A loess-sand section at Zhenbeitai (ZBT) (38°20′22″N, 109°44′12″E) in the transitional zone between the Chinese Loess Plateau and the Mu Us dune field was sampled (Figure 1). Mean annual precipitation (MAP) at the study site is 401 mm, most of which (c. 87%) falls during the growing season (April-September), the annual precipitation variability and monthly precipitation variability of the past 60 years is about 20.63% and 80.12%, respectively. The mean annual temperature (MAT) is 8.4°C.

The ZBT section is composed of distinct alternations of sandy loam soils and eolian sandy loess or sand units (Figure 2). More detailed lithosphere is shown in Table 1. After excavating a 30-40 cm trough to avoid surface contamination, vertical sections were sampled based on stratigraphic observations in the field. In order to minimize the influence of modern vegetation, we begin to sample at a depth of 20 cm. Six OSL samples were collected by hammering stainless steel tubes into cleaned vertical sections and sealing with black plastic bags to avoid light exposure and moisture loss. Samples for organic carbon isotope analysis and magnetic susceptibility analysis were taken at each 10 cm interval from the top to a depth of 17.3 m, for a total of 174 bulk sediment samples.

In the laboratory, each OSL sample was pretreated with H₂O₂ (30%) and HCl (10%) to remove organics and carbonates, respectively, then rinsed and wet-sieved to select the 90-125 μm grain size. Etching using HF (40%) for 40 min to remove feldspars was followed by re-sieving. All OSL paleodose measurements were undertaken at Nanjing University using a Riso TL/OSL-DA-20C/D reader under the Single Aliquot Regenerative protocol (Murray and Wintle, 2000). The environmental dose rate was calculated from the U, Th and K concentrations, measured by neutron activation analysis (NAA) and inductively coupled plasma mass spectrometry (ICP-MS) inductively coupled plasma atomic emission spectrometry (ICP-AES).

For organic carbon isotope analysis, all samples were sieved to <250 μm after being dried for 48 h at 40℃, in order to remove coarse grains and small plant fragments. Each sieved fine portion was ground to 200-mesh with an agate mortar. Carbonate was removed by adding 10% HCl and repeatedly rinsing with distilled water (till pH≈7). The soil samples were finally dried at 50℃ for isotopic analyses. The carbon isotope composition was determined with an isotope ratio mass spectrometer (Delta Plus; Finnigan MAT) in State Key Laboratory of Lake Science and Environment, Chinese Academy of Sciences. Carbon isotopic data is expressed as δ¹³C, with δ¹³C=(R_sample/R_standard)-1, where R is the ¹³C/¹²C ratio and standard is the Vienna Pee Dee Belemnite standard. Magnetic susceptibility (MS) was measured by Bartington MS2 magnetic meter.

Soil organic matter is derived from plants, with little isotopic discrimination (Cerling et al., 1997; Huang et al., 2001). Since the plant composition is mainly controlled by precipitation in this region (Liu et al., 2005a, 2005b; Chen et al., 2006; Yao et al., 2011), precipitation and δ¹³C of soil organic matter should be related. Although a relationship between δ¹³C and rainfall was quantitatively reconstructed for Nussloch (Hatté et al., 1998, 2001), only C₃ plants grew in the cool high latitude setting of that study, and the same relationship cannot be applied to warmer and drier middle latitude regions where C₄ plants are common. Ning et al. (2008) had developed an empirical relationship between δ¹³C_org and precipitation based on the plant and surface soil of the Chinese Loess Plateau, where the mean annual precipitation of the study area has varied between 240 mm and 700 mm during the past ten years. The model was applied to Lantian and Weinan sections in the southern Chinese Loess Plateau, and yielded precipitation of 360-792 mm and 348-845 mm during the last glacial-interglacial cycle, respectively. Our study section is close to the surface soil sampling sites that Ning et al. (2008) studied, and all the sites are dominated by the Asian monsoon circulation, therefore, we initially estimated the precipitation by using this model (Ning et al., 2008):

R = [SQRT(7.17×10²+25.7× δ¹³C_TOC)-6.60]/1.29×10^-2) (1)

where R represents the annual precipitation. The function was originally expressed as follows:

δ¹³C_TOC=Fδ¹³C₄+(1-F)δ¹³C₃ (2)

where δ¹³C_TOC is the δ¹³C value of the soil organic matter (‰); δ¹³C₄ and δ¹³C₃ are the δ¹³C values of C₄ plants and C₃ plant, respectively (‰), and F is the proportion of C₄ plants in a given ecosystem. For δ¹³C₄ and δ¹³C₃, the δ¹³C values of both vary negatively with the mean annual precipitation. The functions can be express as:

δ¹³C₃ = -0.00916x - 22.902 (3)

δ¹³C₄ = -0.0044x - 10.307 (4)

where x is the mean annual precipitation. As for F, it has been demonstrated that there was a linear relation between the C₄/C₃ ratios of soils and annual precipitation (An et al., 2005).

F = 0.00135R - 0.261 (5)

So, equation (1) was got by combining equations (2)-(5). However, when initially applying this function to our isotopic sequence, necessary correction was needed, because our study site is at edge of the region where the above surface sample was taken, the precipitation may be different from that in central and southern Chinese Loess Plateau, thus, we added a constant 0.86 (modern precipitation of ZBT divided by average precipitation of surface sample sites) to modify this function, and obtained a transfer function as following:

R= [SQRT( 7.17×10²+25.7× δ¹³C_TOC)-6.60]/1.29×10^-2 × 0.86 (6)

It should be said that considerable uncertainty remains in this transfer function. It should be kept in mind that the transfer function itself is based on several assumptions, each of which can result in uncertainty (Guiot et al., 2000).

The estimated precipitation is shown in Figure 2. It ranges from 63 to 479 mm from 47 ka to the present, with an average of 312 mm, the maximum value was observed in the last 2.3 ka and a minimum in the deglacial period. Variation of the estimated precipitation shows the same trend as the δ¹³C sequence. Precipitation between 47-26 ka was relatively high and ranged from 181 to 358 mm, with an average of 309 mm. The precipitation between 26 ka and 18 ka ranged from 131 to 418 mm, averaged 338 mm. The lowest precipitation, ranging from 63 to 363 mm, with an average of 194 mm was found between 18 ka to 8 ka. Since ~8 ka, the precipitation is high, with a mean value of 435 mm. Compared with the modern precipitation (401 mm), the average since 8 ka is higher by ~34 mm, the period of 8-18 ka is lower by ~207 mm, the period of 18-26 ka is lower by ~63 mm, and the period between 47-26 ka is lower than the modern by ~92 mm.

There have been several studies that presented quantitative estimates of the past precipitation in the Chinese Loess Plateau, finding that precipitation in the early Holocene was up to 25% higher than at present, and during the last glaciation was lower than the present by ~25% (Maher et al., 1994; Heller et al., 1993). Based on opal phytolith analysis, the precipitation in the Holocene was higher than the modern by ~200 mm, and lower than the modern by ~240 mm during the last glacial maximum (LGM, 26-16 ka) at the Weinan site (Wu et al., 1994). A geochemical transfer function indicates that Holocene precipitation in Weinan was above 700 mm, while it was less than 200 mm in the LGM (Sun et al., 1999). Recently, our research (Yu et al., 2013) found that the precipitation in the LGM decreased by 180-350 mm to the modern, and increased by 100-200 mm to the modern in the Holocene Optimum (HO, 9-5 ka) around our study area, using the sediment analysis and pollen assemblage. All of these results are very close to our results, taking into account the regional precipitation gradient. However, it is difficult to assess which method is the best, because each transfer function has its own drawbacks (Guiot et al., 2000). For instance, the boundary conditions are not always the same between modern conditions and the past, proxies used in the transfer function are not necessarily in equilibrium with the present climate, and the underlying physical processes involved in the proxy data are not direct. Compared with other quantitative methods, the major advantage of the method we used is that the mechanisms linking precipitation and isotopic composition of soil organic matter are relatively clear (Ning et al., 2008). As for all quantitative reconstructions of past climate, further investigation will require comparison of results from a wide variety of methods.

In Figure 3 the record of the ZBT section is compared with other high resolution records in the region influenced by the Asian monsoon circulation, and with Northern Hemisphere (65°N) summer isolation (Berger and Loutre, 1991). This comparison shows that our reconstruction of monsoon precipitation is broadly similar with the δ¹⁸O records of stalagmites and Northern Hemisphere (65°N) summer isolation, although the correlation is not significant, suggesting that the δ¹³C reconstruction is capable of capturing past climatic fluctuations at both the orbital and millennial time scales. However, our record shows some inconsistent with other two records from low latitude, especially before 18 ka, when our record is high and the other two records are low.

The generally good correspondence of precipitation changes with Northern Hemisphere 65°N summer isolation suggests that the precipitation was mainly forced by changes in orbital parameters. Summer insolation variation changed the continent-ocean thermodynamic contrast, and thus enhanced or reduced the summer monsoon intensity. In detail, the humid conditions inferred from the ZBT record in the HO agree with other records in East Asia (Figure 3). For the deglacial period, all records exhibit increased summer monsoon intensity and reduced winter monsoon strength, suggesting a progressive warming. Moreover, some dramatic fluctuations especially the Younger Dryas (YD) cold event can be identified in the ZBT precipitation reconstruction. This clearly supports the conclusion of Liu et al. (2013) that high-latitude temperature changes have forced the wed/dry changes in the monsoon marginal region, and provide more evidences between the high- and low-latitude linkage in Northern Hemisphere.

However, the discrepancies between our result and others should also be noted. At around 18-21 ka, our reconstruction shows that precipitation was relatively high and varied from 131 to 418 mm, with average of 355 mm, while strong winter monsoon activity recorded by other records indicates a dry phase. Inconsistencies can also be observed in the period 10-7 ka, when other records reveal a wet climate but it was dry in our research area. This conflict has been interpreted as the result of enhanced subsidence in drylands north of the monsoon core, together with greater than modern evapotranspiration (Lu et al., 2005; Mason et al., 2009).

In addition, climate simulations also provide some results that help interpret climate changes in East Asia from the LGM and the HO. In LGM, the simulations based on Paleoclimate Modeling Intercomparison Project (PMIP) indicate that the monsoon was strengthened north of 30°N but weakened to the south of that latitude in East Asia, and 14 models chosen in the study consistently simulated a weaker than present East Asian summer monsoon during the LGM, with an average weakening of 25% (Jiang and Lang, 2010). Annual precipitation and evaporation during the LGM were about 25% less than the present in this area (Jiang et al., 2011). Given that the present precipitation of 401 mm, the simulated 25% less yields a value of about 301 mm, which is about 37 mm less than our reconstruction of 338 mm. In the HO, based on the PMIP, the mid-Holocene East Asian monsoon intensity increased by 32% relative to the present level (Jiang et al., 2013). Converted to precipitation, this yields a value of about 529 mm, about 94 mm more than our result of 435 mm during 0-8 ka. These differences reflect the need for additional improvement of the accuracy of both of the data reconstruction and the numerical modeling.

However, the differences between our reconstruction and simulation results are relatively small given the uncertainties of both models and our preliminary results, and quantitatively they are the same in terms of the direction of change at various times, suggesting that our δ¹³C_TOC reconstruction is helpful in understanding the monsoon precipitation behavior in the snady land region.