Introduction

Loess/paleosol sequences in the Chinese Loess Plateau (CLP), especially those with high accumulation rates in the northern and western CLP, are among the best terrestrial archives for late Quaternary paleoclimatic studies1,2. Paleoclimatic records in this area are extremely important for the understanding of the linkage and interplay between the evolution of the ASM at low latitudes3 and temperature variations at high latitudes4. However, until now, a high resolution paleoprecipitation reconstruction using a clear driving mechanism has been lacking in the western CLP.

Fractionation of carbon isotopes by modern C3 plants during CO2 uptake and fixation (Δ) can be described by mathematical models using the formula Δ = a + (ba)(pi/pa), where a is the carbon isotopic fractionation during CO2 diffusion (ca. 4.4‰), b is the net fractionation caused by carboxylation (ca. 29‰) and Pi and Pa are the intercellular and ambient partial pressures of CO2, respectively5. In principle, increasing precipitation will result in more negative carbon isotopic composition (δ13C) values of C3 plants, due to the relatively high stomatal conductance present under a humid environment which will increase Pi and Δ5 and vice versa.

The relations between the δ13C of modern C3 plants and environmental factors have been widely studied and consistent results have been obtained which indicate that the δ13C of modern C3 plants responds principally to precipitation variations6,7,8 (Supplementary Figs. S1, S2 and S3). To a lesser extent, it also reflects more minor changes in temperature, altitude and latitude9,10. Clearly, if the local terrestrial ecosystem was dominated by, or composed entirely of, C3 plants during a specific interval, the sedimentary δ13C data that derives from the local terrestrial biomass should be an indicator of precipitation. In Western Europe, total organic carbon isotopic data (δ13CTOC) of loess indicate a great predominance of C3 plants since the LG and therefore these δ13CTOC data have been used for precipitation reconstruction11,12.

During the past decade, the δ13CTOC record of more than 10 Chinese loess/paleosol profiles has been studied (Fig. 1). The results demonstrate that, along the temporal sequence, the relative abundance of C4 plants increased from the LG to the Holocene (Supplementary Fig. S4); and that, along a spatial gradient, relative C4 plant abundance decreased gradually northwestwards (Supplementary Fig. S5)13,14,15,16,17, as is the case today18. During the LG, δ13CTOC data from the high-temperature eastern CLP (to the east of the Liupan Mts. Fig. 1) indicate that this area was dominated by C3 plants with a minor increase in C4 plants during marine isotope stage 314,17 (Supplementary Fig. S6), which means that the vegetation in the western CLP during the LG could be expected to be predominantly or entirely composed of C3 plants (Supplementary Fig. S6).

Figure 1
figure 1

Locations of the Chinese loess/paleosol profiles that are cited in this paper and their basic climatic background26.

Solid red line in the top left corner shows the distribution of the surface soils for δ13CTOC study in arid central Asia28, adopted in this paper as modern quantitative relations reference27; the original image in the top left corner was downloaded from the public service website of the National Administration of Surveying, Mapping and Geoinformation, China (http://www.tianditu.cn/map/index.html). The solid red dots indicate the locations of the Yuanbao (YB) and Jingyuan (JY) profiles and solid blue dots indicate the locations of the loess profiles in the eastern CLP for δ13CTOC studies13,14,15,16,17. For the codes and δ13CTOC records of these profiles, please refer to Supplementary Fig. S4, S5 and S6.

The Yuanbao profile (YB, 103.63°E, 35.15°N, 2,040 m a.s.l.), close to the Tibetan Plateau, is located on the western edge of the CLP (Fig. 1) with modern mean annual temperature (MAT) and precipitation (MAP) of ca. 6.8°C and ca. 500 mm, respectively. The Jingyuan profile (JY, 104.6°N, 36.35°N, 2,210 m a.s.l.), close to the Tengger Desert, is located on the northwestern edge of the CLP (Fig. 1) with modern MAP and MAT of ca. 238 mm and ca. 5.2°C, respectively. At both sites, precipitation primarily occurs during May to September (ca. 80% of MAP) with higher temperatures (also the main growing season of the local terrestrial vegetation), consistent with the modern climate in this area as controlled principally by the Asian monsoon (Supplementary Fig. S7 and S8). High resolution optically stimulated luminescence (OSL) dating clearly demonstrates that the loess/paleosol sequences of YB19,20 and JY21,22 have accumulated continuously since the LG without significant hiatuses. Previous studies have demonstrated that the loess/paleosol sequences located in the western CLP, including YB and JY, have great potential for high-resolution paleoclimatic reconstruction during the LG21,22,23. The top 25.74 m of YB was sampled for δ13CTOC study at a 4 cm interval (ca. 100 years per sample)24; and the top ca. 30 m of JY was sampled at a 40 cm interval (ca. 1000 years per sample) for δ13CTOC study25.

