High-resolution summer precipitation variations in the western Chinese Loess Plateau during the last glacial

We present a summer precipitation reconstruction for the last glacial (LG) on the western edge of the Chinese Loess Plateau (CLP) using a well-dated organic carbon isotopic dataset together with an independent modern process study results. Our results demonstrate that summer precipitation variations in the CLP during the LG were broadly correlated to the intensity of the Asian summer monsoon (ASM) as recorded by stalagmite oxygen isotopes from southern China. During the last deglaciation, the onset of the increase in temperatures at high latitudes in the Northern Hemisphere and decline in the intensity of the East Asia winter monsoon in mid latitudes was earlier than the increase in ASM intensity and our reconstructed summer precipitation in the western CLP. Quantitative reconstruction of a single paleoclimatic factor provides new insights and opportunities for further understanding of the paleoclimatic variations in monsoonal East Asia and their relation to the global climatic system.

4 Figure S1 The negative correlation between MAP and the δ 13 C of modern C 3 plants. (a) averaged values for 15 sites in the loess region of north China (Wang et al., 2003), the total number of data being 367; (b) averaged values of 7 sites from the central CLP (Zheng and Shangguan, 2007), the total number of data being 121; the number of data for each site is shown by Arabic numerals.
Please note the different correlation coefficients.

Figure S2
The detailed results of the negative correlation between MAP and the δ 13 C of different modern C 3 species. (a) 3 modern C 3 species from northwest China (Liu et al., 2005a); (b) 4 modern C 3 species from northwest China (Wang and Han, 2001a). Please note the differences in the linear relations for different species indicated by different colours.
bungeana, Lespedeza sp. and Heteropappus less) in northwest China displays a completely different linear negative correlation with local MAP, the only significant correlation being with Stipa bungeana (Liu et al., 2005a;a in Fig. S2). Similar results 5 come from 4 modern C 3 species (Plantago depressa, Lepidium apetalum, Chenopodium album and Cirsium leo) also in northwest China (Wang and Han, 2001a; b in Fig. S2), only the correlation with Plantago depressa being insignificant. In a specific location near Baiyin City (also close to the Jingyuan loess profile mentioned in this paper) in northwest China, 7 modern C 3 species were sampled during late June and middle July, 1999. The corresponding weather data is shown in Table S1, and the δ 13 C results are shown in Fig. S3. The results clearly indicate the negative responses of δ 13 C in modern C 3 plants to increasing precipitation and the different sensitivities of the δ 13 C in different C 3 species to identical variation in precipitation (Wang and Han, 2001b).

Table S1
Summer weather data for Baiyin city in 1999 (cited from Wang and Han, 2001b; please note the much higher rainfall in July, and the similar temperature and solar radiation values in June, July and August).

June
July August Rainfall (mm) 32.6 154 13.8 Temperature (°C) 20.2 21.2 21.7 Solar radiation (h) 234.6 255 251 Figure S3 Comparative results of the δ 13 C of 7 modern C 3 species sampled in late June and middle July in 1999 near Baiyin City (modified from Wang and Han, 2001b).
7 Figure S4 Variations in δ 13 C TOC in LC, LA, BJ, WN, JX, YS (Gu et al., 2003), HX, XF, LT (Liu et al., 2005b), and JD loess profile (Vidic and Montañez, 2004) since the last glacial. All the results indicate an increase in C 4 relative abundance since the last glacial to the Holocene, with more positive δ 13 C TOC values occurring in Holocene paleosol (S0) layers. (The locations of these profiles are shown in Fig. 1, and the codes for the loess profiles are identical as in Fig. 1).
Northwest CLP, a decrease in the relative abundance of C 4 plants would be expected . Comparison of the loess δ 13 C TOC data in the 3 profiles (Weinan, Lingtai and Huanxian located in the CLP along a transect from southeast to northwest, respectively), clearly demonstrates the decrease of C 4 relative abundance northwestward during both the last glacial and the Holocene, with the δ 13 C TOC data of both the last glacial loess layer and the Holocene paleosol layer becoming 8 increasingly negative towards the northwest (Fig. S5). Consistent with the decreasing trend of C 4 relative abundance in the CLP towards the northwest during both the last glacial and the Holocene, the loess δ 13 C TOC data from the Yuanbao (YB) profile located in the westernmost CLP indicates that the local terrestrial vegetation during the last glacial was dominated by C 3 plants with only a negligible C 4 contribution (Rao et al., 2005;Chen et al., 2006). Similarly, the loess δ 13 C TOC data from the Jingyuan (JY) profile that located in the northwesternmost CLP indicates that the local terrestrial vegetation since the last glacial was dominated by C 3 plants with only a negligible C 4 contribution (Liu et al., 2011, Fig. S6).

