In summer 2010, we obtained a 1105 cm long sediment core from the deepest part of Wuxu Lake (30 m depth) using a UWITEC piston corer. The core was sub-sampled at 1 cm contiguous intervals and stored at 4 ∘C prior to analysis. The chronology is based on accelerator mass spectrometry (AMS) 14C dates from terrestrial plant macrofossils extracted from the sediment samples. The analyses were made by Beta Analytic Inc. in Miami, USA and the Rafter Radiocarbon Laboratory in the Institute of Geological and Nuclear Sciences, New Zealand. All of the 18 AMS14C dates obtained were calibrated to calendar years before present (0 BP = 1950 AD) using the program Calib 7.1 and the IntCal13 calibration data set (Reimer et al., 2013).

Samples for pollen analysis were determined at 4 cm intervals and treated using standard laboratory methods (Fægri et al., 1989), including treatment with HCl and HF to remove carbonate and silicate, boiling in KOH to remove humic acid, sieving with 10 and 120 µm mesh clothes to remove the fine and coarse fractions, respectively, and mounting in silicone oil. Prior to these treatments, tablets containing a known quantity of Lycopodium spores were added to each sample in order to determine the pollen concentration. At least 500 terrestrial pollen grains per sample were counted. The percentage for each pollen type was calculated based on the sum of total terrestrial pollen; pollen and spores from aquatic plants and ferns were excluded from the calculation.

The pollen diagram was divided into biostratigraphic zones based on constrained incremental sum of squares (CONISS) using the Tilia program (Grimm, 1987). CONISS uses an algorithm based on stratigraphically constrained chord-distance clustering and square-root transformation of the pollen percentage data. Only pollen taxa with a representation > 1 % in at least two samples were included in the zonation.

In order to identify and visualize the main directions of vegetation change, 31 terrestrial pollen types with a representation > 1 % in at least two samples were included in an ordination analysis. Pinus pollen is considered to be transported from the lowest-altitude vegetation zone in the region, or from long-distance sources. Its percentage values are the highest of all of the taxa recorded and they do not exhibit any obvious change; therefore, its weighting was set to 0.1 in the numerical analysis (Xiao et al., 2014a). Detrended correspondence analysis (DCA) yielded gradients of 1.03 standard deviations for the pollen data set, indicating that linear-based methods such as principal component analysis (PCA) are appropriate for the data set. The PCA analysis was applied to the square-root-transformed pollen data for inter-species correlations. The DCA and PCA analyses were performed using the CANOCO program 4.5 (ter Braak and Šmilauer, 2002).

The results of AMS 14C radiocarbon dating of the Wuxu Lake plant remains are shown in Table 1. The results indicate a roughly linear age-versus-depth relationship and therefore that the sediment accumulation rate was relatively constant. A Bayesian model, taking the sediment accumulation rates into account (Blaauw and Andres Christen, 2011), was used to construct the age–depth model (Fig. 2). The model was determined using the default settings for lake sediments at 10 cm intervals implemented using the statistical software package R (R Development Core Team, 2013). The basal age is about 12.3 cal ka BP, yielding an average sediment accumulation rate of 89.5 cm ka-1, and thus the average temporal sampling resolution is about 45 years for the pollen record.

A total of 214 pollen and spore types were identified, including 118 arboreal taxa, 40 herbaceous taxa and 20 fern taxa. The entire pollen record is dominated by arboreal taxa, including Pinus, sclerophyllous Quercus, Picea/Abies and Betula, with contributions of Alnus, Tsuga, Lithocarpus/Castanea, Cupressaceae, deciduous Quercus and Ericaceae. The average percentage of the main herbaceous taxa, including Artemisia, Gramineae, Rosaceae, Ranunculaceae, Thalictrum, Labiatae, Gesneriaceae and Cyperaceae, is 18.4 %. The pollen spectra can be divided into five assemblage zones according to the changes in terrestrial pollen percentages (Fig. 3).

In Zone I (12.3–11.3 cal ka BP), arboreal taxa account for more than 70 % of total terrestrial pollen, among which Pinus, sclerophyllous Quercus and Betula predominate. Other common taxa include deciduous Quercus, and Picea/Abies. The zone is also characterized by the high abundance of herbaceous taxa, including Artemisia, Cyperaceae, Gramineae and Thalictrum, which all exhibit highest percentages throughout the entire record. Carpinus and Picea/Abies maintain at high abundances within the zone, while Betula exhibits a generally increasing trend.

