Introduction
Air pollution is a serious health concern in the world and especially over
Asia (Atkinson et al., 2012). It has been identified as the fifth most
important cause of mortality in India (WHO, 2014). India is among the
countries experiencing an increase in the number of high-pollution events
during this last decade. With a population of 1.3 billion inhabitants, a
density of 420 inhabitants per km2 (12 times the population density of
the United States) and a gross domestic product (GDP) growth of 7.6 % per
year in 2015 (www.worldbank.org), India is one of the fastest growing
economies in the world. Thus, India has many different
challenges to cope with in order to continue its economic development without a negative
environmental impact. Nonetheless, air pollution is progressing up in the
list of policy priorities.
Heavy air pollution results from a combination of high emissions of
pollutants and unfavourable weather conditions. In order to limit air
pollution or to regulate the emissions of pollutants, policy measures are
starting to be implemented in India at a national level (e.g. National
Environment Policy, 2006: http://iced.cag.gov.in/?page_id=1037) or at city level, as in New Delhi, which restricts cars with odd and
even license plate numbers (UNICEF, 2016 and references therein) on alternate
days. In order to meet clean-air standards for reducing the public health
risk and improving air quality in urban areas, the Union Environmental
Ministry of Government of India launched a national Air Quality Index as part of a
major initiative in 2015 for air pollution mitigation (Ghude et
al., 2016).
Changes in air quality are nevertheless not only driven by regulations.
Climate change may also have a non-negligible impact on air quality, by
modifying atmospheric circulation (e.g. wind speed, mixing depth and
transport directions), precipitation, dry deposition, emissions and the
chemical production or loss rates of pollutants (e.g. Jacob and Winner,
2009; Fiore et al., 2015). The impact of climate change on air quality has
been extensively studied in recent years with regional models (e.g. Langner
et al., 2005, 2012; Hedegaard et al., 2008; Simpson et al., 2014; Trail et
al., 2014; Lacressonnière et al., 2016) but to our knowledge, no study
has been focused on India. Climate change is however a main worry in India,
especially in terms of the occurrence and the intensity of extreme events such as
floods and cyclones (e.g. Dash et
al., 2007; Ministry of Environment and Forests, 2010).
Two of the main pollutants having an impact on air quality and health
effects are ozone (O3) and particulate matter with an aerodynamic
diameter lower than 2.5 µm (PM2.5) (e.g. Fann et al.,
2012; Silva et al., 2013; Lelieveld et al., 2013, 2015). Ghude et al. (2016)
showed that around 570 000 and 31 000 premature deaths were due to PM2.5 and
O3 exposure respectively in 2011 – at an economic cost of USD 640 billion, which is a factor of 10 higher than total expenditure on health
by public and private expenditure in India.
O3 is a highly oxidative pollutant formed from precursors. O3
pollution mostly occurs in summer, due to warmer weather driving
photochemical reactions. O3 levels depend on the balance between
reactive nitrogen oxide (NOx) and volatile organic compounds (VOCs). In
the troposphere, the main sink of O3 is the reaction with the hydroxyl
radical (OH) through HOx reactions (e.g. Crutzen et al., 1999). In the
atmospheric boundary layer, dry deposition (uptake by vegetation) is
usually the dominant sink (e.g. Monks et al., 2015).
O3 is known to be associated with respiratory morbidity and mortality
(e.g. Jerrett et al., 2009; Orru et al., 2013) but has increased strongly in
Asia in recent decades with industrialization and urbanization (e.g. Cooper
et al., 2014). Long-term exposure to high concentrations of surface O3
can also damage vegetation with substantial reductions in crop yields and
crop quality (e.g. Morgan et
al., 2006; Mills et al., 2011; Ainsworth et al., 2012). The extent of crop damage
in India has been estimated at 3.5 million tonnes a year (Ghude et al., 2014) – sufficient to
feed about 94 million people living below the poverty line in India.
PM2.5 consists of both primary and secondary components. Primary
PM2.5 components include organic matter (OM), elemental carbon (EC),
dust, sea salt (SS) and other compounds. Secondary PM2.5 comprises
compounds formed through atmospheric processing of gas-phase precursors.
This includes various compounds such as nitrate (NO3-) from NOx,
ammonium (NH4+) from ammonia (NH3), sulfate
(SO42-) from sulfur dioxide (SO2), and a large range of
secondary organic aerosol (SOA) compounds from both anthropogenic and
biogenic VOCs. Important sources of both primary and secondary PM2.5
emissions in India are domestic heating in winter, wood burning (mainly used
for cooking), road transport with contributions from both exhaust (mostly
diesel) as well as non-exhaust emissions from brake and tyre wear, and
industrial combustion. The main sink of PM2.5 is wet deposition,
associated with rain-out and wash-out by precipitation.
Long-term exposure to elevated PM2.5 levels leads to increased risk for
a variety of diseases, such as cardiovascular disease and respiratory
diseases (Lim et al., 2012). The World Health Organization (WHO) states a
guideline value of 10 µg m-3 annual mean concentration (25 µg m-3 for the daily mean) that should not be exceeded in order
to ensure healthy conditions. Moreover, the Global Burden of Disease (GBD)
study (Forouzanfar et al., 2015) ranked exposure to PM2.5 as the
seventh most important risk factor contributing to global mortality,
responsible for 2.9 million premature deaths in 2013. At the
country level, India presents one of the highest population-weighted mean
concentrations in the world for 2013 (Brauer et al., 2016).
This study aims to evaluate the effect of the regional climate change and
future emissions change in realistic air pollutant emission scenarios,
focusing on surface O3 and PM2.5 concentrations. For this
purpose, the EMEP/MSC-W chemical transport model (see Sect. 2) was used,
hereafter referred to as the EMEP model. In this study we conducted a
10-year simulation of air quality in India driven by downscaled
meteorological fields for three periods: 2006–2015 (labelled as the
reference), 2026–2035 and 2045–2055. In this study, the physical and chemical
processes that are responsible for the modelled changes are investigated in
detail.
Section 2 describes the model set-up. Section 3 focuses on the evaluation of
the reference scenario against surface-based measurements. Section 4
highlights the impact of climate change on the level of surface O3
and PM2.5 and Sect. 5 investigates the joint impact of future
emission scenarios. The conclusions are provided in Sect. 6.
Model set-up
The EMEP model is a 3-D Eulerian model described in detail in Simpson et al. (2012). But for global-scale modelling, some important updates have been
incorporated. Although the model has traditionally been aimed at European
simulations, global scale modelling has been possible for many years (Jonson
et al., 2010, 2015a; Wild et al., 2012). These updates, resulting in EMEP
model version rv4.9 as used here, have been described in Simpson et al. (2016) and references cited therein. The main changes concern a new
calculation of aerosol surface area (now based upon the semi-empirical
scheme of Gerber, 1985), revised parameterizations of N2O5
hydrolysis on aerosols, additional gas–aerosol loss processes for O3,
HNO3 and HO2, a new scheme for ship NOx emissions, and the
use of new maps for global leaf area (used to calculate biogenic VOC
emissions) – see Simpson et al. (2015) for details. The value of the
N2O5 uptake coefficient (γN2O5) is
very uncertain, but here we use the “SmixTen” scheme described in 2015,
which seemed to provide the best predictions of O3 for global O3
sites with this model version. In addition, the source function for sea salt
production was updated to account for whitecap area fractions, following the
work of Callaghan et al. (2008).
The domain of each simulation covers the latitudes 5.6–40.7∘ N and the longitudes 56.2–101.7∘ E,
and the horizontal resolution of the simulations follows the resolution of
the meteorological data described in Sect. 2.1. However, the studied
region is more centred over India (see Fig. 4b).
As in the standard EMEP model, the boundary conditions for most PM2.5
components are defined as prescribed concentrations (Simpson et al., 2015),
and O3 boundary conditions (lateral and top) are defined by the
climatological O3 data from Logan (1998). For dust, concentrations from
a global simulation for 2012 (EMEP Status Report 1/2015) have been used as
boundary conditions. The influence of the changes in inflow of O3 or
PM2.5 from outside the Asian domain is not taken into account.
PM emissions are split into EC, OM (here assumed inert) and the remainder,
for both fine and coarse PM. The OM emissions are further divided into
fossil-fuel and wood-burning compounds for each source sector. As in
Bergström et al. (2012), the OM / OC ratios of emissions by mass are
assumed to be 1.3 for fossil-fuel sources and 1.7 for wood-burning sources.
The model also calculates windblown dust emissions from soil erosion, but
these emissions are negligible over our studied domain compared to the dust
transported from the boundary conditions.
Secondary PM2.5 aerosol consists of inorganic sulfate, nitrate and
ammonium, and SOA; the latter is generated from both anthropogenic and
biogenic emissions (ASOA, BSOA respectively), using the “VBS” scheme
detailed in Bergström et al. (2012) and Simpson et al. (2012).
The main loss process for particles is wet deposition, and the model
calculates in-cloud and sub-cloud scavenging of gases and particles as
detailed in Simpson et al. (2012). Gas and particle species are also removed
from the atmosphere by dry deposition. Calculations of O3 deposition in
the EMEP model are rather detailed compared to most chemical transport
models. We make use of the stomatal conductance algorithm (now commonly
referred to as DO3 SE) originally presented in Emberson et al., 2000, 2001),
which depends on temperature, light, humidity and soil moisture. Calculation
of non-stomatal sinks, in conjunction with an ecosystem-specific calculation
of vertical O3 profiles, is an important part of this calculation, as
discussed in Tuovinen et al. (2004, 2009) or Simpson et al. (2003). The
methodology and robustness of the calculations of O3 deposition and
stomatal conductance have been explored in a number of publications
(Tuovinen et al., 2004, 2007, 2009; Emberson et al., 2007; Büker et al.,
2012).
An initial spin-up of 1 year (2005) was conducted, followed by ten 1-year
simulations from 2006 to 2015. Each simulation was used as spin-up of the
following year of simulation. The initial spin-up (2005) was excluded from
the analysis. To conduct the evaluation on the impact of future climate,
similar runs were done with spin-ups of 1 year (2025 and 2045), followed
by ten 1-year simulations from 2026 to 2035 and from 2046 to 2055,
respectively. In this way, short-term (up to 2030) and medium-term
(up to 2050) future climate changes have been analysed. These short-term
and medium-term future climate (FC) scenarios used the same anthropogenic
emissions as the reference scenario. In addition to the climate change, the
impact of the future emission scenarios was investigated by using
anthropogenic emissions for the 2030s and the 2050s. These simulations,
referred to as future climate and emissions (FCE) scenarios, were run for
the same time periods as the FC scenarios, but used emissions for their
respective baseline year (2030 for the 2030s and 2050 for the 2050s). In
order to simplify the reading, the four future scenarios are named as
FC2030, FC2050, FCE2030 and FCE2050.
Downscaled meteorological data
In this work, the EMEP model used meteorological data from the Norwegian
Earth System Model (NorESM1-M, Bentsen et al., 2013). These data were
downscaled using the Weather Research and Forecasting (WRF) model version
3.4 following the RCP8.5 scenario (Riahi et al., 2011) for the years
2006–2060. The RCP8.5 combines assumptions about high population and
relatively slow income growth with modest rates of technological change and
energy intensity improvements, leading in the long term to high energy
demand and greenhouse gas emissions in the absence of climate change policies (Riahi et
al., 2011). The method and the evaluation are further detailed in Jackson et al. (2018).
The domain used follows the CORDEX South Asia domain specifications
(http://www.cordex.org/index.php?option=com_content&view=article&id=87&Itemid=614), yielding 193 by 130
grid points after removal of a 10-grid-point buffer zone in each direction,
on approximately 50 km horizontal resolution and with 30 vertical levels.
The different options used were Thompson microphysics, CAM radiation scheme,
Noah Land-Surface Model, Mellor–Yamada–Janjić TKE scheme and the Kain–Fritsch
cumulus scheme. The evaluation against ERA-Interim for the temperature and
APHRODITE for the precipitation, indicates that the downscaled run has a
cold bias especially over the ocean, but when comparing with seven other
simulations from the CORDEX South Asia ensemble (also using the RCP8.5
scenario), it still performs among the best over the Indian subcontinent
(Jackson et al., 2018). For precipitation, the monsoon season
(July–September) was simulated to be too dry, which may be at least
partially caused by the too cold Indian Ocean and thus less evaporation. The
Western Ghats region receives particularly little precipitation in all
seasons, which can be explained by the relatively coarse resolution
leading to too little orographic precipitation.
Annual emissions (in Gg yr-1) used for the
reference (2010), FC2030 and FC2050 scenarios (dark blue), and for 2030
(dark green) and 2050 (dark red), used for the FCE2030 and the FCE2050
scenarios over India, respectively. The variation of each compound with
respect to the reference scenario is also provided by coloured percentage figure. The
ECLIPSE emissions are also plotted for comparison and represented by light
coloured bars. The variation of each compound with respect to ECLIPSE2010
scenario is also provided by italic black percentage figure given in parentheses.
For the future scenarios, NorESM1-M predicts an increase in temperature
close to the mean of the CORDEX South Asia ensemble. For many areas there is
no consensus concerning the sign of the precipitation change, except during
monsoon and post-monsoon periods (October–November) in the 2050s, for which most
of the models, including NorESM1-M, predict an increase in precipitation
over the major part of India, in comparison with the 2006–2015 period.
During the pre-monsoon period (April–June) in the 2050s, half of the models,
including NorESM1-M, show a decrease in precipitation, which is larger over
the Indo-Gangetic Plains. NorESM1-M also presents this decrease in the
2030s. In winter (December–March), the western coast is characterized by an
increase in precipitation, even if this change is lower in NorESM1-M than
in the other models (not shown).