Figure S5
Variations in loess δ 13 C TOC in WN, LA (Gu et al., 2003) and HX profile (Liu et al., 2005b) along a spatial gradient. The results indicate that the relative abundance of C 4 plants decreased from southeast to northwest in the CLP during both the Holocene and the last glacial, with δ 13 C TOC data in both Holocene paleosol (S0) and last glacial loess (L1) layers decreasing gradually towards the northwest.
9 Figure S6 Comparison of δ 13 C TOC records from the JY profile (a) located in the northwesternmost CLP (Liu et al., 2011),the YB profiles (b) located in the westernmost CLP (this study), and the LT profile (c) located in the southernmost CLP (Liu et al., 2005b), See Figure 1 for site locations. The age series of the JY profile is a linear interpolation of OSL data from Sun et al., 2010Sun et al., , 2012; the YB profile series is a linear interpolation of OSL data from Wintle, 2006 andLai et al., 2007; and the LT profile series is taken directly from Liu et al., 2005b. On a glacial/interglacial timescale, the δ 13 C TOC data for the Holocene paleosol at the JY site were more negative than for the last glacial loess, which is converse to the results from the other profiles (see Figures S4 and S5). This indicates that the δ 13 C TOC data for the JY do not record variations in C 3 /C 4 relative abundance but, rather, record δ 13 C variations of C 3 plants since the last glacial. During the last glacial, δ 13 C TOC data for JY and YB were more negative in the weakly developed paleosol layer formed during MIS3 than in the loess layers accumulated during MIS2 and MIS4, which is the converse of the results from the LT profile. This indicates that the terrestrial vegetation at the JY and YB sites during the last glacial was dominated by, or composed entirely of, C 3 plants. The overall trend of the loess δ 13 C TOC data is shown in bold light blue arrows for clear comparison.
The loess δ 13 C TOC data from the YB profile during the early Holocene are emphasized by the comparative use of red dots indicating the LT profile data and the apparent C 4 contribution.
Clearly, the results shown above demonstrate that local terrestrial vegetation 10 during the last glacial at the YB site and since the last glacial at the JY site were dominated by C 3 plants with only a negligible C 4 contribution. In other words, loess δ 13 C TOC data of the last glacial at the YB site and since the last glacial at the JY site can be used for paleoprecipitation reconstruction by way of the modern relation between the δ 13 C data of C 3 plants and precipitation.

Part 3, modern summer monsoon limit in East Asia
As shown in Figure S7, the YB and JY profiles are located in the frontier area of the modern Asian summer monsoon.

Figure S7
Locations of the YB and JY loess/paleosol profiles and the Hulu cave in southern China.
The yellow arrow shows the approximate winter monsoon path; the brown arrow shows the westerlies and the blue arrow shows the East Asian and Indian summer monsoon paths. The dashed red line indicates the modern Asian summer monsoon limit (Chen et al., 2008). Clearly, the YB and JY profiles are located in the frontier region of the modern Asian summer monsoon. The original map was generated by ESRI ArcGIS (v9.1); for the source of the original data for this map please refer to Amante and Eakins, 2009. 11 Considering that the intensity of the summer monsoon gradually deceased during the late Holocene, as shown by a recently reported stalagmite oxygen isotopic record from Sanbao Cave (Fig. S8, Dong et al., 2010), it seems that the intensity of the most recent summer monsoon is very close to that during the Younger Dryas event (ca. 12 ka B.P.). However, there remains a significant distance between the modern summer monsoon limit and the location of the JY and YB sites, especially the latter (more than 250km), so it seems likely that the summer monsoon was also the major source of summer precipitation in the JY and YB sites during the last glacial.