A notable feature of Zone II (11.3–10.4 cal ka BP) is the abrupt decrease in herbaceous taxa to the benefit of arboreal taxa. Artemisia and Cyperaceae fall from 10 to 5 %, and Gramineae and Thalictrum from 5 to 2 %, respectively. Betula reaches its maximum (generally over 20 %) for the entire record. Pinus, Picea/Abies and Carpinus exhibit similar percentages as in zone I.

The third zone (Zone III, 10.4–4.9 cal ka BP) is characterized by relative high arboreal pollen percentages of the entire record and is divided into three sub-zones.

Sub-zone III-1 (10.4–8.2 cal ka BP): The percentages of total arboreal and herbaceous pollen are relatively constant; however, Tsuga begins to be continuously represented in the pollen spectra. Shrub taxa such as Actinidiaceae and Rubus increase significantly, while Rosaceae, Potentilla, Gesneriaceae, Labiatae and Hypericum increase slightly. Betula, Thalictrum and Cyperaceae decrease gradually.

Sub-zone III-2 (8.2–6.6 cal ka BP): Herbaceous taxa increase compared to the previous sub-zone, generally resulting from increases in Artemisia, Thalictrum, Ranunculaceae and Cyperaceae. The representation of Carpinus and deciduous Quercus are similar to the previous sub-zone. Betula is gradually replaced by sclerophyllous Quercus, which is the dominant arboreal taxon. Picea/Abies decreases slightly from 5 to 2 %, while Tsuga and Taxodiaceae/Cupressaceae exhibit a minor increase.

Sub-zone III-3 (6.6–4.9 cal ka BP): Sclerophyllous Quercus increases slightly at the expanse of Betula, Taxodiaceae/Cupressaceae and Picea/Abies. Actinidiaceae and Rubus return to relatively high values. The percentages of total arboreal pollen increases slightly compared to the previous sub-zone.

The contribution of herbaceous taxa in Zone IV (4.9–2.6 cal ka BP) increases up to 30 %, as the result of higher percentages of Artemisia, Cyperaceae, Gramineae. Arboreal taxa decrease to around 70 %, mainly due to the reduced Betula and deciduous Quercus. Tsuga percentages are the highest in the entire record, and Ericaceae and Hippophae increase significantly. There is a slight increase in the representation of Picea/Abies, Alnus and Carpinus.

Overall, the pollen spectra in Zone V (after 2.6 cal ka BP) are similar to those of Zone IV, but with a slightly increased representation of arboreal taxa. Betula continues to decrease, Carpinus and Tsuga decrease slightly, and sclerophyllous and deciduous Quercus increase slightly, to up to 20 and 5 %, respectively. But Rosaceae, Potentilla, Gesneriaceae, Labiatae and Hypericum increase slightly, while Artemisia, Cyperaceae, Gramineae and Ranunculaceae decrease slightly. It is noteworthy that Sanguisorba increases significantly in this zone.

The PCA analysis, based on 31 terrestrial pollen taxa from 276 samples, indicates that the first two axes capture 45.8 % of the total variance, with the first PCA component capturing over 33.7 % (Fig. 4a). Three assemblages can be distinguished: alpine shrubs and meadows characterized by Cyperaceae, Artemisia, Polygonum, Thalictrum, Ranunculaceae, Ericaceae, Hippophae and Salix (in the top left quadrant); cool–cold mixed forest characterized by Abies/Picea, Betula, Carpinus and deciduous Quercus (in the top right quadrant); and temperate mixed forest characterized by sclerophyllous Quercus, Tsuga, Alnus, Lithocarpus/Castanopsis, Rubus and Actinidiaceae (in the bottom left quadrant). The ordination of pollen taxa along the first PCA axis apparently reflects a transition from warm to cold winter temperature, since cold-tolerant taxa such as Abies/Picea, Betula and other deciduous broadleaved taxa are located on the positive side, while Tsuga, which is sensitive to winter temperature and annual temperature range, is on the negative side (An et al., 2011; Li et al., 2015). The arrangement of the pollen taxa along the second axis separates the major alpine shrub and meadow taxa from the forest taxa, reflecting the degree of openness of the vegetation communities, and can be interpreted as representing a change from dry to more humid conditions. The PCA separates the samples into approximately five groups (Fig. 4b), which generally correspond to the defined pollen zones of the sequence. Samples of zones I, II and III have moderate to high positive scores on the first axis, while samples of zone IV and V have negative scores. Samples of zones I, II and V have high scores on the second axis, while samples of zone III and V have low scores. Spectral analysis was conducted on the PCA axis 2 sample scores using the program REDFIT38 (Schulz and Mudelsee, 2002), and revealed periodicities of 110, 106 and 93 years (significant at the > 90 % confidence level; Fig. 5).

