Black carbon (BC) emissions play an important role in regional climate
change in the Arctic. It is necessary to pay attention to the impact of
long-range transport from regions outside the Arctic as BC emissions from
local sources in the Arctic were relatively small. The task force
Hemispheric Transport of Air Pollution Phase 2 (HTAP2) set up a series of
simulation scenarios to investigate the response of BC in a given region to
different source regions. This study investigated the responses of Arctic BC
concentrations and surface temperature to 20 % anthropogenic emission
reductions from six regions in 2010 within the framework of HTAP2 based on
ensemble modeling results. Emission reductions from East Asia (EAS) had the most
(monthly contributions: 0.2–1.5 ng m
Black carbon (BC) is one of the short-lived climate forcers (SLCFs; AMAP, 2015) and was regarded as the second-largest contributor to global warming, only inferior to carbon dioxide (Bond et al., 2013). BC over the Arctic can perturb the radiation balance in a number of ways. Direct aerosol forcing occurred through absorption or scattering of solar (shortwave) radiation. BC is the most efficient atmospheric particulate species at absorbing visible light (Bond et al., 2013); the added atmospheric heating will subsequently increase the downward longwave radiation to the surface and warm the surface (AMAP, 2011). Radiative forcing by BC can also result from aerosol–cloud interactions that affected cloud microphysical properties, albedo, extent, lifetime, and longwave emissivity (Twomey, 1977; Garrett and Zhao, 2006). BC has an additional forcing mechanism after depositing onto snow and ice surfaces (Clarke and Noone, 1985). The surface albedo of snow and ice could be reduced and further enhanced the absorption of solar radiation at the surface. In the Arctic, surface temperature responses were strongly linked to surface radiative forcing as the stable atmosphere of the region prevented rapid heat exchange with the upper troposphere (Hansen and Nazarenko, 2004).
The Arctic has been warming twice as rapidly as the world in the past 50 years and has experienced significant changes in its ice and snow covers as well as permafrost (AMAP, 2017). Reductions in carbon dioxide emissions are the backbone of any meaningful effort to mitigate climate forcing. But even if significant reductions in carbon dioxide are made, slowdown of the temperature rise in the Arctic and the sea level rise caused by the melting of glaciers may not be achieved in time. Hence, the goal of slowing down the deterioration of the Arctic may best be achieved by also targeting shorter-lived climate forcing agents, especially those that could impose appreciable surface forcing and trigger regional-scale climate feedbacks pertaining to the melting of sea ice and snow. Modeling studies by UNEP/WMO (2011) and Stohl et al. (2015) suggested that the climate response of SLCF mitigation was strongest in the Arctic region. AMAP (2011 and 2015) as well as Sand et al. (2016) demonstrated that the northern areas in the Arctic had the largest temperature response per unit of emission reductions in SLCFs, with the Nordic countries (Denmark, Finland, Iceland, Norway, and Sweden) and Russia having the largest impact compared to other Arctic countries such as the United States and Canada.
The few studies that investigated specific regional aerosol forcing
(Shindell and Faluvegi, 2009; Shindell et al., 2012; Teng et al., 2012)
typically used a single climate model at a time to investigate the climate
response to idealized, historical, or projected forcing. However, different
models varied considerably in the representation of aerosols and radiative
properties, resulting in large uncertainties in simulating the aerosol
radiative forcing (Myhre et al., 2013; Shindell et al., 2013). When
investigating the climate response to regional emissions, such uncertainties
were likely to be confounded even further by the variability between models
in regional climate and circulation patterns and variation in the global and
regional climate sensitivity (the amount of simulated warming per unit
radiative forcing). Hence, the task force Hemispheric Transport of Air
Pollution Phase 2 (HTAP2;
This study aims to investigate the responses of Arctic BC concentrations and surface temperature to 20 % anthropogenic emission reductions from different regions in the Northern Hemisphere (NH). A comparison of six global modeling works within the framework of HTAP2 experiments for the Arctic region in 2010 was presented. The ensemble modeling results were used to apportion the contribution from different source regions to the near-surface and vertical black carbon in the Arctic. In addition, the Arctic surface temperature responses to the emission reductions were estimated.
