasing at the beginning of the zone (Fig. 2b). The internal parameter zone 3 (452–0 cm; 6.6 cal ka BP to present) shows slightly decreasing, still positive axis 1 scores and both positive and negative, but slightly decreasing axis 2 scores (Fig. 2b).

Discussion

Paleoenvironmental indicators from sediment variables

TOC concentrations reflect preserved organic matter content (Fig. 2b), i.e. the balance between allochthonous organic matter input, autochthonous organic matter production, and degradation (Meyers and Lallier-Vergés 1999). The major part of TOC in core KK12-1 was probably formed in the lake, given the high abundance of aquatic plant remains. Sparse terrestrial vegetation and low precipitation in the modern catchment of Lake Karakul support the assumption that terrestrial organic matter did not contribute markedly to organic matter accumulation on the lake floor. Accordingly, no terrestrial macrofossil remains were found in the sediments of the core. Thus, considerable input of terrestrial organic matter, which was recorded in Ahung Co on the south-central Tibetan Plateau, did not occur in the drier setting of Lake Karakul (Morrill et al. 2006). We consider Br to be a complementary high-resolution proxy for organic matter content, consistent with previous studies of marine (Mayer et al. 1981; Ziegler et al. 2008) and lake sediments (Kalugin et al. 2007; Biskaborn et al. 2013). The TOC/TN ratios (Fig. 2b) likely reflect the source of sediment organic matter at the core site. Values between 4 and 10 indicate dominance of phytoplankton, whereas values ≥18 indicate a greater contribution from vascular plant sources (Meyers and Lallier-Vergés 1999). Thus, the TOC/TN ratios of core KK12-1 indicate a mix of sources, including lacustrine algae and submerged vascular plants. Combined interpretation of TOC concentrations and TOC/TN ratios may indicate bioproductivity variations in Lake Karakul.

TIC reflects the carbonate content (Fig. 2b), most likely originating from allochthonous input of detrital sediments from the catchment and/or as dust, and/or through precipitation from the water column (Cohen 2003). Visual inspection of smear slides indicated that the carbonate fraction comprises endogenic aragonite crystals, and shells and shell fragments of ostracods and aquatic gastropods. It cannot, however, be ruled out that detrital carbonate from the southern part of the catchment is transported to the lake during sporadic floods and that fine-grained carbonate particles of aeolian origin also accumulated in the lake basin (Strecker et al. 1995).

The strong correlation between δ13C Carb and δ18O Carb (r = 0.82), measured from the authigenic carbonate fraction, reflects the closed-basin setting of Lake Karakul (Talbot 1990). δ13C Carb values (Fig. 2a.1) are influenced by the productivity in the surface waters of the lake, where phytoplankton preferentially incorporate 12C, which leaves the water column enriched in 13C. δ13C Carb values are also controlled by climate conditions, including precipitation and evaporation (Talbot 1990). Precipitation at Lake Karakul is, however, generally low and accordingly, glaciers and snowfields represent the main water sources. Therefore, we assume that δ18O Carb values mostly reflect evaporation/freshwater input ratios and temperature, rather than changes in the amount of precipitation. The long-term δ18O signal is additionally affected by changes in the δ18O signature of the water source (i.e. the ice-volume effect) and the large temperature change over a glacial-interglacial cycle.

End-member modelling of grain-size data was applied to decipher different transport and deposition processes of the inorganic non-carbonate sediment fraction (Weltje 1997). Taking into account the grain-size distribution in the sediments from core KK12-1 and of modern reference samples (Fig. 3b and Table S2), the end-member modelling results imply that the Lake Karakul core sediments reflect four different sediment populations (Fig. 2a.2). Individual end-members, however, do not necessarily represent a single process, but may result from different sources and transport mechanisms, and can be deposited in mixed composition in different sedimentary environments.

EM1, with its narrow peak in the fine-grained fraction, likely represents very fine aeolian dust that was transported over large distances in high-suspension clouds (Wu et al. 2006) and deposited from the atmosphere via precipitation (Vandenberghe et al. 2006; Dietze et al. 2012). Therefore, EM1 might be indicative of the influence of the Westerlies air stream. Furthermore, EM1 could also be regarded as very fine clastic input derived from glacial meltwater, so-called “glacial milk” (Wang et al. 2015), and point towards increased glacier melt.

