We present a continuous record of fossil diatoms from Huguang Maar Lake (HML) in southeastern China, spanning the time interval 17,500 to 6000 calendar years (cal years) B.P. The seasonal change in relative abundance of the dominant diatom taxa, Aulacoseira and Cyclotella species, can be used as a proxy of the strength of winter monsoon winds (WMW), which is supported by the results of a sediment trap experiment in HML and by an extensive review of the literature on the autoecologies of these species. In the sediment, high C. stelligera abundance and high-diatom concentration, which indicate warm conditions and low wind-driven turbulence of the water column, characterize an interval equivalent to the Greenland Interstadial 1. This is followed by an interval with low-diatom concentration and with assemblages dominated by Aulacoseira species, which suggests high wind-driven turbulence and therefore strong WMW. This interval corresponds with the Greenland Stadial 1. During the early and middle Holocene, another two episodes with strong WMW are evident from the data between 10,000 and 8500 and between 7000 and 6000 cal years B.P. The diatom record implies that strong winter monsoon episodes not only occurred during the last glacial-Holocene transition but also during the Holocene “thermal maximum.”
 The Asian-Australian monsoon is an important component of the Earth's climate system that influences the societal and economic activity of roughly half the world's population. The Asian monsoon consists of summer monsoon and winter monsoon. The summer monsoon transport moisture and heat northward from the north of Australia across the Warm Pool to northern China. The winter monsoon is characterized by cold, dry Siberian airflowing southward across eastern China, ultimately contributing to the Australian summer monsoon [An, 2000]. In addition, the Asian monsoon may affect climate globally, through interactions with ENSO [Z. Y. Liu et al., 2000]. Recently, several studies focused on the relationship between the Asian winter monsoon and El Nino [Wang et al., 2000; W. Zhou et al., 2007]. Archives such as cave deposits have been used successfully to reconstruct changes in the past strength of the rain-bearing East Asian summer monsoon. Such records show that on a millennial scale the monsoon is controlled by changes in oceanic and atmospheric circulation patterns in addition to those forced by solar changes [Wang et al., 2001; Yuan et al., 2004; Dykoski et al., 2005]. However, in most high-resolution studies, reconstruction of the winter monsoon is rarely offered because of the lack of suitable proxie records, especially during the Holocene. Therefore, there is a clear need to develop high-resolution-independent proxies records suitable for the reconstruction of the strength of winter monsoon winds.
 Here we use seasonal changes in diatom composition from the HML as a proxy for tracking changes in the intensity of WMW. Our data (sediment trap experiments, meteorological and lake hydrology physical data) suggest that, over the time interval 17,500 to 6000 calendar years (cal years) B.P., shifts in the relative abundances of the dominant diatom taxa in this lake (heavily silicified, meroplanktonic, species of Aulacoseira against small euplanktonic Cyclotella species) are mainly controlled by changes in seasonal windiness.
 The HML (21°9′N, 110°17′E, Figure 1) is located in Guangdong province, near the South China Sea coast. This lake is very sensitive to hydrological and atmospheric cycles, because it is situated in a zone with seasonal climate, influenced by both the Asian summer and winter monsoons (Figures 1a and 1b). The mean annual temperature in Zhanjiang (15 km from the HML) is 23.1°C and the mean annual precipitation is 1440 mm (Figure 2a). The natural vegetation is that of a tropical semievergreen seasonal rain forest [Zheng and Lei, 1999]. The areas of the lake and its catchment are 2.3 km2 and 3.5 km2, respectively. The lake has no surface inflow or outflow. It has a maximum depth of 22 m, and is warm monomictic, being stratified from March to October. Human impact on the lake is small as only two small temples, built during the Sui (between A.D. 581 and 618) and Song dynasties (between A.D. 960 and 1200), are located within the catchment of HML. Moreover, agricultural activities have been stopped since the year 2000, when HML became part of a protected natural reserve.