Figure S8
Stalagmite oxygen isotopic record for the last ca. 14, 000 years from the Sanbao Cave located in central China (Dong et al., 2010). Different colors represent data from different stalagmite samples. Note the horizontal grey bar, which provides a comparison of the most recent data with that of the YD event.
Part 4, surface soil δ 13 C results from arid central Asia As above mentioned in Supplementary Part 1, the relation between the δ 13 C values of modern C 3 plants and precipitation cannot be used directly for paleoprecipitation reconstruction. Therefore, relations between surface soil δ 13 C values and precipitation from an area with a full vegetation cover dominated by C 3 plants is a better choice as a modern reference for paleoprecipitation reconstruction.
Given that such surface soil δ 13 C values can represent the carbon isotopic signal of the overlying vegetation at an ecosystem level, the influence of different sensitivities of different C 3 plants to variations in precipitation can be largely avoided.
In arid central Asia, the δ 13 C in 196 surface soil samples along a south to north transect were measured and reported (Lee et al., 2005;Feng et al., 2008;Fig. S9).
Owing to the proximity of our profiles (YB and JY) to the surface soil transect ( Fig. 1) and the considerable spatial gradient of the surface soil transect (very representative), we chose the relation between these surface soil δ 13 C values and precipitation as a modern reference for paleoprecipitation reconstruction.
It has been widely recognized that temperature is the most important climatic factor controlling the growth of C 4 plants (Long, 1983;Rao et al., 2012). An investigation of modern plants on Gengga Mountain in southwestern China indicated that almost no C 4 plants have been observed above an altitude of ca. 2100m with a MAT of 9.4°C and a summer temperature of 15.3°C (Li et al., 2009). A similar investigation on Lingshan Mountain near Beijing city in north China demonstrated that almost no C 4 plants have been observed above an altitude of ca. 1800m (Wang et al., 2010). According to the results of Long, C 4 plants are extremely rare in areas with summer temperatures lower than 16°C (Long, 1983). Comparative analyses of surface soil δ 13 C values at a continental scale from eastern China to Australia and the Great Plains of North America, with a MAT of ca. 12°C, has been found to be the "threshold 13 Figure S9 Relations between surface soil δ 13 C TOC data from arid central Asia and precipitation and temperature. (a) δ 13 C TOC in 196 modern surface soils plotted against MAP (Feng et al., 2008); (b) δ 13 C TOC in 196 modern surface soils plotted against MAT (Feng et al., 2008). A line is not shown because a linear relation is not apparent; (c) averaged δ 13 C TOC values of 19 sites (solid blue dots) plotted against summer precipitation (Lee et al., 2005); the linear negative correlation is represented by the purple line, and the black lines represent the 95% confidence interval of the linear correlation; (d) averaged δ 13 C TOC values of 19 sites plotted against summer temperature (Lee et al., 2005). Horizontal and vertical light blue bars in (c) and (d) represent the 1 σ standard deviations of the averaged surface soil δ 13 C TOC values and corresponding averaged climatic data (summer precipitation and temperature) respectively. See Fig. 1 for distribution of the surface soils.
The quantitative relation between summer precipitation and averaged surface soil δ 13 C TOC data in (c) has been used as the modern reference for summer precipitation reconstruction in this work, with the assumption that the loess δ 13 C TOC data were systematically 1‰ more positive after longterm decomposition, and the 95% confidence interval of the linear correlation in (c) has been used to estimate the uncertainties of the summer precipitation reconstruction.
temperature" for the growth of C 4 plants (Rao et al., 2010 Although we cannot completely preclude a contribution from C 4 plants in the δ 13 C dataset from arid central Asia, and considering the significant influence of temperature on the growth of C 4 plants and the distribution of the δ 13 C dataset along MAT in arid central Asia (b in Fig. S9), we firmly believe that the relative abundance of C 4 plants in the local biomass, or the contribution of C 4 plants to the surface soil δ 13 C data, are extremely limited, and therefore negligible.
It should also be noted that our survey in the Linxia Basin in which the YB site is located (MAT of 6.8°C) demonstrates that there are only few C 4 species mainly around croplands and residential areas. Therefore, considering the lower temperature during the last glacial, it is reasonable to conclude that the contribution by C 4 plants to the YB loess δ 13 C during the last glacial was negligible.