High resolution δ13CTOC data (Fig. 2) from YB indicated that the study site was dominated by C3 plants during the LG and the relative abundance of C4 plants slightly increased during the Holocene period24,26 (Supplementary Fig. S6). δ13CTOC data from JY (Fig. 2) indicated that the vegetation in the study site was dominated by C3 plants during both the LG and the Holocene25 (Supplementary Fig. S6). These results are consistent with the previous finding that the relative abundance of C4 plants decreased northwestwards across the entire CLP13,14,15,16,17,18 (Supplementary Fig. S4, S5). All of this evidence demonstrates that the δ13CTOC data from YB during the LG and those from JY since the LG are suitable for paleoprecipitation reconstruction and that the results from both sites can be directly compared.

Figure 2
figure 2

Loess δ13CTOC data of the JY (a) and YB (b) profiles plotted against ages.

Red dots represent the OSL data of the JY21,22 and YB19,20 profiles, with errors represented by light blue bars. The age series of both profiles were obtained by linear interpolation of these OSL data. Bold grey arrows indicate the same trend in loess δ13CTOC data of the JY and YB profiles during the LG. The most important and opposite trend occurs during the transition from the LG to the Holocene, which indicates a relative increase in abundance of C4 plants towards the Holocene at the YB site (also in Supplementary Fig. S6).

For paleoprecipitation reconstruction, the quantitative relations between δ13C of modern C3 plants and precipitation are established directly from modern case studies6,7,8,9. However, although almost all of the relevant studies have shown a negative correlation between the δ13C of C3 plants and precipitation, the magnitude of the correlation varies between different C3 plant species and in different regions6,7,8,9 (Supplementary Fig. S2, S3 and Table S1). Here we have used the results of δ13CTOC measurements of 196 modern surface soils from arid central Asia27,28 as our point of reference (Supplementary Fig. S9). Our selection is based on: (i) the close location of these surface soils to the JY and YB sites; and (ii) the fact that both investigation on modern plants29 and the δ13CTOC data of these surface soils27,28 (most of them < −24‰) demonstrate that the modern terrestrial ecosystem in this area is greatly dominated by C3 plants (Supplementary Fig. S9). Considering that the δ13C of modern C3 plants in arid central Asia principally responds to summer precipitation and the constrained uncertainties of the linear correlation between averaged surface soil δ13CTOC values of 19 weather stations and the summer precipitation amount (May to Sept.) as recorded by the corresponding weather stations, we chose the correspondingly quantitative relation27 for the reconstruction of summer precipitation (Supplementary Fig. S9). Because the carbon isotopic composition of organic matter becomes more positive after long-term decomposition30, we have selected 1‰ as the magnitude of the baseline value for the δ13CTOC data of loess/paleosol samples relative to their original values at the time of deposition. The secondary effect of temperature and atmospheric CO2 concentrations on C3 plant δ13C has been neglected. Based on these assumptions our final expression for summer precipitation reconstruction is: summer precipitation (mm) = −58*(δ13CTOC*1000 − 1) − 1266.5.

Correspondingly, the 95% confidence interval of the relevant linear correlation27 is used for the uncertainty estimation of the summer precipitation reconstruction (Supplementary Fig. S9).

Results

In spite of the different resolutions, a high variability occurs in summer precipitation reconstructions of both the YB and JY sites, apparently consistent with the characteristics of modern climate in the study area (Supplementary Fig. S10); this indicates the high sensitivity of carbon isotopes of local terrestrial vegetation to variations in summer precipitation.

