Human impact is a well-known confounder in pollen-based quantitative climate
reconstructions as most terrestrial ecosystems have been artificially
affected to varying degrees. In this paper, we use a “human-induced” pollen
dataset (H-set) and a corresponding “natural” pollen dataset (N-set) to
establish pollen–climate calibration sets for temperate eastern China (TEC).
The two calibration sets, taking a weighted averaging partial least squares
(WA-PLS) approach, are used to reconstruct past climate variables from a
fossil record, which is located at the margin of the East Asian summer
monsoon in north-central China and covers the late glacial Holocene from
14.7 ka BP (thousands of years before AD 1950). Ordination results suggest
that mean annual precipitation (
Pollen analysis was initially developed 100 years ago for inferring past changes in vegetation and climate (von Post, 1916). Since the 1970s, quantitative reconstructions from biological proxies have made a revolutionary change to studies of the past climate (Imbrie and Kipp, 1971; Juggins, 2013). Numerical methods, such as the modern analogue technique (MAT; Overpeck et al., 1985), weighted averaging partial least squares (WA-PLS; ter Braak and Juggins, 1993), and others (Birks et al., 2010; Juggins and Birks, 2012), are widely used in pollen-based quantitative reconstructions (Guiot, 1990; Markgraf et al., 2002; Seppä et al., 2004; St. Jacques et al., 2008; Tarasov et al., 2011; Xu et al., 2010b). Pollen-based palaeoclimate reconstructions relay on modern pollen–climate relationship studies (Li et al., 2009; Markgraf et al., 2002; Seppä et al., 2004; Shen et al., 2006) and pollen–climate data compilations (Bartlein et al., 2010; Prentice et al., 2000; Tarasov et al., 2005; Whitmore et al., 2005; Yu et al., 1998; Zheng et al., 2014). Thousands of modern pollen samples from bioclimatic regions all over the world have been collected and analysed, for example, in the framework of the BIOME6000 Project (Prentice et al., 2000). These pollen data have been used for testing the biome reconstruction method and regional sensitivity (e.g. Tarasov et al., 1998; Yu et al., 1998) and for quantitative climate reconstructions using statistical approaches. A methodological assumption that the ecological response of species does not change during the Quaternary based on Lyell's uniformitarianism (Scott, 1963) is implicit in these studies, and they require modern organism–environment relationships as calibration models (Birks et al., 2010; Juggins and Birks, 2012).
There are several types of uncertainties in reconstructing palaeoclimate from pollen data using calibration models (Guiot et al., 2009; Marquer et al., 2014; Parnell et al., 2016; Xu et al., 2016b). In China and other regions with long-term human occupation, biomes can be strongly modified rather than natural (Ren and Beug, 2002; Zhang et al., 2010). The question of how well modern samples reflect the natural vegetation thus needs to be addressed (Xu et al., 2010a). It is most likely that modern pollen–climate relationships in such regions after a long-lasting human impact are different from what they were in the past. For example, comparing the performance of pre-disturbance (1895–1924) and modern (1961–1990) pollen–climate calibration sets from Minnesota, St. Jacques et al. (2008) found that the pre-settlement model performs better than the modern one in reconstructing past climate. The human impact on the terrestrial vegetation over the past 150 years in the American Mid-west is thus apparent in the modern calibration set. Such a distortion in the modern pollen dataset can generate bias in the climate reconstruction for those regions (Li et al., 2014; St. Jacques et al., 2015, 2008; Tian et al., 2017). Palynologists therefore have to face this challenge in vegetation and climate reconstructions when using pollen data from densely populated regions (Juggins and Birks, 2012; Seppä et al., 2004; Tarasov et al., 1999; Xu et al., 2010a).