EM2 shows a main peak in the fine-moderate silt fraction. Thus EM2 (and grain size) is also likely to reflect remote sources of wind-transported material (Tsoar and Pye 1987; Sun et al. 2002). Loess deposits are not common in the Pamir-Alay region above 3000 m asl, so EM2 most likely represents aeolian sediments transported over long distances (Vandenberghe 2013). The fine-grained sediments of EM2, however, may also include the suspension load of glacial outwash material or rock flour (Boulton 1978), which is unidentifiable through grain-size measurement and EMMA.

EM3 covers a wide grain-size range and its dominant grain size can also be found in the reference samples of the modern pond sediment, slack-water silt and dry-river-channel sediment (Fig. 3b and Table S2). The wide range and polymodal distribution are likely indicative of fluvial deposits (Walling and Moorehead 1989). Hence, its downcore distribution probably represents fluvial input to the lake. Accordingly, high portions of EM3 sediments in core KK12-1 are tentatively interpreted as proxies for moister conditions at the study site, although meltwaters from glacier tongues approaching the lake, and local aeolian sediments from alluvial fans and plains, may have also contributed to the accumulation of the relatively poorly sorted silts and fine sands of EM3 (Vandenberghe 2013).

EM4 shows similarities with the reference sample of well-sorted aeolian sand. Its coarse grain size indicates a local source area of hundreds of metres to a few kilometres distance. Although the source of this coarse grain-size fraction is possibly influenced by glacial activity, through formation of unconsolidated sediments by glacial erosion in the area, glaciers are probably not the main means of transport into the lake and to the core location. Because of the distance of the core location from the modern shoreline, and the paleo-shorelines above the lake, we favour transport by strong local winds, especially during winter and under snow-free conditions, which enabled saltation and creep of grains and transport onto the frozen lake surface (Tsoar and Pye 1987; Sun et al. 2002; Mischke et al. 2010).

Quantitative, single-sample XRF measurements over the core enabled the calibration of XRF-scanner data, which resulted in semi-quantitative data (Weltje and Tjallingii 2008; Biskaborn et al. 2013). We used Titanium (Ti) and its oxide (Fig. 2a.1) as a proxy for fine clastic input to the lake (Fey et al. 2009; Kylander et al. 2011). Sr/Rb and Zr/Rb were interpreted as high-resolution grain-size indicators (Biskaborn et al. 2013; ESM). Fe/Mn ratios (Fig. 2b) are generally regarded as a proxy for redox conditions at the lake bottom (Boyle 2001) and thus reflect oxygen concentration at the bottom of Lake Karakul. We furthermore consider Br to be a complementary, high-resolution proxy for organic matter content, consistent with previous studies of marine (Mayer et al. 1981; Ziegler et al. 2008) and lake sediments (Kalugin et al. 2007; Biskaborn et al. 2013; for details on XRF interpretation, see the ESM).

The PCA biplot displays relationships among the external lake proxies (Fig. 4a). Along the first axis, showing the external parameters, plotted proxies indicate coarser grain size (e.g. Zr/Rb as a detrital input and mean grain-size indicator) and imply fluvial input, with δ13C Carb and δ18O Carb reflecting the evaporation/freshwater input ratio on the negative end. Proxies reflecting fine grain size, such as EM1 and TiO2 plot on the positive end of the axis. On the positive end of the second axis, plot EM4 (reflecting minerogenic matter moving through saltation on ice during periods of dry climate) and TiO2, which is assumed to enter the lake mainly as fine dust particles. EM3, which was identified as a fluvial indicator, and EM2, which might reflect glacial outwash or wet-deposited dust, plot on the negative end of the second axis. Although the PCA external axes scores show a more similar trend during the Holocene, they run opposite to one another during the late Pleistocene and LGM. We assume this is a consequence of glacial influence, which affects the transport mechanisms for the grain sizes represented by the end-members. EM1 and EM2 especially, are probably, in part, glacially derived sediments and reached the lake as “glacial milk” during the late Pleistocene.

