Source influence on emission pathways and ambient PM2.5 pollution over India (2015–2050)

India is currently experiencing degraded air quality, and future economic development will lead to challenges for air quality management. Scenarios of sectoral emissions of fine particulate matter and its precursors were developed and evaluated for 2015–2050, under specific pathways of diffusion of cleaner and more energy-efficient technologies. The impacts of individual source sectors on PM2.5 concentrations were assessed through systematic simulations of spatially and temporally resolved particulate matter concentrations, using the GEOS-Chem model, followed by population-weighted aggregation to national and state levels. We find that PM2.5 pollution is a pan-India problem, with a regional character, and is not limited to urban areas or megacities. Under present-day emissions, levels in most states exceeded the national PM2.5 annual standard (40 μg m−3). Sources related to human activities were responsible for the largest proportion of the present-day population exposure to PM2.5 in India. About 60 % of India’s mean population-weighted PM2.5 concentrations come from anthropogenic source sectors, while the remainder are from “other” sources, windblown dust and extra-regional sources. Leading contributors are residential biomass combustion, power plant and industrial coal combustion and anthropogenic dust (including coal fly ash, fugitive road dust and waste burning). Transportation, brick production and distributed diesel were other contributors to PM2.5. Future evolution of emissions under regulations set at current levels and promulgated levels caused further deterioration of air quality in 2030 and 2050. Under an ambitious prospective policy scenario, promoting very large shifts away from traditional biomass technologies and coal-based electricity generation, significant reductions in PM2.5 levels are achievable in 2030 and 2050. Effective mitigation of future air pollution in India requires adoption of aggressive prospective regulation, currently not formulated, for a three-pronged switch away from (i) biomass-fuelled traditional technologies, (ii) industrial coal-burning and (iii) open burning of agricultural residue. Future air pollution is dominated by industrial process emissions, reflecting larger expansion in industrial, rather than residential energy demand. However, even under the most active reductions envisioned, the 2050 mean exposure, excluding any impact from windblown mineral dust, is estimated to be nearly 3 times higher than the WHO Air Quality Guideline.

. Spatial proxies used to distribute emissions..  Table S1. Spatial proxies used to distribute emissions Source Category/Activity Proxies Reference

Brick production
Distributed at district level using district-wise no. of household built using burnt bricks, excluding the cities with high population densities and distributing the emissions from those city grids in the surrounding grids.

Census, 2011
Food and agro processing (Jaggery making, cashewnut processing unit, tea and coffee drying, spices drying, silk reeling and dairy processing) Jaggery makingdistrict level sugarcane produced Ministry of Agriculture; Cashewnut, tea, coffeedistributed to specific districts with production as proxy Cashew Manufacturer's Association; Spices dryingdistrict level spices produced Indian Tea Association; Silk reelingdistributed to specific states (production) carrying this activity with rural population further at district level  Uncertainties in the activity rates were calculated analytically, assuming normal distribution for the underlying uncertainties in all input quantities. For each input: (a) the mean and standard deviation calculated from a set of available (three or more) data points; (b) upper and lower bounds assumed based on two data points; or (c) a representative uncertainty assumed from similar data, where only one data-point exists. Uncertainty in the emission factors was estimated from the standard deviation in the set of compiled emission factors of a particular pollutant from a particular fuel technology combination. If the emission factor being used was taken from a single reported source, the reported rating was quantified using the percentage errors cited in IPCC (2006a,b) and EMEP (2009). The measured emission factors with unspecified uncertainties were assigned the highest-known uncertainty for the same pollutant and those from similar technologies. Wherever emission factor measurements for a technology were not available an emission factor from a similar technology was chosen and assigned 100% uncertainty (<5% of the technologies fall under this category, including fluidized bed combustors and sponge-iron kilns). A spreadsheet-based approach was developed for combining uncertainties in activity rates and emission factors. A normal/lognormal distribution was assumed for when standard deviation was less/greater than 30% of the mean. Uncertainty propagation in the product of two variables was followed using the sum-ofquadrature rule, calculated analytically. The upper and lower emission bounds were calculated using the resultant lognormal parameters (geometric mean and geometric standard deviation).