Our results indicate that summer precipitation varied from ca. 20 mm to 150 mm with uncertainties of ca. 70 80 mm at the JY site during the past 70 ka with most values during the LG of less than 100 mm. Summer precipitation decreased gradually from MIS3 to an extremely arid MIS2 and then increased towards the Holocene (Supplementary Fig. S11). Considering the location of the JY profile so close to the Tengger Desert (Fig. 1), an extremely arid climate at the JY site during MIS2 is reasonable. Although there are only a few data values for the Holocene, the results indicate that the maximum summer precipitation during the Holocene was more than 150 mm (with uncertainty of ca. 70 mm), which is close to the modern summer precipitation in the JY area (ca. 190 mm, averaged from 1961 to 1990) (Supplementary Fig. S10).

Reconstructed summer precipitation at the YB site during the LG varied mainly from ca. 150 mm to 350 mm with uncertainties of ca. 70 mm (Supplementary Fig. S11, Table S2). Modern summer precipitation in the YB area is ca. 400 mm (averaged from 1961 to 1990, Supplementary Table S3), greater than that of the JY site. During the LG, reconstructed summer precipitation at YB was generally higher than that at JY (Supplementary Fig. S10), suggesting that the modern climatic gradient between the JY and YB sites is consistent with that during the LG, which supports our methodology.

Loess δ13CTOC data of the YB profile during the Holocene are apparently contaminated by C4 plants, especially during the early Holocene (Fig. 2, Supplementary Fig. S6) and this is the essential reason why a reconstruction of summer precipitation for the whole Holocene at the YB site has not been completed. However, if loess δ13CTOC data of the topmost 2 samples (−28‰ and −27.5‰ respectively, Supplementary Table S2) of the YB profile were to be used for the calculation of summer precipitation, then summer precipitation levels of ca. 415 mm and 385 mm (with uncertainties of ca. 70 mm) respectively would be obtained, consistent with modern summer precipitation at the YB site of an averaged value of ca. 400 mm; these findings also support our methodology.

A series of rapid climatic events during the LG has been recorded at the YB site using a grain size proxy for winter monsoon intensity23. These events may be associated with temperature variations at high latitudes of the Northern Hemisphere via a relation between the intensity of the winter monsoon and the north westerlies1,21,22,23. Based on the corresponding North Greenland Ice Core Project (NGRIP) ice-core ages (GICC0531) of Heinrich and interstadial events, we adjusted the ages of the YB winter monsoon record based on their correlation with NGRIP4 oxygen isotopic (δ18O) record (Supplementary Fig. S12). There is an overall consistency between the OSL dating results of the YB profile and the transferred NGRIP ice-core ages, with relatively younger (ca. 4 5 ka) OSL ages19,20 occurring between 20 ka to 60 ka (Supplementary Fig. S13).

Summer precipitation along the adjusted NGRIP ice-core age series at the YB site shows a high degree of consistency with the stalagmite δ18O record from Hulu Cave3 in southern China. Both data sets record Heinrich events 1 to 6 and interstadial events 1 to 19 (Fig. 3); in general, arid events, represented by sharply decreased summer precipitation in the western CLP, broadly correspond to a weakened ASM recorded in stalagmite δ18O data from southeastern China3 and to cold events in high northern latitudes recorded in the NGRIP δ18O data4. More importantly, from the H4 to H1 events, the stalagmite δ18O data gradually became more positive and the summer precipitation in the YB area also decreased gradually (Fig. 3). Assuming that the stalagmite δ18O data from southern China have recorded variations in the intensity of the ASM3 and based on the comparison of the YB and Hulu records, it appears that the decrease in the intensity of the ASM was the direct cause of the decrease in summer precipitation in northwest China.

Figure 3
figure 3

Comparison of the summer precipitation reconstruction results of the YB profile and other records.

(a) Hulu stalagmite δ18O record3 (purple-PD, green-MSD, red-MSL) from southern China; (b) reconstructed summer precipitation at the YB site during the LG (blue dots represent the calculated values of summer precipitation, the light blue bars represent the uncertainties of the corresponding reconstructed summer precipitation, the solid red line represents the 500-years-time-window running averaged values and the original data is shown in Supplementary Table S2); (c) grain size data of the YB profile (green dots represent the original values, the solid red line represents the 500-years-time-window running averaged values and the original data is shown in Supplementary Table S2); (d) the Northern Hemisphere solar insolation34 (NHSI, black line, 65°N, average value of June); (e) NGRIP ice-core δ18O record4 (blue line, with GICC05 chronology31). Black Arabic numerals indicate the interstadial events and the yellow vertical bars indicate the YD and Heinrich events. Gray arrows demonstrate the overall trends. Age series of the YB records were transferred from the NGRIP ice core (Supplementary Fig. S12).