The main data structure of the internal lake proxies is shown in Fig. 4b. TOC, TOC Br and TOC/TN are indicators for bioproductivity and plot along the first PCA internal axis. The second PCA internal axis reflects lake-bottom ventilation and possibly lake-level changes, indicated by the strong correlation with the Fe/Mn ratio and TIC. Whereas Fe/Mn is often used as an indicator for redox conditions, carbonate precipitation as aragonite crystals from surface waters in a Mg2+- and SO4 2−-dominated, closed-basin lake, normally occurs during periods of sufficient runoff and relatively warm conditions, both facilitating the establishment of a more stratified water column with oxygen deficiency at the lake floor (Mischke et al. 2010). Thus, axis 2 of PCA internal possibly indicates lower lake levels with oxidizing conditions at the lake bottom clustering on the negative end, and higher lake levels on the positive end of the second PCA axis. The bioproductivity proxy cluster is uncoupled from the lake-level proxies, which supports our interpretation for the lake internal proxies.

Implications of the Lake Karakul sediment record

Pre-LGM to late glacial (28.6–13.3 cal ka BP)

The pre-LGM phase of the KK12-1 record (zone 1; Fig. 2a.2) is characterised by a high input of EM4 sediments, probably indicating strong local cold-season storms. Moderate values of EM1 and low values of EM2 probably reflect wet deposition of fine dust transported over long distances, but a generally low accumulation of coarser-grained dust and/or low glacial sediment input during this time. Moderate Zr/Rb and Sr/Rb ratios possibly result from the presence of feldspars and other detrital minerals (Kalugin et al. 2007; Kylander et al. 2011), whereas medium peaks coincide with EM4 peaks, suggesting an input of coarser-grained sediments. High δ18O Carb values could result from marked aridity and strong evaporation from the lake surface, which was apparently not sealed by long-lasting ice cover.

EM4 reaches high proportions between 22 and 20 cal ka BP, pointing towards dry and stormy winters. Relatively arid and cold conditions that prevailed during the LGM are consistent with the dominance of EM2 sediments in the core. Generally decreasing δ18O Carb values during the LGM presumably result from global cooling, which led to isotopically depleted precipitation in the Karakul region, and decreasing evaporation effects from the lake surface, as a consequence of prolonged periods of ice cover.

High values of EM1 during the late glacial could point towards persistent long-distance transport, presumably recording a Westerly influence and fine-dust accumulation with precipitation (Wu et al. 2006; Vandenberghe et al. 2006) or increased sediment-rich glacial meltwater (Liu et al. 2014; Wang et al. 2015). Increasing EM3 values suggest marked fluvial input to the lake, likely caused by increasing temperatures and glacial meltwater discharge, which might also be reflected by slightly increasing Zr/Rb and Sr/Rb values and increased mean grain size of detrital input (Biskaborn et al. 2013; ESM). Accordingly, high δ18O Carb values probably result from relatively warm temperatures, isotopically less depleted precipitation in the Lake Karakul region, and enhanced evaporation from the lake surface during a longer ice-free period.

Zech et al. (2005) claimed there were glacial advances around 27 ka BP in the southern Pamirs. Dortch et al. (2013), investigating glacial deposits in the western Himalayan range using 10Be surface exposure ages, proposed several glacial advances, termed semi-arid western Himalayan–Tibetan stages (SWHTS). One SWHTS, named 2F, occurred around 30 ± 3 ka BP (Fig. 5) and three other local glacial stages in the Himalayan–Tibetan mountains were dated between 17 and 14 ka BP. They were apparently synchronous with the Oldest Dryas and Heinrich event 1 of the Northern Hemisphere and were likely controlled by precipitation from the Westerlies. Owen and Dortch (2014) compare the SWHTS with Monsoonal Himalayan–Tibetan stages (MOHITS; Murari et al. 2014) and argued that at times SWHTS and MOHITS were concurrent and probably had similar moisture sources, whereas at other times only one of the stages is evident, implying differing climate controls. Komatsu and Tsukamoto (2015) recorded a late glacial glacier advance in the Karakul catchment at 15 ka BP and suggested that glaciers responded to enhanced moisture availability from the Westerlies. Their suggestion is in agreement with increased Westerlies precipitation (EM1), higher fluvial input or increased glacial meltwater discharge (EM3) to the lake, inferred from the KK12-1 core.