S2.
Future emission pathways S2.1. Methodology Figure S1. Methodology for estimation of future sectoral activity, apportionment to technology mix and related scenario based emissions. Figure S2. Sectoral Growth between 2015-2050. Growth rates were computed based on analysis of existing data and reviewed literature.  [Ray et. al., 2013]

S2.3 Evolution of technology mix
In 2015, power generation was almost entirely from subcritical pressure thermal power plants with an average gross efficiency of 30.5% (IEP, 2006;IESS, NITI Aayog 2015). A switch to more efficient technologies such as supercritical (SC), ultra-supercritical (USC), and integrated gasification combined cycle (IGCC) is expected in future. For 2030 and 2050, respectively, the non-fossil shares were assumed to be 30% and 40% in REF, 40% and 60% in S2, and 75% and 80% in S3. The assumed technology mix in S2 follows the NDC's proposed non-fossil share of 2030. In S3, it is consistent with high efficiency-low carbon growth cases in earlier studies (Anandarajah and Gambhir 2014;Shukla and Chaturvedi 2012;Level 4, IESS, Niti Aayog, 2015). The transition of thermal power plants sub-critical boiler technology to more efficient technologies like super-critical, ultra-super-critical and integrated gasification combined cycle (IGCC) is based on published scenarios (IEP, 2006;IESS, Niti Aayog, 2015).
Emissions from on-road vehicles are based from a previous study (Pandey and Venkataraman, 2014). The detailed list of vehicle category is included in the study (Table 3, Pandey and Venkataraman, 2014). Twowheelers contribute the most to the fleet of private vehicles with approximately 82% share, followed by passenger cars (15%) and three-wheelers (3%). For present day, all vehicles are assumed to be compliant with BS III standards with 2 wheelers having the highest emission levels for PM2.5 followed by three wheelers (0.5 times lower) and gasoline cars (0.1 times lower). Future shifts to BS IV and BS VI emission standards lead to reductions in emission levels by 80% and 90% respectively. The emissions standards for vehicular emissions in India are based on European standards. The values of emission limits as prescribed for BS IV and BS VI are equivalent to those in EURO IV and EURO VI, while the US standards are more stringent than BS standards (DieselNet, 2018). In the transport sector, current technology shares are 81% private vehicles (two-wheeler, three-wheeler and cars) and 19% public vehicles (buses and taxis) (Pandey and Venkataraman, 2014). The share of private vehicles is projected to increase in a reference scenario till 2030, especially for two-wheelers and cars (NTDPC, 2013;Guttikunda and Mohan, 2014). However, beyond 2030, as GDP stabilizes, no further increase in private vehicle share is assumed, with a greater demand for public transport. Therefore, in the S2 scenario, private vehicle share is assumed as 75% and 70% in 2030 and 2050, respectively. For S3 private vehicle share is assumed to decrease rapidly to 60% in 2030 and 40% in 2050 in consistent with Level 2 of IESS (NITI Aayog, 2015) (Table S5). For future emissions, Auto Fuel Policy (Auto Fuel Policy Vision 2025, 2014) recommendations were applied, wherein 2/3-wheelers were proposed to have Bharat Stage (BS)-IV standards from 1st April 2015, light and heavy duty diesel vehicles to have BS-Va and BS-Vb. There is a recent proposal to leapfrog directly to BS-VI for all on-road vehicle categories by 2020 (MoRTH, 2016). However, scenarios used here, do not reflect such a quick change, keeping the share of BS-VI at modest levels owing to expected delays in availability of BS-VI compliant fuels and/or difficulties in making the technologies adaptive to Indian road conditions as well as cost-effective (ICRA, 2016), along with the use of non-BS-VI compliant vehicles in peri-urban and rural areas. In the transport sector, engine efficiency improvements are not foreseen to have significant increases across technologies (e.g. across BS-III to BS-VI) as these standards primarily govern the control of emissions of air pollutants. Until 2015 there