Tree pollen includes Pinus, Quercus, Cupressus, Ulmus, Betula, and others. Mesophilous herb pollen includes Compositate, Artemisia, Gramineae, Urticaceae, Polygonum, Liliaceae, Leguminosae, and others. Xeromorphic herb pollen includes Nitraria, Zygophyllaceae, Ephedra, Chenopodiaceae and others. Wetland plant pollen includes Typhaceae, Cyperaceae, Plantago and Sparganiaceae. Shrub pollen includes Elaeagnaceae and Rosaceae (Figure 2).

All modern pollen samples contain relatively high percentages of Haploxylon type Pinus pollen and Diploxylon type Pinus pollen, ranging between ~10% and ~40% (Figure 3), although, except for a few pine trees planted in this area, there are no pine forests or pine-associated forest-steppes occurring around the lake or in the watershed. The Pinus pollen has most likley probably been transported by wind from the pine-dominated forest-steppes. It is apparent that the Pinus pollen is over-represented in the pollen spectra, confirming the widely-reported over-representation of Pinus pollen in the semi-arid and arid regions of China and Mongolia [18] .

Despite the dominance of grasses in the steppes and desert steppes in the Ordos Plateau, Gramineae pollen percentages are rather low, being below 10% in seven of the ten surface samples, confirming the under-representation of Gramineae pollen previously reported for the Mongolian steppes [18] . Chenopodiaceae and Artemisia pollen percentages are relatively high. Although there are only scattered Quercus trees but no Quercus forest in the Ordos Plateau, Quercus accounts for most of the broadleaf tree pollen.

The basal section (530 - 435 cm) of the core is slightly dark sand. It is overlain by lacustrine sediments consisting of a dark clay (435 - 350 cm), a slightly dark silty-clay (350 - 300 cm), la Carbonate-enriched and clayed varves (300 - 230 cm), and a reddish-brown silty-clay (230 - 80 cm). The top 80 cm is grayish sand (Figure 3).

Fifteen AMS ¹⁴C dates were obtained (Table 1). All material is bulk. The basal sand (435 - 530 cm) was dated at three depths with ¹⁴C ages from 13,910 ± 70 to

27,850 ± 280 ¹⁴C yr BP. The large age difference between 487 cm (13,910 ± 70 ¹⁴C yr BP) and 506 cm (27,850 ± 280 ¹⁴C yr BP) perhaps results from a depositional hiatus or an erosion event during the last glacial maximum. The age reversal between 506 cm (27,850 ± 280 ¹⁴C yr BP) and 527 cm (24,290 ± 537 ¹⁴C yr BP) may be a result of the upward movement of sediments, either by sand reactivation or bioturbation. Three slight age reversals occur at 133 cm, 327 cm, and 385 cm.

A carbon reservoir age could not be directly calculated. Because the top 80 cm of the core consists of aeolian transported sand, the common linear regression method could not be used, while the complete lack of terrestrial material eliminated the possibility of employing the comparison method. Therefore, the AMS dates were corrected by applying previously published regional data, which show that the radiocarbon reservoir in the arid-semiarid western China is generally high, though variable [19] [20] [21] , with values ranging from 1000 - 2000 yr in the Ordos plateau [19] and values ranging from 2590 to 4340 yr in the Qaidam Basin of the Tibetan Plateau [22] . Closer to the site, we recently obtained a radiocarbon reservoir of 1950 years from Qigai Nuur, only 1 km from Bahar Nuur. This is consistent with reservoir values of 2000 yr from Daihai [23] about 200 km to the north, and 1935 yr from Lake Gouchi about 100 km to the west [24] . Accodingly, assuming a reservoir effect of 1950 yr for the Bahar Nuur, we subtracted 1950 yr from all the AMS dates obtained from the core. All corrected AMS dates were converted into calibrated ages using the INTCAL program [25] , and a linear regression, best-fit calibrated age-depth model was constructed after eliminating the three reversed age data. A linear relationship (R² = 0.9521) exists among the 9 AMS dates within the lacustrine section of the core (435 - 80 cm), suggesting that the age model adequately captures the chronology. The linear relationship is also supported by the relatively uniform particle size distribution within the section of lacustrine sediments. The average sediment rate for this section is 0.081 cm/yr.

Because pollen is absent or rare in the sand intervals at the bottom and at the top of the core, fossil pollen data are only presented for the lacustrine section from 435 - 80 cm. Seven pollen assemblage zones, including the top and bottom non- polleniferous zones, are recognized for the core, based on visual inspection and CONISS results (Figure 4).