Fig. 5
figure 5

Comparison of the Lake Karakul record with a summer and winter insolation at 40°N (Laskar et al. 2004); b Kesang cave (Cheng et al. 2012); c lake level changes of Lake Van (Çağatay et al. 2014); d semi-arid western Himalaya–Tibetan Stages (Dortch et al. 2013); e lake level changes from Lake Issyk Kul (Ricketts et al. 2001; Romanovsky 2002); f PCA axis scores for lake internal and lake external parameters from Lake Karakul, Tajikistan, this study; g pollen-derived aridity Index from Lake Son Kul (Mathis et al. 2014); h mean effective moisture from monsoonal Central Asia (Herzschuh 2006)

Full size image

End of the late glacial and early to middle Holocene (13.3–6.6 cal ka BP)

Indicators that mainly reflect fine aeolian input to the lake, EM1 and 2, in combination with TiO2, display low values between the end of the late glacial and the early Holocene (zone 2; Fig. 2a.2), and high values of EM3 imply increased fluvial input. This could be a consequence of increased precipitation and surface runoff or glacier melt. No evidence, however, has yet been found, which suggests that glaciers came close to the lake during the early to middle Holocene. A maximum of EM4 and the residual EM at ca. 8.5 cal ka BP may indicate the occurrence of especially dry and stormy winters, analogous to the LGM, or more likely, the formation of a single event layer of aeolian sand. Maximum Sr/Rb and Zr/Rb ratios probably reflect increased grain-size associated with higher input of EM3 and EM4, and signal coarser detrital input. The δ18O Carb values are high at the end of the late glacial and during the early and middle Holocene, most likely reflecting rising global temperatures and the arrival of less depleted precipitation in Lake Karakul’s catchment. The high δ18O Carb values could have been caused by increased evaporation that resulted from higher temperatures and shorter ice-cover periods. A marked change in moisture availability towards wetter conditions in the study region is reflected in this zone.

Global temperature rose strongly during the Bølling/Allerød and early Holocene until approximately 9.5 ka BP, and stayed high until about 5.5 ka BP (Marcott et al. 2013). Seong et al. (2009) inferred several glacial advances in the Muztag Ata and Kongur Shan Mountains during the early Holocene and argued for moisture transport from the middle latitude Westerlies, supported by the pollen-derived aridity index from Lake Son Kul (Fig. 5; Mathis et al. 2014).

6.6 cal ka BP until present

From 6.6 cal ka BP onward (zone 3; Fig. 2a.2) EM1 and EM2 show an increasing trend, reflecting enhanced influence of drier air masses. EM4 shows a more sporadic pattern towards the present day, which likely reflects reduced winter storms and the weakening influence of the Siberian High. EM3 displays slightly declining values that accelerated in the last 2 cal ka, also implying decreasing fluvial input, caused by either less moisture availability or a lower input of glacier-transported material, given the great distances of glaciers from the lake and coring location. Decreasing Sr/Rb and Zr/Rb ratios correspond to a decreasing mean grain size and a steady increase in TiO2, emphasizing the increasing long-distance dust input. Declining δ18O Carb values are probably recording the global cooling trend since the middle Holocene, with 18O-depleted incoming moisture and reduced evaporation (Taft et al. 2014).

Liu et al. (2014) investigated glacial activity in the westernmost Tibetan Plateau using multi-proxy analyses of glacio-lacustrine sediments from Lake Karakuli (~150 km SE of Lake Karakul) and found a general correlation between local glacier advances and Holocene ice-rafting events in the northern Atlantic. Sediments from Lake Son Kul (~340 km NE of Lake Karakul) in the Tien Shan Mountains show a moderate drying trend, with negative water balance, and only a few decadal to centennial reversals between 8.2 and 2 cal ka BP (Huang et al. 2014; Lauterbach et al. 2014). Ricketts et al. (2001) infer a wet period preceding 6.9 cal ka BP for Lake Issyk-Kul (~480 km NE of Lake Karakul), suggesting increased moisture transport from the west or possible Asian Monsoon influence. This is in accordance with findings by Mathis et al. (2014) at Lake Son Kul, whose aridity index implies moister conditions from ca. 8.4 to 4.5 cal ka BP, and drier conditions thereafter (Fig. 5).

