Holocene linkages between char, soot, biomass burning and climate from Lake Daihai, China


  • Y. M. Han,

    1. Key Laboratory of Aerosol Science and Technology, State Key Laboratory of Loess and Quaternary Geology, Institute of Earth Environment, Chinese Academy of Sciences, Xi'an, China
    Search for more papers by this author
  • J. R. Marlon,

    Corresponding author
    1. Department of Geography, University of Wisconsin–Madison, Madison, Wisconsin, USA
    • Corresponding author: J. R. Marlon, Department of Geography, University of Wisconsin–Madison, 550 N. Park St., Madison, WI 53706, USA. (marlon@wisc.edu)

    Search for more papers by this author
  • J. J. Cao,

    1. Key Laboratory of Aerosol Science and Technology, State Key Laboratory of Loess and Quaternary Geology, Institute of Earth Environment, Chinese Academy of Sciences, Xi'an, China
    Search for more papers by this author
  • Z. D. Jin,

    1. Key Laboratory of Aerosol Science and Technology, State Key Laboratory of Loess and Quaternary Geology, Institute of Earth Environment, Chinese Academy of Sciences, Xi'an, China
    Search for more papers by this author
  • Z. S. An

    1. Key Laboratory of Aerosol Science and Technology, State Key Laboratory of Loess and Quaternary Geology, Institute of Earth Environment, Chinese Academy of Sciences, Xi'an, China
    Search for more papers by this author


[1] Black or elemental carbon (EC), including soot and char, are byproducts of anthropogenic fossil-fuel and biomass burning, and also of wildfires. EC, and particularly soot, strongly affects atmospheric chemistry and physics and thus radiative forcing; it can also alter regional climate and precipitation. Pre-industrial variations in EC as well as its source areas and controls however, are poorly known. Here we use a lake-sediment EC record from China to reconstruct Holocene variations in soot (combustion emissions formed via gas-to-particle conversion processes) and char (combustion residues from pyrolysis) measured with a thermal/optical method. Comparisons with sedimentary charcoal records (i.e., particles measured microscopically), climate and population data are used to infer variations in biomass burning and its controls. During the Holocene, positive correlations are observed between EC and an independent index of regional biomass burning. Negative correlations are observed between EC and monsoon intensity, and tree cover inferred from arboreal pollen percentages. Abrupt declines in temperature are also linked with widespread declines in fire. Our results 1) confirm the robustness of a relatively new method for reconstructing variations in EC; 2) document variations in regional biomass burning; 3) support a strong climatic control of biomass burning throughout the Holocene; and 4) indicate that char levels are higher today than at any time during the Holocene.

1. Introduction

[2] Large increases in atmospheric black carbon (BC) have had substantial impacts on radiative forcing and the hydrologic cycle since industrialization [Ramanathan et al., 2001; Hansen and Nazarenko, 2004]. Soot and char are light-absorbing components of BC that can affect climate, but soot is more widely distributed and has a particularly strong effect on areas covered by extensive snow and ice, such as the Arctic and Tibetan glaciers. Soot darkens such surfaces, lowering their albedo and increasing warming [Ramanathan and Carmichael, 2008]. Today, black carbon emissions are likely the second strongest contribution to global warming (after carbon dioxide emissions [Ramanathan and Carmichael, 2008]). Aerosols including soot have also been shown to have a negative effect on radiative forcing, however, and thus their net effect on climate remains highly uncertain [Intergovernmental Panel on Climate Change, 2007]. Pre-industrial BC levels, which originate primarily from vegetation (biomass) burning are estimated to be far smaller than those occurring from today's fossil-fuel burning [McConnell et al., 2007], but BC remains critically important for understanding climate system dynamics because of its large contribution to the land-atmosphere carbon flux [Patra et al., 2005] and to the global methane cycle [Mischler et al., 2009].