were no fuel economy standards for India. However, energy efficiency improvement are assumed over the years in the S3 scenario keeping in mind the recently proposed fuel economy targets (MoP, 2015) In the brick sector, currently 76% of total bricks are produced by Bull's trench kilns (BTK) and 21% by clamp kilns. Clamp kilns are highly polluting, with sun-dried bricks, stacked alternately with layers of powdered fuel, allowed to smolder until the bricks are baked. The demand for non-fired-brick walling materials is currently negligible, but expected to rise (10-25% in REF, 30-45% in S2 and 40-75% in S3 for 2030-2050), from increased availability of hollow-block technology and the governmental incentives for fly-ash bricks (UNDP, 2009). For fired bricks, cleaner technologies include a retrofit to existing Bull's trench kilns, called zig-zag firing, or significantly more capital intensive, vertical shaft brick kilns (VSBK) which have increased efficiency. For small clamp kilns, it is believed that regulation may not be effective, so a constant activity level, but a decreasing share was assumed in future, with new cleaner technologies filling growing demand (personal communication, Maithel, 2015).
Evolution of technologies in informal industry from say traditional wood furnaces, presently supplying all energy requirements, to gasifier and LPG based technologies is assumed to increase in 2030 and 2050 respectively, to 20% and 35% in S2 and 65% and 80% in S3 (Table S5).
India's rural population largely depends on biomass fuels for cooking and lighting (Venkataraman et al. 2010). Although India has introduced improved biomass cook-stoves to improve fuel efficiency and to reduce smoke exposure using chimneys or combustion improvements, further technological improvements or alternatives are required to reach LPG-like emission levels to reduce disease risk due to household biomass burning. The REF scenario assumes an increasing penetration rate of liquefied petroleum gas (LPG) and piped natural gas (PNG) typical of 1995-2015 (Pandey et al. 2014). In the S2 and S3 scenarios, assumed future switch in residential energy to use of LPG/PNG or low-emission biomass gasifier stoves and biogas, is consistent with energy efficiency increases proposed in Levels 2 and 4 of the IESS (NITI Aayog 2015). We use lower rates of clean technology adoption in the residential sector in both the REF and S2 scenarios, because no current legislation or standards target this sector, but a complete switch away from traditional biomass fuels in S3. In case of lighting, 37% usage is of highly polluting kerosene wick lamps and lanterns, which emit large amounts of black carbon (Lam et al. 2012), while the balance is of electricity, with less than 1% solar lamps. Residential lighting is assumed to shift from a modest present-day dependence on kerosene to a complete switch to electricity and solar lamps in 2030 and 2050 (National Solar Mission 2010), a change expected with a national promotion of renewable energy.
In the agricultural sector it is assumed, based on satellite active fire cycles in agricultural land-use areas (Venkataraman et al., 2006), that residues of cereal and sugarcane are burned in field. Gupta (2014) indicated greater mechanization of agriculture, with decrease in amounts of residue, but increase in incidence of field burning, needed to clear the rubble consisting of 6-12 inch stalks, before sowing. Mulching technology was reported to allow sowing even through rubble and loosely spread residue, thus avoiding burning for field clearing. The present work applies different levels of mulching, replacing field burning, in future years (Table S5).  , 2015). Under this scheme, every industry (includes power plants and heavy industries, referred to as "designated consumers" in the scheme) must meet a certain energy efficiency target by implementing appropriate and timely technological reforms. Thus, for industries also, the specific energy per unit activity is representative of the level of penetration of the PAT scheme across different industries over time under each scenario.

