Introduction
Atmospheric pollution is a major environmental problem in China, with
substantial adverse health effects (Yang et al., 2013; Apte et al., 2015).
Between 2006 and 2012, approximately 1.1 billion people (82 % of the
nation's population) live in areas where the yearly average mass
concentrations of fine particulate matter (PM2.5) exceeds
35 µgm-3 (Geng et al., 2015) – above interim target 1 for
annual average exposure set by the World Health Organization (WHO, 2005). In
turn, this magnitude of exposure has had large impacts on public health and
economic output. In 2010, PM2.5 pollution alone was linked to 1.2
million premature deaths in China (Yang et al., 2013), or ∼35 % of
all such deaths worldwide (Apte et al., 2015; Lim et al., 2012), with
associated economic losses equivalent to more than 6 % of China's GDP
(Matus et al., 2012). The distribution of air pollution and attendant impacts
vary across Chinese provinces due to differences in physical geography,
meteorology, population density, level of economic development, production
structure, and available technologies (Geng et al., 2015; Jiang et al., 2015;
Ma et al., 2014). For example, annual average PM2.5 concentrations
in northern China are roughly 1.5 times greater than the national average and
2 times greater than concentrations in southern China (Geng et al., 2015). In
light of these differences, the central and local governments have
established various goals, strategies, and measures for reducing air
pollution, with varying degrees of success (Lin et al., 2010).
Effective and efficient control of air pollution relies upon an
understanding of the pollution sources and their relative
environmental impacts. This has led to an increasing number of studies
aimed at attributing pollution to sources at high spatial, temporal,
and sectoral resolutions (e.g., Chambliss et al., 2014; Zhang et al.,
2015; Turner et al., 2015; Li et al., 2015). An important finding of
these studies is that regional air quality is in many cases strongly
influenced by pollution produced in other regions and transported in
the atmosphere across regional boundaries (e.g., Hu et al., 2015; Li
et al., 2015). For instance, a recent study found that during the
month of January 2013–2015, roughly half of the PM2.5
present in Beijing and Tianjin (47 and 55 %, respectively) was due
to emissions produced in other regions (Li et al., 2015). Recent work
has further investigated the effect of trade on air pollutant
emissions (Guan et al., 2014; Huo et al., 2014; Zhao et al., 2015;
Meng et al., 2016) and related impacts on air quality, health, and
climate (Takahashi et al., 2014; Jiang et al., 2015; Lin et al., 2014,
2016; Li et al., 2016b; Wang et al., 2017; Zhang et al., 2017). Here,
we assess for the first time the health impacts of trans-boundary
PM2.5 pollution and trade within China. Our results reveal
with greater detail than previously the health impacts of specific
economic activities (e.g., the production of raw materials and
intermediate goods, production of final goods, and consumption of
final goods) by region in China. This information may be used by
policymakers in the design and evaluation of control strategies that
account for cross-regional pollution.
Our analysis entails a novel coupling of physical, economic, and
epidemiological models that use the latest available data, from 2010.
Together, these models allow us to estimate premature deaths in China
due to local and trans-boundary anthropogenic PM2.5
pollution associated with three different economic activities
(production of raw materials/intermediate goods, production of final
goods, and consumption of final goods) for each of seven regions
(North, Yangtze River Delta, Southeast, Central, Northwest, Southwest,
and Northeast China; see Table A1 for region definitions).
Materials and methods
In this study, four state-of-the-art models were integrated to analyse
the drivers of PM2.5-related deaths across seven regions in
China. Figure 1 depicts the integrated assessment
framework in four steps. Below we describe in details for each step.
Schematic of the methodology used in this study.
Estimation of PM2.5-related premature deaths
Satellite-based ground-level PM2.5 mass concentrations at
a 0.5∘×0.667∘ resolution used in this study
were derived from our previous work (Geng et al., 2015). It was
estimated by using the aerosol optical depth (AOD) derived from
satellite instruments and conversion factors between AOD and
PM2.5, simulated by the GEOS-Chem chemical transport model
(Bey et al., 2001). The satellite-based AOD was generated by
combining results from MODIS and MISR instruments onboard the Terra
satellite, after being filtered by using ground-based AOD
measurements. The conversion factors between AOD and PM2.5
were calculated by the nested GEOS–Chem model over China at
a resolution of 0.5∘×0.667∘ (Chen et al.,
2009). In addition, for this simulation, anthropogenic emissions over China
were taken from the Multi-resolution Emission Inventory of China
(MEIC: http://www.meicmodel.org/), an updated version of the
technology-based, bottom-up pollution inventory developed by Tsinghua
University (Zhang et al., 2009; Lei et al., 2011; Liu
et al., 2015). Note that, in this study, the simulated proportion of
mineral dust in surface PM2.5 was subtracted to exclude the
impact of natural mineral dusts on premature deaths, and we assumed
that the contribution of dust to PM2.5 concentrations was
proportional to their previously estimated air pollution disease
burden (Chafe et al., 2014; Bhalla et al., 2014).