[3] Sedimentary charcoal records are the primary source of information about century- to millennial-scale variations in biomass burning [Whitlock and Larsen, 2002], and networks of these records have been combined to produce regional and global syntheses [Carcaillet et al., 2002; Power et al., 2008]. There are large uncertainties in paleofire reconstructions however, and large gaps in the network, including in Asia. Also, most charcoal records reflect local burning (within tens of kilometers), which results in high inter-site variability [Gavin et al., 2006]. Ice core records have been used to reconstruct regional and broad-scale changes in biomass burning through analyses of vanillic acid and CO, for example [McConnell et al., 2007; Wang et al., 2010], and further efforts are underway to develop additional fire proxies [e.g., Eckmeier and Wiesenberg, 2009] on longer temporal scales, but these approaches are still new and uncertain, and thus far do not span the Holocene. Pre-industrial BC is also produced from biomass burning, but long-term BC records typically span centuries rather than millennia [Elmquist et al., 2007; Thevenon et al., 2009], and the longest records often have limited sampling resolution records [Bird and Cali, 1998; Thevenon et al., 2003]. Importantly, the wide variety of analytical methods for determining BC often produce conflicting results, due not only to measurement error but also to differences in the combustion products identified and the definitions of these products [Elmquist et al., 2006; Hammes et al., 2007; Conedera et al., 2009; Bird and Ascough, 2012].

[4] BC is not a single chemical compound or a group with well-defined characteristics, but a suite of compounds occurring along a “combustion continuum” [Goldberg, 1985; Masiello, 2004]. BC may contain both elemental carbon (EC), which is thermally refractory pure carbon with a graphitic structure, as well as organic carbon (OC) and both efficiently absorb light [Chow et al., 2009]. Two primary types of BC are combustion residues from pyrolysis (char) and combustion emissions formed via gas-to-particle conversion (soot) [Masiello, 2004; Elmquist et al., 2006]. Char is produced during pyrolysis by heating organic substances, or as an impure form of graphitic carbon obtained as a residue when carbonaceous material is partially burned or heated with limited access to air. Char is composed mainly of micrometer-sized particles, is less condensed than soot and often retains the morphology of its source material. The countable char particles identified under a microscope (generally >10μm) are called charcoal. Soot refers only to those carbon particles that form at high temperatures via gas-phase processes. Soot is composed mainly of submicron particles formed by the condensation of hydrocarbon radicals at high temperatures (>600°C). Because soot is more condensed and recalcitrant than char, previous studies have had better success distinguishing soot than char from the broader spectrum of BC compounds [Elmquist et al., 2006].

[5] Char and soot have different chemical and physical properties [Masiello, 2004] and thus light-absorbing characteristics [Bond et al., 2002; Kirchstetter et al., 2004; Andreae and Gelencser, 2006; Alexander et al., 2008], but there is no absolute or standard measurement technique for them because of their complex characteristics [Schmidt and Noack, 2000]. As a result, comparing BC results from different studies has been challenging and artifacts have been reported for each of the different methods currently in use [Hammes et al., 2007]. The primary method for analyzing black carbon in soils and sediments is a chemothermal method (CTO-375), which measures BC in soot form [Gustafsson et al., 1997, 2001]. In aerosol studies, thermal methods are also common [e.g., Cachier et al., 1989; Fung, 1990], and more recently, thermal-optical methods [e.g.,Chow et al., 1993; Birch and Cary, 1996; Chow et al., 2001b; Cavalli et al., 2010]. As just one example of the complications, in the thermal method, some catalysts and ions in samples may decrease the activation energy of BC in the analysis process, leading to negative results (underestimation) [Novakov et al., 2005; Han et al., 2009b]. On the other hand, some water soluble organic carbon (WSOC) not evolved in organic carbon (OC) oxidization steps [Yu and Ito, 2002; Schauer et al., 2003] can result in positive results (overestimation). As a result, discrepancies between comparisons of BC measurements using different methods (and potentially different BC or EC definitions) can be more than a factor of 500 [Schmidt, 2001; Hammes et al., 2007].

[6] Recently, Han et al. [2007b] evaluated the use of thermal optical reflectance (TOR) from the Interagency Monitoring of Protected Visual Environments (IMPROVE), a method commonly used for analyzing carbonaceous aerosols [Chow et al., 2001a; Gelencsér, 2004] to differentiate char and soot from EC, which is defined operationally below. The results from this study indicate that pure char material always peaks at the low temperature stage (i.e., EC1, evolving at 550°C in a 2% O2, 98% He atmosphere), whereas pure soot samples peak at the high temperature stage (i.e., EC2 and EC3, evolving at 700°C and 800°C in a 2% O2, 98% He atmosphere, respectively) [Han et al., 2007a]. Comparison of the TOR and chemothermal (CTO-375) methods [Gustafsson et al., 1997, 2001] demonstrate that soot determined by the CTO-375 method corresponds to EC2 and EC3 as found by the TOR method [Han et al., 2007b], suggesting that the TOR method can differentiate between char and soot. This differentiation has since been used in aerosol studies [e.g., Han et al., 2010], as well as in the study of soil and road dusts to assess local air pollution in urban areas [Han et al., 2009a, 2011]. In aerosol studies, the presence of WSOC and certain ions (especially Cl and NO3) can influence the reliability of EC quantification as well as the differentiation between char and soot, however, with sedimentary samples, pretreatment procedures (described below) are employed to remove the WSOC and ions.