2. Material and Methods
 In 1997, seven sediment cores (HUG-A to HUG-G) from three different sites in HML were drilled with a high-precision piston coring system [Mingram et al., 2007] (Figure 1). The Core HUG-B used for this study has a length of 24.28 m and was recovered from a water depth of 13.4 m [Mingram et al., 2004] (Figure 1c). The age model is based on 9 AMS (accelerator mass spectrometry) 14C dates, 5 of leaves and 4 of bulk sediment. Details of the age model are given in [Yancheva et al., 2007], and all the ages given here are calibrated ages. The events identified in the HML diatom sequence are compared with the GRIP isotope stratotype [Lowe et al., 2008], now recognized as a template for climatic events during the Last Termination. For the readership convenience we also refer to the widely use terms such as “Younger Dryas,” “Bølling,” “Allerød,” and “Preboreal.” Trend in magnetic susceptibility and titanium content, which were derived from the same core and interpreted as proxy records for aeolian input [Yancheva et al., 2007], will be compared to our diatom data. The wind speed data come from the China Meteorological Data Sharing Service System.
 In order to investigate the autoecology of the diatom species that dominate the present-day assemblages as well as the core samples, sediment traps were deployed in the lake in 2007. Cylindrical sediment traps were built according to the recommendations of Blomqvist and Hakanson . The traps were deployed for one month between 9 August and 9 September and between 9 October and 9 November at about 16 m water depth in the center of the lake, respectively. The solid material together with water in the cups was transported to the laboratory for diatom analysis. Lake water temperature profiles were measured at 1-m intervals in the field using a water collector equipped with a thermometer.
 Diatom slides were prepared in test tubes from about 0.05 g of dry sediment using hot H2O2 followed by HCl to remove organic matters and carbonates, respectively [Battarbee et al., 2001]. The 300 to 600 diatom valves were counted from almost all samples, except for 4 samples with low-diatom concentration for which only 200 valves could be counted. Diatom taxonomy followed the guidelines of Krammer and Lange-Bertalot  and valves were identified to the lowest taxonomic level possible. The diatom data are shown as relative abundance. The S/G ratio (S/G stands for Cyclotella stelligera/Aulacoseira granulat) and a, two most abundant species in the diatom sequence, is a (nonlinear) estimate of the abundance of C. stelligera to that of A. granulata. In the HML, a low S/G ratio indicates strong WMW (see below for explanations). Diatom concentration (diatom valve number per gram) was estimated in each section by adding divinyl benzene microspheres to the cleaned suspension [Battarbee and Kneen, 1982]. The stratigraphical data were divided into diatom assemblage zones (DAZ) using the program ZONE (A. F. Lotter and S. Juggins, unpublished data, 1994), on the basis of a numerically optimal sum-of-squares partitioning with stratigraphical constraints [Birks and Gordon, 1985]. The resulting zonation was compared with the “broken stick” model [Bennett, 1996], and four statistically significant zones were established using the unpublished program BSTICK. In addition, we also distinguished nonsignificant subzones within DAZ 1 and 4.
3.1. Sediment Trap and Thermal Stratification
 Sediment trap samples were collected and water temperature profiles were recorded twice in August and November 2007. Our data show that thermal stratification was developed in August with the temperature of the upper mixed layer reaching 31°C, while the bottom water temperature was 26°C (Figure 2b). This difference in temperature may seem small but because it occurs at high temperature, it causes large difference in water density and strong stratification [Wetzel and Likens, 2000]. In contrast, in November with higher wind speed inducing turbulent mixing, the water column was nearly isothermal (Figure 2b), at 23–24°C.
 The diatom assemblage composition of the two trap samples of August and November is significantly different. In August, C. stelligera dominated (91%), the relative abundance of A. granulata was very low (only 2.7%). In October, C. stelligera relative abundance decreased to 70% while that of A. granulata increased to 25%.
3.2. Diatom Stratigraphy
 135 diatom taxa were recorded in the studied sediment core HUG-B. Figure 3 shows the most common diatom taxa, expressed as percent relative abundances.
 DAZ 1 (between 17,500 and 16,800 cal years B.P.) is dominated by heavily silicified, meroplanktonic Aulacoseira species (A. ambigua, A. granulata) and planktonic Cyclotella radiosa and small, planktonic C. stelligera. The most common benthic diatoms are Achnanthes lanceolata, Fragilaria brevistriata, Fragilaria construens var. venter, and Fragilaria pinnata. Within this zone, two subzones can be distinguished. DAZ 1a characterized by dominant C. radiosa and DAZ 1b characterized by dominant A. ambigua.
 In DAZ 2 (between 15,100 and 14,400 cal years B.P.) A. granulata dominates while C. radiosa decreases.