Figure S3. Energy Evolution in Scenarios REF, S2 and S3
Much of the energy demand in S1 is from electricity generation which is majorly fossil fueled, industry (coal and biomass fueled), in residential biomass is dominantly used as fuel. In scenarios S2 and S3 use of energy efficient technologies like Non-carbon fuel use thermal power, PAT implementation in industries and cleaner technologies in brick production, LPG use in residential and energy efficient standards in transport can help to lower the energy demand.

S2.5. Technology linked emission factors
For thermal power, emission factors (Table S8) assumed a mean 38% ash content coal, typical of India, with electrostatic precipitators (ESP) working at 99.98% while more efficient supercritical, ultra-super critical and IGCC technologies, had emission reductions in proportion with increased energy efficiency. In December 2015, the Indian Ministry of Environment and Forests issued new norms for thermal plants with emission standards for SO2 and NOx (MoEFCC,2015). Reported barriers to quick adoption of desulphurization and de-NOx technologies (CSE, 2016), lead to assumptions here of low rates of flue gas desulphurization technology adoption. Preliminary surveys show little progress in the implementation of new standards, mainly due to insufficient knowledge in advanced pollution control technologies and lack of i) space for installation, ii) storage for raw materials and iii) clarity on cost recovery (CSE, 2016). Similarly, in heavy industries like cement, iron and steel, fertilizer and non-ferrous, 90% (S1 and S2) and 100% (S3) operation of existing controls are considered while emission factors for PAT technologies were reduced below non-PAT values using their increase in efficiency (Table S8).
It was assumed that non-fired brick production, which uses cement, involves no use of fuel for firing or drying purposes, hence produces no emissions at the stage of brick production, to avoid double-counting of emissions related to feedstock, which are accounted in cement production. In informal industry, the use of traditional biomass technologies for major thermal and drying operations was assumed shift to cleaner gasifier or LPG technologies, hence, emission factors similar to those for residential cooking were considered. In the residential sector, available measurements (reviewed in Pandey et al. 2014) were used to derive emission factors for wood, dung-cake, crop residue combustion in cook stoves, as also for kerosene and LPG cook stoves, which are also used for biomass fired water-heating and space-heating. Diesel generator sets, for residential use and for mobile towers have been included, whose emission factors are set similar to measured factors for agricultural diesel pumps.
In the agriculture sector, emissions from field burning of cereal straw and sugarcane residue were included.
Here, emission factors (Table S8) for cereal and sugarcane burning were used, with zero emissions allocated, in cases of future shifts to deep sowing-mulching technology (Gupta, 2014). The distributed diesel category included diesel use in agricultural tractors and pumps, and in diesel generator sets used for non-grid electricity supply. Emission factors for distributed diesel sources are used, with zero emission allocation for a shift to electric or solar technologies.    Overall emissions from ECLIPSE were found to be in good agreement with those from our inventory, with the difference in total emissions lying within 20%. However, major differences are found in power generation sector, industry and residential. The differences can be attributed to emissions from extraction processes of fuels, commercial activities, and quantification of process emissions from industries. HTAP agree well with PM and its constituents but is nearly a factor 1.5-2 greater for NOx, NMVOCs and SO2. The differences can be majorly attributed to emissions from extraction process in the power sector and difference in control for NOx and SO2. Similar to HTAP_v2, REAS 2.1 also agrees well for BC and OC while has 0.7 times lower PM and nearly 1.5 times higher emissions of NOx, NMVOCs and SO2 as compared to our inventory. The differences mostly come from inclusion of agricultural emissions (such as fertilizer application and manure management of livestock), non-combustion emissions primarily from solvent use, paint use, evaporative emissions from vehicles, emissions from fuel extraction processes and emissions released from soil in REAS 2.1. Other causes of difference include use of different emission factors and methodologies for emissions estimates, particularly for the residential biomass combustion and transportation. In other inventories, activity data are primarily taken from energy consumption estimates by International Energy Agency (IEA), where as in our inventory the activity data is calculated using food consumption at the state level and end-use energy for cooking (Habib et al., 2004) and vehicular sales to arrive at on-road vehicular population considering age of the vehicles (Pandey and Venkataraman, 2014).   (combustion & processing), Surface Transportation and Agriculture waste burning in fields. Gridded emissions in kg m-2 s-1 are summed over the Indian landmass and converted to million tonnes y-1 (Table S9). The present estimates do not include emissions from soils and animal rearing or from shipping and aviation, rather they focus on energy use and solvent based activities. Therefore, corresponding sectors in the RCP database were excluded from the evaluation.

Evaluation with ECLIPSE V5a-CLE and GAINS
Across RCP scenarios, SO2 emissions from India are well bounded: 4-9.5 MT/yr in 2030 and 3-7.5 MT/yr in 2050. Emissions of SO2 estimated here for the highest-control scenario, S3, agreed with those from RCP 4.5 in 2030 and RCP 8.5 in 2050, due to similar assumptions of over 80% non-coal electricity generation. However, the S2 and REF scenarios estimated much larger emissions, respectively, exceeding RCP8.5 by 1.5 to 2 times in 2030 and 3 to 5 times in 2050. This results from our assumption of low levels (max 25%) of deployment of flue gas desulphurization, as only four coal-fired TPPs in India operate flue gas desulphurization (FGD) units and among those to be commissioned through 2030, only 7 TPPs are listed to have FGD (CAT andUrban Emissions, 2014, Prayas Energy Group, 2011), which differs from assumptions of greater SO2 emission control in the RCP scenarios. These assumptions would reflect in higher secondary sulphate contribution to PM2.5 concentrations from thermal and total coal sectors under the REF andS2 scenarios, in 2030 and2050. For NOx emissions as well, there is similar agreement of the S3 scenario here with RCP4.5 in both 2030 and 2050, but significantly larger emissions estimated in the S2 and REF scenarios. The emissions shares are dominated by thermal power and transport sector and grow with sectoral growth under the first two scenarios. Under the S3 scenario, shifts to tighter emission standards for vehicles and a greater share of CNG in public transport, and to non-fossil power generation, reduce NOx emissions. A non-negligible ~20% share is from residential, agricultural field burning and brick production sectors, which is reduced in magnitude by the adoption of mitigation based largely on cleaner combustion technologies. Similar to emissions of SO2, those of NOx in S1 and S2 grow well beyond magnitudes in the RCP database for future years, while those in S3 agree with RCP emission magnitudes, consistent with differences in assumptions in the thermal power sector.
For NMVOC, there is close agreement of S3 scenario emissions with those of RCP6.0 and of S2 scenario emissions lying between those of RCP4.5 and RCP8.5, both in 2030 and 2050. The REF scenario, which assumes negligible shifts away from residential biomass fuel use and agricultural field burning, calculates somewhat larger NMVOC emissions. Present day NMVOC emissions are dominated by residential energy use, largely from traditional biomass fuel stoves, followed by fugitive emissions from energy extraction (coal mining and oil exploration), and open burning of agricultural residues in fields.
Emissions of BC in the S3 scenario agreed best with RCP6.0 in 2030 and RCP8.5 in 2050, while REF and S2 scenario BC emissions exceeded those of the RCP8.5 by factors of 1.5 to 3, from inclusion of new sources like residential lighting (with kerosene wick lamps) and water and space heating (with biomass fuels). Emissions of OC in the S2 scenario closely matched those in RCP4.5, while those in REF matched RCP8.5, in both 2030 and 2050; however, those in S3 were a factor of 3 lower than the lowest RCP6.0 emissions.   The uncertainty represented by the bars is based on uncertainty in the GBD estimates of ambient PM2.5 concentrations. It is estimated by sampling 1,000 draws of a distribution for each grid cell based on the model output mean and standard deviation (GBD MAPS Working Group, 2018).