[7] Here we employ the TOR method of EC analysis, which has the potential to distinguish between char (ECchar) and soot (ECsoot) [Han et al., 2007b, 2009b], to reconstruct Holocene variations in EC from Lake Daihai in Inner Mongolia. We compare the EC reconstructions with an independent reconstruction of regional biomass burning based on a composite record of 36 sedimentary charcoal records from eastern Asia [Marlon et al., 2012]. The EC records are also compared with changes in climate, vegetation and human activities to identify potential influences on EC variations.

2. Methods

2.1. Data Collection and Chronology

[8] A 12.08-m long sediment core (DH99-A) was recovered from the center of Lake Daihai (112°40′E, 40°32′N) in 1999 (seeauxiliary materialFigure S1b in Text S1) using a piston corer driven by a TOHO drilling rig (Model D1-B, Japan). DH99-A contains irregularly laminated, gray-black silt clay. A total of 523 samples from the upper 10.92 m of DH99-A, including a surface sample which was collected in 2001 from the same place, were used for elemental carbon analysis. Multiple chronologies have been developed for the DH99-A and for a second core (DH99-B) from Lake Daihai based on a variety of dates from both cores (Table S1 inText S1). In total, sixteen 210Pb dates [Jin et al., 2001], twenty-one accelerator mass spectrometry (AMS) radiocarbon dates [Jin et al., 2004; Xiao et al., 2004] and two aragonite U-Th dates analyzed by Thermal Ionization Mass Spectrometry (TIMS) [Sun et al., 2001] are available from both cores (Table S1 in Text S1). In order to maximize the information available from all the available dates, including those that appear too young or old, we used Bayesian age-depth modeling software [Blaauw and Christen, 2011] to estimate ages and uncertainties for each sample (Table S2 in Text S1). Sediment samples were pretreated with HCl and HF acids to remove carbonates, minerals and metal oxides before carbon analysis using a thermal optical reflectance (TOR) method detailed in Han et al. [2007b] and summarized below to differentiate between char and soot [Han et al., 2009b].

2.2. Carbon Analysis

[9] All samples were dried in an oven at 40°C for 2 days. The dried samples were ground in an agate mortar and pestle and passed through a 63-μm sieve. The TOR method was employed for carbon measurement after acid pretreatment [Han et al., 2007b, 2009b] following the IMPROVE protocol. Approximately 0.15 ± 0.05 g of each sample was weighed (1/10,000 balance) and pretreated with HCl and HF acids to remove carbonates, minerals and metal oxides. The residues were filtered onto pre-fired (850°C, 3 h) 47-mm quartz filters for carbon analysis using a DRI Model 2001 Thermal/Optical Carbon Analyzer. Carbon was allowed to evolve through programmed, progressive heating resulting in eight carbon fractions: four organic carbons (OC, comprising OC1, OC2, OC3 and OC4 at 120, 250, 450 and 550°C in pure helium), one pyrolyzed organic carbon (POC, produced in the heating process, and monitored by assessing the return to initial values of laser reflectance) and three EC (EC1, EC2, and EC3 at 550, 700, and 800°C in 2% oxygen and 98% helium). OC is defined as OC1 + OC2 + OC3 + OC4 + POC and EC as EC1 + EC2 + EC3 − POC. ECchar is defined as EC1 minus POC and ECsoot as the sum of EC2 and EC3 [Han et al., 2007b]. The analyzer was calibrated daily using known quantities of CH4. Replicate analyses were performed at the rate of one per group of 10 samples. The difference in comparison with the average values from replicate analyses was <10% for ECchar and ECsoot. In addition to EC, total organic carbon concentrations (TOC) were obtained from Jin et al. [2004]who determined the values using a USA CE-440 CHON elemental analyzer after HCl acid pretreatment.