 In DAZ 3 (between 14,400 and 13,400 cal years B.P.) C. stelligera dominates while A. granulata, A. ambigua, and C. radiosa virtually disappear from the assemblages.
 In DAZ 4 (between 13,400 and 6,000 cal years B.P.) four subzones were identified according to the alternating dominance of A. granulata and C. stelligera. A. granulata is most abundant in DAZ 4a and 4c (i.e., between 13,400 to 11,000 and 10,000 to 8,500 cal years B.P., respectively) while C. stelligera dominates DAZ 4b and 4d (i.e., between 11, 00 and 10,000 and between 8500 to 6000 year ago, respectively). Benthic diatoms are particularly abundant in DAZ 4c and are mainly represented by Cymbella gracilis, Cymbella minuta and Gomphanema gracile. Additionally, a short interval at the top of DAZ 4d could also be distinguished although not identified as statistically significant by the program ZONE (probably because it includes too few samples). In this interval, C. stelligera decreases significantly while A. granulata increases, and the benthic component of the assemblages also changed, being dominated by Achnanthes saccula and F. pinnata.
4.1. Relationship Between Aulacoseira and Cyclotella Species and Wind Strength
 Previous studies on diatom ecology have shown that many Aulacoseira species are meroplanktonic species. Meroplanktonic organisms enter the plankton when mixing conditions are such that they can be suspended and maintained in the water column [Kilham, 1990]. Aulacoseira species can adapt to low-light levels that cells experience in a well-mixed water column by increasing the amount of chlorophyll per cell and thus their light-harvesting capacity [Talling, 1957]. These species also have resting stages that allow them to survive in the dark when they sink to the sediment surface at the lake bottom [Schelske et al., 1995]. Aulacoseira species, such as A. ambigua and A. granulata, also appear to have rather high nutrient requirements [Kilham and Kilham, 1975]. During periods of strong wind, the lake thermal stratification breaks down, causing nutrient concentrations to increase and therefore offering conditions for meroplanktonic diatoms to thrive in abundance. Therefore, the abundance of Aulacoseira species, such as A. granulata and A. ambigua, is an indirect paleoenvironmental indicator of the persistence of strong, seasonal wind stress and resultant turbulent water column mixing and nutrient upwelling conditions [Pilskaln and Johnson, 1991].
 By contrast, euplanktonic species such as Cyclotella more commonly occur when lakes are thermally stratified during warm periods with weak wind [Battarbee et al., 2002; Sorvari et al., 2002; Rühland et al., 2003]. C. stelligera, in particular, has been found to be the dominant algae in the deep chlorophyll layer that develops at the bottom of the epilimnion during thermal stratification in summer in Lake Superior [Fahnenstiel and Glime, 1983]. Euplanktonic species occur when lakes nutrients are more likely to be limiting [Tilman et al., 1982]. Cyclotella species are also commonly observed in higher abundances in lakes with low nutrients (i.e., oligotrophic) [Rühland et al., 2003]. During warm periods with weak wind, lakes stratify, causing nutrients (P, N and Si) to become depleted in the epilimnion [Petrova, 1986], as euplanktonic diatoms compete for limiting resources among each other and with other algae [Tilman et al., 1982].
 In our sediment trap, A. granulata and C. stelligera are the main species. The water column temperature (Figure 2b) and the diatom assemblages found in sediment trap samples from the HML support previous findings on the ecology of Aulacoseira species and C. stelligera. In August, C. stelligera dominates the trap samples, but in October, the relative abundance of C. stelligera decreased and that of A. granulata significantly increased.
 It is well established that the stability of thermal stratification in lakes is greatly affected by wind speed. In particular, when wind speed exceeds a threshold of roughly 3 m s−1, Langmuir cells circulation can develop, which has a strong effect on the vertical distribution in the epilimnion of nonmotile planktonic organisms such as diatoms [Reynolds, 2006].
 The meteorological data for the Zhanjiang area in 2007 show that in November the daily mean wind speed often exceeded 3 m s−1 (70% of the daily mean), and in October the daily mean wind speed often exceeded 3 m s−1 (54% of the daily mean). In contrast in August the daily mean wind speed seldom exceeds 3 m s−1 (26% of the daily mean) (Figure 2c). Over the whole period covered by the meteorological data (1954–2007), there is still a significant difference between the daily mean wind speed of winter months (December, January, and February) and summer months (June, July, and August), although it is less pronounced (58 and 35% exceeding 3 m s−1, respectively). The data suggests that C. stelligera dominates during summer because of well-developed lake thermal stratification and decreases with increasing wind strength, while A. granulata has opposite requirements (Figure 2c). Therefore, these data suggest that wind speed is an important factor in the mixing regime of this lake and is consistent with the diatom shifts recorded in the trap, indicating that relative abundance of A. granulata and C. stelligera is a good indicator of WMW and thermal stratification.