Emissions
Anthropogenic emissions of SOx, NOx, CO, PM and non-methane volatile organic compounds (NMVOCs) over India
were taken from Sharma and Kumar (2016). These data have a resolution of
36 km × 36 km and are available for 2011 (used for the reference, the
FC2030 and the FC2050 scenarios) and for 2030 and 2050 (used for the FCE2030
and the FCE2050 scenarios, respectively).
For NH3 (not available from Sharma and Kumar, 2016), and for all areas
outside India, anthropogenic emissions from the ECLIPSEv5a baseline data set
(http://www.iiasa.ac.at/web/home/research/researchPrograms/air/Global_emissions.html) were used (2010 for the reference, FC2030 and FC2050
scenarios; 2030 for the FCE2030 scenario; 2050 for the FCE2050 scenario).
The ECLIPSEv5a baseline emission data set was created with the GAINS model
(Greenhouse gas–Air pollution Interactions and Synergies;
http://www.iiasa.ac.at/web/home/research/researchPrograms/GAINS.en.html)
(Amann et al., 2011), which provides emissions of long-lived greenhouse
gases and shorter-lived species in a consistent framework.
The anthropogenic emissions used for India are presented in Fig. 1. These
future scenarios are characterized by sharp increases in all emissions, even
if the CO and the NH3 emissions increase somewhat less in relative
terms (close to 30 % by 2030 and 60 % by 2050) in comparison to the
other components. Indeed, the predicted increases between 2011 and 2050 are
very large, amounting to 304 % (SOx), 287 % (NMVOC), 162 %
(NOx and PMcoarse) and 100 % (PM2.5).
The scenario estimating the emissions used by Sharma and Kumar (2016) only
incorporates the policies which were already implemented before 2014/15.
Thus future road maps of stringent standards in transport and power sectors
have been taken into account, but not in the industrial sector. For example,
there are no standards for NOx and SO2 for many coal-consuming
industries. Similarly, despite reduction in biomass-based combustion, there
are limited controls over the fugitive NMVOC emissions – which are expected to
grow immensely in future. Consequently, the increase in these gases is
larger than pollutants like PM2.5, which shows much lesser increase due
to interventions taken/planned by the Government of India. Although current
policies have likely led to reductions in emission intensities, this may not
be enough for controlling absolute emissions in future. This explains the
large increase in emissions in contrast to other scenarios described for
example in the recent report from the International Energy Agency (IEA,
2016). Indeed, the IEA (2016) forecasts that existing and planned policies in
India will help contain pollutant emissions growth in the New Policies
Scenario. Thus SO2 and NOx emissions each grow by only 10 % by
2040, and by 7 % for the PM2.5 emissions. In their pessimistic
scenario, i.e. in the absence of policy efforts, the IEA estimated that
SO2 and PM2.5 emissions would roughly double by 2040 and NOx
emissions would grow almost 2.5 times.
While the NOx and PM2.5 emissions used hereafter follow the same
trend as in the IEA report, the SOx emissions are projected to increase
more, by around 4 times from 2011 to 2050. It is noteworthy that there are
differences in economic growth rates assumed in the IEA report and the
assessments used in Sharma and Kumar (2016). Sharma and Kumar (2016) assumed
higher growth rates for India than in the IEA report. This comparison shows
that the emissions used in this work reflect a pessimistic scenario. The
emissions will continue to grow if no stringent standards are adopted and our
FCE scenarios highlight the air quality issue in India without policy
effort.
For comparison, the ECLIPSEv5a emissions are also plotted in Fig. 1 since
the NH3 emissions from ECLIPSEv5a were used as complement of the
emissions from Sharma and Kumar (2016). The emissions used in this study
show larger increase, and the amount of pollutants is also higher for all
compounds compared to ECLIPSEv5a, except for NOx in 2050. It is also
interesting to note that the emissions used in the FCE scenarios are higher
than the emissions used in the RCP8.5 scenarios for all species over India,
except NH3 (not shown). One of the drawbacks of these RCP8.5 emissions
is that only elemental carbon and organic carbon emissions are reported and
not PM2.5 and PMcoarse emissions (e.g. Zhang et al., 2016).
Moreover, the RCP scenarios were not developed with a primary focus on air
pollution concerns but for greenhouse gases (e.g. Amann et al., 2013).
For the other emissions, biogenic emissions of isoprene and monoterpene are
calculated in the model by emission factors as a function of temperature and
solar radiation (Simpson et al., 2012). The land-cover data underlying these
calculations are from GLC-2000
(http://forobs.jrc.ec.europa.eu/products/glc2000/glc2000.php).
The forest fire emissions used correspond to the mean of “Fire INventory
from NCAR version 1.5” FINNv1.5 emissions (Wiedinmyer et al., 2014) from
2005 to 2015.
(a) Monthly surface
O3 mean concentrations for the 22 stations (red) and
EMEP (averaged over the period of simulation) (blue). EMEP concentrations
are collocated to each station. The shade error corresponds to the standard
deviation. The correlation coefficient (r), the mean bias (MB), the
normalized mean bias (NMB), the root mean square (RMS) error, and the mean
normalized gross error (MNGE) are provided. (b) Correlation coefficient for each site.
(c) Normalized mean bias for each site. The type of
station is given by a letter in parentheses (u, urban; s, suburban; r, rural).
Monthly surface O3 mean
concentrations for the urban (a), suburban (b) and rural (c) stations shown
in Fig. 2 (red) and EMEP (averaged over the period of simulation) (blue).
EMEP concentrations are collocated to each station. The number of stations
is given. The shade error corresponds to the standard deviation. The
correlation coefficient (r), the mean bias (MB), the normalized mean bias (NMB),
the root mean square (RMS) error, and the mean normalized gross error (MNGE) are provided.
(a) Scatterplot between the surface
PM2.5 concentrations from EMEP (averaged over the
period of simulation) and the concentrations from WHO in µg m-3. Data are represented by a different symbol
for the corresponding year. The correlation coefficient (r), the mean bias (MB),
the normalized mean bias (NMB), the root mean square (RMS) error and
the mean normalized gross error (MNGE) are provided. (b) Distribution of the
mean surface PM2.5 concentrations for the period
2006–2015 (reference scenario). The WHO measurements from 2009 to 2013 are
superimposed on the map and represented by coloured symbols following the
symbols shown on the scatterplot.
Evaluation of the reference simulation with measurements
In this section, we evaluate the levels of the simulated surface O3 and
PM2.5 for the reference scenario to ensure the validity of this
scenario. The pollutant concentrations were averaged over their respective
decade of simulation. It is important to do this evaluation in order to
identify the biases or the errors of the reference runs, and give confidence
in the model's ability to analyse future air quality projections. It should
be noted that many factors can affect such evaluations, including accuracy
of the emissions, model processes, the quality of the observations, the
resolution and the quality of the downscaled meteorological fields, but good
agreements found with the reference scenario increase our confidence in
predicted concentrations. The details of the statistical numbers are
provided in the Appendix.
O3
Surendran et al. (2015) presented an evaluation of surface O3 mixing
ratios simulated by the global atmospheric chemistry and transport model
MOZART-4 against surface-based measurements. We have used an updated version
of this catalogue of surface observations. In total, 22 stations were
available for this comparison with different periods of measurements as
shown in Fig. S1 in the Supplement. This data set corresponds to monthly means over their
corresponding period. The discrepancies between the periods of all the
stations may have an impact on the evaluation, since the measurements do not
necessarily match the emissions year used for the reference scenario. The
observations compiled by Surendran et al. (2015) are a mixture of data from
the Modelling Air Pollution and Networking (MAPAN) observational network of
the Ministry of Earth Sciences (MoES) and from the Indian Institute of
Tropical Meteorology (IITM) over urban, suburban and rural sites, with 11, 4
and 7 stations respectively (the individual time-series are shown in Fig. S2).
Averaging the concentrations over all these sites, the simulated O3
shows a high temporal correlation (r= 0.9) with the data set (Fig. 2a).
This shows that EMEP captures rather well the seasonal variation of the
surface O3 over the different sites but it overestimates the mean
value. The mean overestimation is 35 % (11 ppb) but it varies from site to
site, between -1.4 % and around 130 %. There is no clear geographical
pattern to this overestimation or for the temporal correlation (Figs. 2b–c) but the comparison shows the lowest bias for rural sites
(15 %) and the highest biases for the urban and suburban sites (Fig. 3),
as expected due to the coarse scale of the model and the titration effect
discussed below. The overestimation in O3 found in this work is in
agreement with previous studies (e.g. Kumar et al., 2012; Chatani et al.,
2014; Sharma et al., 2016), although of course there are many differences in
both emissions and models between these studies. It has also been noted that
the EMEP model slightly overestimates O3, especially with the global
version of the model in spring and in winter (e.g. Jonson et al., 2015b).
This bias can however be impacted by the parameters used – such as for example the
boundary conditions and the emissions. Stadtler et al. (2017), who used
PANHAM anthropogenic emissions, also reported an overestimation in O3
over different regions such as Asia.
Several hypotheses could explain the overestimation in monthly averaged
surface O3. These include general uncertainties in anthropogenic and
biogenic emissions, an overestimation in the transported O3 from the
boundary conditions (including stratospheric–tropospheric exchange),
inadequate accounting for the impacts of the large PM concentrations on
gas–aerosol interactions, or systematic biases in the deposition estimates.
There is also very likely a misrepresentation of the NOx–O3
equilibrium. Under titration conditions (typically when fresh urban NO
emissions are reacting with incoming O3 to create NO2) an
underestimation in NO2 is associated with an overestimation in O3.
Sharma et al. (2016) and Chatani et al. (2014) also show overestimation in
O3 by the models mainly due to coarser resolutions which are not able
to account for titration chemistry at the local scales. Titration of O3
with NO can occur over Indian cities (e.g. Sinha et al., 2014; Sharma and
Khare, 2017) and is difficult to reproduce in regional models (e.g. Engardt,
2008). There were unfortunately no co-located NO2 or NO measurements
available for this O3 data set over India. However, a comparison was
attempted with NO2 and O3 measurements provided by
https://openaq.org for 2016 over Indian cities and shown in Fig. S3. We only
used sites measuring both compounds simultaneously and continuously during
all months. Moreover, https://openaq.org archives worldwide real-time air
quality measurements without validating the data. This highlights the
difficulty in evaluating the model results without reliable co-located
measurements of trace gases and meteorological parameters. For India, the
source of these data is the Central Pollution Control Board of India (CPCB,
http://www.cpcb.gov.in/CAAQM/frmUserAvgReportCriteria.aspx). As in the
comparison with the updated version of O3 data from Surendran et al. (2015), these observations reflect the O3 peak around April–May. It
also illustrates the underestimation by EMEP in NO2 surface
concentrations and the clear overestimation in O3 over urban sites.
Figure S3 may also suggest that Ox (NO2+ O3) concentrations are over-predicted. As Ox is conserved under titration
reactions, this suggest an overestimate of photochemical activity in the
region. Some possible reasons for this might be problems with the
anthropogenic and/or biogenic emissions, or over-active chemistry, e.g.
over-predictions of photolysis rates for Indian conditions (as EMEP
photolysis calculations assume standard atmospheric conditions, and thus do
not account for attenuation of radiation due to enhanced aerosols over
polluted regions) or problems with heterogeneous reactions. However, it is
important to remember that the observations are provided without quality
assurance, so data quality may also play a role.
The dilution of the urban emissions into large grid boxes for the urban scale
could also partly explain the overestimated O3 (e.g. Sillman et al.,
1990; Pleim and Ching, 1993), especially by considering that downscaled
meteorological fields were used at a coarse resolution (50 km) for a
comparison at city level. This statement needs however to be tested, because
an increased grid resolution does not necessarily lead to a better
simulation of O3 or NO2, as explained by Pleim and Ching (1993).
Sharma et al. (2017) also concluded that improving the models' resolutions
leads to better performance only to an extent, and may not always show
improvement with finer resolutions.
PM2.5
In contrast to the O3 evaluation, three different data sets were
available for the evaluation of the surface PM2.5 concentrations. Two of the
data sets correspond to the means over a specific period over Indian cities
and are from in situ observations from the CPCB of India. Among these
two data sets, one corresponds to the WHO database
(http://www.who.int/phe/health_topics/outdoorair/databases/cities/en – database 2015). This is a database
containing annual means from 2009 to 2013. The other of these two data sets corresponds to
averaged concentrations over the period from 2000 to 2010 published by Dey
et al. (2012). The third data set corresponds to hourly measurements at the
US embassy and consulates in India available for 2014 (i.e. over New Delhi,
Chennai, Kolkata, Mumbai, Hyderabad; available at
https://in.usembassy.gov/embassy-consulates/new-delhi/air-quality-data/).
As for O3, this evaluation remains challenging due to the location of
each site, i.e. downtown, without information about the representativeness
of the measured concentrations for a larger area. Despite the difficulty of
comparing urban stations with simulations from a regional model, a fair
agreement (spatial correlation of 0.5 and a bias of 4 %) with the data
from WHO was found with the simulated surface PM2.5 concentrations
(Fig. 4a). A better agreement is found for the coastal sites, especially in
the south and the east of India (Fig. 4b).
The agreement between the simulated concentrations with observations is
largely improved in the comparison with the data provided in Dey et al. (2012) (Fig. 5). The correlation is around 0.8 and the bias is about 6 %.
It is worth noting that a few discrepancies are observed between the data
sets provided by WHO and by Dey et al. (2012). For example, Dey al. (2012)
presented higher concentrations for the city of Patna than the value published
by WHO. It is also probable that a change in the emissions and thus in the
observed PM2.5 concentrations between the periods of both data sets has
an impact on the comparison. Similar patterns are also noted in the
measurements. For example, the city of Delhi is characterized by higher observed
concentrations in both data sets than the value simulated by the model. The
bias from the model can be expected, given its resolution.
(a) Scatterplot between the surface
PM2.5 concentrations from EMEP (averaged over the
period of simulation) and the concentrations listed in Dey et al. (2012) in
µg m-3. The correlation coefficient (r), the
mean bias (MB), the normalized mean bias (NMB), the root mean square (RMS) error
and the mean normalized gross error (MNGE) are provided. (b) Distribution
of the mean surface PM2.5 concentrations for the period 2006–2015 (reference scenario). The
measurements from Dey et al. (2012) are superimposed on the map and
represented by coloured dots.
Time series of monthly surface
PM2.5 mean concentrations in µg m-3 for the observations in 2014 (red) and EMEP for
the reference scenario (averaged over the period of simulation) (blue) over
New Delhi, Chennai, Kolkata, Mumbai and Hyderabad. The red shaded error
corresponds to the standard deviation of the measurements. The correlation
coefficient (r), the mean bias (MB), the normalized mean bias (NMB), the
root mean square (RMS) error, and the mean normalized gross error (MNGE) are
provided.
Despite the differences in the data sets, comparison with
observations shows limited biases from EMEP (even though the mean normalized
gross errors are large) and good correlations.
Compared to the five urban sites provided by the US Embassy and consulates,
a limited agreement is found (Fig. 6), with an underestimation in PM2.5 by EMEP for all sites, especially in winter. This comparison shows however
a fair agreement, especially given the large variability in the
observations, as over New Delhi on 16 July 2014 with a PM2.5 surface
concentration ranging from 5 to 955 µg m-3. Our reference
simulation has also been compared with the data provided by
https://openaq.org for 2016 (Fig. S4). The observations show a large
variability within each month, making the interpretation of this comparison
difficult. A chemical speciation in the measurements will be helpful to
interpret the biases found over these cities. Indeed, the EMEP model
predicts a large contribution from primary particulate matter (PPM) to
PM2.5, reaching 50 % in December and in January, mainly composed of
primary organic matter (not shown), over the sites presented in Figs. 6 and S4. The model also predicts a main natural contribution to PM2.5 from
May to September over these sites. For example, the site of Hyderabad
reaches up to 70 % in dust in July. An evaluation of the source
attribution of the PM2.5 simulated by the EMEP model would be
instructive.
Finally, it should be noted that for these simulations, the EMEP model is
driven by climate–model meteorology. Such meteorology is more statistical in
nature than the assimilated Numerical Weather Prediction meteorology
normally used with the EMEP model, and by its nature (non-assimilated), such
climate meteorology cannot reproduce actual meteorology for the periods
studied. It is also important to recall that, even with the use of recent
inventories, uncertainties in emissions may persist (e.g. Saikawa et al.,
2017).
Overall, however, the results suggest that the PM2.5 concentrations
simulated by the EMEP model with this setup provide a fair representation of
the surface concentrations observed at the Indian monitoring sites, even if
the model tends to underestimate the highest concentrations and overestimate
the lowest ones.
(a) Distribution of surface O3
mixing ratios (in ppb) for the reference scenario. Distribution of the relative difference and absolute difference
in surface O3 between the reference scenario and
the FC2030 scenario (b) and the FC2050 scenario (c).
The relative differences are calculated as ([FC – reference] / reference) × 100 %, and the absolute differences as [FC – reference]. Grey
dots mark grid points that do not satisfy the 95 % level of significance.
Impact of climate
In this section, we analyse the differences between the FC scenarios (at
short-term and medium-term, i.e. FC2030 and FC2050) and the reference
scenario. All meteorological fields and pollutant concentrations were
averaged over their respective decade of simulation. It is important to
recall that uncertainties in the representation of meteorological conditions
can impact our chemical results even if consistencies in the projections
were simulated, especially during the monsoon and pre-monsoon periods, as
explained in Sect. 2.1.
O3
The reference scenario shows large surface O3 over Tibet, east India
and over the Bay of Bengal along the Indian coast (Fig. 7). The large values
seen over Tibet are mainly the result of topographical effects, since
O3 values generally increase with altitude (e.g. Loibl et al., 1994).
High O3 near coastal areas is also expected, since the deposition
velocity of O3 is very low over sea (e.g. Ganzeveld et al., 2009), thus
minimizing the near-surface sink which usually affects land areas.
Increased temperatures associated with climate change would be expected to
coincide with a rise in surface O3 due to the correlation
between O3 production and temperature in polluted areas as explained by
Jacob and Winner (2009), although such relationships are often weak (Langner
et al., 2005, 2012) and less clear in background areas. This correlation is
not obvious in our simulated projections, presumably due to the large number
of other factors which change, such as humidity levels, mixing heights,
other meteorological changes, and biogenic emissions which are affected by
climate change. As our model does not include any CO2 inhibition effect
on isoprene emissions (e.g. Guenther et al., 1991; Arneth et al., 2007), or
potential changes in vegetation in a different climate, these biogenic
emissions are simply a function of temperature and increase in the FC
scenarios. The uncertainties associated with these assumptions are however
difficult to quantify. For example, Hantson et al. (2017) found global
isoprene emissions for the period 2071–2100 to be 544 TgC yr-1 without
CO2 inhibition, but only 377 TgC yr-1 with this effect (i.e -31 %). For
monoterpenes, the equivalent figures were 35.7 and 24.8 TgC yr-1 (also
-31 %). Young et al. (2009) estimated even bigger changes for isoprene,
from 764 to 346 TgC yr-1, and showed that this uncertainty can indeed
have strong effects on surface O3 levels. The largest changes were
found in South America and Africa, though annual changes over India were
only around 5–10 %. Although significant, these changes are model
estimates only. The experimental data behind the CO2 inhibition effect
are extremely limited, and as noted in Simpson et al. (2014) and references
therein, current knowledge is insufficient to make reliable predictions on
this issue.
(a) Seasonal distribution of O3 and
relative difference between the reference scenario and the FC2050 scenario. (b) Seasonal distribution of O3 deposition
velocity and relative difference between the reference scenario and the
FC2050 scenario. The relative differences are calculated as
([FC2050 – reference] / reference) × 100 %. Regions discussed in
the text are noted on the distributions of relative difference. Grey dots
mark grid points that do not satisfy the 95 % level of significance.
While the regions with a change in O3 when using the FC2030 scenario
are relatively scattered, the use of the FC2050 scenario highlights a
clear north–south difference over land (Fig. 7). This is
characterized by an increase in surface O3 concentrations over the
northern part of India (by up to 4.4 %, +2 ppb) and a decrease over the
southern tip of India reaching -3.4 % (-1.4 ppb) (Fig. 7). The changes
are statistically significant at the 95 % level for both FC scenarios
showing a robust effect due to the climate change.
(a) Distribution of surface PM2.5
concentrations (in µg m-3) for the reference scenario. Distribution of the
relative difference and absolute difference in surface
PM2.5 concentrations between the reference scenario
and the FC2030 scenario (b) and the FC2050 scenario (c). The relative differences are calculated as ([FC – reference] /
reference) × 100 %, and the absolute differences as [FC –
reference]. Grey dots mark grid points that do not satisfy the 95 % level
of significance.