[10] Uncertainty in the estimation of the EC data (Table S2 in Text S1) is calculated as:

display math

where CV = coefficient of variance; N = number of samples; ci = concentration of initial analysis; ci,r = concentration of sample “i” replicate analysis; Unci = uncertainty; MDL = minimum detection limit. The replicate analyses in this study are at the rate of one per group of 10 samples, and the reproducibility is better than 8% for EC and better than 10% for ECchar and ECsoot. The average CV for EC, ECchar and ECsoot from the replicate analyses is used as the common CV for the corresponding carbon content for all samples in this study. The MDLof the DRI Model 2001 carbon analyzers is based on the analyses of 214 blank quartz-fiber filters, and is 0.45μg cm−2 for total carbon and organic carbon, and 0.06 μg cm−2 for elemental carbon. Here we use 0.06 μg cm−2 for ECchar and 0.03 μg cm−2 for ECsoot, respectively, which have been tested in our lab based on the analyses of 20 blank quartz-fiber filters. The MDL inμg cm−2is converted to mass concentrations in mg g-1 based on the weighed amount for each sample. Concentrations were converted to influx values by multiplying by the sedimentation rates calculated from the chronology (Figure 1).

Figure 1.

Posterior age–depth model of the sediment cores from Lake Daihai overlaying the calibrated distributions of the individual dates (purple) developed using the Bayesian software “Bacon” [Blaauw and Christen, 2011]. Grey dots indicate the model's 95% probability intervals.

2.3. Comparison of EC Trends With Changes in Climate, Vegetation and Human Activities

[11] An independent published record of regional biomass burning in eastern Asia Multiple was obtained along with data sets on potential factors influencing EC variations for comparison with the EC data, including 1) variations in arboreal pollen percentages from Lake Daihai [Xiao et al., 2004] that indicate a shift from forest to non-forest vegetation, which affected fuel type and flammability; 2)δ18O data from Greenland [Stuiver et al., 1995] used to infer widespread features of northern hemisphere temperature variations (e.g., a cold interval ca. 8.2 ka); 3) δ18O data from Dongge [Dykoski et al., 2005; Wang et al., 2005] and Sanbao [Dong et al., 2010] caves used to infer regional monsoon intensity; 4) estimated changes in population and cultivated area (the sum of estimated crop and pastureland) from the HYDE data set [Klein Goldewijk et al., 2010] for 10°–45°N latitude, 65°–150°E longitude; and 5) changes in dynasties in China [Zhang et al., 2008] during the mid- and late- Holocene.

[12] To identify trends in these data and facilitate inter-comparisons, the EC data were transformed using a standardization and normalization process described inPower et al. [2008]designed to stabilize the variance in the time series and allow statistical analysis. The high-resolution data (i.e., the EC, charcoal, GISP2 and Dongge Cave records) were also smoothed using locally weighted regression (lowess) applied to a 500-year moving window (constant time step) rather than a constant proportion of data (which can vary substantially through time in sediment records) typical of conventional lowess algorithms. For the correlation analyses, transformed but unsmoothed EC data, composite charcoal data, climate, and pollen data were re-sampled to regular 20-year intervals. Determining the significance of the correlations between the EC, biomass burning and climate proxy data is complicated by temporal autocorrelation in the records that violates an assumption of parametric significance tests. To address this, we used a block-bootstrap method to determine the significance of the correlations [Gavin et al., 2011]. Each bootstrap sample is obtained by resampling (with replacement) the time series in blocks of time corresponding to the autocorrelation structure of the data, determined using the partial autocorrelation function in the R statistical software package [R Development Core Team, 2008]. We tested multiple block sizes (from 10 to 30 lags) and report the lowest significance levels found using this range.

3. Results and Discussion

3.1. Sampling and Chronology

[13] The chronology used for the EC data are presented in Figure 1. Uncertainties in dating are small for the past 6000 years as the results from Xiao et al. [2004] and Jin et al. [2004] are generally consistent for the upper 9.0 m. Dating discrepancies occur for the lower 3.08 m, however, with the ages determined by Xiao et al. [2004] a little greater than those reported by Jin et al. [2004], but those of Jin et al. [2004] and Li et al. [2004] are similar for this interval (Table S1 in Text S1). Additionally, two aragonite U-Th dates between 11 and 12 m [Sun et al., 2001] analyzed by Thermal Ionization Mass Spectrometry (TIMS) showed much younger ages than those reported by both Jin et al. [2004] and Xiao et al. [2004]. 210Pb and 137Cs analysis were also previously conducted for the upper part of the DH99-A [Jin et al., 2001], and the sedimentation rates aged by the 210Pb dating, averaging 1.6–1.8 mm a−1 are consistent with the youngest AMS 14C dates [Jin et al., 2001].