4.2. Interpretation of the Diatom Sequence
 DAZ 1b and 2, covering 16,800–14,400 cal years B.P., is most likely equivalent to the cold period defined in the Greenland ice core δ18O record as Greenland Stadial (GS) 2a [Lowe et al., 2008]. A. granulata, and A. ambigua dominate (Figure 3). In zone 4a (between 13,400 and 11,000 cal years B.P.), that includes the GS 1 cold period [Lowe et al., 2008] (still widely referred to as the Younger Dryas), the relative abundance of A. granulata in the sediments is high, while that of C. stelligera is low. Therefore, these diatom data indicate that the WMW was strong during the GS 2a and GS 1cold periods.
 In contrast, during 14,400 and 13,400 cal years B.P. (DAZ3), equivalent to the warm Greenland Interstadial (GI) 1 (widely known as the Bølling Allerød), and between 11,000 and 10,000 cal years B.P. (DAZ 4b), an interval equivalent to the Preboreal [Lowe et al., 2008], the relative abundance of C. stelligera in the sediments is high, while that of A. granulata is low and A. ambigua disappear (Figure 3). The diatom data from these zone suggested that strength of WMW was reduced during these warm periods.
 Following GS 1, another two intervals with relatively strong WMW occurred between 10,000 to 8500 and between 7000 to 6000 cal years B.P., as evident from the curve of A. granulata abundance (Figure 3). Although we cannot quantify how strong the WMW were during these two intervals, high relative abundances of A. granulata suggest that the WMW was quite strong.
 An alternative interpretation for the variations in diatom relative abundance observed in the HML sediment sequence would be to consider changes in precipitation instead of the WMW. It is generally assumed that increased precipitation during summer could bring more nutrients to the lake from its surrounding catchment because of increased runoff and groundwater supply. Such catchment-mediated process would favor the development of A. granulata, which is an eutrophic species [Kilham and Kilham, 1975], instead of Cyclotella, which is considered as oligotrophic species. However, our trap data show that the dominant diatom species is not A. granulata but C. stelligera in summer. This indicates that the physical process, especially windiness and its effect on the stability of the water column in the case of HML, is likely to be more important here for explaining the dynamic shifts between these two diatoms than changes in nutrient concentrations. Indeed, Aulacoseira taxa are thickly silicified and form filamentous colonies which make them relatively heavy and more likely to sink out of the photic zone as stratification develops unless turbulent conditions help them to remain in suspension [Kilham et al., 1996; Pannard et al., 2008]. On the other hand, Cyclotella species such as C. stelligera are small-sized diatoms that do not form colonies and have therefore much lower sinking rates compared with Aulacoseira which allow them to remain in suspension in the lake column much longer [Rühland et al., 2008]. Cyclotella are also known to tolerate high-light conditions and to be able to regulate their buoyancy to prevent sinking [Saros et al., 2003]. Interestingly, similar shifts between Cyclotella and Aulacoseira, linked to thermal stratification, have been observed elsewhere [Pannard et al., 2008; Rühland et al., 2008]. In a lake system such as HML, with a small and forested catchment, surface runoff from precipitation would bring only little nutrients to the lake water. By contrast, wind-driven turbulence, disturbing the sediment at least in the littoral zone of the lake (Figure 1c), can cause large release of nutrients. Therefore, if nutrient still plays a role in the diatom succession at HML, it is most likely controlled by wind-driven turbulence and has the same effect on the composition of the assemblages (i.e., increased relative abundance of A. granulata).