The integrated exposure–response (IER) model developed by Burnett
et al. (2014) is used in this work to describe the
concentration–response relationship between long-term exposure to
PM2.5 (annual mean values in this study) and premature
deaths for various leading causes. It is fitted by incorporating
information from cohort studies of ambient air pollution, second-hand
tobacco smoke, household solid, cooking fuel, and active smoking
(Burnett et al., 2014). The relative risk (RR) was
calculated as
RRl(C)=1+αl1-e-γlC-C0δl,ifC>C01,else,
where C is the annual mean PM2.5 concentrations in 2010;
C0 is the counterfactual concentration; l represents a given
health effect; and αl, γl, and δl are
parameters used to describe the shape of the concentration–response
curve (Burnett et al., 2014).
The RR was then converted to the attributable fraction (AF):
AF=RR-1RR.
The health outcomes or mortality attributable to PM2.5 was
then estimated:
M=AF×B×P,
where B is the death incidence of a given health effect derived from the
national average data in GBD2013 (Forouzanfar et al., 2015); P is the size
of the exposed population obtained from the LandScan global population
database (Bright et al., 2011).
Following the previous studies (Jiang et al., 2015; Lee et al., 2015),
in this work we mainly focus on four leading causes of the
PM2.5-related premature deaths: ischemic heart disease
(IHD), stroke, chronic obstructive pulmonary disease (COPD), and lung
cancer (LC). In addition, according to the Global Burden of Disease (GBD)
projects (Forouzanfar et al., 2015), we assume that these
PM2.5-related health impacts are source and composition
independent. Data for C0 and B can be found in Table A2.
Contribution of individual source emissions to regional premature deaths
Following Lee et al. (2015), the GEOS-Chem adjoint (version 3.5, driven
by MEIC inventory) model combined with the IER model were applied over
East Asia (11∘ S–55∘ N,
70–150∘ E) at a resolution of 0.5∘×0.667∘ to calculate the contributions of location- and
species-specific emissions to PM2.5-related premature deaths
in individual regions (see region definitions in Table A1).
Firstly, we defined Eq. (3) as the adjoint cost function (or
concentration-dependent function) of a given region (e.g., region r),
and the total value of Mr is with respect to the
satellite-based PM2.5 in region r obtained from Geng et al.
(2015). We then used the adjoint model to calculate partial
derivatives of this cost function with respect to anthropogenic
emissions of individual species (∂Mr∂Ei,j,k), which we referred to as the sensitivity of
region r's premature deaths (Mr) to gridded emissions
(Ei,j,k; i, j, and k are indices for longitude, latitude,
and species) in the simulation domain. A single adjoint simulation
provided sensitivities of Mr with respect to emissions
at all species, locations, and times (Lee et al., 2015; Pappin and
Hakami, 2013; Turner et al., 2015). After computing the model
sensitivities, we multiplied the emission sensitivity by the amount of
emissions to obtain a semi-normalized sensitivity (SS) (Henze et al.,
2007, 2009; Turner et al., 2015; Dedoussi and Barrett, 2014), which
means the contribution of species- and location-specific emissions to
the premature deaths (Turner et al., 2015; Dedoussi and Barrett,
2014):
SSi,j,kr=∂Mr∂Ei,j,k×Ei,j,k.
Then, a normalized SS (hereafter P), which represents the percentage
contribution of source-specific emissions to premature deaths was
calculated as
Pi,j,kr=SSi,j,kr∑i∑j∑kSSi,j,kr×100%.
The normalization process minimizes the effects of nonlinear relation
between emissions and pollutant concentrations and between
concentrations and mortality. A similar approach was taken by Li
et al. (2016a) for attributing ozone radiative forcing to individual
countries.
For this work, we calculated responses to absolute changes in
NH3, SO2, NOx, BC, OC, and anthropogenic
PM2.5 dust (Zhang et al., 2015). Moreover, a total of seven groups
of GEOS-Chem adjoint model simulations were conducted, one group for
each receptor region. In order to reduce the computation costs, 4
months (January, April, July, and October of 2010) of simulations were
conducted for each group. Results for these 4 months are averaged
to represent the annual mean SS in 2010.