Linking lake internal development to climate change

Pre-LGM and LGM (28.6–19.2 cal ka BP)

Low PCA internal axis 1 sample scores and low TOC suggest low bioproductivity and poor preservation of organic matter (Meyers and Lallier-Vergés 1999). The bioproductivity reaches a minimum during the LGM. Moderate TIC and PCA internal axis 2 values during the pre-LGM show that inflow to the lake and resulting carbonate precipitation were still significant before the onset of the LGM. TIC and PCA internal axis 2 values decreased rapidly towards the LGM. Thus, decreased inflow, reduced carbonate precipitation and low lake level prevailed during the LGM. Low Fe/Mn ratios, as a redox proxy, imply a well-mixed waterbody with sufficient dissolved oxygen at the lake bottom (Boyle 2001; Cohen 2003). This inference is consistent with the reconstructed cold and windy climate. Accordingly, climate is surmised to have been the main driver of the lake’s internal conditions during the pre-LGM and LGM. Very low TOC values have also been recorded on the Tibetan Plateau, suggesting low bioproductivity and cold temperatures during the LGM (Zhang and Mischke 2009; An et al. 2012; Zhu et al. 2015). Although only few dates are available at Lake Issyk-Kul in Kyrgyzstan, situated NE of lake Karakul, lacustrine terrace sediments 33 m above current lake level indicate a period of lake-level high-stands starting around 26 ka BP and lasting until the late glacial (Fig. 5; literature in Russian quoted in Romanovsky (2002) and Ricketts et al. (2001)). Similar dates for high-stands were recorded for Lake Van in eastern Turkey (Fig. 5), west of our study region, at approximately 26 cal ka BP and the end of the LGM (Çağatay et al. 2014). Moderate to low lake levels were reconstructed for Lake Karakul during this period.

Late glacial and early to mid-Holocene (17.5 to ca. 6.9 cal ka BP)

A slight increase in bioproductivity is indicated by the continuously increasing PCA internal axis 1 and TOC values (Meyers and Lallier-Vergés 1999). TIC values rise rapidly and reach maximum values during the late glacial, indicating high runoff and enhanced carbonate precipitation. The Fe/Mn ratios show a marked increase during the late glacial and culminate during the onset of the Holocene, implying reducing conditions at the lake bottom (Cohen 2003). The PCA internal axis 2 scores are high during the late glacial and early to middle Holocene, probably reflecting a high lake level (zone 2; Fig. 2b). This inference is supported by the study of Komatsu and Tsukamoto (2015) who proposed a maximum lake level of 35 m above the modern stage during the late glacial and provide field evidence of glaciers reaching the lake at 15 ka BP. This coincides with several glacier advances (SWTHS 2D-2A) during the late glacial recorded by Dortch et al. (2013) (Fig. 5) and Seong et al. (2009), who investigated moraines in the Muztag Ata and Kongur Shan massifs 170 km SE of Lake Karakul, suggesting glacial advances around 17 and 13.7 ka BP. In contrast, lakes farther west of Lake Karakul, such as Lake Van, experienced a rapid drop in lake level after 21 cal ka BP (Landmann et al. 1996; Çağatay et al. 2014).

Early Holocene TOC values remained relatively low in contrast to rapidly increasing Fe/Mn ratios (zone 2; Fig. 2b). The high TIC values indicate that runoff to the lake was high. In consequence, a relatively unchanged bioproductivity and rapidly increasing lake level is inferred for the early Holocene, which led to the establishment of poorly oxygenated bottom waters in Lake Karakul. The inferred moderate bioproductivity in the lake possibly resulted from relatively low lake-water temperatures, in comparison to air temperatures, a consequence of enhanced input of glacial meltwater. We thus suggest that climate was the main driver of the internal lake development.

Ricketts et al. (2001) examined trace elements and δ18O values of ostracod shells from Lake Issyk-Kul and inferred high freshwater input to the lake from the beginning of their record at 8.7–8.3 cal ka BP. They speculated about a northward extension of the Asian Monsoon in the early Holocene as a result of the higher summer insolation. It remains unclear, however, whether or not the northward extension of the Asian Monsoon in the early Holocene existed and reached the study region. A relatively stable, high lake level with sufficient inflow prevailed at Lake Issyk-Kul until 6.9 cal ka BP, consistent with our findings for the early and middle Holocene.

6.6 cal ka BP until present

High TOC values and PCA internal axis 1 scores signal high bioproductivity during the last 6.6 cal ka BP (zone 3; Fig. 2b). Values for TOC/TN between 10 and 16 reflect a shift in organic matter origin from purely planktonic to more mixed sources: planktonic, macrophytes, macroalgae, and/or terrestrial. This shift was most likely caused by an increase in the amount of submerged macrophytes, of which abundant remains are recorded in the upper part of the core. Continuously decreasing Fe/Mn ratios point towards increasing ventilation, likely caused by decreasing lake level and better oxygenation of bottom waters. The inferred lower lake level after 6.6 cal ka BP likely resulted from a shift from moister to drier conditions, and from reduced meltwater inflow. The decreasing TIC values probably reflect decreasing inflow. Hence, internal lake development was mainly related to climate aridification during the last 6.6 cal ka.

A similar development is inferred at Lake Issyk-Kul, where the δ18O values of ostracod shells suggest a major change in lake hydrology between 6.9 and 4.9 cal ka BP (Ricketts et al. 2001; Fig. 5), which led to a shift from an open-lake system to a mostly closed-lake system related to increasing aridity compared to the middle Holocene. Ricketts et al. (2001) suggested increased biological productivity at Lake Issyk-Kul during the second half of the Holocene, after the shift to a closed system, corresponding to our findings from Lake Karakul for the last 6.6 cal ka. Sedimentological studies from Lake Son Kul in the Tian Shan Mountains recorded a similar drying trend, interrupted by short intervals of moister conditions (Huang et al. 2014; Lauterbach et al. 2014), while pollen-derived data indicate a shift from moister to drier conditions at 5.0–4.5 cal ka BP (Mathis et al. 2014; Fig. 5). A trend towards cooler conditions from about 3.5 to 1.9 cal ka BP was reported for Lake Karakul by Mischke et al. (2010) who investigated the last 4.2 ka BP at high resolution. Low TOC values and low Fe/Mn ratios in core KK12-1 support these interpretations, implying decreased bioproductivity and increased mixing and ventilation of the lake floor. After approximately 1.9 cal ka BP, a warmer phase led to increased meltwater supply (Mischke et al. 2010; Taft et al. 2014), reflected by increased Fe/Mn values in core KK12-1 until 0.6 ka BP. Similarly, a warm, but not moist period was observed elsewhere in Central Asia, such as at Lake Karakuli (Aichner et al. 2015), which could be indicative of the Medieval Warm Period (MWP). Afterwards, drier conditions prevailed, possibly reflecting the Little Ice Age, in accordance with the inferences of Mischke et al. (2010).

Conclusions

We identified three main climate phases over the last 29 ka, using multiple variables in a sediment core from Lake Karakul, Tajikistan. The Karakul region was influenced by relatively dry conditions and long-distant transport of aeolian sediments during the pre-LGM, LGM and late glacial. Moister conditions prevailed towards the end of the glacial and in the early and first half of the middle Holocene, until ca. 6.6 cal ka BP. During the second half of the middle Holocene, a drying trend set in, most likely caused by drier air masses transported by the Westerlies, and arid conditions in the Pamir Mountains prevailed during the late Holocene and lasted until today.

Reconstructed changes within the lake were mainly related to changes in the productivity and redox conditions at the lake bottom. Low lake level, coinciding with high ventilation and low productivity, prevailed during the pre-LGM and LGM at Lake Karakul, followed by a lake-level high-stand during the late glacial and early to middle Holocene (17.5 until ~6.6 cal ka BP). Productivity increased rapidly, with enhanced ventilation and declining lake levels from 6.6 cal ka BP onwards. Comparisons of lake internal signals from Lake Karakul with those from other relatively deep lakes in the Westerlies region suggest that the sedimentation regime in Lake Karakul was driven largely by climate, which is not unexpected, as it is mainly fed by glacial meltwater.