Discussion

The assumption that the Chinese stalagmite δ18O data record the intensity of the ASM has been widely debated32,33. For example, in a recent comparative simulation study of the Last Glacial Maximum (LGM) and Heinrich event 1 (H1), changes in the Indian monsoon that are controlled by the sea surface temperature (SST) of the Indian Ocean have been suggested as the main cause of the stalagmite δ18O variations in China33. Due to the lack of direct evidence that the Indian Ocean SST during H1 was lower than during the LGM and based on the high degree of consistency between the variations in our reconstructed summer precipitation and those in the southern Chinese stalagmite δ18O data (Fig. 3), our results demonstrate that, at least during the LG period, variations in the Chinese stalagmite δ18O data records are a valid indicator of the intensity of the ASM.

During the transition from the LG to the Holocene, summer precipitation in the YB area and the intensity of the ASM3 decreased gradually. Subsequently, an enhancement of both the ASM and summer precipitation occurred around the time of the H1 event (Fig. 3). The onset of the rise of temperature at high latitudes recorded by the NGRIP δ18O data4 occurred at around the time of the H2 event or earlier (Fig. 3). Grain size data from the YB profile indicate that the onset of the decrease of the East Asian winter monsoon was also coeval with the H2 event23 (Fig. 3), which is also consistent with grain size records from the Luochuan1 and JY22 profiles. During the transition from the LG to the Holocene, it seems that both high latitude temperatures4 and the mid latitude East Asian winter monsoon1,22,23 responded rapidly to increasing insolation34 (Fig. 3). However, the response of the low latitude ASM was apparently delayed3. All of this evidence indicates more complex driving mechanisms for the low latitude summer monsoon than for the mid and high latitude climatic components.

In conclusion, our results are significant in the following respects: 1) because previous studies have already demonstrated that the local biomass on the western edge of the CLP24,25,26 and in the loess area in Western Europe11,12 was dominated by C3 plants since the LG, sedimentary δ13CTOC derived therefrom may be a valid indicator of paleoprecipitation; and therefore our work provides a new paradigm for paleoprecipitation reconstruction in this vast region between Western Europe and the western CLP and the regions to their north; 2) the paleoclimatic significance of the Chinese stalagmite δ18O data has been widely debated32,33; however, we demonstrated that at least during the LG, the Chinese stalagmite δ18O data are a credible indicator of the intensity of the ASM. Further work needs to be done to test the validity of this finding over longer timescales; and 3) the combination of future independent paleotemperature reconstructions and our paleoprecipitation reconstruction results may constitute a major advance in paleoclimatic studies of monsoonal East Asia.

Methods

The YB profile was sampled at 2 cm intervals (ca. 50 years per sample) for the top 25.74 m. After removal of organic matter and carbonate with HCl (10%) and H2O2 (10%), the grain size of all the samples was measured using a Malvern Master Sizer 2000 laser diffraction analyser. The measurement range of this equipment is 0.02–2000 μm. In this paper we use the percentage of the grain size fraction > 40 μm as an indicator of the east Asian winter monsoon intensity and the strength of the north westerlies1,21,22,23. A 4 cm interval was used for the δ13CTOC measurements. The samples were pretreated as follows: 10% hydrochloric acid (HCl) was used to remove carbonates, followed by washing with distilled water until the suspension was neutral. The wet samples were then sieved at 120 μm to remove tiny sand particles and gravels. After sieving, the samples were dried at 70°C. The gas collection method involved static combustion. The evolved CO2 was analyzed for δ13C using a Thermo Finnigan Delta Plus mass spectrometer. The standard materials used for the measurements were international standard tree-rings (Corundum balls, IAEA-C5). The mean value of 29 repeat measurements was −25.7‰ with a standard deviation of ± 0.13‰ (the reported value of the standard is −25.49 ± 0.72‰). Repeated measurements on both the standard materials and samples (81 times in total) showed that the experimental error is less than ± 0.2‰. (See references 19 and 20 for the OSL dating methods used on profile YB and references 21 and 22 for those used on profile JY. See reference 25 for the organic carbon isotopic data for profile JY.)