Regional pollutant emissions attributed to regions
producing final goods and regions consuming the final goods
The production of a specific product or service represents one stage
in a supply chains because such production requires material and
energy inputs and may in turn supply other production processes (i.e.,
the products are intermediate) or final sales (i.e., the products are
finished goods ready for final consumption) (Davis et al.,
2011). Using the 30-province, 30-sector multi-regional input–output
(MRIO) model of China compiled by Liu et al. (2014), we attribute the
emissions released in a region (i.e., the producer) to both final
produces in supply chains (who produced the finished products using
intermediate inputs made locally or imported from other regions, here
we call this regions as “assembler”) and final consumers (who
ultimately consume the finished products).
The MRIO analysis starts with the monetary flows between sectors and regions:
x1x2x3⋮xm=A1,1A1,2A1,3⋯A1,mA2,1A22A2,3⋯A2,mA3,1A3,2A3,3⋯A3,m⋮⋮⋮⋱⋮Am,1Am,2Am,3⋯Am,mx1x2x3⋮xm+∑sy1,s∑sy2,s∑sy3,s⋮∑sym,s,
where xr is a vector of the total economic output
of each sector in province r, yr,s is a vector of
the finished products by each sector produced in region r and consumed
in region s, Ar,s is a normalized matrix of
intermediate coefficients in which the columns reflect the input from
the sectors in region r required to produce one unit of output from
each sector in region s, and m is the total province number (here m=30). Solving for total output, the equation can be written as
x=I-A-1y,
where I is identity matrix, A is the block
matrix in Eq. (6), and (I-A)-1 is the
Leontief inverse matrix.
Under this framework, pollutant emissions embodied in the trade flow
can be calculated as
e=f^I-A-1y,
where f^ is the diagonalization of the vector of region-specific
pollutant emissions for unit output of each sector.
Region- and sector-specific emissions attributed to assembler region s
can be calculated as follows:
easses=f^I-A-10⋮∑rys,r⋮0
where
easses=(easse1,seasse2,seasse3,s…eassem,s)', and
easser,s is a sector-specific vector
for emissions occurred in region r caused by producing intermediate
products to be assembled (as finished products) in region s.
Region- and sector-specific emissions attributed to consumer region s
can be calculated as
econss=f^I-A-1y1,sy2,sy3,s⋮ym,s,
where
econss=(econs1,secons2,secons3,s…econsm,s)'; and
econsr,s is a sector-specific vector
for emissions occurred in region r caused by production of
intermediate or final products to be consumed in region s.
In this section, sector-specific emissions to produce f in
Eqs. (8) to (10) were derived from mapping process between MEIC
model and sectors defined in the MRIO model for each provinces, which
can be found in our previous studies (Huo et al., 2014; Zhao et al.,
2015). Within each region, emissions attributed to each activator
(assembler or consumer) can be allocated to individual locations (grid
cells) based on the sector-spatial distribution in MEIC and the
attributed ratios:
Rk, asser,s=ek, asser,s/ekr,Rk, consr,s=ek, consr,s/ekr,
where ekr is the sector-specific emissions
vector (species k) produced in region r, and Rk, asser,s and Rk, consr,s are
sector-specific ratios of emissions occurred in region r but allocated
to region s from assembler and consumer perspectives, respectively. As
part of our calculation, we aggregated the interregional emission
impact of 30 provinces into 7 regions, as defined in Table A1.
Anthropogenic PM2.5-related premature deaths in
China for 2010 at the 0.5∘×0.667∘
horizontal resolution (a), and regional premature deaths
attributed to regions where emissions were produced (b) and
regions where products were ultimately consumed (c). Datasets at the end of each bar mean the percentages of regional
premature mortality attributed to local source and the percentages
attributed to other regions. Y.R.D. is the Yangtze River Delta.
Premature deaths attributed to regions producing final
goods and regions consuming the final goods
Results from above three steps were integrated to attribute regional-
and source-specific PM2.5 deaths to specific economic
activities (i.e., the production of final goods by the “assembler”,
and the ultimate consumption of those goods) in specific regions along
supply chains as
Masses=∑rMr∑t∑k(P(i,j)∈t,kr×Ri,j∈t,k,asset,s),Mconss=∑rMr∑t∑kP(i,j)∈t,kr×R(i,j)∈t,k,const,s,
where Masses and
Mconss mean premature deaths attributed to
region s from assembler and consumer perspectives, respectively;
Ri,j∈t,k,asset,s and
Ri,j∈t,k,const,s are
sector average ratios of emission occurred in grid (i, j)
relocated to region s from assembler and consumer perspectives,
respectively.
Results
National and regional mortality attributed to anthropogenic PM2.5
In 2010, China's population-weighted PM2.5 concentrations
caused by anthropogenic emissions reached 53 µgm-3,
leading to 1.02 (95 % CI: 0.64–1.22) million premature deaths,
which accounted for about 35 % of the global total mortality from
ambient PM2.5 (Apte et al., 2015). Adding another 0.23
(95 % CI: 0.14–0.27) million premature deaths from windblown
natural dusts, our estimate of premature deaths is within 2 % of
the result of GBD 2010 for China (1.27 million premature deaths;
Lim et al., 2012).
