1
Global climate forcing of aerosols embodied in international trade
Jintai Lin, Dan Tong, Steven Davis, Ruijing Ni, Xiaoxiao Tan, Da Pan, Hongyan Zhao,
Zifeng Lu, David Streets, Tong Feng, Qiang Zhang, Yingying Yan, Yongyun Hu, Jing
Li, Zhu Liu, Xujia Jiang, Guannan Geng, Kebin He, Yi Huang, Dabo Guan
Table of Contents
Supplementary Information Methods
............................................................................
2
S1. Production-based emissions
.................................................................................
2
S2. Consumption-based emissions
.............................................................................
5
S3. Gridded monthly emissions
..................................................................................
7
S4. Atmospheric evolution and transport simulated by GEOS-
Chem
.......................
7
S5. Radiative forcing calculations using RRTMG
.....................................................
9
S6. Uncertainties and limitations
..............................................................................
11
References
....................................................................................................................
14
Supplementary
Information Tables
..............................................................................
24
Supplementary Information Figures
............................................................................
26
Global climate forcing of aerosols embodied in
international trade
SUPPLEMENTARY INFORMATION
DOI: 10.1038/NGEO2798
NATURE GEOSCIENCE
| www.nature.com/naturegeoscience
1
© Macmillan Publishers Limited
, part of Springer Nature
. All rights reserved
2
Supplementary
Information
Methods
S1. Production
-
based emissions
Emissions of species other than NH
3
Production
-
based emissions represent pollutants physically released in each
region, and they are calculated as the product of
emission factors and activity rates. A
country
-
specific E
p
inventory in 2007 for SO
2
, NO
x
, CO, BC and POA is built for this
study. The inventory uses a detailed technology
-
based methodology as in previous
studies
32
-
34
, and it covers 65 sectors and 228 coun
tries/regions worldwide. Global
anthropogenic emissions in 2007 are estimated at 101.7 Tg for SO
2
, 95.3 Tg for NO
x
,
532.1 Tg for CO, 5.8 Tg for BC, and 28.8 Tg for POA. [Here emitted POA is 2.1 times
as much as organic carbon, after accounting for the oxyg
en atoms contained, consistent
with the assumption in GEOS
-
Chem.] After the emission data are derived, they are
further mapped to the 129 countries/regions and 57 sectors defined in the Global Trade
Analysis Project version 8 (GTAP8)
24
, in order to facilit
ate the subsequent calculation
of consumption
-
based emissions. Emission factors and activity data are described as
follows.
Activity data
:
We take the country
-
based fuel consumption data
from the
International Energy Agency (IEA)
3
5,36
for 46 sub
-
sectors and 51 fuels in four major
sectors
(residential, industry, power, and transportation)
. We further aggregate these
fuels into
19
types, considering that emissions related to certain fuels in the
IEA
database are small and their emission
factors are not available
27,
3
2
. For Greenland, there
© Macmillan Publishers Limited
, part of Springer Nature
. All rights reserved
3
are no fossil fuel data in the IEA database, thus we use the data compiled in the United
States Energy Information Administration (http://www.eia.gov/). We then divide the
fuel use in each sector by diff
erent technologies (four technologies in the power sector,
10 in industry, 21 in transportation, and 11 in the residential sector)
3
2
-
3
4
,
37
.
The
technology distributions for various vehicle types follow previous studies
38,39
.
Biofuel
combustion technologies in
the residential sector follow
the
Greenhouse Gas and Air
Pollution Interactions and Synergies (GAINS, http://gains.iiasa.ac.at/models/) model.
In addition, we include 16 non
-
combustion industrial process sectors, taking production
data
from
the
United
Sta
tes
Geological
Survey
statistics
(USGS,
http://minerals.usgs.gov/minerals/pubs/myb.html) and United Nations data (UNdata,
http://data.un.org/).
Emission factors: We compile emission factors from a wide variety of literature
and our previous works, includi
ng using data from
reliable regional inventories
to
calibrate the emission factors
for China, India, Southeast Asia, Canada, the United
States, and Europe.
Emission factors for SO
2
from
fuel combustion
follow our previous
works
27,40
-
43
.
For the non
-
combust
ion sources, we take the unabated SO
2
emission
factors from the public databases
44,45
, and then we follow our previous studies
27,37,41,42
to employ the flue gas desulfurization
application rates and corresponding SO
2
removal
efficiencies. We use r
egional
e
mission inventories
27,37,40,46,47
to calibrate the SO
2
emission factors
for China, India, Southeast Asia, Canada, the United States, and
Europe
. Emission factors of NO
x
and CO follow
Yan et al.
39
for on
-
road vehicles and
several public databases
44,45,48
fo
r other sources; we further replace the global defaults
© Macmillan Publishers Limited
, part of Springer Nature
. All rights reserved
4
by regional emission databases where
available
3
3
,49
-
53
. For
BC and POA, we take the
emission factor
s
from Bond et al.
3
2
,54
, except that we follow Yan et al.
38,39
for on
-
road
vehicles, Lam et al.
55
and Huang et al.
56
for residential kerosene, and
our previous
works
27,46
for all e
mission factors of China and India.
Emissions of NH
3
An additional country
-
based E
p
data base is built here for NH
3
in 2007. Emissions
are calculated for 129 countries/regio
ns and 57 sectors (13 for agricultural activities)
defined in GTAP8
24
. We combine the global EDGAR inventory and several existing
regional inventories that often have more detailed sectoral information to facilitate a
global supply chain analysis (see Supp
lementary Information Table 2). For example,
although agriculture accounts for 96% of global anthropogenic NH
3
emissions, there
are 2
–
3 agricultural sectors only in EDGAR and other global inventories, whereas much
more information is available in the regio
nal inventories for the United States, China,
and Europe. For regions other than the United States, China and Europe, agricultural
emissions are often sorted in the inventories according to sources (e.g. fertilizer,
compost, and manure) instead of sectors.
In this case, we map the source
-
based
emissions to individual agricultural sectors, using as weighting functions the regionally
aggregated (over regions other than China, the US, and Europe) contributions of
individual agricultural sectors from MASAGE.
C
omparison with HTAP v2.2
© Macmillan Publishers Limited
, part of Springer Nature
. All rights reserved
5
The scatterplot in Supplementary Information Fig. 2 compares the E
p
inventory
built here with the HTAP v2.2 inventory for 2008
57
. HTAP v2.2 was recently developed
from an internationally collaborative project, and it combines the E
DGAR inventory
58
with regional inventories in Asia, North America, and Europe. HTAP v2.2 is thus
expected to be more updated than EDGAR and other older global inventories.
Supplementary Information Fig. 2 shows that total E
p
emissions in our inventory are
in
line with HTAP v2.2. Our inventory is spatially consistent with HTAP v2.2 with a
correlation coefficient of 0.99
–
1.00 for all the six species. The bias relative to HTAP
v2.2 is within 8% for NO
x
, CO, SO
2
and NH
3
, 11% for BC, and 18% for POA. The
differe
nces for China, India, the United States and other large emitters are generally
small. Although the differences are larger for small emitters, as expected, they are
normally within the uncertainty of current emission inventories
2,59
. The E
p
are very
small
for Greenland (the outlier region shown in the left of each panel), but the values
here are much higher than HTAP v2.2; this is because the IEA fuel database for
Greenland used in HTAP v2.2 contains missing values for fossil fuels, which issue is
corrected
here by taking the EIA data.
S2. Consumption
-
based emissions
Consumption
-
based emissions represent pollutants released along the global
supply chain as a result of certain region’s consumption of final products and services.
For example, a cell phone purc
hased in the United States may be assembled in China
with iron ores mined in Australia, Steel made in Japan and high
-
end assembling
© Macmillan Publishers Limited
, part of Springer Nature
. All rights reserved
6
mechanics manufactured in the United States. And consumption
-
based emissions
attribute the pollutants consequently released
in these countries to the United States.
Here we use the multi
-
regional input
-
output model (MRIO) from GTAP8
24
, based
on monetary flows, to trace the economic interconnections among sectors and regions.
We then combine the MRIO analysis with the productio
n
-
based emission inventory to
obtain the consumption
-
based emissions on a country and sectoral basis. The above
method has been used to calculate consumption
-
based emissions of CO
2
and air
pollutants
15,20,21,60
-
63
, and used to evaluate export
-
related envir
onmental and health
impact
13,26
. Detailed descriptions of the MRIO approach are provided in previous
studies
18,21,60
. Below is a brief introduction of this approach.
퐱
=
퐀퐱
+
퐲
=
(
퐈
−
퐀
)
−
ퟏ
퐲
(1)
퐄
′
=
퐟
̂
(
퐈
−
퐀
)
−
ퟏ
퐲
′
(2)
Equation 1 shows how final consumption is supplied through the supply chain
across 129 countries and 57 sectors. Here
퐱
is a vector for country
-
and sector
-
specific
monetary outputs to supply the associated final consumption
퐲
(e.g., supplied by any
give
n country and sector to all countries and sectors),
퐀퐱
is the intermediate outputs,
(
퐈
−
퐀
)
−
ퟏ
is the Leontief inverse matrix,
퐀
is the direct requirement coefficient
matrix, and
퐈
is the unit matrix. Equation 2 calculates region
-
and sector
-
specific
con
sumption
-
based emissions
퐄
′
associated with final consumption
퐲
′
(e.g.,
supplied by all countries and sectors to any given country and sector). Here
퐟
̂
is the
diagonalization of a vector representing region
-
and sector
-
explicit emissions per
monetary out
put, as derived by dividing production
-
based emissions by monetary
© Macmillan Publishers Limited
, part of Springer Nature
. All rights reserved
7
outputs
퐱
. Values of
퐲
,
퐲
′
and
퐀
are available in the MRIO model, and
region
-
and
sector
-
explicit
production
-
based emissions are derived in this study.
S3. Gridded monthly emissions
Gridded emissions are required to drive the chemical transport modeling. We
convert the country
-
based annual emissions to a monthly 0.1º long. x 0.1º lat. gridded
dataset, based on the horizontal and monthly distribution of the HTAP v2.2 emission
inventory
for 2008
57
. To support the atmospheric simulations, in the model world, E
c
of any region is released in countries producing goods to supply that region
–
for
example, a portion of China’s emissions is related to consumption in Western Europe,
and in simul
ating the effect of E
c
of Western Europe, this portion is released within the
Chinese territory and is gridded following China’s E
p
. Supplementary Fig. 3 gives an
example of how Western Europe’s E
p
and E
c
of BC are distributed horizontally.
Prior to the co
nversion, we map our emissions from 57 sectors to five main sectors
designated in HTAP v2.2 (power generation, industry, transportation, residential use,
and agriculture).
S4. Atmospheric evolution and transport simulated by GEOS
-
Chem
We use the global GE
OS
-
Chem CTM version 9
-
02 to simulate the atmospheric
evolution of aerosols and precursor gases. A series of model simulations are conducted
to derive the individual effects of E
p
and E
c
of the 11 aggregated regions on the
atmospheric distribution of SIOA,
POA and BC.
© Macmillan Publishers Limited
, part of Springer Nature
. All rights reserved
8
GEOS
-
Chem is driven by the GEOS
-
5 assimilated meteorology from the
NASA
Global Modeling and Assimilation Office (GMAO). The model is run on a 2.5º long. x
2º lat. grid with 47 vertical layers, w
ith full O
x
-
NO
x
-
VOC
-
CO
-
HO
x
gaseous
chemistry
64,65
and online aerosol calculations. Simulated aerosols include SIOA
66,67
,
POA, BC
67,68
, dust
69,70
, and sea salts
71,72
. POA is simulated as primary organic carbon
with less mass by a factor of 2.1; the remainin
g mass accounts for the oxygen molecules
contained
73
. SIOA is assumed to be in thermodynamical equilibrium following
ISOROPIA
-
II
74
.
Wet scavenging of soluble aerosols and gases in convective updrafts,
rainout, and washout follows Liu et al.
75
, with updates
for BC by Wang et al.
76
. Dry
deposition of gases and aerosols follow Wesely
77
and Zhang et al.
78
, respectively.
Model advection uses the TPCORE algorithm of Lin and Rood
79
, convection follows a
modified Relaxed Arakawa
-
Schubert scheme
80
, and mixing in the
boundary layer
follows a
non
-
local scheme
81,82
.
Global anthropogenic emissions of NO
x
, SO
2
, NH
3
, CO, BC and POA are derived
in this study. Other emissions are set as follows. Global anthropogenic emissions of
non
-
methane volatile organic compounds (NMVOC)
are taken from the RETRO
dataset for 2000 as described by Hu et al.