Zone 1 (530 - 435 cm; before 8.6 ka BP), corresponding with the basal sand, contains little or no pollen. Here the TOC is less than 2% and the δ¹³C of organic matter is relatively low (mostly −24‰ to −26‰). Carbonate values are <8%.

Zone 2 (435 - 350 cm; 8.6 - 8.2 ka BP), which roughly corresponds with the clay unit, is characterized by low pollen percentages for Pinus pollen and high percentages of Gramineae. Pinus pollen are <80%. Quercus is <8%. Chenopodiaceae is high with average value 5%. Ephdra fluctuate within 1% - 5%. Gramineae pollen percentage peaks in this zone. Pollen concentration is less than 20,000

grains/g. TOC increases abruptly to 4% - 6% and remains high throughout the lacustrine section. Mean grain size is uniformly low (<50 μm) in this clay unit, whereas δ¹³C_org is generally high (−22‰ to −24‰). Carbonate increases sharply to 20% - 30%.

Zone 3 (350 - 300 cm; 8.2 - 7.8 ka BP), corresponding with the silty-clay unit, is marked by a distinct increase in Quercus pollen at the base. The pollen percentages of Ephedra remain high, while the pollen percentage of Pinus increases slightly. Quercus fluctuate within 2% - 10%. Chenopodiaceae is around 3% - 4%. Ephdra is about 2% - 3%, Gramineae pollen percentage <10%. Pollen concentration fluctuates within 8000 - 22,000 grains/g. Mean grain size increases slightly and Carbonate remains high.

Zone 4 (350 - 220 cm; 7.8 - 6.3 ka BP), which roughly corresponds with the laminated calcareous clay, is marked by stable pollen percentages. Pinus pollen values are around 75% - 80%. Quercus fluctuate within 2% - 10%. Chenopodiaceae and Ephdra reach their highest values, while Gramineae values decline slightly. Pollen concentration fluctuates within 8000 - 18,000 grains/g. Mean grain size is very small and carbonate is high. Charcoal concentration is around 4000 - 6000 grains/g. δ¹³C_org values fluctuate in this zone but on the whole decline slightly.

Zone 5 (220 - 100 cm; 6.3 - 4.4 ka BP), corresponding with the silty-clay, is characterized by increasing pollen percentages of Pinus, while Quercus pollen reaches its maximum abundance. Rosaceae increases, reaching 2%. The steppe component (Chenopodiaceae, Ephdra, Atermisia and Gramineae) pollen percentage decline while wetland taxa (i.e. Spargania) increase. Pollen concentration values are highest in this zone. Charcoal concentration fluctuates within 4000 - 10,000 grains/g. Mean grain size is slightly larger than in the previous zone but is variable. δ¹³C_org drops sharply near the top of this zone while TOC increases. Carbonate fluctuates in this zone but remains relatively high.

Zone 6 (100 - 80 cm; 4.4 - 4.3 ka BP) roughly corresponds with the silty-clay. Pinus pollen increases to maximum percentages while almost all other pollen taxa abruptly decrease or disappear, with only Gramineae remaining stable. Pollen concentrations also decline sharply in this zone. Mean grain size increases dramatically while TOC decreases sharply. Charcoal concentration declines. δ¹³C_org recovers from the crash but remains low and Carbonate is low.

Zone 7 (80 - 0 cm; after 4.3 ka BP), corresponds with the top eolian sand unit. Pollen grains were not plotted due to their scarcity.

The non-polleniferous sand deposited before 8.6 ka BP suggests that no lake existed and that the vegetation was locally very sparse before 8.6 ka BP. The climate was probably dry, or the dry lakebed was not conducive to pollen deposition and preservation. Although an early Holocene climate optimum was documented for Midiwan [6] during 11.5 - 8.3 ka BP following a humid phase from 16.3 and 8.87 ka BP in Yanhaizhi [8] , a number of studies have shown that along the northern margin of the East Asian monsoon precipitation was low in the early Holocene, increasing after 8 ka BP [4] [26] .

At ~8.6 ka BP, the core lithology changes from sand to clay, and pollen becomes much more abundant. Charcoal concentrations also increase significantly. The abrupt onset of lacustrine sedimentation suggests that the lake formed rather quickly. This indicates a rapid climatic change towards wetter conditions, and denser vegetation. This is supported by dramatic increases in TOC and a sharp decrease in mean grain size.

The presence of a lake from ~8.6 to 4.3 ka BP, as indicated by contonious lacustrine sedimentation, implies that the mid-Holocene was a more humid than the present. Between ~8.6 and ~7.8 ka BP (Zone 2 and Zone 3), during the early phase of this wet period the vegetation around Bahar Nuur was a grass dominated steppe with some xerophytic herbs and shrubs such as Chenopodiaceae and Ephedra.

From ~7.8 to ~6.3 ka BP (Zone 4), although steppe vegetation still dominates, many other types of plants appear and persist. This sediment consists of alternating varves, composed of black (organic matter-rich) and white (carbonate) couplets. The formation of varves requires strikingly different seasonal conditions and lake depth deep enough to permit stable bottom water [27] . The white couplet suggests that the summer temperatures were high enough to drive evaporation-generated carbonate deposition. The fine particle size, indicating high lake levels, suggests that warm and humid conditions prevailed during this period.

Quercus (Quercus liaotungensis) populations reach their greatest abundance between ~6.3 and ~4.4 ka BP (Zone 5), indicating humid conditions. It is more humid than zone 4. The slight but sustained increase in Rosaceae pollen indicats that shrubs such as Caragana also increased in abundance. The increase in trees and shrubs occurred at the expense of the xerophytic plants, as indicated by a concomitant decrease in the pollen percentages of Gramineae, Chenopodiaceae, and Ephedra. The general decrease in δ¹³C_org values from zone 2 to zone 5 also supports this scenario, as C3 plants (including most trees) have much lower (i.e., more negative) δ¹³C values than C4 plants (including many desert grasses and herbs). As a result, the vegetation reached its highest density during this period, as evidenced by maximum pollen concentration values for this zone. High charcoal concentrations in this zone are also consistent with maximum biomass, which provide more fuels for fires. The increase in wetland plants such as Spargania also indicate humid environment. The pollen evidence indicates that the Holocene Optimum occurred in this area during Zone5 (~6.3 to ~4.4 ka BP).

Quercus pollen declined abruptly after ~4.4 ka BP, as did charcoal concentrations and pollen for most other taxa. These changes signify a return to harsher climate conditions, leading to a more open vegetative landscape. The change from lacustrine sedimentation to deposition of eolian sand after ~4.3 ka BP indicates that the lake either dried up or became ephemeral during the late Holocene, thereby ending the pollen record. Most records from area dominated by the Asian monsoon show markedly lower effective moisture in late Holocene due to the weak summer monsoon [28] .

To put the reconstructed vegetation history of Bahar Nuur in a regional perspective, we reviewed other records in or near the Ordos Plateau. At the Dadiwan and Sujiawan sections in the western part of the Chinese Loess Plateau, deciduous forest of high density and diversity dominated by oaks and birch that existed from 6.56 to 5.79 ka ¹⁴C yr BP were replaced by pine-dominated forest-steppe during 5.79 - 4.95 ka BP (¹⁴C age), then degraded into elm-dominated sparse forest-steppe during 4.954 ka BP (¹⁴C age). The vegetation and climate conditions deteriorated after 4 ka BP (¹⁴C age), as forest was replaced by steppe. Thus, the period of maximum warmth and humidity prevailed between 6.56 and 4.95 ka BP (¹⁴C age) in the western Loess Plateau [29] . However, the pollen- based proxy records from Midiwan [5] [6] suggest that a desert-grassland interval occurred between 7.5 and 4.5 ka (¹⁴C age) with a relatively warm climate in Ordos Plateau. Corresponding lithology is silty peat with a large amount of plant residues and freshwater fossil mollusks [6] which indicates humid environment. Furthermore, organic carbon content is relatively high compared with next zone. Although pollen concentration is low, Chenopodiaceae and Ephedra pollens are much less than Artemsia and Composite pollens. At the TYG section to the southeast of Bahar Nuur, a warm and humid Holocene Optimum occurred during between ~7.7 and ~5.6 ka BP, as inferred from dune building as well as pollen assemblages [30] . Our data from Bahar Nuur show lush mid-Holocene vegetation in the central Ordos Plateau, with the most mesic conditions occurring around 6.3 ~ 4.4 ka BP. To the west, records from Lake Yanhaizi suggest that the mid-Holocene (8.0 ~ 4.3 ka BP) was arid in Inner Mongolia, with low TOC and high sand sediment [9] . Further to the northeast, pollen records and pollen- based paleoclimatic reconstruction from Daihai in Inner Mongolia suggest that the mean July temperature was 2˚C ~ 3˚C higher and the precipitation was 100 ~ 180 mm higher than present from ~7.9 ~ 4.45 ka BP [26] . At Diajiaohaizi in Daiqingshan, pollen records suggest a generally warm and humid mid-Holo- cene. Temperature was 1˚C - 2˚C higher than today and precipitation was 150 - 200 mm more than today during 6.9 ~ 6.3 ka BP [31] .

To date, only the Yanhaizi site clearly documents a dry mid-Holocene, contra- dicting records from most other sites in the Ordos Plateau and Chinese Loess Plateau. Since lakes Yanhaizi, Barhar Nuur and Diahai are located in the northwestern margin of Asian monsoon, precipitation levels in this area are closely related to variations in the strength of Asian summer monsoon. We compare our records with Northern Hemisphere insolation and the δ¹⁸O record from Dongge cave, which is an excellent proxy for Asian monsoon strength.

The Asian monsoon is strong before ~4.4 ka BP and weak after ~4 ka BP as indicated by the Dongge cave δ¹⁸O proxy record, corresponding to lush vegetation from 6.3 - 4.4 ka BP and sparse vegetation after ~4.2 ka BP at both Barhar Nuur and Daihai. Corresponding to the dry event that occurred at ~4.2 ka BP, vegetation deteriorated sharply in both Barhar Nuur and Daihai while a weak Asian monsoon event occurred in Dongge cave. This severe climatic deterioration was synchronous with a climatic event during the late third millennium BC identified in many Northern Hemisphere sites [32] , marking the ending of Holocene Optimum. Wu and Liu [32] suggest that this climatic change may have played a very important role in the collapse of Neolithic cultures around the Central Plain of China. A humid mid-Holocene can be inferred from Figure 5, as the Asian monsoon was relatively strong, and broadleaves tree pollen and pollen concentration are high in both Barhar Nuur and Daihai during 6.3 ~ 4.4 ka BP, reflecting denser vegetation and a humid mid-Holocene. This humid mid-Holocene is also echoed in Qinghai Lake which is also located in north margin of Asian monsoon. According to multi-proxy analysis, Shen presents a warm and wet period between 14.1 ka BP and 4.5 ka BP in Qinghai Lake area. After that, the climate gradually became colder and drier [20] .

It has been suggested that variation in the strength of the East Asian monsoon is controlled by changes in summer solar radiation in the Northern Hemisphere on glacial-interglacial timescales [33] . Holocene solar insolation peaked at ~10 ka BP in northern semi-sphere [34] . However, the existence of substantial ice

masses in the northern high latitudes and remnant ice sheets may have hampered the Asian continent from warming up during the early Holocene, resulting in a cold, dry early Holocene in the Ordos Plateau. Oceans need much more energy to warm up than the land. Tropical Holocene SST increased steadily from ~10 ka to ~6 ka BP [36] . According to the δ¹⁸O proxy record from Dongge cave, the Asian monsoon which was relatively strong until about 4.4 ka BP, would have kept monsoon precipitation levels high during mid-Holocene, particularly given the enhanced evaporation on west Pasific which effectively injected more water vapor into the East Asian monsoon system during ~10 ka to ~6 ka BP [3] . The warmer and wetter climate associated with the Holocene Climatic Optimum lead to denser vegetation cover, which might have further promoted the Optimum through the feedback mechanism related to the air-vegetation-soil coupling under the denser vegetated wetter conditions [3] . It is plausible that the humid mid-Holocene is a lagged response to Asian monsoon and peak summer insolation in the Northern Hemisphere. When summer insolation largely decreased after 6 ka BP and Asian monsoon was weak in late Holocene, it results in a dry late Holocene in the Ordos Plateau.

The dry mid-Holocene record from Yanhaizi Lake is inconsistent with the wet mid-Holocene extensively documented in other regional sites. The effective soil moisture during the mid-Holocene was probably low in Yanhaizi area where the increase in evaporation resulting from rising temperatures might have exceeded the overall increase in precipitation, leading to an increase in climatic aridity and a decrease in effective soil moisture. However, we find no evidence for a dry mid-Holocene in the Ordos Plateau. More work is required to understand these regional differences.