Table 1 and Fig. 2a show details of regional anthropogenic
PM2.5 concentrations and related mortality. As figures and
table shown, atmosphere pollution and related health impact vary
substantially across the seven Chinese regions. Dominated by heavy
industries, the North region suffered the most severe pollution, and its
population-weighted mean PM2.5 concentrations reached 82
µgm-3, followed by Central (67
µgm-3), Southwest (52 µgm-3), and
Yangtze River Delta (50 µgm-3). However, considering
the total population exposed to pollution, the Central region had the
highest mortality (302 200 premature deaths; 95 % CI:
187 400–359 900) and a high mortality ratio (90 deaths per
105 people; 95 % CI: 56–107), followed by Southwest
(195 200 premature deaths; 95 % CI: 121 900–234 600) and North (182 200 premature deaths; 95 % CI: 115 700–213 600).
Regional populations, PM2.5 concentrations, mortality, and
mortality ratios within China.
Region
Population
Population weighted
Mortality
Mortality ratio
(millions)
PM2.5 concentrations
(thousands of deaths)
(deaths per 105 persons)
(µgm-3)
North
192
82
182 (116–214, 95 % CI)
95 (60–111.95 % CI)
Yangtze River Delta
144
50
116 (76–153)
81 (50–99)
Southeast
143
27
78 (52–98)
55 (35–67)
Central
337
67
302 (209–415)
90 (56–107)
Northwest
157
34
80 (62–118)
51 (31–61)
Southwest
254
52
195(132–254)
77 (48–92)
Northeast
114
28
65 (42–82)
57 (35–70)
National average/total
1354
53
1018 (636–1222)
76 (47–91)
Effects of atmospheric transport of air pollution on
regional mortality
Regional atmospheric pollution and related health impacts can be
attributed to emissions from both local and other regions as a result
of atmospheric transport. Further, emissions in a given region can
also be attributed to regions who consuming the related products due
to trade; thus, pollution-induced mortality can finally be attributed
to the consuming regions. Table 2 (and Fig. 2b and c) presented the
source attribution of regional PM2.5 exposure and related
premature deaths from both production and consumption perspectives.
Regional population-weighted mean PM2.5 concentration in 2010
and related premature deaths from production and consumption perspectives.
Each number in the cell shows the population-weighted mean PM2.5
concentration or premature deaths in the region indicated by the column due
to pollution emitted or goods consumed in the region indicated by the row.
Numbers in the parentheses represent the fraction of population-weighted mean
PM2.5 concentration or premature deaths.
Region
North
Y.R.D
Southeast
Central
Northwest
Southwest
Northeast
Population-weighted
81.6
50.2
27.4
67.0
34.0
51.9
27.6
mean PM2.5
concentration
(µgm-3)
Region
North
57.1(70 %)
8.9(17.6 %)
1.4(5 %)
9.4(14 %)
2.4(7.1 %)
1.4(2.7 %)
4.1(14.7 %)
where
Y.R.D
5.3(6.5 %)
30.1(60 %)
2.4(8.8 %)
6.2(9.2 %)
0.3(0.9 %)
0.5(1 %)
0.6(2.1 %)
pollution
Southeast
0(0.1 %)
0.6(1.2 %)
18.7(68.4 %)
0.8(1.2 %)
0(0 %)
0.7(1.3 %)
0(0 %)
emitted
Central
10.9(13.4 %)
6.2(12.4 %)
3.3(11.9 %)
44.8(66.8 %)
5.8(17.1 %)
6.4(12.3 %)
0.7(2.5 %)
Northwest
5.5(6.8 %)
2.6(5.1 %)
0.7(2.4 %)
4(6 %)
21.8(64.2 %)
2.7(5.3 %)
2.2(8 %)
Southwest
0.2(0.3 %)
0.2(0.4 %)
0.5(1.8 %)
1.2(1.8 %)
3.2(9.5 %)
40(76.9 %)
0(0.1 %)
Northeast
2(2.5 %)
1.2(2.4 %)
0.2(0.8 %)
0.5(0.8 %)
0.2(0.7 %)
0.1(0.1 %)
19.6(70.9 %)
Out of China
0.4(0.5 %)
0.5(1 %)
0.2(0.9 %)
0.2(0.3 %)
0.2(0.5 %)
0.2(0.3 %)
0.5(1.7 %)
Region
North
35.5(43.5 %)
7.4(14.7 %)
1.6(5.8 %)
8.5(12.7 %)
3.8(11.2 %)
2.9(5.5 %)
3.8(13.8 %)
where
Y.R.D
9(11 %)
17.8(35.5 %)
2.5(9.3 %)
7.6(11.4 %)
2.3(6.8 %)
2.1(4.1 %)
1.7(6.3 %)
goods
Southeast
1.9(2.3 %)
1.6(3.1 %)
10.4(38.2 %)
2.6(3.9 %)
1(3 %)
2.4(4.5 %)
0.4(1.6 %)
consumed
Central
9.5(11.6 %)
5.6(11.1 %)
3.1(11.3 %)
30(44.8 %)