Given the close proximity of Wuxu Lake to the modern tree-line, the vegetation around the catchment should react sensitively to climate change. However, lake surface pollen assemblages from the region indicate that large amounts of arboreal pollen, including Pinus, Picea/Abies, Betula, deciduous Quercus, Tsuga and evergreen Quercus from the lower vegetation zones, are introduced into sub-alpine and alpine lakes by anabatic winds (Kramer et al., 2010; Xiao et al., 2011). This makes it difficult to use the pollen data to trace past fluctuations in the tree-line and the vegetation composition of the catchment. Fortunately, these studies also indicate that the lake surface pollen spectra from different vegetation types still closely correlate with the environmental gradients (Kramer et al., 2010; Xiao et al., 2011). In addition, theoretical models of pollen transport show that the proportion of the non-local pollen component deposited in lake sediments increases with increasing lake size (Jackson and Lyford, 1999; Sugita, 1994). Thus, the pollen assemblages from Wuxu Lake, which is relatively small, should reliably reflect the response of the regional vegetation composition to changes in climate. The inferred changes in vegetation and climate are summarized below.

The pollen spectra between 12.3 and 11.3 cal ka BP are characterized by high percentages of Gramineae, Cyperaceae, Artemisia, Polygonum, Thalictrum and Ranunculaceae, with relatively high percentages of Salix, Hippophae and Ericaceae. The high shrub and herbaceous pollen percentages indicate the expansion of alpine shrubs and meadows and open vegetation cover around Wuxu Lake, reflecting weak summer rainfall during the late Younger Dryas (YD). The gradually decreasing herbaceous representation also indicates that the ISM had begun to strengthen. During this period, the surrounding arboreal vegetation was dominated by broadleaved deciduous forest, together with Picea/Abies forest and sclerophyllous Quercus. The dominance of cold-tolerant species in the forest vegetation suggests lower winter temperatures and gradually increasing precipitation in summer.

From 11.3 to 10.4 cal ka BP, the decreases in herbaceous pollen, Salix and Ericaceae, and significant increases in Betula, reflect the replacement of shrubland and meadow by Betula woodland. Pinus, Picea/Abies, Carpinus, deciduous and sclerophyllous Quercus were common. These changes indicate that the vegetation around Wuxu Lake gradually became closed and that the climate became more seasonal, with warmer and wetter summers and cold winters.

The gradual decrease of Betula and Carpinus, and the slight increase of Tsuga, Actinidiaceae, Rubus, Rosaceae, Potentilla, Gesneriaceae, Labiatae and Hypericum until 8.2 cal ka BP, indicate that the vegetation cover was closed. The deciduous broadleaved forest began to retreat and conifer and broadleaved mixed forest with Tsuga appeared within the vertical vegetation belts. Actinidiaceae and Rubus replaced Salix and Ericaceae, forming the understory. These vegetation changes indicate that the climate was very humid in summer and gradually became warmer in winter.

The continuous increase of Tsuga and sclerophyllous Quercus, and the gradual decrease of Betula and Picea/Abies between 8.2 and 6.6 cal ka BP, suggest that mixed forest continued to expand towards Wuxu Lake. These vegetation changes indicate that the summers were rather dry and that there was reduced seasonality of temperature.

From 6.6 to 4.9 cal ka BP, the relatively high representation of sclerophyllous Quercus, increased Actinidiaceae and Rubus and steadily decreasing Betula and Picea/Abies suggest the presence of sclerophyllous Quercus forest with a dense understory gradually replacing deciduous broadleaved forest and Picea/Abies forest. The summers were humid and the winters were warm.