[14] The Bayesian age-depth analysis yields a nearly linear model for the first 10 m of sediment, with an average Holocene deposition time during of 7.6 yr/cm. The bottom 2 m of sediment, which records early Holocene conditions, show a slower deposition time of about 26.7 yr/cm. While slower sedimentation rates and increased dating uncertainties (observed in both the radiocarbon and U-Th dates) during the earliest period are unfortunate, they are not uncommon in mid- and high-latitude lake sediments. Furthermore, the consistency in estimated ages fromJin et al. [2004] and Li et al. [2004]from the bottom 2 m of sediment suggests the chronology is robust even in the early Holocene interval, and despite the young estimates from the two aragonite U-Th dates, which may have resulted from detritism.

3.2. Holocene Variations in Soot and Char

[15] Long-term trends in ECsoot and ECchar showed large variations during the Holocene (Figures 2a and 2b). ECsoot fluxes varied from 0.002 to 0.71 mg cm−2 yr−1 and ECchar varied from 0.002 to 2.63 mg cm−2 yr−1 since 10 ka. In general, mean ECsoot and ECchar values were low prior to 8 ka and higher after, although a marked decline occurred in the past 2000 years. ECsoot levels increased from very low values prior to 8 ka to relatively high values ca. 5 ka. A brief decline in ECsootoccurs from 5 - 4 ka, and then levels stabilize until about 1.5 ka. After 1.5 ka, ECsoot declines sharply to a local minima ca. 600 and 300 years ago, and then increases rapidly toward present. ECcharremains low during the early Holocene prior to 8 ka, and then increases gradually from 8 - 1.5 ka. After 1.5 ka, ECchar declines rapidly (like ECsoot) to a local minima between ca. 600 and 300 years ago and then increases sharply in the past few centuries.

Figure 2.

Comparison of (a) standardized ECsoot values with (b) standardized ECcharvalues from Lake Daihai; (c) total organic carbon; (d) the composite standardized and smoothed (250- and 500-yr windows) biomass burning record from 36 sites in eastern Asia [Marlon et al., 2012]; (e) GISP2 δ18O data [Stuiver et al., 1995] used to infer northern hemisphere temperature variations; (f) δ18O data from Dongge Cave [Dykoski et al., 2005] and (g) Sanbao Cave [Dong et al., 2010] used to infer regional moisture availability; (h) estimated changes in population and (i) crop and pastureland from the HYDE data set [Klein Goldewijk et al., 2010] for 10°–45°N latitude, 65°–150°E longitude; (j) tree pollen from [Xiao et al., 2004]; (k) changes in dynasties [Zhang et al., 2008]; and (l) deposition times (yr cm−1) based on the age-depth model. High-resolution data (i.e., the EC, Dongge Cave and GISP2 records) were smoothed with a lowess curve using a 500-year moving window.

[16] High-frequency variations in ECsoot and ECchar also show several shifts during the Holocene. Variability is low, for example, prior to 8 ka, but this is likely due in part to low sedimentation rates (high deposition times) during this period (Figure 2l). High-frequency changes in both ECsoot and ECcharare relatively consistent from 8 - 2 ka, with the exception of a reduction in variability from about 5 - 4 ka, and an abrupt decline in overall EC values ca. 2.9 ka. Variability in ECsoot and ECchar appears to increase after 2 ka, and modern ECsoot and ECchar values in particular are among the highest observed during the Holocene. Holocene ECsoot levels surpassed modern values only twice–once ca. 1750 years ago (AD 200) and once ca. 600 years ago (AD 1350) (Figure 2a). The modern ECchar value is the highest in the Holocene and surpasses the mean by more than six standard deviations (Figure 2b).

[17] ECchar accounts for the largest component of total EC concentrations (86%) and is strongly correlated with total EC (r = 0.99; S3 in SI). ECsoot is less correlated with total EC (r = 0.49; S3 in SI), but ECsoot and ECchar fluxes are strongly positively correlated (r = 0.89). Despite this high correlation, several noticeable differences exist between the standardized ECsoot and ECchar flux data (Figures 2a and 2b). For example, after 8 ka, ECsoot increases immediately, whereas ECchar remains relatively low until 7 ka. ECcharalso shows a steady increase in its mean values from the mid- to late-Holocene (e.g., between 6 and 2 ka), whereas mean ECsoot values remain similar during these two periods. ECcharalso shows a stronger decline between 2 - 1 ka. Thus, although ECsoot and ECchar appear to reflect a similar underlying process (e.g., biomass burning), differences in ECsoot and ECchar production, transportation or deposition may account for the observed differences [Elmquist et al., 2006; Han et al., 2010]. A negative correlation between ECchar and TOC (r = −0.64; P < 0.001) and ECsoot and TOC (r = −0.42; P < 0.05) suggests that charring in the analysis has little impact on the EC and EC fractions, and non-pyrogenic matter is not included in the analysis processes using the thermal optical method. ECchar and ECsoot, which are indicators of biomass burning in this region, are mainly associated with dry conditions (see details of the comparison of EC with the speleothem record below).

3.3. Controls on Holocene Trends in Biomass Burning

[18] Biomass burning is controlled predominantly by climate on regional- to global-scales [Carmona-Moreno et al., 2005]. Seasonal changes in temperature and precipitation, for example, produce a distinct cycle of burning in the northern and southern hemispheres each year [van der Werf et al., 2006]. Vegetation changes also have a strong influence on millennial-scale trends in fire through their effects on fuel characteristics (e.g., abundance, distribution, and flammability). Human activities may have contributed to carbon accumulations in Lake Daihai, but the extent and timing of such human impacts during the Holocene are largely unknown. The charcoal-based biomass burning index as well as data for each of the primary controls on biomass burning–climate, vegetation and human activities–show distinct trends during the Holocene that may help explain the observed changes in EC at Lake Daihai.

[19] The regional biomass burning index (Figure 2d) shows large variations prior to 6 ka, and an upward trend from 6 - 1.6 ka. Biomass burning declines sharply from 1.6 - 1 ka, then rises to AD 1950, and finally declines again to present. Monsoon intensity inferred from the Dongge and Sanbao caveδ18O values (Figures 2f and 2g) is relatively strong prior to 6 ka, decreases from 6 - 3 ka, and subsequently remains relatively low. There are no regional temperature reconstructions for eastern Asia for the Holocene. Theδ18O data from Greenland indicate a slight and gradual decline in temperatures since about 8 ka at high latitudes, but may have varied from midlatitude temperature trends [Grafenstein et al., 1999]. Daniau et al. [2012], for example, suggest that midlatitude temperatures increased during the Holocene and that such changes are consistent with increased biomass burning. Aside from the long-term temperature trends, however, theδ18O data from Greenland indicate several large short-term changes that are known to be widespread, including a decline in temperatures associated with the “8.2 ka event” [Alley and Agustsdottir, 2005] and the Little Ice Age ca. 550 - 250 years ago [Mann et al., 2009].

[20] Vegetation changes near Lake Daihai were reconstructed from changes in the relative abundance of pollen taxa [Xiao et al., 2004] and are representative of broader trends in the region [Zhao et al., 2009]. The primary vegetation types shift from steppe in the early Holocene to steppe forest in the mid-Holocene (8 - 3 ka), to desert steppe in the late Holocene (after ∼3 ka). Specifically, arid steppe (high Artemisia, Chenopodiaceae and Ephedra pollen and low arboreal pollen [AP] values) existed from ca. 10 - 8 ka due to a colder-than-present climate and low effective moisture. There is an apparent contradiction between the climate inferred from the vegetation at this time and lowδ18O values in the speleothem records thought to reflect increased monsoon intensity. This discrepancy is widely recognized and may be attributable to low North Atlantic sea-surface temperatures (SSTs) and high-latitude air temperatures that affect the availability, amount and transport of water vapor [Chen et al., 2008]. Between 8 - 3 ka, tree pollen is relatively high and regional climate was generally warmer and wetter than previously. The transition from the mid- to late-Holocene was marked by a shift toward drier conditions, evidenced both by the Dongge Caveδ18O record and the decline of tree pollen (Figures 2c and 2f). Xiao et al. [2004]infer that forest steppe shifted to steppe vegetation during this time due to effectively drier conditions. Today the region is characterized by desert steppe vegetation due to its transitional semi-arid/semi-humid climate.

[21] The population estimates from the HYDE data set show a steady increase during the Holocene (Figure 2h) with the exception of two important events: the Mongol Invasions starting in AD 1211, when China lost about a third of its population, and the fall of the Ming Dynasty around AD 1644, when it lost about a sixth of its population [McEvedy and Jones, 1978; Pongratz et al., 2008]. The land-use area estimates from HYDE show an initial increase beginning 7000 years ago, with variations during the past 2000 years that reflect the changes in population estimates.