4.8(14.1 %)
6.1(11.7 %)
1.3(4.6 %)
Northwest
6.8(8.4 %)
3.4(6.8 %)
1.2(4.5 %)
4.9(7.4 %)
14.1(41.5 %)
3.2(6.2 %)
2.2(7.9 %)
Southwest
2.1(2.5 %)
1.2(2.4 %)
1.3(4.6 %)
2.8(4.2 %)
3.2(9.5 %)
29.7(57.1 %)
0.4(1.6 %)
Northeast
4.3(5.3 %)
2(4.1 %)
0.5(1.8 %)
1.8(2.7 %)
1.1(3.2 %)
0.7(1.3 %)
13.9(50.2 %)
Out of China
12.1(14.8 %)
10.7(21.4 %)
6.5(23.7 %)
8.5(12.7 %)
3.5(10.2 %)
4.8(9.2 %)
3.4(12.3 %)
Premature mortality
1822
1160
785
3022
796
1952
649
(100 person)
Region
North
1258(69.1 %)
182(15.7 %)
44(5.7 %)
381(12.6 %)
64(8.1 %)
69(3.6 %)
97(15 %)
where
Y.R.D
130(7.1 %)
727(62.7 %)
79(10.1 %)
315(10.4 %)
8(1 %)
33(1.7 %)
14(2.2 %)
pollution
Southeast
1(0.1 %)
24(2.1 %)
516(65.8 %)
63(2.1 %)
0(0 %)
49(2.5 %)
0(0 %)
emitted
Central
202(11.1 %)
135(11.6 %)
93(11.8 %)
2008(66.5 %)
130(16.3 %)
343(17.6 %)
17(2.6 %)
Northwest
135(7.4 %)
44(3.8 %)
21(2.7 %)
145(4.8 %)
480(60.3 %)
82(4.2 %)
54(8.4 %)
Southwest
5(0.3 %)
5(0.5 %)
15(2 %)
76(2.5 %)
91(11.4 %)
1356(69.5 %)
1(0.1 %)
Northeast
74(4.1 %)
30(2.6 %)
7(0.9 %)
24(0.8 %)
16(2 %)
4(0.2 %)
453(69.8 %)
Out of China
17(0.9 %)
13(1.2 %)
8(1.1 %)
9(0.3 %)
7(0.8 %)
15(0.8 %)
13(2 %)
Region
North
777(42.6 %)
155(13.4 %)
48(6.2 %)
355(11.8 %)
84(10.6 %)
112(5.8 %)
91(14 %)
where
Y.R.D
204(11.2 %)
426(36.7 %)
78(9.9 %)
360(11.9 %)
53(6.6 %)
93(4.8 %)
42(6.4 %)
goods
Southeast
41(2.2 %)
43(3.7 %)
294(37.5 %)
137(4.5 %)
22(2.8 %)
109(5.6 %)
10(1.6 %)
consumed
Central
188(10.3 %)
124(10.6 %)
88(11.2 %)
1349(44.7 %)
109(13.7 %)
294(15.1 %)
31(4.7 %)
Northwest
158(8.7 %)
71(6.1 %)
36(4.6 %)
202(6.7 %)
326(40.9 %)
109(5.6 %)
52(8.1 %)
Southwest
44(2.4 %)
28(2.4 %)
35(4.5 %)
141(4.7 %)
85(10.7 %)
1001(51.3 %)
10(1.6 %)
Northeast
116(6.4 %)
47(4 %)
15(1.9 %)
78(2.6 %)
31(3.8 %)
26(1.3 %)
322(49.6 %)
Out of China
277(15.2 %)
253(21.8 %)
182(23.2 %)
389(12.9 %)
79(10 %)
193(9.9 %)
77(11.9 %)
As shown in Table 2 and Fig. 2b, in the year 2010, 33 % of total
premature deaths due to outdoor PM2.5 exposure were caused
by trans-boundary pollution, and ratios for specific regions vary from
30 % in Northeast to 40 % in Northwest. Among these, less than
1 % was caused by pollution transported from region out of
China. Figure 3a further shows the effect of atmospheric transport on
premature deaths in each Chinese region due to PM2.5 air
pollution produced in other regions, with particularly large
interregional impacts highlighted by arrows. The red shading in
Fig. 3a corresponds to regions (e.g., North, Yangtze River Delta,
and Northwest) whose emissions caused a greater number of deaths in
other regions than pollution in other regions caused in that region –
a net export of premature mortality. In contrast, the blue-shaded
regions (Southwest, Southeast, and Central) experienced greater
numbers of deaths due to extra-regional emissions than their emissions
caused in other regions. Regionally, pollution from the North region
that was transported in the atmosphere to the populous Central and
Yangtze River Delta regions is particularly harmful and causes the
most premature deaths, with 38 100 (95 % CI: 23 600–45 400)
and 18 200 (95 % CI: 11 200–22 300) premature deaths related
to these trans-boundary flows, respectively. Perhaps due to its
substantial emissions and central location in the country, premature
deaths occurred in the Central region by emissions produced elsewhere
(101 400; 95 % CI: 62 900–120 700) and deaths caused by the
Central region's emissions transported elsewhere (91 900; 95 % CI: 57 500–110 400) are approximately equal, and Southwest
experienced
the most premature deaths from emission in Central. Nationally, the
net flows of trans-boundary PM2.5-related health impact
mainly caused by pollution transported from north to south and from
east to west.