Since 4.9 cal ka BP, the significantly high representation of herbaceous pollen taxa (including Artemisia, Gramineae and Cyperaceae), Hippophae and Ericaceae indicates that the regional vegetation cover became more open compared to the early Holocene. Increased sclerophyllous Quercus, Tsuga and decreased Betula suggest an expansion of Tsuga forest, accompanied by the retreat of Betula forest and a slight expansion of Carpinus forest. The summers were relatively dry and the winters were warmer, compared to the preceding interval.

After 2.6 cal ka BP, Betula forest was further replaced by sclerophyllous Quercus. Tsuga and Alnus remained at a similar level as during the preceding stage. The slight decrease in Artemisia, Cyperaceae, Gramineae and Ranunculaceae indicates that the alpine meadows retreated, whiles increase in Rosaceae, Potentilla, Gesneriaceae, Labiatae and Hypericum suggests that the forest was relatively closed. With humid summers and warm winters the climate was more favorable compared to the preceding interval. The minor increase in Sanguisorba, a grazing indicator (Kramer et al., 2010), suggests the influence of human activity in the region.

The YD is the last millennial-scale cooling event before the beginning of the Holocene in the Northern Hemisphere (Stuiver et al., 1995). In the ISM region, a roughly contemporaneous cold and dry event has been observed in numerous records but in general they are of low resolution. At about 11.3 cal ka BP, the abrupt decrease of PCA 2 axis sample scores may reflect the termination of the YD cold event in the region. A high-resolution stalagmite δ18O record from Moomi Cave in Yemen exhibits a sharp fall at about 11.4 ka BP, marking the onset of the Holocene (Shakun et al., 2007). A pollen and stoma record from Tiancai Lake in southwestern China also suggests that the age of the termination of the YD was about 11.5 cal ka BP (Xiao et al., 2014a). However, other records, such as pollen records from Erhai Lake and Naleng Lake (Kramer et al., 2010; J. Shen et al., 2006), and stable carbon isotope record from Muge Co (Sun et al., 2015) show relatively large uncertainties due to the bulk sediment dated. Thus the timings in the ISM region are generally consistent with the age of the YD termination in the Greenland ice core record (Stuiver et al., 1995).

Several factors may be responsible for the 200-year time lag in the Wuxu Lake record. Firstly, the stage for vegetation succession: e.g., the Abies/Picea form the climax forest in the sub-alpine ecotone after glacier retreat in northwestern Sichuan took about 100 years (Cheng and Luo, 2004). Pollen records from North America and Europe also show that vegetation may lag climate change by 100–200 years (Williams et al., 2002). Secondly, the influence of a centennial-scale event may have hindered our pollen record to distinguish the short event as the YD (Rasmussen et al., 2006; Shakun et al., 2007). Thirdly, errors in the AMS 14C dates could also be responsible for the 200-year time lag (the 95 % confidence limit of the point ranges from 11.0 to 11.6 cal ka BP).