The correlation between the temporal change in O3 (ΔO3)
and ΔT over land is limited in FC2030 and FC2050 scenarios. This
shows that for both FC scenarios, even though the change in temperature is
statistically significant (not shown), other processes are occurring which
impact on the thermal influence on the photochemical production of O3.
Figure S5 shows the change in one important process, the O3 deposition
velocity, Vd(O3). The distribution of relative difference in O3 is
linked to the distribution of relative difference in Vd(O3) for both FC
scenarios, especially in the FC2050 scenario. Wu et al. (2012) already
identified a slight increase in O3 deposition in the south of India and
over the Western Ghats due to an increase in the leaf area of broadleaf
forests but such processes are not included in our model. Instead, the
changes in Vd are due to the factors which control stomatal conductance
(gs) in the EMEP model, namely temperature, humidity (vapour pressure
deficits), radiation and soil moisture (Emberson et al., 2001; Simpson et
al., 2012). In northern European conditions, an increase in temperature will
usually result in an increase in gs, but in India, temperatures are
often above the optimum values, and increases in temperatures may decrease
gs. The other factors will also affect the sign of changes in gs –
such as soil moisture, shown in Fig. S6. Figure S6 shows the large impact of
changes in soil moisture on the variation in Vd(O3) for both FC
scenarios. The monthly variation in soil moisture matches the variation in
Vd(O3) rather well.
With regard to seasonal changes and focusing on the FC2050 scenario (Fig. 8), where the signatures in the change in O3 are more significant
(similar plots for the FC2030 scenario shown in Fig. S7), the impact of
Vd(O3) is clearly visible. Exceptions are modelled over three regions
as annotated in Fig. 8, where they are labelled as (A), (B) and (C) in the
distribution of the relative differences. For these regions, the deposition
velocity is correlated with the surface O3, in contrast to the
anti-correlation found over the rest of the domain.
During the pre-monsoon period, region (A) is characterized by a high level of
NMVOCs and NOx. During the winter, the regions (B) and (C) are
characterized by a high level of NMVOCs and a low level of NOx (Fig. S8).
During the pre-monsoon period, a decrease in NOx and NMVOC is simulated
over region (A) (Fig. S8). The reduction of these two precursors may explain
the decrease in O3. The two other regions, regions (B) and (C), are both
characterized during winter by a decrease in NOx and an increase in
NMVOCs. Combined with the increase in O3, this result gives an
indication of the presence of a VOC-sensitive regime. This contrasts with
the NOx-sensitive regime otherwise prevailing in India as calculated by
Sharma et al. (2016) and observed by Mahajan et al. (2015). It is however
interesting to note that the presence of a VOC-limited regime over region
(A) during the pre-monsoon period and over region (B) in winter, was already
observed by satellite measurements (Mahajan et al., 2015).
The NMVOCs for the reference scenario over region (C), corresponding mainly
to Myanmar, are probably from biomass burning as the peak forest fire season
in this region occurs in winter (e.g. van der Werf
et al., 2010 or Pommier et al., 2017).
For the FC2030 scenario, an identical pattern is observed with an
anti-correlation between the relative difference in O3 and the relative
difference in Vd(O3), also with the exception of three other regions
(A′, B′ and C′) as shown in Fig. S7. This shows that the change in O3 is
related to the change in Vd(O3), except over three regions, as for the
FC2050 scenario. Over these three regions, the complementary effect of
NOx–NMVOCs is also obvious in this scenario (Fig. S9). The change in
location of the three regions between the 2030s and the 2050s shows that the
local meteorology has an impact on the change in the chemistry, such as the
surface temperature. Indeed, the changes in temperature are not homogeneous
over the domain and vary with the seasons.
PM2.5
In the reference scenario, the largest surface PM2.5 concentrations are
located over the Indo-Gangetic Plain (Fig. 9), known to be a highly
populated area (e.g. Chowdhury and Maithani 2014; or
http://www.census2011.co.in/states.php) and as a large source of pollutants
emissions (e.g. Clarisse et al., 2009; Mallik and Lal, 2014; Tiwari et al.,
2016).
According to these calculations, climate change has a larger impact, in
terms of absolute values, on PM2.5 than on O3. Climate change is
predicted to lead a fairly homogeneous rise in surface PM2.5 levels
over India, especially for the FC2050 scenario, by up to 6.5 % (4.6 µg m-3) (Fig. 9). This maximum increase is located over the
Indo-Gangetic Plain, where a decrease in surface wind speed is predicted (not
shown). The decrease in wind speed may limit the emission of dust and the
dispersion of the PM2.5 emitted over this area. In both FC scenarios,
an increase in surface PM2.5 concentrations is predicted for the
western part of the domain (Arabian Sea) and a decrease over the eastern
part of the domain (Bay of Bengal). It is worth noting that with a mesoscale
model, Glotfelty et al. (2016) also simulated an increase in PM2.5 over
India. However, a proper comparison with other studies remains difficult, as
different models or scenarios were used. It is also noteworthy that the
changes in PM2.5 are statistically significant at the 95 %
confidence level.
The distribution of the relative difference in PM2.5 is roughly
homogeneous in the FC2050 scenario over India (Fig. 9) but it does not match
the pattern of precipitation change (Fig. S10). As PM2.5 is highly
sensitive to wet scavenging, we would expect an impact of changes in
precipitation on the change in PM2.5, but this relationship is not
shown in these distributions (Figs. 9 and S10).
The composition of PM2.5 is mainly dominated by dust, OM and
secondary inorganic aerosol (SIA). SIA includes SO42-,
NO3- and NH4+. The seasonal distribution of their
contribution on PM2.5 provides complementary information on the
composition of PM2.5 (Fig. S11). Generally speaking, dust dominates
during the pre-monsoon and monsoon periods over India, while the amounts of
OM and SIA are large during the post-monsoon period and in winter. It is also worth
noting that PM2.5 over the Arabian Sea and Tibet are mostly
influenced by dust for each season. Dust over the Arabian Sea originates
from the Sahara Desert, while the Tibet Plateau is a known regional source
of dust (e.g. Xu et al., 2015; Xin et al., 2016). PM2.5 over the
Bay of Bengal is largely impacted by dust during the monsoon and OM during
the winter.
Seasonal distribution of surface
PM2.5 concentrations (in µg m-3) for the reference
scenario, and seasonal composition of PM2.5 (in
µg m-3) for the three
regions highlighted by black boxes on the map for the reference and the
FC2050 scenarios. The black percent corresponds to the relative difference
in PM2.5 between both scenarios for each region. Note
the different y axis for Region 3.
Distribution of the relative difference (a, c) and
absolute difference (b, d) in surface O3 between
the reference and the FCE2030 scenario (a–b) and the FCE2050 scenario
(c–d). The relative differences are calculated as ([FCE–
reference] / reference) × 100 %, and the absolute differences as
[FCE – reference]. The black box delimits the region described in the
text.
Seasonal distribution of the relative difference in
surface O3 between the reference scenario and the
FCE2050 scenario. The relative differences are calculated as ([FCE2050 –
reference] / reference) × 100 %. Regions discussed in the text
are noted on the distributions for their respective season.
Distribution of the relative difference (a, c) and absolute difference (b, d) in surface PM2.5 between the reference scenario and the FCE2030 scenario
(a–b) and the FCE2050 scenario (c–d). The relative differences
are calculated as ([FCE – reference] / reference) × 100 %, and
the absolute differences as [FCE – reference]. The black box delimits the
region described in the text.
The simulated OM is mainly composed of SOA. It is also interesting to note
that the OM over Myanmar (region C in Fig. 8) is strongly influenced by
primary emissions from fires and spatially coincides with the O3
production seen previously in Fig. 8. SOA is predicted to increase, by up to
19 % for FC2030 and up to 33 % for FC2050 over India. This rise is
probably due to an increase in biogenic VOCs as suggested by Heald et al. (2008) (see also Fig. S8b) as a result of temperature increases. As noted
above though, isoprene emissions might actually be inhibited by CO2
effects in a future climate, and neither Heald's model nor ours accounts for
such effects.
In order to better interpret the seasonal process, more detailed examples
over India for the FC2050 scenarios with three regions are shown in Fig. 10.
The results with the FC2030 scenario (not shown) lead to similar
conclusions. The composition of PM2.5 over these regions coincides
with the overall description provided by Fig. S11, i.e. there is a large
amount of dust during the pre-monsoon and the monsoon; and OM and SIA during
the post-monsoon and the winter. Wind speed is also higher during the
pre-monsoon and the monsoon for these three regions, explaining the large
amount of dust during these seasons. The budget of dust is sensitive to
precipitation, while OM and SIA are also highly related to chemistry, as
described hereafter.
Indeed, region (1), representing mainly a rural area, is subject to a large
decrease in PM2.5 by 8 % during the monsoon. This is mainly due to
the reduction in dust, representing 55 % of PM2.5, largely scavenged
by the increased precipitation (+36 %) (as explained in Section 2.1).
The increase in PM2.5 during the pre-monsoon and during the winter is
linked to the increase in dust by 15 % and in OM by 10 %, respectively.
This increase in dust depends on the change in precipitation (10 %
decrease) and probably also on the increase in wind speed by 3 %. The
augmentation in OM is related to the increase in biogenic emissions as
isoprene (+14 %) and monoterpene (+11 %). During the post-monsoon,
the slight rise in PM2.5 is mainly due to the increase in OM and SIA.
The impact of dust is also still high for a region located far from the
desert as region (2), but the change in the PM2.5 level is also largely
related to the change in SIA and OM in all seasons. Region (2) experiences a
larger change in PM2.5 during the monsoon (-5 %) related to the
increased precipitation (+35 %) and the post-monsoon (+7 %), probably
linked to the increase in isoprene and in monoterpene emissions (+13 %
and +11 %, respectively). The reduction in precipitation by 25 %
during the pre-monsoon probably explains the increase in PM2.5.
For region (3), located within the Indo-Gangetic Plain and which includes
Delhi, the largest variation in PM2.5 by 20 % is modelled during the
post-monsoon. This shows that this region is affected by a large penalty
from the climate change on surface PM2.5 concentrations during the
post-monsoon. This increase is caused by the rise in both SIA (+29 %)
and OM (+21 %) and probably by the reduction of the dispersion as
predicted by the decrease in the surface wind speed by 5 %. The
augmentation in SIA and OM may be related to the large increase in isoprene
and in monoterpene emissions (+19 % both), explained by increased
temperature. Among all the seasons and among the three selected regions, the
larger increase in temperature (+0.6 %) occurs in this case. It is also
worth noting that it coincides with the larger growth in O3 among these
three regions (+6 %). The changes during the pre-monsoon and the winter
are mainly due to the variation in SIA, and the joint changes in SIA and OM,
respectively. The decrease in PM2.5 during the monsoon is linked to the
reduction in dust and in SIA (by 5 % for both), which are linked to the
increase in precipitation (+16 %) over this wet region (2.8 mm day-1).
In addition to confirming the seasonal variation in the composition of
PM2.5 over India as shown in Fig. S11, these three
cases show that the main parameters influencing the changes in the main
components (SIA, OM and dust), are the precipitation, the biogenic emissions
and the wind speed.
Impact of future emission scenarios combined with climate change
By combining the climate effect with future changes in emissions, we explore
the differences between the FCE scenarios (2030 and 2050) and the reference
scenario. As in the previous sections, the simulated fields were averaged
over their respective period of simulation.
O3
For both FCE scenarios, a substantial increase in O3 over India is
modelled, as shown in Fig. 11. This augmentation in O3 reaches 13 % or
5 ppb in the 2030s (mean of 3 % or 1 ppb) and reaching 45 % or 18 ppb
in the 2050s (mean of 13 % or 6 ppb) within the domain defined by the
black box in Fig. 11 (latitudes 8–38∘ N and the longitudes
68–90∘ E). The increase in O3 is noticeable during the four
seasons but it is more intense during the monsoon as shown by Fig. 12. It is
worth noting that there is a decrease in O3 over the Western Ghats
during the monsoon – e.g. region α in Fig. 12: -12 % in 2030 (not
shown) and -4 % in 2050 – while the rise in O3 over the rest of the
country is larger than for the other seasons. This contrast between the
Western Ghats and the rest of India is more pronounced in the FCE2030
scenario. Another region (labelled as β) in winter, is also
characterized by a decrease in O3 – -11 % in 2030 (not shown), -4 %
in 2050 (Fig. 12). Both reductions can be explained by the NOx–VOC
chemistry. Both precursors largely increase in the FCE2030 and FCE2050
scenarios as shown by the large relative differences presented in Fig. S12.
However, regions (α) and (β) present a decrease in their
NMVOC / NOx ratio in the future (Fig. S12). This ratio is already lower
in the reference scenario for both regions (≤ 16, Fig. S12) than in the
rest of India since the mean ratio over land covering the area defined in
Fig. 11 is close to 60. This means that NOx increases more for these
regions than NMVOC, probably developing a NOx-saturated regime and
causing the O3 depletion. Thus both regions, for their respective
season, suggest a VOC-sensitive regime for the FCE2030 and FCE2050
scenarios.
This substantial increase in O3 leads to a large increase in the ozone
health indicator, SOMO35. The SOMO35 metric is defined as the annual sum of
daily maximum running 8 h average O3 concentrations over 35 ppb. The
SOMO35 levels for the reference scenario are already higher (Fig. S13) than
over Europe (e.g. van Loon et al., 2007; EMEP Status Report 1/2017), probably
related to the warmer climate and the large emissions of O3 precursors
over India, and the overestimation in O3 from the model as shown in
Sect. 3.1. SOMO35 is predicted to significantly increase for both FCE
scenarios (Fig. S13).
PM2.5
Climate change has a non-negligible impact on surface PM2.5 concentrations, but this impact is small compared with the effects of
emissions in the FCE scenarios. Looking at the PM2.5 in Fig. 13, a
large increase is simulated throughout the domain. This rise in surface
concentrations is larger in the FCE2050 scenario than in the FCE2030
scenario. Within the region delimited by the black box in Fig. 13 (same as
Fig. 11), the mean rise in PM2.5 is equal to 37 % (13 µg m-3) in the 2030s and to 67 % (23 µg m-3) in the 2050s. These
increments alone are comparable to the annual threshold that WHO recommends
not to exceed, i.e. 10 µg m-3, for the FCE2030 scenario, and
double that for the FCE2050 scenario. This increase in concentrations is also
large for each season (Fig. S14). It has a maximum during the post-monsoon
in both scenarios, reaching 117 % (119 µg m-3) in the 2030s
(not shown) and 172 % (168 µg m-3) in the 2050s. These huge
numbers prefigure an enormous increase in fine particulate matter over
India, as already suggested by Amann et al. (2017), and imply serious health
issues for the population, especially children (UNICEF, 2016).
As expected by the large increase in emissions as for SOx and NOx
presented in Fig. 1, the future concentrations of PM2.5 are influenced
by SO42-, NO3- and NH4+ for each season.
These compounds also show the largest increase during the post-monsoon
season. This is particularly obvious for the three selected regions of Fig. 10 since SIA increases by at least 100 % in the FCE2030 scenario and by at
least 200 % in the FCE2050 scenario (Fig. S15). The larger increase in
PM2.5 is simulated over region (2) for both FCE scenarios during the
post-monsoon; by 75 % in the 2030s and 132 % in the 2050s (Fig. S15).
Region (3), characterized by the large impact of climate on its PM2.5
during the post-monsoon as shown previously in Fig. 10, has an increase in
PM2.5 by around 69 % in FCE2030 and 112 % in FCE2050.
While the surface PM2.5 over the land region delimited in Fig. 13 is
composed on average by 5.1 % of NH4+, 6.8 % of NO3-, and 9.7 % of SO42- for the reference
scenario; their mean contribution grows and becomes respectively 6.7,
7.2 and 13.6 % in the 2030s and 7.8, 7.5 and 16.8 % in the
2050s. OM and the dust remain two major components of surface PM2.5 but in the 2030s, SIA becomes the second largest component since it
represents 28 % of PM2.5 (29 % for dust and 19 % for OM) and
the main component in the 2050s with 32 %, while dust represents 25 %
and OM corresponds to 18 % of PM2.5. It is also worth noting that
even though the PPM are high for the three scenarios (close to 20 % of
PM2.5), the amount of EC within these PPM remains low, around 15 %.
It is interesting to note that even under increasing anthropogenic emissions
a significant fraction of PM2.5 comes from sources (dust and some
fraction of SOA) that are challenging to control through policy measures.
Still, even biogenic SOA is partly the product of anthropogenic emissions
(and certainly land-use policy, e.g. Tsigaridis and Kanakidou, 2007;
Ashworth et al., 2012), and dust is also partly a function of land-use and
climate change. But such interactions are beyond the scope of our study.
Conclusions
Driven by downscaled meteorological fields, the EMEP model was applied to
investigate the impact of changes in regional climate and emissions on
surface O3 and PM2.5 over India. The evaluation of the reference
scenario with surface-based observations suggests a fair simulation of the
seasonal variation of O3 and a good representation of surface
PM2.5 concentrations over Indian cities. Additional information on the
chemical components in PM2.5 will be helpful to interpret the
differences and confirm the large part of primary organic matter simulated
in winter by EMEP and the high ratio of dust during the monsoon. EMEP
overestimates O3 by 11 ppb and we suspect that NOx titration over
cities, unresolved by a rather coarse grid (ca. 50 km), and possibly
uncertainties in the emissions, are the main cause, especially in winter.
However, there is a lack of reliable available measurements of NOx and
O3 to fully validate this assumption.
The O3 change due to regional climate change for the medium-term
(FC2050) scenario highlights a clear north–south gradient over India,
with an increase over the north, by up to 4.4 % (2 ppb) and a decrease
over the south, by up to -3.4 % (-1.4 ppb). This O3 budget is highly
impacted by the change in O3 deposition velocity due to the change in
soil moisture, and over a few areas by the biogenic NMVOCs. Climate change
leads to increases in the PM2.5 levels at short and medium-terms,
reaching a maximum of 6.5 % (4.6 µg m-3) over the Indo-Gangetic
Plain by the 2050s. The PM2.5, mainly composed of dust, OM and SIA, is
mainly controlled by change in precipitation and biogenic emissions. For
example, over the Indo-Gangetic Plain, an increase of 20 % during the
post-monsoon is predicted, related to a rise in isoprene and in monoterpene
emissions, while a rural region is characterized by a 8 % decrease in
PM2.5 during the monsoon, linked to the increased precipitation in
2050.
A large increase in anthropogenic emissions is predicted if no further
policy efforts are made. Combined with climate change impacts; these
emissions are predicted to lead to large changes in surface O3 and
PM2.5. For surface O3, these changes reach 45 % over some
regions in 2050. This augmentation is substantial for each season, with the
exception of two regions – as e.g. over the Western Ghats during the monsoon
characterized a decrease in O3 around -12 % in 2030 (-4 % in 2050)
related to the dependence of O3 production on the NOx and VOC
concentrations.
India is predicted to suffer large increases in PM2.5 levels due to the
increases in anthropogenic emissions in this no-further control scenario.
The increase in PM2.5 will occur rapidly since the mean rise is close
to 37 % for the short-term scenario (2030s) and 67 % for the medium-term
scenario (2050s) over the main part of the country. The PM2.5 levels
are predicted to reach very high levels, up to a maximum of 117 % (119 µg m-3)
increase in the 2030s and 172 % (168 µg m-3)
in the 2050s during the post-monsoon season. In the 2030s, the SIA will
become the second largest component of PM2.5 over India, exceeding the
amount of OM by reaching a ratio close to 28 % and the main component in
the 2050s with 32 %.
Finally, we note that this is the first serious attempt to use the EMEP
model over the Indian subcontinent, and there are likely many improvements
needed before modelling skill achieves the same level as obtained in European
simulations. For example, the vegetation characterization used in the EMEP
model was focused on European vegetation, and is probably not fully suitable
for India, which may affect the response in temperature over India. Many
issues affect any modelling study for this region. For example, emission
rates of biogenic VOC from vegetation over India are also largely unknown;
Guenther et al. (2006) show only one site in or near the Himalayas – and
nothing over the rest of the Indian sub-continent. Emissions of other
compounds are also rather uncertain. Proper model evaluation in this region
would require quality-assured measurements of a range of compounds in rural
as well as urban areas. Still, given the rapidly increasing emission and
pollution levels in India, it is clear that further efforts are warranted,
and increasing attention will improve the basis for future model
verification and hence for a sounder basis for emissions policy assessments
in future.