The effect of atmospheric transport (a) and trade
(b–d) on each region's PM2.5-related premature
deaths. Panel (a) compares the number of premature
mortality related to emissions produced in each region and deaths
that occurred in that region. Panel (b) and (c)
compare regional production-related premature deaths with deaths
related to production of final goods in that region, and deaths
related to consumption of goods and services in that region,
respectively. Panel (d) compares the number of premature
deaths occurred in each region with deaths related to consumption of
goods and services in that region. Deaths in other regions due to
Chinese pollution and deaths due to emissions in other nations are
not included in any of the maps, and international export on
regional premature deaths are not included in map (c) and
(d). Arrows between regions denote the largest
interregional transfers, with numbers of displaced premature deaths
shown in thousands.
Effects of interregional trade on regional mortality
Compared to the physical atmospheric transport, trade leads to more broad cross-regional impact (Table 2 and Fig. 2c), as the
production of emissions can occurred far from where the products were
finally consumed. Nationally, 56 % of PM2.5-related
premature mortality in China in 2010 was linked to consumption in
a different region through both interregional (within China) and
international trade activities. Among these, 42 % of premature
mortality was associated with domestic consumption in other regions, and the ratios vary
from 38 % in Northeast to 49 % in Northwest. International
export accounted for approximately 14 % of total
PM2.5-related premature mortality in China in 2010,
comparable to 12 % in 2007 reported by Jiang et al. (2015).
For a finished product or service, it may experience different stages
before being sold to final consumers, such as material production and
products assemble, these may occurred in different regions. Figure 3b
shows the effect of trade between the region producing a raw material
or intermediate good and the region producing the final good ready for
consumption. This is important because in many cases the region
assembling or otherwise preparing the final good is able to capture
a large fraction of the final good's value without undertaking more
energy- and pollution-intensive processes that were required to
produce the raw materials and intermediate goods (Prell et al., 2014;
Liu et al., 2016). This distinction is particularly relevant in China
because previous studies have shown that more affluent coastal
provinces in China are increasingly importing intermediate goods and
materials from less-developed provinces (Feng et al., 2013; Jiang
et al., 2015; Zhao et al., 2015). Here, we find that premature deaths
related to final goods produced in red-shaded regions are
substantially greater than the deaths due to the emissions produced in
those regions. In particular, final goods assembled/manufactured in
the Yangtze River Delta and Southeast regions led to 56 200 (95 % CI: 35 100–67 300) and 33 600 (95 % CI: 20 900–40 200)
deaths due to emissions in other regions, respectively. In contrast,
blue-shaded regions like Central, Northwest, and Southwest are
those which disproportionately produce and export raw materials and
intermediate goods (e.g., mineral ores and metals) and therefore
suffer health impacts to support the manufacture of final goods in
other regions. For example, 15 % of deaths caused by emissions in
the Central region are related to final goods manufactured in the
North, Yangtze River Delta, and Southeast regions.