[22] Analyses of correlations between the EC data and its potential controls, including the biomass burning and monsoon indexes, as well as temperature proxies and vegetation changes indicate that both ECsoot and ECchar fluxes have significant positive correlations with biomass burning and negative correlations (although not significantly) with monsoon intensity, TOC content, tree pollen and the Greenland temperature record (Table 1). In particular, ECsoot and ECchar are both positively correlated with the regional biomass burning index, and the relationship is slightly stronger for ECchar (r = 0.51, P < 0.01) than for ECsoot (r = 0.41, P < 0.05). ECchar is also more strongly correlated with the δ18O values from Dongge Cave (r = 0.88, P < 0.001) than is ECsoot (r = 0.66, P < 0.001) (higher δ18O values reflect a weaker monsoon, so there is a negative relationship between biomass burning and monsoon intensity). Finally, ECchar (but not ECsoot) is negatively correlated with the tree pollen data (r = −0.35, P < 0.05).

Table 1. Statistical Correlations Between EC and Potential Explanatory Variables
EC DataPotential Explanatory VariableCorrelation
ECsootRegional charcoal compositer = 0.41; P < 0.05
ECsootDongge Cave δ18Or = 0.66; P < 0.001
ECsootArboreal Pollen %r = −0.04; P > 0.05
ECsootGISP2 δ18Or = −0.16; P > 0.05
ECcharRegional charcoal compositer = 0.51; P < 0.01
ECcharDongge Cave δ18Or = 0.88; P < 0.001
ECcharArboreal Pollen %r = −0.35; P < 0.05
ECcharGISP2 δ18Or = −0.23; P > 0.05

[23] Low ECsoot and ECchar levels prior to 8 ka imply low biomass burning and are consistent with cool, dry climate conditions (and potentially reduced convection and lightning). Limited vegetation productivity and fuel abundance inferred from the pollen data also likely reduced fire spread (Figure 2j). High charcoal levels from 10 - 9 ka primarily reflect increased burning to the northeast of Lake Daihai [Li et al., 2005], although high burning is common elsewhere in arid steppe environments during the late glacial; this may be partly reflected by moderately high EC values ca. 10 ka (Figures 2a and 2b).

[24] A comparison of ECsoot with the GISP2 δ18O record (Figures 2a and 2e) indicate that both were reduced during the abrupt and widespread cooling that occurred during the 8.2 ka event [Alley and Agustsdottir, 2005]. The maximum likely age of the EC samples currently dated to 8 ka is 8.35 ka, and thus it is possible that the minimum in the EC data was synchronous with the 8.2 ka event. Because the independent regional biomass burning record also reaches a minimum ca. 8.2 we argue that the climate changes reduced fire and thus ECsoot and ECchar levels at 8.2 ka (Figures 2a, 2b, 2d, and 2e). Elsewhere in the world, climate changes often resulted in an opposite effect (i.e., increased burning), however [Marlon et al., 2012].

[25] Human activities may be expected to have had an increasing influence on ECsoot and ECchar trends and/or variability as the Holocene progressed, particularly in the late Holocene with the advent of dynasties in China (Figure 2k) [Expert Group of the Xia-Shang-Zhou Project, 2000]. The location of Lake Daihai near the northern limit of intensive agriculture also made it an area of frequent conflict between nomadic tribes and settled farming societies [Huang and Su, 2009]. Early Holocene hunter gatherer populations were low, however, ECsoot and ECchar were low, and charcoal levels were declining, suggesting very limited ecological effects from people on fire prior to 8 ka (Figures 2a, 2b, 2d, and 2h). The earliest large increase in ECsoot, ECchar, and charcoal occurs between 8 - 7 ka and is associated with a large increase in AP (Figures 2a, 2b, 2d, and 2j), which is also not consistent with widespread human impacts on burning.

[26] Two short-term increases in ECchar at 7 ka and 4.1 ka coincide with declines in AP that may reflect localized human impacts on burning through deforestation, but the first local evidence for agricultural activity does not occur until 6.3 ka [Tian, 2000] and there is no major shift in ECsoot or ECchar at that time. The second increase in ECchar and decline in tree pollen ca 4.1 ka occurs during a period of declining farming and cultural activity following the demise of the Laohushan culture [Tian, 2000]. The gradual increase in ECcharfrom 7 - 2 ka parallels a gradual increase in regional human activity (Figures 2h and 2i), but increases in Pinus (pine), Quercus (oak) and Ostryopsis(birch) during the mid-Holocene indicate an expansion of forest steppe vegetation and a shift to generally warm, wet conditions [Xiao et al., 2004] that do not support a strong human impact on burning then. It is possible that fire gradually increased during more frequent but brief intervals of drought during this interval.