In China, rich archaeological evidence suggests that crop domestication may have taken place in the early Holocene or even earlier (Bestel et al., 2014; Lu et al., 2009; Zhao and Piperno, 2000), and enhanced farming practices have been reported since 8000 years ago (Liu et al., 2015; Lu et al., 2009; Zhao, 2011). Early agriculture was usually accompanied by slash-and-burn clearance of forest patches (Ruddiman, 2003), and pollen-inferred anthropogenic impacts on natural vegetation are noted from as early as 6000 years ago in eastern China (Ren and Beug, 2002; Wang et al., 2010). Due to growing demand for land, construction materials, and fuel, disturbances to the natural vegetation over the last 2 millennia occurred widely and are commonly detected in the pollen records (Cao et al., 2010; Ni et al., 2014; Xu et al., 2016a; Zhao et al., 2010, 2009). Consequently, human impact in both the modern reference datasets and the fossil pollen records needs to be considered when reconstructing past climate from pollen. In China, in contrast to North America, it is not possible to establish a calibration set consisting of pre-settlement pollen and climate data. However, it is still important to estimate what kind of bias may appear in pollen-based quantitative climate reconstructions using Chinese pollen data.
In the past 2 decades, a number of modern pollen studies have been conducted in China to investigate regional pollen–vegetation–climate relationships (Herzschuh et al., 2010; Li et al., 2009; Lu et al., 2011; Luo et al., 2009; Shen et al., 2006; Xu et al., 2007; Zhang et al., 2012; Zheng et al., 2008) and human impact on vegetation (Ding et al., 2011; Liu et al., 2006; Pang et al., 2011; Wang et al., 2009; Yang et al., 2012; Zhang et al., 2014, 2010). At the same time, representative modern reference datasets (Cao et al., 2014; Xu et al., 2010a; Yu et al., 2000; Zheng et al., 2008, 2014) and fossil pollen datasets (Cao et al., 2013; Ren and Beug, 2002; Sun et al., 1999) have been assembled, which make it possible to reconstruct the vegetation and climate for individual sites, regions, or the of whole China (Chen et al., 2015; Ni et al., 2014; Tian et al., 2016; Wang et al., 2014; Xu et al., 2010b). Despite the aim of these studies to use modern surface samples from natural (i.e. likely undisturbed) vegetation communities for establishing their calibration datasets and to exclude samples representing human-disturbed vegetation communities (Zheng et al., 2014), the presence in the datasets of some samples from eastern China referred to as “troublesome” due to intense human impact surrounding the vegetation patches (Xu et al., 2010a) suggests that the problem was not completely resolved. Li et al. (2014) assessed the reference pollen data in the currently available datasets from central-eastern China using a human influence index (HII) and concluded that surface samples are biased due to significant human impact on the natural vegetation. Pollen-based climate reconstructions for the late Holocene in this region would thus also be biased.
Obtaining reliable climate reconstructions in regions with a long history of human activities is indeed a big challenge but also an important scientific task. For example, reconstructing rainfall in temperate eastern China (TEC) is not only necessary for understanding the East Asian summer monsoon (EASM) variations (Chen et al., 2015; Guiot et al., 2008; Wen et al., 2013; Xu et al., 2010b), but also very important when studying human adaptation to climate change and the origins of agriculture and cultural evolution (Lu et al., 2009; Mu et al., 2015; Tarasov et al., 2006). In this paper, we (1) compiled a human-induced training set with 791 surface pollen spectra and a corresponding natural training set with 806 spectra from TEC, (2) compared the pollen–climate model performances of the two calibration sets, (3) investigated the deviations in the reconstructed results for a fossil record based on each calibration set, and (4) discuss the mechanism of bias caused by human impact.