With supply chains or trade extending, finished products may finally
be consumed by another region. Figure 3c further shows the full effect
of trade from the producer of emissions-related deaths to the final
consumer in map. As this study does not include premature deaths
caused by international imports, in this figure we only present
regional production-related deaths caused by domestic consumption,
premature deaths induced by goods and services produced in China for
international export are shown in Fig. 4 separately. As Fig. 3c shows,
deaths related to consumption in red-shaded regions are substantially
greater than the deaths number caused by emissions produced in those
regions. Note that, in this figure, the Central region shows net export of
production-related premature deaths with all other six regions, this
can be attributed to its abundance interregional export, severe
pollution, and high population density. Moreover, Fig. 3c also
highlights the case of Northeast. Even though Northwest shows net
pollution export with other regions (Zhao et al., 2015), but its
exported emissions cause fewer deaths than the relative small emissions
occurred on other regions to support consumption in Northeast, just
because that population and production intensities in Northeast are
far less than those in other regions, such as the Central and North
regions.
Flow map of premature deaths connecting producers and
international exporter of finished products. The percentage values
are relative to national total anthropogenic PM2.5-related
premature deaths (1.02 million).
Source of PM2.5-related premature mortality in each
region (a) and their “spillover” source by producing
regions (b) (i.e., the gray bars in (a): deaths due to emissions
in other regions related to goods and services consumed in other
regions).
Figure 3d shows the combined effect of atmospheric transport and trade
on each region. As in Fig. 3c, deaths caused by international export
were excluded in this map. Here, premature deaths related to final
goods consumed in red-shaded regions are substantially greater than
the deaths occurred in those regions. Even though similar with
Fig. 3c, Fig. 3d shows that atmospheric transport aggravated the
premature deaths transferred from North and Y.R.D to the Central region.
Figure 4 attributed China's PM2.5-related premature deaths
embodied in international export to seven Chinese regions where the
emissions were produced and where the final products were exported. As
shown, of these international exports, roughly three-quarters
(76 %) of the related deaths are associated with the exports from
the east coast regions (North, Yangtze River Delta, and
Southeast). However, only 59 % of deaths related to exports from
these coastal regions are caused by emissions actually produced in
those regions, and even fewer (49 %) of associated deaths actually
occurred in those regions (Fig. 2b). These results emphasize that
international exports commonly entail intermediate inputs from
less-developed regions of China (e.g., the Central region; Feng
et al., 2013; Zhao et al., 2015).
Summing from the previous sections, Fig. 5 integrates the results
related to both atmospheric transport and trade to show
PM2.5-related premature mortality in each region due to
local and other regions' manufacturing and consumption activities, and
separating the effects of locally produced and
atmospherically transported pollution. For a given region, emissions
produced in the region to supply either local or other regions'
consumption accounted for the largest share of deaths in the region
(60–70 %; purple and light-blue bars in Fig. 5a), followed by the
“spillover impact” of emissions produced in other regions that are
not related to the local region's manufacturing or consumption
activities (27–37 %; gray bars in Fig. 5a). Emissions produced in
other regions and related to the local region's consumption
contributed 1–3 % of each regions' mortality, via atmospheric
trans-boundary transport (dark-blue bars in Fig. 5a). Finally, the
effect of atmospheric transport from other countries contributed only
1–2 % of deaths in any Chinese region (light-purple bars in
Fig. 5a).
Figure 5b further breaks down the regions involved in each region's
spillover impacts according to where the emissions were produced. The
magnitude of spillover deaths depends largely on a given region's
population and the extent of emissions in their upwind regions. For
instance, 75 and 66 % of spillover deaths in the Yangtze River
Delta and Southwest regions are linked to emissions in upwind regions
(primarily the Central and North regions), respectively. As the most
populated region, the Central region suffered the most spillover
deaths (96 100; 95 % CI: 59 600–114 400), 69 % of which
were related to emission produced in the North and Yangtze River Delta
regions.
Comparisons between the simulated and satellite-derived
PM2.5 concentrations over the seven China regions.
Uncertainties and limitations
The calculation of premature deaths caused by atmospheric transport
and trade is subject to a number of uncertainties and
limitations. Bottom-up emission inventories are uncertain due to
incomplete knowledge of activity, technology distribution and emission
factors. The uncertainties in China's emission inventory were
estimated to be -14–13, -13–37, -17–54, -25–136, and
-40–121 % for SO2, NOx,
PM2.5, black carbon (BC), and organic carbon (OC),
respectively (Zhao et al., 2011). Although the quantitative
uncertainties are not provided by the MEIC inventory, it has been
widely used in chemical transport models and validated against surface
and satellite observations (e.g., Chen et al., 2015; Geng et al.,
2015; Li et al., 2015; Zheng et al., 2015; Hu et al., 2016).