83
; emissions in China, the rest of Asia and the
United States are further replaced by the regional inventories MEIC for 2008
(www.meicmodel.org),
INTEX
-
B
for
2006
35
and
NEI05
for
2005
(ft
p://aftp.fsl.noaa.gov/divisions/taq/emissions_data_2005), respectively. Biogenic
emissions of NMVOC follow the MEGAN model
84
. Soil emissions of NO
x
follow
Hudman et al.
85
. Lightning emissions of NO
x
follow the Price and Rind scheme with a
© Macmillan Publishers Limited
, part of Springer Nature
. All rights reserved
9
satellite
-
based a
djustment and a backward ‘C
-
shape’ vertical profile
86
-
88
. Biomass
burning emissions use the monthly GFED
-
3 data for 2007
89
.
We conduct three sets of model simulations for 2007 with a spin
-
up period of 6
months. The first set contains a control simulation
(S1) with all emissions unperturbed
and a second simulation (S2) with anthropogenic emissions of NO
x
, SO
2
, NH
3
, CO, BC
and POA removed globally. The second set of simulations (S3 to S13) tests the
contributions of production
-
based emissions from the 11 reg
ions, by removing
anthropogenic emissions produced within their territories, one region at a time. The
third set of simulations (S14 to S24) is the counterpart of the second set. It turns off
global anthropogenic emissions related to consumption of each of
these 11 regions. For
each simulation, GEOS
-
Chem outputs 3
-
hourly 3
-
dimensional mass concentrations of
SIOA, POA and BC for further radiative forcing calculations.
S5. Radiative forcing calculations using RRTMG
We use the RRTMG RTM for shortwave (RRTMG_SW
version v3.9)
90
to
calculate the all
-
sky top
-
of
-
the
-
atmosphere RF of SIOA, POA and BC, based on the
atmospheric distributions of aerosols simulated by GEOS
-
Chem. The RF accounts for
scattering and absorption of solar radiation in the atmosphere, i.e., the
RF from aerosol
-
radiation interactions. It does not account for rapid adjustments or feedbacks of clouds
and the hydrological cycle. The longwave RF is negligible
73
and not calculated here.
Aerosols are assumed to mix externally to facilitate a species
-
sp
ecific RF calculation,
and each aerosol type has a prescribed dry size distribution. Aerosol microphysical
properties follow Heald et al.
73
, including dry size distributions, hygroscopic growth
© Macmillan Publishers Limited
, part of Springer Nature
. All rights reserved
10
factors, and refractive indices. Following Hansen et al.
3
, we
further scale the RF of BC
by a factor of two to account for enhanced absorption by internal mixing with other
aerosols
2
-
4
.
The spatially and temporally varying aerosol mass concentrations are obtained
from the GEOS
-
Chem outputs. Ancillary meteorological a
nd surface albedo data are
taken from the GEOS
-
5 dataset, including cloud fraction, liquid water content, ice water
content, air temperature, relative humidity, tropopause pressure, and air pressure
profiles. The effective droplet radius is assumed as 14.2
μm for liquid clouds and 24.8
μm for ice clouds
73
.
Three sets of RTM calculations, with a total of 70 runs, are conducted for 2007 in
correspondence to the sets of CTM simulations. The first set contains a run (R1)
including all anthropogenic aerosols gl
obally and three subsequent runs (R2
–
R4) that
are similar to R1 but removing global anthropogenic SIOA, POA and BC one by one.
The second and third sets contain 33 (3 species x 11 regions per species) runs each, in
which an aerosol species related to a giv
en region’s production or consumption is
removed. To reduce the computational costs, the 3
-
hourly CTM aerosol data are
averaged for each month to produce monthly mean 3
-
hourly datasets (i.e., the monthly
mean diurnal cycle is preserved). The difference in
RF between this monthly
-
mean
based calculation and a calculation based on daily data is very small, according to our
initial test.
The change from Ri (i = 2 to 70) to R1 gives the RF of an aerosol species globally
(i = 2 to 4), related to a region’s produc
tion (RF
p
, i = 5 to 37), or related to a region’s
© Macmillan Publishers Limited
, part of Springer Nature
. All rights reserved
11
consumption (RF
c
, i = 38 to 70). For any aerosols (SIOA, POA and BC), the globally
cumulated RF responds quite linearly to emission perturbations, as revealed by the fact
that the RF
c
is the same as the RF
p
if the contributions of all regions are summed
(Supplementary Information Table 1). The RF of SIOA contributed by individual
regions may respond more nonlinearly to emission perturbations due to changes in the
atmospheric oxidative capacity, dependence of
sulfate and nitrate formation on the
amount of NH
3
91
, and additional nonlinearity in radiative transfer calculation. This
nonlinearity is reduced here since emissions of all species are perturbed simultaneously
in the CTM simulations, including CO that af
fects the oxidative capacity.
S6. Uncertainties and limitations
Our estimated global RF
p
and RF
c
(summed across the contributions of all regions)
are both about 0.32
W
/m
2
for BC,
-
0.10
W
/m
2
for POA, and
-
0.48
W
/m
2
for SIOA,
comparable to the mean values estimated in the IPCC AR5 (0.40
W
/m
2
,
-
0.09
W
/m
2
and
-
0.51
W
/m
2
, respectively)
1
. [Note that although the IPCC AR5 values are for the
anthropogenic aerosol changes from 1750 to 2011, the anthropogenic emissions ar
e
negligible in 1750
73,92
, and the changes from 2007 to 2011 are very small
1
.] Here we
provide a general discussion of errors in emissions, CTM and RTM relevant to the
global and regional RF and the relative difference between regional RF
c
and RF
p
. All
err
ors are referred to as 2σ uncertainties that correspond to a 95% confidence interval
(CI)
–
for example, an error of 10% for a best estimate of 1.0 means a 95% CI at [0.9,
1.1].
© Macmillan Publishers Limited
, part of Springer Nature
. All rights reserved
12
The calculation of E
p
is subject to errors in national production data and emi
ssion
factors
13,14,58
. The HTAP assessment report
91
suggests a lower bound of errors in the
global total E
p
to be 10
–
30% for the species studied here. Regionally, E
p
may contain
larger errors in the developing countries due to less accurate data inputs; th
is additional
error is estimated here to be within 30%, by comparing E
p
for the 11 individual regions
between the HTAP v2.2 inventory and our results.
Regionally, E
c
shares most errors with E
p
, although E
c
contains an additional error
from the MRIO calcul
ation
13,18
associated with inaccuracies in national economic
statistics, sectoral details and data harmonization
93,94
. Peters et al.
31
showed that
regional E
p
and E
c
of CO
2
have comparable variability across studies that use different
MRIO models, suggesti
ng a very small MRIO
-
related error compared to the error in E
p
.
The study on China’s trade with a detailed statistical analysis by Lin et al.
13
showed
that the uncertainty in the input
-
output analysis contributes ~ 10% of total errors in
export
-
related emi
ssions of pollutants, with the remaining 90% from the calculation of
E
p
. Considering the MRIO
-
related error, here we assume a 10% additive error for E
c
on
top of the error translated from E
p
. This leads to the 2σ error values (calculated as 10%
* E
c
/ E
p
)
presented in Fig. 1.
Given the amount of emissions, the RF calculation is subject to errors in the CTM
-
simulated atmospheric processes
65,91
and the RTM
-
simulated radiative transfer
processes. The atmospheric loadings and vertical profiles of aerosols are r
elatively well
simulated by the CTM here
73,95
. Larger uncertainties exist in the current understanding
of aerosol optical properties
2,3,73
, such as the extent of absorption enhancement of BC
© Macmillan Publishers Limited
, part of Springer Nature
. All rights reserved
13
through internal mixing
2
and the absorption capability of POA
96
.
The CTM and RTM
related errors together are on the order of 30% for SIOA, 50% for POA, and 100% for
BC
2,73,95,97
.
It is computationally prohibitive to perform systematic Monte Carlo or sensitivity
analyses that integrate all errors associated with emission
s, CTM and RTM. Here we
give a rough estimate. Globally accumulated RF
p
and RF
c
share the same errors, and
we estimate an error of 40% for SIOA, 60% for POA, and 150% for BC (i.e., by a factor
of 2.5), based on the uncertainties adopted in the IPCC AR5
1
. T
he errors for regional
RF
p
and RF
c
may be larger for less
-
studied developing regions. Nevertheless, most
errors in regional RF
p
and RF
c
are common and do not affect their relative difference
13
,
except for the effect of MRIO
-
related error on RF
c
(inherited
from E
c
). To account for
the MRIO
-
related error, we assume a 10% additive error for regional RF
c
on top of the
error translated from RF
p
. This leads to the 2σ error values (calculated as 10% * RF
c
/
RF
p
) presented in Fig. 3.
Due to lack of data, we do not
consider the impact of trade on aerosols related to
international aviation or shipping. Nor do we include secondary organic aerosols
because of considerable difficulties and uncertainties in emission calculations and
chemical simulations. Although some por
tions of POA may absorb the solar radiation
and partly (by 27%) offset the negative RF by POA scattering
96
, we do not account for
this absorption due to large uncertainties in determining the absorbing POA, consistent
with the IPCC AR5 assumption. We also
do not quantify the indirect RF of aerosols.
Inclusion of these aspects would reveal additional effects of trade on aerosol RF.
© Macmillan Publishers Limited
, part of Springer Nature
. All rights reserved
14
References
3
2
Bond, T. C.
et al.
A technology
-
based global inventory of black and organic
carbon emissions
from combustion.
Journal of Geophysical Research
109
,
D14203
(2004).
3
3
Zhang, Q.
et al.
Asian emissions in 2006 for the NASA INTEX
-
B mission.
Atmospheric Chemistry and Physics
9
, 5131
-
5153 (2009).
3
4
Streets, D.
et al.
An inventory of gaseous and primary aerosol emissions in Asia
in the year 2000.
Journal of Geophysical Research
108
,
D21,
8809 (2003).
3
5
IEA.
Energy
S
tatistics
and
B
alances
of OECD
C
ountries
, 2007
-
2008
(International
Energy Agency, 2010).
3
6
IEA.
Energy
S
tatistics
and
B
alances
of Non
-
OECD
C
ountries
, 2007
-
2008
(International Energy Agency, 2010).
3
7
Liu, F.
et al.
High
-
resolution inventory of technologies, activities, and
emissions of coal
-
fired power plants in China from 1990 to 2010.
Atmospheric
Chemistry and Physics
15
, 13299
-
13317
(2015).
38
Yan, F., Winijkul, E., Jung, S., Bond,
T. C. & Streets, D. G. Global emission
projections of particulate matter (PM): I. Exhaust emissions from on
-
road
vehicles.
Atmospheric Environment
45
, 4830
-
4844
(2011).
39
Yan, F.
et al.
Global emission projections for t
he transportation sector using
dynamic technology modeling.
Atmospheric Chemistry and Physics
14
, 5709
-
5733
(2014).
© Macmillan Publishers Limited
, part of Springer Nature
. All rights reserved
15
40
Lu, Z.
et al.
Sulfur dioxide emissions in China and sulfur trends in East Asia
since 2000.
Atmospheric Chem
istry and Physics
10
, 6311
-
6331
(2010).
41
Streets, D. G., Wu, Y. & Chin, M. Two
-
decadal aerosol trends as a likely
explanation of the global dimming/brightening transition.
Geophysical
Research Letters
33
,
L15806
(2006).
42
Streets, D. G.
et al.
Anthropogenic and natural contributions to regional trends
in aerosol optical depth, 1980
-
2006.
Journal of Geophysical Research
;
Atmospheres
114
,
D00D18
(2009).
43
Streets, D. G.
et al.
Aerosol trends over China, 1980
-
2000.
Atmospheric
Research
88
, 174
-
182
(2008).
44
EMEP. EMEP/EEA
A
ir
P
ollutant
E
mission
I
nventory
G
uidebook
2013:
Technical
G
uid
ance
to
P
repare
N
ational
E
mission
I
nventories. (2013).
45
USEPA. Compilation of Air Pollutant Emission Factors (AP
-
42). U.S.
Environmental Protection Agency. (1999).
46
Lei, Y., Zhang, Q., He, K. B. & Streets, D. G. Primary anthro
pogenic aerosol
emission trends for China, 1990
-
2005.
Atmospheric Chemistry and Physics
11
,
931
-
954
(2011).
47
Lu, Z. & Streets, D. G.
The Southeast Asia Composition, Cloud, Climate
Coupling
Regional
Study
Emission
Inventory,
available
at:
http://bio.cgrer.uiowa.edu/SEAC4RS/emission.html (last access: 26 March
2014), (2012)
.
© Macmillan Publishers Limited
, part of Springer Nature
. All rights reserved