Temperate eastern China (TEC; 30–53
Due to the large climatic and topographic gradients, several large-scale
natural vegetation regions have been described for the study area (Fig. 1):
(I) cold temperate needleleaf deciduous forest region, (II) temperate mixed
needleleaf and deciduous broadleaf forest region, (III) warm temperate
deciduous broadleaf forest region, (VI) temperate steppe region,
(IVAi) northern subtropical broadleaf evergreen-deciduous forest zone,
(VIIBi) temperate semi-shrub and shrub desert zone, and (VIIIAi) subalpine
scrub and alpine meadow zone (Editorial Committee of Vegetation Map of China,
2007; Wu et al., 2013). However, many natural ecosystems have been intensely
modified by settlements and agricultural land use. For example, in 2015,
forest coverage in the supposedly densely forested north-east of China was about
41 % and only 26 % in the warm temperate forest region, according to
data from the China National Bureau of Statistics
(
Maps of the study region showing the distributions
of
We use pollen data from a number of studies attempting to detect human-induced changes in surface pollen assemblages, including 43 spectra from the Anyang area in the central China Plain (Wang et al., 2009), 12 from the eastern Hexi Corridor (Ma et al., 2009), 78 from warm temperate hilly areas (Ding et al., 2011), 88 from the Hebei Plain and adjacent mountain area (Pang et al., 2011), 13 from south-east China (Yang et al., 2012), and 105 spectra from north-east China (Li et al., 2012, 2015). Additionally, 70 unpublished spectra from the coastal plain between the Yellow River and the Yangtze River were generated for the purpose of this study. The samples were mostly collected from croplands, abandoned croplands, economic gardens and forests, pasturelands, and roadside scrub and woodlands. The field sampling strategies, laboratory procedures, analytical methods, pollen taxa, and other detailed information are described in the aforementioned studies. Additionally, we make use of some reference pollen datasets partly or entirely covering the study area (Wen et al., 2013; Xu et al., 2010a, 2007; Zheng et al., 2008), which were used to represent natural vegetation communities.
Samples from the different natural vegetation communities were integrated
into a natural dataset (N-set), while samples from human-induced
vegetation or vegetation patches obviously disturbed by human activities were
integrated into a human-induced dataset (H-set; Fig. 1b). We used
reference samples from an approximate extent of 31–51
Simplified pollen percentage diagram of core GH09B from Lake Gonghai. The local pollen zone (and subzone) boundaries are based on the results of a constrained cluster analysis with the CONISS programme in Tilia software (Grimm, 1987, 2011) used to make the pollen diagram. The detailed information on vegetation succession was presented in Xu et al. (2016a).
Mean monthly climate averages were derived from the latest available
observation data (1981–2010) from 1208 well-distributed meteorological
stations across the study area (Fig. 1a). The original data can be accessed
from the China National Meteorological Information Center
(
Sanderson et al. (2002) developed a human influence index (HII) dataset for
mapping the areas with and without a human footprint. The HII dataset
quantifies human influence on terrestrial ecosystems based on four proxies
(nine datasets), including human population pressure (population density),
land transformation (land use/cover, roads and railways, built-up centres,
settlements), accessibility (roads and railways, coastlines, navigable
rivers), and electrical power infrastructure (night-time lights). Each of the
nine datasets assigns a score from 0 to 10 according to a rating (or
alternatively in a single score and 0) to assess human influence on 1 km
Lake Gonghai (38
Vegetation succession in the area around Lake Gonghai has experienced five
major stages during the last 14.7 kyr (Fig. 2). Open forests and upland
meadows dominated during the last deglaciation (14.7–11.1 ka), and abrupt
strengthening of
Ordination results of redundancy analysis (RDA) for 15 major pollen
taxa with climate variables (
Relationships between surface pollen spectra and climate variables are
assessed by ordination techniques. To stabilise the variance and optimise the
signal-to-noise ratio in the data, pollen taxa which occur in at least three
samples and contribute
HII was also analysed in the same way to evaluate the human impact on the pollen data and the quality of the training set. As HII is an environmental variable with certain stochastic features in locations and intensity, there is no robust ecological basis to estimate the optima and tolerance for a pollen taxon using any pollen–HII calibration set. Therefore, we used an indirect method to assess the potential bias induced from the training set due to human impact on surface samples. At first, we found five closest modern analogues for each fossil sample using MAT (Simpson, 2007) and then used the mean HII value at the analogue sites to examine the human influence on the analogue samples and to evaluate the bias in climate reconstruction for that given fossil sample.