Uncertainties from the simulation of GEOS-Chem and its adjoint subject
to their limitations or errors in chemical and physical
representation, such as the chemical conversion, diffusion,
deposition, and advection transport. Here we conduct a comparison of
the modeled and the satellite-derived PM2.5 concentration
and use the normalized mean error (NME) between these two datasets
over China seven regions to represent the overall model errors. As
shown in Fig. 6, the NME varies among seven regions, ranging from
the lowest in North (30 %) to the highest in Northwest
(71 %). Note that the two datasets agree reasonably well, with
R ranging from 0.67 to 0.95 for seven regions. This provides
confidence that the results of this study are based on realistic
simulation. The adjoint model may introduce additional uncertainties
due to lack consideration of the nonlinear response of the predicted
concentration to perturbation of emission input. However, due to its
complex in backward calculation and integration with the forward
model, there are very few statistical quantification for its
uncertainties so far. Lee et al. (2015) used ±40 % to
represent the total uncertainties caused by the GEOS-Chem adjoint
model.
Uncertainties in satellite-derived PM2.5 map is ±5 %
on average according to GBD 2013 (Brauer et al., 2016), as it has been
calibrated by satellite-based and surface observations. Uncertainties
in IER model are relatively high, mainly arising from the model
itself, as it is fitted by limited information on actual exposure to
PM2.5 for source-specific relative risks. Burnett
et al. (2014) estimated the uncertainties from
IER model by using simulation approach, and they fitted out 1000 sets
parameters for the IER function to represent its possible
shape. Additionally, the IER model is limited to several assumptions,
e.g., PM2.5-related health impact is independent of exposure
period, PM2.5 composition, and toxicity for particles from
different sources (Burnett et al., 2014; Jiang
et al., 2015; Lee et al., 2015).
Additional uncertainties originate from MRIO analysis when linking
trade among different regions. MRIO model inherit all uncertainties in
its source (survey) data and data manipulation (Peters, 2007; Weber,
2008; Wiedmann et al., 2011; Wiedmann, 2009). In addition, MRIO
analysis is limited to sector detail, region coverage, and the number
of environmental extensions (Tukker and Dietzenbacher,
2013). Moreover, the China domestic MRIO model, which considers no
effect from international import (Hummels et al., 2001), can also
introduce some uncertainties. The study conducted by Lin et al. (2014)
concluded that the uncertainties in Chinese input–output model
contributed to ∼10 % of total errors in export-related
pollutant emissions.
A comprehensive uncertainty analysis combining all affecting factors
above is difficult due to the limitations of the computational
loads. The uncertainty ranges presented in previous sections only
represent the uncertainties in IER function, which is obtained by 1000
sets runs of IER parameters fitted by Burnett
et al. (2014) to calculate the possible
distribution of regional premature mortality.
Discussion and conclusions
Patterns of atmospheric PM2.5 pollution and resulting premature
deaths in China are the result of complex interacting physical transport
processes and economic activities (Lin et al., 2014). We found that, in 2010,
about one-third of PM2.5-related premature mortality in China was
caused by regional air pollution transport. In the meanwhile, large numbers
of premature deaths are caused by economic activities in a different region
from where the deaths occurred. For the year 2010, 42 and 14 % of
PM2.5-related premature deaths were associated with
domestic consumption in a different region and international
trade respectively. More economically developed regions (e.g., the Yangtze
River Delta, Southeast, and North regions) tend to externalize their
emissions and related health impacts by importing goods from less
economically developed regions (e.g., the Central region), and the frequent
wind from north to south and from east to west further aggravated the
cross-regional impacts. Thus, relocating emissions within the nation will not
completely alleviate the environmental and health burden; atmospheric
transport of pollution often leads to health impacts in downwind regions. To
reduce pollution and relative health impacts effectively, regions should
promote interregional technology cooperation, including both production and
emission control technologies. Further, as main final consumers, the east
coast regions can lead a “greening supply chains” action by importing more
green products, thus exerting a cleaning effect on its upstream production
chains (Skelton, 2013).
As a main driver of China's production, international export accounted
for 14 % of China's anthropogenic PM2.5-related
premature deaths in 2010. Moreover, its impact was not evenly
distributed among regions, as the developed eastern coastal regions
partly transfer their export-related premature mortality to the less
developed central and west regions by importing raw material from the
less developed central and west regions (Figs. 3b and 4). This
exerts disproportionally life loss and economic gains from
international exports among regions (Jiang et al., 2015). Moreover,
the added pollution results from international export can further
affect other countries atmospheric environment (Lin et al., 2014,
2016) or even premature deaths (Zhang et al., 2017) through
cross-continental atmospheric transport. Thus, a jointed pollution
mitigation action among regions, nations, and even production chains is
in urgent needed, not only for domestic equality in development but
also for global human health.
Our results represent the most detailed analysis of air pollution
mortality in China, its sources, and its underlying economic
drivers. Based on these findings, future measures to alleviate these
health impacts could be prioritized according to the source and
location of emissions as well as the type and economic value of the
emitting activities and related patterns of consumption.