[27] A sharp decline in both ECsoot and ECchar occurs ca. 3 ka coincident with a reduction in regional biomass burning and also with increased effective moisture (Figures 2a, 2b, 2d, and 2g). Subsequently, available moisture declines, AP declines sharply, regional biomass burning increases, and both ECsoot and ECcharshow some of their highest Holocene values; all of this occurs despite a reduction in estimated population growth and land-use (Figures 2a, 2b, 2d, 2f, 2g, 2h, 2i, and 2j). Reduced burning during wet periods and increased fire during dry periods suggests that changes in monsoon intensity or effective moisture continued to exert a strong influence on ECsoot, ECchar and biomass burning throughout the late Holocene.

[28] The large decline in ECsoot and ECcharfrom ca. 2 - 1 ka is matched by a similar decline in biomass burning (Figures 2a, 2b, and 2d) and a final decline in AP (Figure 2j). The lowest late-Holocene EC levels occur around the Little Ice Age (LIA, 550 - 250 years ago [Mann et al., 2009]), but the regional fire decline starts well before the LIA. In central eastern Inner Mongolia [Huang et al., 2005] and the southern loess plateau, burning is thought to be primarily human-caused since 3.1 ka [Huang et al., 2006]. A large fire decline during the past ∼600 years in the loess plateau is thought to result from the complete transformation of the landscape to agricultural land and to a lack of vegetation left to burn [Huang et al., 2006], but the HYDE data at least are not consistent with this idea. There is not a unique shift in EC, for example during the Mongol invasions, and changing dynasties also do not coincide with major EC fluctuations (Figures 2a, 2b, and 2k). Shifts in moisture availability seem a more plausible explanation for the large changes in late-Holocene biomass burning, which are similar across the northern hemisphere [Marlon et al., 2012].

[29] The large variations in biomass burning of the past millennium have implications for debates beyond eastern Asia. The EC minimum 500 years ago in particular parallels a global decline in fire activity at that time, which is roughly coincident with both the onset of the LIA and with the arrival of Europeans in America that led to widespread indigenous population collapse. The fire decline in the tropical Americas is often argued to have reduced global atmospheric CH4 concentrations from biomass burning [Ferretti et al., 2005; Houweling et al., 2008; Mischler et al., 2009; Finkelstein and Cowling, 2011], but whether the decline was triggered by population collapse or reduced temperatures is debated [Marlon et al., 2008; Nevle and Bird, 2008; Dull et al., 2010; Power et al., 2012]. The occurrence of a marked regional decline in soot and char in eastern Asia provides important evidence that the fire decline was not limited to the Americas, and thus pandemics associated with European colonization alone cannot account for the decline in Asia. Rather, climate changes or other shifts in human activities must be invoked to explain the widespread fire decline.

[30] In the past 50 years, ECsoot and ECchar reach very high levels (Figures 2a and 2b). The sources of EC during this interval shift from biomass burning to a combination of biomass and fossil-fuel combustion products, partly reflecting the industrialization of Asia during the latter half of the 20th century. As the Holocene EC records show, similar ECsoot levels were only attained about 800 and 2000 years ago, and current ECchar levels are higher than at any other time in the Holocene.

3.4. Concluding Discussion

[31] The similarity in char, soot and charcoal trends during the Holocene provide evidence that all three types of data reflect components of biomass burning. The distinction between ECsoot and ECchar is subtle, but ECchar shows consistently higher levels during the late Holocene, and shows a stronger relationship to both climate changes and regional biomass burning. This result strengthens our confidence in the sedimentary EC reconstructions and also suggests that degradation of charcoal, for example, due to differences in production conditions or environmental exposure does not bias the results [Ascough et al., 2011]. In summary, our results 1) confirm the robustness of a relatively new method for reconstructing variations in EC; 2) document variations in regional biomass burning; 3) support a strong climatic control of biomass burning throughout the Holocene; and 4) indicate that char levels are higher today than at any time during the Holocene.


[32] We thank Patrick Bartlein, Jessica Blois, and Daniel Gavin for software and R scripts used in the data analysis. This study is supported by the National Natural Science Foundation of China (41073102, 40925009 and 41273140), the Knowledge Innovation Program of the Chinese Academy of Sciences (KZCX2-YW-QN113), the National Basic Research Program of China (2010CB833403), the “Strategic Priority Research Program” of the Chinese Academy of Sciences (XDA05100402), and a U.S. National Science Foundation (NSF) Postdoctoral Fellowship to JRM (EAR-0948288).