The WA-PLS approach combines the virtues of the WA method to model ecological
optima of species and the PLS method to select linear components from biological
assemblages (ter Braak and Juggins, 1993). It has been tested along with the WA,
MAT, and pollen response surface method (PRS) for eastern China data and
demonstrated to give better results (Cao et al., 2014; Xu et al., 2010b) due
to its generally good performance under non-analogue situations and its ability
to cope with spatial autocorrelation (Cao et al., 2014; Juggins and Birks,
2012). The optimal number of WA-PLS components was selected using a
randomization
The significance of the obtained reconstructions was also tested. The proportion of variance in the fossil sequence explained by 999 transfer functions trained with random data was calculated from a constrained ordination (Telford and Birks, 2011). To help understand the bias mechanism of the human impact on pollen assemblages, we estimated the WA optima and tolerances (Birks et al., 1990; ter Braak and Looman, 1986) of selected climate variables for major taxa. All numerical analyses were performed using vegan version 2.3-5 (Oksanen et al., 2016), analogue version 0.17-0 (Simpson, 2007), rioja version 0.9-5 (Juggins, 2015), and palaeoSig version 1.1-3 (Telford, 2015) in the R 3.2.4 environment (R Core Team, 2016).
Scatter plots of pollen-based predicted annual precipitation
(
Summary statistics for redundancy analysis (RDA) with pollen species
and climate variables (annual precipitation
Ordinations are based on square-root-transformed pollen data of 99 taxa in
the N-set and 93 taxa in the H-set after noise reduction. DCA showed that the
length of the first axis is 2.65 SD (standard deviation units) in the N-set
and 2.36 SD in the H-set, suggesting that linear ordination techniques (e.g.
RDA) are appropriate to present the distribution of pollen taxa along the
climate gradients in our datasets. When using each of the climatic variables
as a sole predictor,
Summary performance statistics of the first three components of the
weighted averaging partial least squares regression (WA-PLS) for annual
precipitation (
A two-component WA-PLS model performed best with the lowest RMSEP and highest
Proportion of variance (solid lines) in Gonghai Lake fossil pollen
data explained by annual precipitation (
Fossil pollen record of Lake Gonghai:
We applied the pollen–
The WA optima and tolerances of 15 major pollen taxa in both the N- and
H-sets were estimated (Fig. 7). The optima for
Caterpillar plot of weighted average (WA) optima and tolerances for
15 major pollen taxa in response to annual precipitation (
The pollen record is a complex and non-linear function of vegetation, which in
turn is a function of climate based on some key assumptions (Birks et al.,
2010). The big challenge for pollen-based climate reconstructions is that
this indirect pollen–climate relationship can be affected by several other
(non-climatic) factors, for example, by human activities (Birks and
Seppä, 2004; Ren, 2000; Xu et al., 2010b). RDA results show that the
ability of
Selected pollen taxa can be separated into two tree and herb groups by
contrasting their WA optima in the N- and H-sets (Fig. 7). The inferred
optima of most woody taxa in the H-set are shifted towards drier conditions
and their tolerances compressed. This means that a fossil pollen assemblage
with a high proportion of woody taxa would be assigned a lower
In short, human impact obscures the climatic signals in pollen spectra by
distorting the response relationship between pollen abundance and climate
(Birks et al., 2010; Seppä et al., 2004), thus influencing the assumed
climatic optima and tolerances of pollen taxa in the model (Fig. 7). When
such a human-influenced calibration set is applied to a fossil record, which
represents mostly natural vegetation, a more or less serious bias in the
reconstructed past climate should be expected (St. Jacques et al., 2008; Xu
et al., 2010a). Using the H-set in this study, significantly lower
(
Modern HIIs capture basic characteristics of human influence on ecosystems
and allow for a quantitative evaluation of the human impact on the land surface
(Sanderson et al., 2002). Li et al. (2014) innovatively employed an HII to
establish a calibration set with pollen data and applied it to a 6200-year
fossil record from Lake Tianchi in central China (Zhao et al., 2010). The
pollen–HII calibration model (