The onset of warm and humid conditions around Wuxu Lake occurred after 10.4 cal ka BP and was maintained until 4.9 cal ka BP, resulting in a prolonged Holocene Optimum except for a relatively cold pulse between 8.2 and 6.6 cal ka BP (Fig. 6a). The δ18O and δD values of rainfall reflect changes in isotopic composition in moisture source areas and by transport distance, and are not correlated with seasonal rainfall amount. However, in the ISM region these isotope ratios are suggested to reflect monsoon intensity over time spans longer than the annual scale (Breitenbach et al., 2010; Contreras-Rosales et al., 2014). High-resolution stalagmite δ18O records from Qunf Cave in southern Oman (Fig. 6b) and Tianmen Cave in southern QTP (Fig. 6e) indicate an interval of strong ISM in the early Holocene, followed by a progressive weakening trend at about 6–7 ka BP (Cai et al., 2012; Fleitmann et al., 2003). Records of carbonate δ18O and plant wax δD from lake and marine sediments, which reflect the isotopic composition of the precipitation, reveal a similar trend (Fig. 6d, Bird et al., 2014; Contreras-Rosales et al., 2014; Sarkar et al., 2015). Thus, the traditional view suggests that a warm and humid climate with a strong summer monsoon occurred during the first half of the Holocene in the ISM region (Fig. 6f, Wang et al., 2010; Zhang et al., 2011), coinciding with gradual changes in Northern Hemisphere summer insolation (Fig. 6c, Berger and Loutre, 1991). The abrupt monsoonal intensification and the early- to- mid-Holocene climatic optimum around Wuxu Lake are in accord with this view. In detail, the climatic optimum exhibits two peaks, at 10.4–8.2 and 6.6–4.9 cal ka BP, with a slight reduction between 8.2 and 6.6 cal ka BP. However, in the ISM region only the early stage of the Holocene monsoonal maximum is well documented in paleoclimatic records with reliable age control. In the Hajjar Mountain range in northern Oman, sediment accumulation rates based on optically stimulated luminescence dating show that the early Holocene humid period began at 10.5 ka BP, and reached a maximum at 9.0–8.0 ka BP (Fuchs and Buerkert, 2008). Sedimentation data from Paru Co from the southern QTP suggest that the ISM precipitation maximum occurred during the early Holocene, between 10.1 and 7.1 cal ka BP (Fig. 6h, Bird et al., 2014). In addition, reconstructed monsoon precipitation based on pollen assemblages from Xingyun Lake in Southwest China reached a maximum during the interval 7.8–7.5 cal ka BP (Fig. 6g, Chen et al., 2014). This general pattern of the Holocene climatic optimum is also observed in several other records from the QTP, but is affected by the carbon reservoir effect. A pollen record from Tso Kar in northwestern India indicates a rapid increase in summer monsoon precipitation from 10.8 to 9.2 cal ka BP, a moderate reduction in precipitation between 9.2 and 6.8 cal ka BP, and a second precipitation pulse from 6.9 and 4.8 cal ka BP (Fig. 6i, Demske et al., 2009). Similarly, the record from Lake Naleng in the southeastern QTP indicates relatively stable, warm and humid conditions from 10.7 to 4.4 cal ka BP, except for the interval between 8.1 and 7.2 cal ka BP (Fig. 6j, Kramer et al., 2010). In addition, reconstructed total solar irradiance based on cosmogenic radionuclides indicates significantly weakened solar activity between 8 and 7 ka BP (Steinhilber et al., 2012). Furthermore, a ∼ 90-year periodicity in the pollen record from Wuxu Lake has also been documented in the stalagmite δ18O record from Qunf Cave in southern Oman (Fleitmann et al., 2003), and is close to the significant 87-year periodicity of the Δ14C record (Stuiver and Braziunas, 1993). This correspondence suggests a link between solar irradiance and ISM variability during the Holocene.

Most of the records from the QTP indicate that the climate became cold and dry in the late Holocene, suggesting that the environment of the QTP and the adjacent region was predominantly influenced by the ISM (Sun et al., 2015). However, the inconsistency in the timing and duration of the Holocene climatic optimum indicates the occurrence of local variations in rainfall amount in response to the ISM (Bird et al., 2014), which is compatible with the complex terrain of the margin of the QTP. The topography effect of the Tibetan Plateau affects the moisture transfer path and establishes unstable potential energy stratification (Chen et al., 2007; Houze, 2012). The steep terrain of the margin of the QTP strengthens ascending air motions, promoting the release of latent heat and the rapid development of strong convection. Because of their high elevations, the mountains block low-level airflows to the windward sides and significantly reduce moisture transport to the interior. Until now, the long duration of the Holocene climatic optimum has only been observed in records from the margin of the QTP, suggesting that local topography and rain-shadow effects may also have played an important role in the Holocene moisture evolution of the QTP.