HTAP2 developed a harmonized emissions database covering all countries and
the major sectors for global and regional modeling from 2008 to 2010. The
emissions database was obtained from the nationally reported emissions
(e.g., National Emission Inventory for the United States), the regional
scientific inventories (e.g., the European Monitoring and Evaluation Programme
(EMEP) and the Netherlands Organisation for Applied Scientific Research (TNO) for
Europe and the Model Inter-Comparison Study for Asia (MICS–Asia III)), and the
Emissions Database for Global Atmospheric Research data (EDGARv4.3) for the
rest of the world (mainly South America, Africa, Russia, and Oceania).
Biomass burning emissions were not prescribed in HTAP2. Temporal resolution
of data sources was monthly, and thus the HTAP2 emission inventory provided
harmonized emission data with monthly resolution for all the air pollutants
including BC. It should be noted that the emissions of international
shipping and international aviation in HTAP2 were considered constant over
the year. It was recommended that modeling groups use the Global Fire
Emissions Database (GFED4;
Emission perturbations were conducted in sensitivity simulations to investigate the response of various air pollutants in a given region to different source regions. In this study, the Arctic region was the targeted receptor region of interest. Six source regions in HTAP2 experiments, namely East Asia (EAS), Europe (EUR), Middle East (MDE), North America (NAM), Russia–Belarus–Ukraine (RBU), and South Asia (SAS), were selected to demonstrate their influences on the BC concentrations over the Arctic region (Fig. 1a). Two emission scenarios were designed for the HTAP2 simulation to explore the source–receptor relationships, i.e., the base scenario (BASE) with no emission reduction and the control scenario (EASALL, EURALL, MDEALL, NAMALL, RBUALL, and SASALL) with 20 % reduction in all anthropogenic emissions in six regions, respectively.
Anthropogenic BC emission sectors included power plants, industries,
transportation, shipping, aviation, agriculture, and residential sectors.
The emission inventory had a monthly temporal resolution and a spatial
resolution of 0.1
Total anthropogenic emissions and 20 % emission reductions in BC
in different regions of HTAP2 in 2010 (unit: Gg yr
Figure 1b–g illustrate the spatial distribution of the 20 % reductions in annual BC emissions in six source regions in 2010. It can be found that the most intense reductions in BC emissions in EAS and SAS were concentrated in East China and India, respectively, which were mainly attributed to emissions from residential sectors, followed by transportation and industries. The BC emission reductions in EUR were widely distributed, with high values in central Europe and residential and transportation sectors accounting for the largest proportion. The reductions near the Arctic circle could be found in the north of EUR, NAM, and RBU. For MDE, most BC was emitted from Iran, which is located in the northeast of this region. Overall, the spatial pattern of BC emission reductions in six regions was closely related to the spatial distribution of the human population.
Considering that the simulations should cover all months of 2010 and all
emission source regions, five global models (i.e., CAMchem,
CHASER_re1, GEOS-Chem, GOCART–v5, and Oslo CTM3–v2) were
incorporated to simulate the responses of BC concentrations in the Arctic to
20 % BC emission reductions from EAS, EUR, MDE, NAM, RBU, and SAS,
respectively. The brief information of model configurations is listed in
Table 2. As required by HTAP2, all simulations should include a spin-up time
of 6 months prior to the period of interest. The outputs from all models are
available upon request from
Configurations of models used in this study.
The climate effects of air pollutants have been the focus of climate change
research since the last century (IPCC, 1990, 2001). In the last few
years, the metrics for estimating this kind of effect have been constantly
improving (Shindell et al., 2012; Bond et al., 2013; Smith and Mizrahi,
2013; Stohl et al., 2015). The Intergovernmental Panel on Climate Change
(IPCC) used the global warming potential (GWP) as a method for comparing the
potential climate impact of emissions of different greenhouse gases (IPCC,
1990). GWP is the time-integrated radiative forcing due to a pulse emission
of a given species over some given time horizon (commonly 20, 100, or 500 years) relative to a pulse emission of carbon dioxide. GWP does not purport
to represent the impact of air pollutant emissions on temperature. Although
a short-lived climate pollutant (SLCP) could have the same GWP as a
long-lived climate pollutant, identical (in mass terms) pulse emissions
could cause a different temperature change at a given time because
long-lived climate pollutants accumulate in the climate system, while
short-lived climate pollutants can be broken down by various processes.
Consequently, warming caused by long-lived climate pollutants is determined
by total cumulative emissions to date, while the warming due to short-lived
climate pollutants is determined more by the current rate of emissions in
any given decade and depends much less on historical emissions. This means
the importance of SLCP emissions is often overstated based on GWP. Shine et
al. (2005) proposed the global temperature change potential (GTP) as a
replacement for GWP to represent the global-mean surface temperature change
for both a pulse emission (GTP
ARTPs is more suitable for this study to calculate the temperature response,
considering that the research object is BC with a short lifetime and a focus on
the regional impact of the BC emission reductions on temperature changes in the
Arctic. For SLCFs with atmospheric lifetimes much shorter than both the time
horizon of the ARTP and the response time of the climate system, the general
expression for the ARTP following a pulse emission of BC (
Regional temperature responses at time
To evaluate the model performance from all five models, the monthly
simulated surface BC concentrations of the BASE scenario were compared with
the observations at four monitoring sites in the Arctic Circle in 2010. The
locations of the four sites, including Alert (82.5
Metrics (Text S1 in the Supplement) including correlative coefficient (COR), normalized mean bias (NMB), normalized mean error (NME), mean bias (MB), and mean absolute error (MAE) were selected for evaluating the model performance in this study (US EPA, 2007). In addition to the evaluation for each single model, the multi-model ensemble mean (calculated as the average of all participating models) was also evaluated. The statistical results are listed in Tables 3 and S1. A comparison between the monthly variations in simulated and observed BC concentrations is shown in Fig. S2a.
The correlations of the simulated BC concentrations among different models were moderate to high, with CORs ranging from 0.33 to 0.98 (Table S1), suggesting that the temporal variations in different models were relatively consistent. Overall, CAMchem, GEOS-Chem, GOCART–v5, and Oslo CTM3–v2 underestimated the near-surface BC (Fig. S2a), which may be attributed to an underestimation of BC emissions, e.g., gas flaring (Huang et al., 2014, 2015; Stohl et al., 2013) and shipping emissions (Marelle et al., 2016). Also, appropriate temporal allocation of BC emissions from residential combustion was another important factor governing the model performance (Stohl et al., 2013). However, the simulated BC surface concentrations from CHASER_re1 were higher than those of the other four models and observations (Fig. S2a), which was mainly due to their slow BC aging rate in remote and polar regions (Sudo et al., 2015).
Table 3 shows the model performances at the four Arctic sites. No single
model could reproduce the BC concentrations in the Arctic well, and models
performed differently at different monitoring sites. Relatively good
agreement between the observation and models was found at Zeppelin, with
CORs, NME, MB, and MAE of 0.59–0.83, 38.59 %–142.64 %,
The vertical profiles of simulated BC concentrations of the BASE simulation were also compared with aircraft measurements from HIAPER Pole-to-Pole Observations (HIPPO) during 24 March–16 April 2010 (Fig. S2b). Different from comparison between observed and simulated BC concentrations near the surface, the vertical profiles of BC concentrations were overestimated by most models. As the aircraft ascended and descended along each flight track, BC concentrations from HIPPO varied with time, latitude, longitude, and altitude. However, most of the simulation results of HTAP2 were provided in monthly temporal resolution, and simulation and observation results cannot be exactly matched. This partly explained the difference between the simulations and observations. Overall, currently no single model could reproduce the BC concentrations over different regions of the Arctic well. There are a number of reasons responsible for this. First, the BC emission inventory in the Arctic is not well understood due to a lack of local activity data and emission factors, e.g., gas flaring in the oil and gas production fields, biofuel combustion, non-road transportation, etc. Secondly, the lifetime of BC in the atmosphere is sensitive to its wet deposition rates. However, different models have divergent treatment of wet scavenging parameterizations, which may not be representative in the Arctic region and could result in the simulated BC concentrations ranging between several magnitudes. The mechanism of BC sinks is still not well understood in the Arctic. Last but not least, almost all the global models used the latitude–longitude projection, which has very large distortions over the polar regions, and this may also affect the ability of global models to simulate the air pollutants over the Arctic region.
Although the single model did not reproduce the BC concentrations in the Arctic well, the consistency of the model ensemble mean with the observation was improved to some extent. The NME and MAE of the model ensemble mean was closer to zero compared with the single model. Therefore, to reduce the bias from one single model, the multi-model ensemble mean was used for further analysis.
Comparison of the simulations and observations of monthly surface BC concentrations at Alert, Barrow, Tiksi, and Zeppelin in 2010.
Before analyzing the responses of Arctic BC to emission reductions, it is necessary to understand the spatiotemporal distribution of BC concentrations in the Arctic region. In this study, the months from May to October were defined as summer, and November to April were defined as winter due to the special geographical location of the Arctic (Aamaas et al., 2017).
Spatial distributions of Arctic near-surface BC concentrations in summer and
winter simulated from each model are shown in Fig. S4. BC simulated by
CHASER_re1 showed relatively high concentrations over the
whole Arctic, followed by GEOS-chem and GOCART–v5, while those simulated by
Oslo CTM3–v2 and CAMchem were lower. The difference in simulated BC
concentrations between land and ocean was more obvious in summer than that
in winter, especially for GEOS-chem and GOCART–v5. The mean BC
concentrations from the ensemble models near the surface Arctic
(66–90
Spatial distribution of near-surface BC concentrations in
The response of the Arctic near-surface BC to 20 % emission reductions from different source regions was analyzed through emission perturbation simulations. Figure 3 shows the spatial distribution of the response referred to above in summer and in winter of 2010 based on multi-model ensemble mean results. The source region contributions to the surface BC concentrations exhibited significant seasonal variability with higher values in winter. The BC emission reductions in EAS almost affected the whole Arctic, especially in winter, indicating the significance of the intercontinental transport of BC. The spatial distribution of the Arctic near-surface BC response to SAS emission reductions was similar to that of EAS, but the extent was much weaker. The emission reductions from EUR, NAM, and RBU mainly affected the local and nearby areas, which was generally consistent with the spatial pattern of emissions (Fig. 1). The contribution from MDE emission reductions was very little.
Spatial distribution of contribution of 20 % emission reductions in different source regions to Arctic near-surface BC in summer and in winter in 2010.
The monthly variations in the response of the Arctic near-surface BC
concentrations to 20 % emission reductions from six source regions are
presented in Fig. 4. Results from the ensemble simulations are averaged
over the Arctic, covering latitudinal areas from 66
Monthly mean reduced concentrations of the near-surface Arctic BC due to 20 % emission reductions from six source regions in 2010.
The annual contribution of 20 % emission reductions from EAS, EUR, MDE,
NAM, RBU, and SAS to the Arctic near-surface BC concentrations reached 0.70,
0.54, 0.01, 0.20, 0.29, and 0.09 ng m
Figure S6 compares the contributions of 20 % emission reductions to Arctic near-surface BC concentrations simulated by different models. All five models showed similar monthly variations, of which CHASER_re1 simulated high BC concentrations compared to the other models due to slow aging speed (Sudo et al., 2015). All models showed the major source regions of Arctic BC from EAS, EUR, and RUB. NAM and SAS contributed moderately, while the contribution from MDE was negligible.
To assess the contributions from various source regions to the BC profiles based on the model ensemble mean, the vertical stratification needed to be unified as most participating models had different vertical settings. Since CHASER had a relatively coarse vertical resolution of 32 layers, the other models were unified to the same vertical stratification, as detailed in Table S2.
As shown in Fig. 5, the contributions of regional emission reductions to BC exhibited strong vertical gradients over the Arctic. In general, the BC profiles displayed a bimodal pattern in summer, showing peaks at around 1.0–1.6 km a.s.l. (4th and 5th layers) and 8.0–8.9 km a.s.l. (13th and 14th layers), while in winter, the BC profiles showed a unimodal pattern with peaks around 0.6–1.6 km a.s.l. (3rd–5th layers). Long-range transport of air pollution may occur near the planetary boundary layer (Eckhardt et al., 2003; Stohl et al., 2002). High contributions in the low layers (e.g., 3rd–5th layers) were consistent with the height of the planetary boundary layer in the Arctic (Zhang et al., 2018; Cheng, 2011).
Contribution of 20 % emission reductions from six source regions
to BC concentrations in different vertical layers
It has been summarized that there were several major transport pathways for
BC into the Arctic troposphere (Stohl, 2006). (i) BC transported rapidly at
a low level, followed by uplifting at the Arctic front when it is located far
north. Significant deposition of BC in the Arctic occurs mostly north of
70
As shown in Fig. 5, BC can also be transported into the upper troposphere
of the Arctic. Air masses preferably kept their potential temperature almost
constant during transport as the atmospheric circulation can be well
described by adiabatic motions in the absence of diabatic processes related
to clouds, radiation, and turbulence. The potential temperature was low
within the polar dome area, and thus only air masses that experienced diabatic
cooling were able to enter the polar dome (Stohl, 2006). That is to say, the
air masses from SAS and low-latitude regions of EAS could not easily penetrate the polar dome but can be lifted and transported to the Arctic in
the middle and upper troposphere along the isentropes (AMAP, 2011; Barrie,
1986; Law and Stohl, 2007; Stohl, 2006). This agreed well with the previous
study of Koch and Hansen (2005) and Stohl (2006). The contribution
from SAS to the Arctic BC concentrations peaked at about 9.7 km a.s.l. (0.4 ng m
To further analyze the response of the Arctic BC concentrations to emission
reductions in six source regions in HTAP2, the contribution of 20 %
emission reductions to BC concentrations at different latitudes of the
Arctic were calculated (Figs. 6 and 7). In regard to the different
horizontal resolution of participating models, the Arctic region
(66–90
Contributions of 20 % emission reductions in different regions to near-surface BC concentrations in each latitudinal band of the Arctic. The results of summer and winter correspond to the left and right panel in the figure.
Contributions of 20 % emission reductions from all six source
regions to the vertical BC concentrations of the Arctic in different
latitude bands vary with vertical layers in
The response of the Arctic BC concentrations to emission reductions in six
source regions became weaker with the increase in the latitude due to the
continuous loss of BC during transport (e.g., dry and wet depositions)
(Fig. 6). The difference in contributions between two adjacent latitudinal
bands became smaller closer to the north pole. The contributions of
20 % emission reductions to the Arctic BC concentrations near the surface were
the highest between 66–69
The contributions from EAS and EUR were higher than those from the other
four regions in each latitudinal band. In detail, the contributions from EUR
(0.8 ng m
The downward trends of the response of the Arctic near-surface BC to
emission reductions with the increase in latitude from EUR and RBU were more
obvious than those of other regions (Fig. 6). Dry and wet depositions of BC
decreased with the increase in transport distance, and the decreasing rates
became slower (Fig. S7). The changes in dry and wet depositions caused by
emission reductions from EUR and RBU were still obvious in the Arctic region
(66–90
Figure 7 further depicts the response of the vertical Arctic BC profiles in
different latitudinal bands to 20 % emission reductions. The contributions
of eight latitudinal bands showed a typical bimodal pattern in summer with
peaks at 0.6–1.6 km a.s.l. (3rd–5th layers) and 8.0–8.9 km a.s.l. (13th and 14th layers), while the contribution displayed a
single peak at 0.4–1.0 km a.s.l. (2nd–4th layers) in
winter. Similar to Sect. 3.3.2, the peak value of the contribution at the
low layers was due to the transport of EAS, EUR, NAM, and RBU emission
reductions to the Arctic through different pathways both in summer and
winter. The peak value in the high layers in summer was due to the transport
of EAS and SAS. However, a high contribution of 20 % emission reductions
to BC concentrations in SAS was found in the high layers, while the
contribution was low in other regions, leading to a single peak in winter.