A good correlation of the HII data with cereal-type Poaceae pollen in
northern China (Li et al., 2015) suggests that the HII can be seen as a
surrogate of indicator pollen taxa for human activities. However, cereal-type
Poaceae pollen generally has a very low abundance in a fossil sequence. For
example, the cereal-type Poaceae in a natural profile close to the
archaeological sites from Anyang, the centre of agricultural and societal
development during late Shang Dynasty, comprises only around 2 % of the
total pollen during the last 3400 years (Cao et al., 2010). In the sequence
from Lake Gonghai used in the current study, it contributes about 2–4 %
during the last 2 millennia (Fig. 2). Therefore, HII explains only
1.12 % of the variance in the N-set and 2.29 % in the H-set after
removing the cereal-type Poaceae and other distinct cultivars (Table 1). In
addition to cereal-type Poaceae, taxa such as
Reconstructing human influence quantitatively from fossil pollen data with a
direct pollen–HII calibration set might not be an easy task in most cases (Li
et al., 2014), but we can still use HII as an assessment tool in a
broad spectrum way. The reconstructed climate of a certain fossil sample is
mostly determined by its closest modern analogues even though different
approaches may have been used for the reconstruction (e.g. WA-PLS; Birks et
al., 1990). By examining the mean HII values at sites of the best modern
analogues, we can evaluate the bias in the climate reconstruction of the
corresponding fossil sample. A high analogue HII value indicates greater
potential bias in the reference samples. As shown in Fig. 6d, analogue HII
values in the H-set are usually higher than in the N-set, suggesting a higher
bias in the
Agriculture became the dominant subsistence strategy in today's (potentially) warm temperate forest region (III, including the central China Plains) and northern subtropical mixed forest zone (IVAi, including the Yangtze Plains) from about 6.5–5 ka BP (Crawford, 2011; Zhao, 2010). Potential human disturbance to the vegetation in eastern China since 6 ka BP has been inferred from many pollen studies (Ren and Beug, 2002; Wang et al., 2010), not to mention historical times (Cao et al., 2010; Zhao et al., 2010). Our analogue HII assessments indicate that the bias in the climate reconstruction induced from human impact via reference pollen samples or via changes in the fossil pollen assemblages (particularly during historical times; Li et al., 2014) is real. This raises the question of whether the Holocene climate could be quantitatively reconstructed using pollen data from eastern China. The answer is not a simple yes or no (Ren and Beug, 2002), although we keep an optimistic view based on the comparison results presented in this study.
Reconstructed
Regarding the potential human-induced bias in fossil records, the challenge
for pollen-based quantitative climate reconstructions is more from the lack
of natural surface samples in regions with intensive agricultural
activities. In eastern China, a calibration set only including preferenced
pollen samples from lake surface sediments with low human disturbance is
still not available (Liu et al., 2013), and surface samples are mostly
collected from mountains and steppe areas (Xu et al., 2010b). This means that
our current modern pollen datasets (Cao et al., 2014; Zheng et al., 2014)
still contain relatively few samples from central-eastern China. Collecting
extra samples from natural vegetation in mountain areas, such as
Luzhong, Qin, Dabie, and Qian, would help to improve the
pollen-based climate reconstructions for the region. The Qin Mountains, for
example, have a large and well-forested range (ca. 57 000 km
This paper attempts to assess the extent of bias induced from human impact in
pollen-based quantitative climate reconstructions. Numerical analyses suggest
that
Supplementary data are available at
The authors declare that they have no conflict of interest.
We are grateful to Chunhai Li for his pollen analyses of samples from the Jiangsu coastal plain and Cathy Jenks for her linguistic help. This work was supported by the key programmes of the National Natural Science Foundation of China (40730103 and 41630753). The doctoral research of Wei Ding at Freie Universität Berlin in the working group of Pavel E. Tarasov was funded by the China Scholarship Council (2011813072). Edited by: Dominik Fleitmann Reviewed by: two anonymous referees