It was suggested above that the first PCA axis may reflect winter temperature. Since the winter temperature in China is negatively correlated with the intensity of the EAWM (Guo, 1994; Ren, 1990), it can be assumed that the sample scores on PCA axis 1 are a proxy for the EAWM intensity. The record from Wuxu Lake suggests that the EAWM was strong from the late YD to the early Holocene, and that it gradually weakened in the late Holocene (Fig. 7a). The overall trend of the EAWM during the past 12.3 ka probably followed gradual changes in Northern Hemisphere winter insolation (Fig. 7b, Berger and Loutre, 1991). A strong EAWM in the early Holocene is consistent with other records from the Chinese monsoonal region. For example, the diatom record from Huguangyan Lake in southern China indicates that the water column was well mixed in the early Holocene, mainly as the result of cold, windy conditions during winter (Fig. 7c, Wang et al., 2012). This hypothesis is further supported by the records of total organic carbon content and δ15N from the same lake (Jia et al., 2015). The larger sea surface temperature (SST) gradients over the South China Sea reveal a strengthened EAWM during the early Holocene (Fig. 7d to f, Huang et al., 2011; Steinke et al., 2010, 2011). However, the grain-size record of Chinese loess deposits also indicates that the EAWM winds gradually weakened from the early Holocene to the mid-Holocene, and then gradually strengthened in the late Holocene (Fig. 7g, Sun et al., 2012), a similar pattern to that recorded by geochemical parameters from Gonghe Basin in the northeastern QTP (Liu et al., 2013). The discrepancies may be due to the fact that the grain-size of loess and dune mobility were also influenced by the advance or retreat of deserts in northern China (Mason et al., 2008; Yang and Ding, 2008).

Interestingly, the pollen record from Wuxu Lake suggests that the EAWM was weaker in the late YD than in the early Holocene. However, this finding is in conflict with the diatom record from Huguangyan Lake, which indicates that the EAWM intensified significantly in response to abrupt climate change in the North Atlantic Ocean (Fig. 7c, Wang et al., 2012). The records from the South China Sea also indicate an intensified EAWM during this interval, in response to the slowdown of the Atlantic meridional overturning circulation (Fig. 7d to f, Huang et al., 2011; Steinke et al., 2010, 2011). However, it should be noted that the marine records are poorly dated and are of low temporal resolution. The anomaly may be explained by the climate in the tropical eastern Pacific. Observation data show that a strong EAWM usually occurs when there is a negative SST anomaly in the tropical eastern Pacific (La Niña), while a positive anomaly (El Niño) is usually accompanied by a weak EAWM (Chen et al., 2000; Wang et al., 2000). A model study indicates a significant enhancement of the El Niño–Southern Oscillation (ENSO) amplitude during the YD (Liu et al., 2014), which accords with the weak EAWM revealed by the Wuxu Lake record (Fig. 7h). Furthermore, a gradual intensification of ENSO during the Holocene also accords with a weakened EAWM, suggesting that low-latitude climate processes also played an important role in the EAWM evolution during the past 12.3 ka.

Previous studies of the dust deposits of the Chinese Loess Plateau indicate that the winter monsoon is negatively correlated with the summer monsoon on orbital and millennial timescales (Porter, 2001; Sun et al., 2012). As mentioned above, the grain-size of loess is controlled by both the winter wind intensity and the summer precipitation. Comparison of the EAWM proxy record (Fig. 8a) with the stalagmite δ18O record from Qunf Cave in southern Oman (Fig. 8c, Fleitmann et al., 2003) and with the plant wax δD record from the northern Bay of Bengal (Fig. 8d, Contreras-Rosales et al., 2014), which are ISM intensity records, reveals a broadly in-phase relationship between the EAWM and ISM in the past 12.3 ka and suggests a stronger seasonal contrast during the early Holocene than during the late Holocene. This stronger seasonal contrast during the early Holocene clearly tracks solar insolation differences between the winter and summer seasons (Fig. 8b and e, Berger and Loutre, 1991). During the Holocene, increases in winter insolation and in winter temperature at high latitudes of the Northern Hemisphere reduced the intensity of the Siberian High and resulted in a weak EAWM; however, decreased summer insolation caused the southward migration of the intertropical convergence zone and resulted in a weak ISM (Wang et al., 2012). In addition, solar insolation in the Southern Hemisphere was relatively low and El Niño strength was relatively weak during the early Holocene (Fig. 8e and f, Berger and Loutre, 1991; Liu et al., 2014), which would probably have promoted both a strong EAWM and ISM (Chen et al., 2000; Kumar et al., 1999; Wang et al., 2000). Based on historical documents from eastern China, a relationship between the frequency of cold winters and summer rainfall during AD 700–900 further supports the notion that the strength of the winter monsoon is in-phase with the summer monsoon (Zhang and Lu, 2007).