The statistical results of SAS indicated that the contribution in the vertical
appeared in one peak in the 15th layer (9.7 km a.s.l.), with values of 0.45
and 0.48 ng m
The same as the whole Arctic region (Sect. 3.3.1 and 3.3.2), the
contributions of 20 % emission reductions to BC concentrations in eight
latitude bands were higher in winter than in summer, whether near the surface or
in the vertical. The contribution of 20 % emission reductions from all six
source regions to BC concentrations in eight latitude bands of the Arctic
near the surface was 0.7–1.9 ng m
The impact of BC emission reductions on decreasing the Arctic
(60–90
Arctic surface temperature response to 20 % regional BC emission
reductions in
Global and Arctic surface temperature responses to 20 % regional
BC emission reductions in
In addition, the impacts of BC emission reductions from six source regions
on the Arctic and global surface temperature were compared in this study
(Fig. 9). Due to the BC emission reductions from the six source regions,
the surface temperature in the Arctic decreased 27–780
It should be noted that the estimation of temperature response was subject to
large uncertainties for the following reasons. On the one hand, even though
the HTAP2 emissions database was all constructed by bottom-up methods, the
different inventories and spatiotemporal distributions were constructed with
sub-regional (country, state, county, or province level) activity data and
emission factors, which lead to inconsistencies at the borders between two
adjacent inventories. Version 5 of Evaluating the Climate and Air
Quality Impacts of Short-Lived Pollutants (ECLIPSEv5;
On the other hand, the time evolution of
Although the HTAP2 emissions database contains uncertainties, and ARTP calculations are simplifications, these emission metrics are useful, simple, and quick approximations for calculating the temperature response in the different latitude bands for emissions of BC. It should be noted that the estimated responses of Arctic surface temperature to 20 % emission reductions were only valid for the comparison among different source regions but cannot be used to reflect the actual change in temperature. On the one hand, in reality, not all emissions sectors of a specific source region can be reduced by 20 % at the same time. On the other hand, there were many other factors (e.g., greenhouse gases, sea ice coverage) that can affect the temperature change in the Arctic besides BC.
CAMchem, CHASER_re1, GEOS-Chem, GOCART, and Oslo CTM3 in the HTAP2 experiment were used in this study to estimate the responses of Arctic BC to multi-region emission reductions in 2010. Six regions (e.g., EAS, EUR, MDE, RBU, NAM, and SAS) were selected as the source regions, and the Arctic was the receptor region. HTAP2 set up the base scenario with all BC emissions and also simulated BC concentrations with 20 % reduction in anthropogenic emissions. The ARPT was further used to calculate the benefit of BC emission reductions to the decrease in Arctic temperature.
The statistical results of 20 % BC emission reductions showed that
emission reductions in EAS were the largest, with values of 355.6 Gg yr
The temporal variations in simulations from different models were relatively consistent as the correlations of the simulated BC concentrations among different models ranged from 0.33 to 0.98. However, the simulated BC concentrations did not agree so well with observations at monitoring sites, except Zeppelin. In order to reduce the difference in simulation performance of each model in different areas of the Arctic, the model ensemble mean was used for analysis.
The contribution of 20 % BC emission reductions from EAS, EUR, MDE, NAM,
RBU, and SAS to the Arctic near-surface BC concentrations reached 0.70,
0.54, 0.01, 0.20, 0.29, and 0.09 ng m
The response of Arctic near-surface BC concentrations to 20 % emission
reductions from EAS and EUR was larger than the other four source regions, with
a monthly value of 0.2–1.5 and 0.2–1.0 ng m
The response of Arctic BC to emission reductions from source regions in
winter was higher than that in summer. The contributions of 20 % emission
reductions to the Arctic BC concentrations near the surface were the highest
between 66–69
The response of Arctic temperature to BC emission reductions was the most
significant at the timescale of 10 years and then gradually decreased with
the passage of time. The Arctic had benefited the most from emission
reduction in EAS, with more than 300 and 660
Overall, this study provided insights on the source regions and seasonal contributions of Arctic BC from the most recent international ensemble modeling efforts. The discrepancy between model results and observations and the spread among different HTAP models may be attributed to various factors such as emissions in the remote Arctic, physical parameterizations, and convection and deposition processes. This would subsequently result in large uncertainties in the climatic effects of air pollutants. More observation sites for the typical transport pathways from source regions to the Arctic should be planned to improve the model capability of simulating the transport behavior of black carbon.
All data used in this paper can be obtained through the AeroCom servers and
web interfaces, accessible at
The supplement related to this article is available online at:
KH and JSF designed this study. ML, KS, DH, TK, MC, and ST performed modeling. NZ analyzed data and wrote the paper. All have commented on and reviewed the paper.
The authors declare that they have no conflict of interest.
We are sincerely thankful for the HTAPv2 international initiative. We also thank the handling editor and two reviewers for providing the insightful comments and suggestions. Kan Huang also acknowledges Jiangsu Shuangchuang Program through Jiangsu Fuyu Environmental Technology Co., Ltd.
This work was supported by the National Key R&D Program of China (grant no. 2018YFC0213105), the National Natural Science Foundation of Shanghai (grant no. 18230722600), and the National Natural Science Foundation of China (grant no. 91644105).
This paper was edited by Frank Dentener and reviewed by two anonymous referees.