There seems to be an exponential relation between surface soil δ 13 C values and 15 MAT in arid central Asia (b in Fig. S9). However, the positive surface soil δ 13 C values around MAT of ca. 2.5°C mainly range from -20‰ to -24‰. The corresponding MAP for these positive δ 13 C values (-20‰ to -24‰) mainly range from 100mm to 300mm (a in Fig. S9), falling into the climatic range of desert and Gobi in arid central Asia, and located in the middle of the studied transect ( Fig. 1, Lee et al., 2005;Feng et al., 2008). Therefore, the exponential relation may just reflect the significant effect of precipitation on the plant δ 13 C values from another aspect, rather than the significant effect of temperature. More importantly, the influence of temperature on plant δ 13 C values is very complicated. Generally speaking, temperature can affect plant δ 13 C via the effect of the stomatal conductance of the leaves and the bioactivity of the photosynthetic enzymes. Temperatures that are either too low or too high will restrain the stomatal conductance of the leaves and the bioactivity of the photosynthetic enzymes. Normally, the "transform temperature" lies between 20°C and 30°C or higher. Apparently, even considering that summer temperature is higher than MAT, the observed MAT of ca. 2.5°C (the corresponding summer temperature is 8°C to 10°C, Lee et al., 2005) is too low to be treated as the "transform temperature". Therefore, the exponential relation between MAT and the surface soil δ 13 C values from arid central Asia (b in Fig. S9) is just presentational, not logical.  (Lee et al., 2005). Therefore, the relation between averaged surface soil δ 13 C values close to the 19 weather stations and corresponding summer (May to September) precipitation recorded by the weather stations is much more stable with constrained uncertainties (Lee et al., 2005;c in Fig. S9). In this paper, we select the quantitative relation between the averaged surface soil δ 13 C values and summer precipitation (c in Fig. S9) as the modern reference for summer precipitation reconstruction with the assumption that loess δ 13 C TOC data were systematically 1‰ more positive after long-term decomposition. Based on the original data including the averaged δ 13 C values and summer precipitation amount from the 19 weather stations, we calculated the 95% confidence interval (CI) of the linear relation (c in Fig. S9). Also, the quantitative estimation of the summer precipitation with a prediction interval (PI) at the 95% level was calculated using this relation and our δ 13 C data during the last glacial derived from the YB site and since the last glacial derived from the JY site. The linear fitting, CI calculation, summer precipitation estimation, and PI calculation were all performed using the statistical package R, High variability also existed in the loess δ 13 C TOC data from the YB and JY profiles and the corresponding reconstructed summer precipitation, especially in the highresolution data of the YB profile during the last glacial (Fig. S10), indicating the high The most recent reconstructed summer precipitation for the JY site is ca. 150mm with an uncertainty of ca. 70mm, close to the modern averaged summer precipitation of the Jingyuan station of ca. 190mm (1961~1990). The Holocene loess δ 13 C TOC data 19 in the YB profile apparently contain a C 4 signal as shown in Fig. S6, especially during the early Holocene, that's why we abandoned the summer precipitation reconstruction of the entire Holocene at the YB site. The loess δ 13 C TOC data in the topmost 2 samples are -28‰ and -27.5‰, respectively (Table S2). If these two data are used to calculate the summer precipitation, results of ca. 415mm and 386mm, respectively, are obtained also with uncertainties of ca. 70mm , which is very close to the modern averaged summer precipitation at the Linxia station of ca. 400mm (1961~1990; Table S3).
Comparison of the most recent calculated summer precipitation and the corresponding modern averaged value, the relatively greater difference in the JY profile apparently resulted from its relatively positive loess δ 13 C TOC data which fall into the most positive end of the surface soil δ 13 C TOC values in arid central Asia (Fig. S9), thus raising uncertainty. All this evidence validates our summer precipitation reconstruction method.
Detailed comparison of the reconstructed summer precipitation of the YB and JY profiles with the age sequences is impossible, due to the data resolution in the JY profile being too low. However, both datasets show higher summer precipitation from 30ka to 60ka (marine isotope stage 3, MIS3), followed by a decrease towards MIS2 ( Fig. S11).