4.3. Comparison of the Diatom Data With Other Paleoclimatic Records
 Our interpretation of the diatom record as a proxy for the variations in WMW is supported by independent paleoclimatic data. Changes in the magnetic susceptibility (χ) and titanium content of the sediment from the HML (Figure 4) are interpreted as a measure for dust input and lake mixing, which are both intensified in times of stronger winter monsoon winds [Yancheva et al., 2007]. During cold climates, before 14.400 cal years B.P. and during GS 1, Aulacoseira species (granulata and ambigua) are abundant and S/G ratio is low while Ti content and χ are high (Figure 4). Pollen analysis from this same core also supports our interpretation of the diatom results as it shows that subtropical and temperate trees were dominant during GS 1, indicating a relatively cool climate. In contrast, during GI 1 and Preboreal, tropical evergreen trees dominated [Wang et al., 2007].
 During the time interval 10,000 to 8500 cal years B.P. although Ti content and χ fluctuations are not as significant as during GS 1 (Figure 4), the diatom data and S/G ratio indicating strong WMW (Figure 4). The increasees in benthic diatoms in zone 4c (Figure 3) are also consistent with strong WMW, as wind-induced turbulence would bring more benthic diatom from the littoral zone where they grow into the center of the lake where the sediment core was retrieved (Figure 1c). At high latitudes in the Northern hemisphere, the atmospheric circulation over the North Atlantic and Siberia strengthened during the interval between 9000 and 8000 cal years B.P. [Mayewski et al., 2004]. At least one large pulse of glacier meltwater into the North Atlantic probably enhanced production of sea ice, providing an additional positive feedback on climate cooling. In low latitudes, this is a period of widespread aridity [Mayewski et al., 2004]. In the arid regions of northern China, the temporal distribution of calibrated 14C dates of palaeosol units from numerous stratigraphic sections investigated can be used to distinguish episodes of soil formation (mild and moist climate) and episodes of loess and sand deposition (under a cold and dry climate, when no soil was formed and therefore from which no radiocarbon dates were yielded). Within the Holocene, one of the dry periods identified in this way occurred between 9,000 and 8500 years B.P. [Guo et al., 2000]. However, some records from southern China indicated that the summer monsoon was also strong during the early Holocene (usually in the literature, this interval has been referred as the “early Holocene optimum” or the Holocene “thermal maximum”) [Wang et al., 2005; Wang et al., 2007]. A possible explanation for this apparent discrepancy is that both winter and summer monsoons were strong during the early Holocene, as high summer insolation enhanced the seasonal contrast [Kutzbach, 1981]. It is worth mentioning that particularly strong storminess in summer would also benefit the development of A. granulata [Miyajima et al., 1994] while being detrimental to taxa adapted to a more stable water column such as C. stelligera.
 Between 7000 to 6000 years B.P., the abundance of A. granulata, the S/G ratio and high percentages of benthic species all suggest that WMW was strong. This is consistent with the Ti and χ records (Figure 4). In the paleosol records of northern China, the sparsity of 14C dates for this time interval also suggests that the interval was characterized by severe dry events [Guo et al., 2000]. Lake geomorphological and lithological evidence from the Alashan Plateau indicates strong lake desiccation during the mid-Holocene around 5000 to 7000 cal years B.P. [Chen et al., 2003]. We should also consider that a stronger winter monsoon might not have been the only cause of the dry events recorded in North China. These events may have been caused by lower precipitation (weaker summer monsoon), or a combination of both a strengthening of the winter monsoon and a weakening of the summer monsoon.
 The results of this study highlight the overall good correspondence we observed between the geochemical and diatom data. This suggests that the diatom S/G ratio and the HML titanium content are a reliable proxy record of the strength of the winter monsoon over the time interval studied.
 The relative diatom abundance of Aulacoseira species and Cyclotella species from the HML can be used as a proxy of the strength of winter monsoon winds. The HML sedimentary diatom record shows dramatic changes implying four episodes with strong WMW over the time interval from 17,500 to 6000 cal years B.P. Remarkably, diatom assemblages imply that strong winter monsoon episode not only occurred during the last glacial-Holocene transition, but also during the Holocene thermal maximum.
 The authors thank Yang Xiangdong, Deng Chenglong, Hao Qingzhen, Qin Xiaoguang, and Li Fengjiang for constructive comments. The authors also thank Li Xianzhong, Peng Zhiming, Wu Wenrong, and Lin Weiguo for their help in the field work. We are grateful to three reviewers for their very helpful comments on the manuscript. This research was supported by the National Natural Science Foundation and the Chinese Academy Sciences of China through grants 40572105, KZCX3-SW-145, and KZCX2-YW-17.