Part 6, refined age-model of the YB profile
The relatively huge errors of the OSL data from the YB profile, especially those during the last glacial (as shown in Fig. 2 in the main text) preclude the comparison of 20 high-resolution YB records with other records. For a long time (e.g. Porter and An, 1995), grain size data in the Chinese loess have been related to temperature variations at high latitudes in the northern hemisphere by way of variations in the intensity of the East Asian winter monsoon and the vigor of the westerly winds. During cold phases such as the Heinrich events, the enhanced winter monsoon and the northern hemisphere westerlies transported more coarse dust grains to the CLP, especially to its western sector because of its proximity to the deserts (Chen et al., 1997;Sun et al., 2010Sun et al., , 2012. This relation allows us to transfer the ice core ages from high latitudes in the northern hemisphere to the loess profile in the western CLP. The grain size data in the YB profile (> 40μm, %, mainly reflecting the intensity of the winter monsoon) have been compared with the oxygen isotopic record of NGRIP (NGRIP, 2004; indicating the temperature variations in the high latitudes of the northern hemisphere) as a means of selecting appropriate age control points, mainly dependent on the cold events ( Fig. S12). After that, the transferred age series in the YB profile was obtained by linear interpolation of the selected age control points. The results indicate that, between ca. 10 and 20ka, the OSL data were generally consistent with the transferred NGRIP ages; between ca. 20 and 60ka, it seems that the OSL data are systematically younger than the transferred NGRIP ages with an average offset of ca. 4~5ka (Fig. S13). This comparison further confirms the validity of transferring the NGRIP ice core ages to the YB loess profile based on comparison of the loess grain size and the NGRIP oxygen isotopic data (NGRIP, 21 Figure S12 Comparison of high-resolution grain size data (blue lines) from the YB profile plotted against depth and oxygen isotopic data (red lines) from the NGRIP ice core (NGRIP, 2004) plotted against the age series of GICC05 (Svensson et al., 2008). Based on the comparison, 9 age control points (represented by the vertical grey bars) were selected in order to match the NGRIP age series to the YB profile by interpolation. Arabic numerals indicate interstadial events. YD is the Younger Dryas event and H1-H6 marks the Heinrich events. Please note the reversed grain size scale.
2004). Although the transferred age series is not accurate, and considering the widely accepted control mechanism of Chinese loess grain size, we chose the transferred age series from the NGRIP ice core as the final age model of the YB profile for comparing the YB records with others ( Fig. 3 in main text).
22 Figure S13 Comparison of the OSL data (solid red and purple dots with error bars, Lai and Wintle, 2006;Lai et al., 2007) and the selected 9 age control points (solid blue triangles) transferred from NGRIP plotted against depth of the YB profile. The purple line is based on part of the OSL data represented by solid purple dots to show the approximately systematic difference between OSL data and NGRIP data between 20ka and 60ka.

Part 7, original data of the YB profile
For an intuitive presentation, original data and reconstructed summer precipitation from the YB profile are plotted against depth and age series as shown in Figs.S14 and S15, respectively. Correspondingly, all relevant data from the YB profile are shown in Table S2.
23 Figure S14 The original data from the YB profile plotted against depth. Red series are grain size data (>40μm, %); blue series are magnetic susceptibility (SI units); purple series are loess  13 C TOC data (‰, VPDB); black dots with error bars are the OSL dating results (same as in Fig. S13); calculated summer precipitation is shown by solid red dots with uncertainties represented by light blue vertical bars.

Figure S15
The original data from the YB profile plotted against the age series transferred from the NGRIP ice core. Red series are grain size data (>40μm, %); blue series are magnetic susceptibility (SI units); purple series are loess  13 C TOC data (‰, VPDB); calculated summer precipitation is shown by solid red dots with uncertainties represented by light blue vertical bars.

Part 8, climate data from Jingyuan and Linxia (1961~1990)
The climate data from Jingyuan and Linxia (close to the YB profile) stations, including the monthly precipitation and temperature data from 1961 t0 1990, are shown in Table S3 as following: