GEOS-Chem is a 3-dimensional atmospheric chemical transport model (http://acmg.seas.harvard.edu/geos/) driven by assimilated meteorological data from the Goddard Earth Observing System (GEOS). We simulated PM_2.5 concentrations over India using the 0.5° × 0.625° Asian nested grid in GEOS-Chem version v12.1.1 (11°S–55°N, 60°E–150°E) for the year 2010, which uses MERRA2 meteorological data. Boundary conditions for the nested simulation were generated every 3 h from the global 4° × 5° version of the model [11]. Fully coupled simulations included O₃-NOx-hydrocarbon chemistry, aerosols and gas-aerosol phase partitioning, and dry and wet deposition schemes [12–18].

Global anthropogenic emissions were from the Community Emissions Data System (CEDS) except for NH₃ emissions, which were from the Global Emission Initiative (GEIA) inventory [19]. Anthropogenic emissions over the Asian region (nested grid) were from MIX inventory [20]. The MIX emission inventory for 2010 considers ten species [SO₂, NOx, NH₃, CO, non-methane volatile organic carbon (NMVOC), PM₁₀, PM_2.5, BC, OC and CO₂] from five different sectors (residential, industry, power, transport and agriculture) [20]. Total anthropogenic and natural emissions over the Indian region are presented in the Supplementary Material (Table S1 and Fig. S1). Each simulation was made for 18 months, with the first 6 months as spin-up for both global (4° × 5°) and the nested (0.5° × 0.625°) models [21].

Model outputs were saved at monthly resolution and included total PM_2.5 concentrations in each model grid, and the concentrations of different components of PM_2.5 (SO₄²⁻, NO₃⁻, NH₄⁺, BC, OC, SOA, mineral dust and sea salt aerosols). From this, the percent contributions by chemical components (SO₄²⁻, NO₃⁻, NH₄⁺, BC and OC) to PM_2.5 concentrations were estimated because understanding the variation in the concentration of different components and their relationship with the emissions, helps in generating policies for the future [1]. The premature mortalities in each model region containing UAs were also calculated.

A population-weighted annual mean PM_2.5 concentration for the entire country was calculated as:

Population-weighted concentration (for individual grid i ranging from 1 to N) is the mean concentration to which the entire country population (P) may be exposed to PM_2.5 (μg m⁻³) [22] and accounts for the variation in spatial and temporal distribution of PM_2.5 [23].

We compared GEOS-Chem simulated PM_2.5, SO₂ and NO₂ concentrations with observed annual mean concentrations of PM.2.5 (at eleven sites, Table S2), SO₂ and NO₂ (at 48 UAs) in 2010 by the National Ambient Air Quality Monitoring Programme, NAMP [24]. Details on locations and observed concentrations of SO₂ and NO₂ are presented in the Supplementary Material (Table S3 and S4). As ground-based observations for PM_2.5 were scarce, re-gridded satellite derived concentrations obtained from the aerosol optical depth (AOD) values retrieved from Moderate Resolution Imaging Spectroradiometer (MODIS), Multi-angle Imaging Spectroradiometer (MISR) and Sea-Viewing Wide Field-of-View Sensor (SeaWiFS) [25] were also used. Normalized mean bias (NMB) (Eq. (2)) and coefficients of correlation (r) of model (M) predicted PM_2.5, SO₂ and NO₂ concentrations with the corresponding ground-based or satellite-based observations (O) were determined.

Four diseases IHD, COPD, stroke and LC were used as endpoints because of their direct link with ambient PM_2.5 exposure [26–28]. Relative risks of these diseases were calculated for each model grid using a non-linear Integrated Exposure Response (IER) function [5, 29–31]:

Where RR_i,j represents the relative risk for a disease j at a model grid i. ΔPM_2.5 is the difference between the modeled PM_2.5 concentration (μg m⁻³) in the i^th grid and the counterfactual PM_2.5 concentration, 5.8 μg m⁻³ [31]. The counterfactual PM_2.5 concentration is the amount of PM_2.5 below which no risk exists (i.e. RR ≤ 1) [31]. A range of 1,000 sets of disease specific constants (α_j, γ_j, δ_j) were obtained from the Monte Carlo simulation performed by Burnett et al. [30]. The median and 95% confidence limits (CLs) of these constants are presented in Table S5. Premature mortality per year, ΔM_i,j, is given by [30, 32]:

Where y_i,j is the baseline mortality (unitless), RR_i,j is the relative risk for specific disease j at modelled grid (i to N) and P_i is the exposed population for 2010. The 2010 total population information was obtained from 2011 census data [33]. The 0.5° × 0.5° gridded population for the year 2010 was obtained from the Shared Socioeconomic Pathways (SSP) scenarios (https://tntcat.iiasa.ac.at/SspD, last accessed 8 October, 2020) and regridded to the simulation resolution (0.5° × 0.625°). The baseline mortality was quantified using the number of deaths in India for 2010 from the Global Health Database Exchange [http://ghdx.healthdata.org, last accessed 03 July, 2020; data is provided by the Office of the Registrar General and Census Commissioner, India (2014)].

The impact of individual chemical components of PM_2.5 on the premature mortality was estimated by first subtracting the individual component from the actual PM_2.5 concentration in the respective UA location. Later the premature mortality by 4 different diseases (COPD, IHD, lung cancer and stroke) at these UAs of India from the obtained PM_2.5 concentration was determined by using IER function. The total premature mortality due to a specific chemical constituent is estimated by deducing the premature mortality estimated in absence of the specific chemical component from the total premature mortality value.

Overall country-wide results were similar to other previous studies. Population weighted mean concentration of PM_2.5 was 66.2 μg m⁻³, approximately 12% lower than Ghude et al. [7], 2016 (75 μg m⁻³ in 2011), David et al. [6], 2019 (75.5 μg m⁻³ in 2012), Venkataraman et al. [34], 2018 (~75 μg m⁻³ in 2015) and 10% higher than Cohen et al. [35], 2017 (60 μg m⁻³ in 1990), but within the percent variation (~7–15%) reported from multiple studies [6]. 99.5% of the total population in India was exposed to an annual mean PM_2.5 concentration greater than the WHO air quality guidelines (annual mean concentration of 10 μg m⁻³); [36], similar to earlier results [6, 34, 37].

The highest mean surface concentrations of PM_2.5 were observed during DJF (December-January-February) followed by SON (September-October-November), MAM (March-April-May) and JJA (June-July-August) (Fig. 1). A decrease in the concentration of PM_2.5 and its precursors during JJA occurs due to precipitation and washout or wet deposition of pollutants in the atmosphere whereas a peak in the concentrations during DJF in India occurs due to low temperatures and low mixing heights [38] limiting the dispersion of pollutants.

The annual average concentrations of NO₃⁻, NH₄⁺, SO₄²⁻, BC, OC, SOA, mineral dust, and sea salt aerosols over India for the year 2010 were estimated as 6.90 μg m⁻³, 3.92 μg m⁻³, 5.21 μg m⁻³, 1.7 μg m⁻³, 5.1 μg m⁻³, 7.2 μg m⁻³, 10.38 μg m⁻³ and 0.21 μg m⁻³, respectively. The percent contribution of sulfate-nitrate-ammonium (SNA) to PM_2.5 over India was observed to be 40% (with NO₃⁻ contributing by as 16%, SO₄²⁻ by 14% and NH₄⁺ by 10% to PM_2.5 concentration). OC concentration contributed 21% to PM_2.5 followed by mineral dust (18%), secondary organic aerosols (17%), BC (3%) and sea salt aerosols (1%). The northern Indian region (mainly the IGP region) has the highest ambient PM_2.5 concentration, similar to other studies [6]. Fig. S2 and S3 present the annual spatial and monthly mean variations in the concentrations of various chemical constituents (NO₃⁻, NH₄⁺, SO₄²⁻, BC and OC) of PM_2.5 in India. The Indo-Gangetic plains were observed to have high concentrations of different chemical constituents of PM_2.5.

Further discussions on comparison of model results with field-observed and satellite-derived concentrations and related discussions are presented in Section S1 and Fig. S4 and, on premature mortality in India as a whole in Section S2. The total premature mortality in India was 0.54 (0.45–0.63) million (Table S6), comparable to previous reports [0.57 (0.32–0.73) million [7] and 0.567 million [39]]. Maximum premature mortality was observed in the Indo-Gangetic region (Fig. S5).

The annual average concentration of PM_2.5 in all the UAs were > 10 μg m⁻³ (range 26.06–144.33 μg m⁻³). Fig. 2(a) and Table S7 show the premature mortalities obtained in the UAs for 2010. The total number of premature mortality due to PM_2.5 exposure in these UAs was 0.10 (0.09–0.125) million for the year 2010. Delhi had the highest mortality (11945), followed by Kolkata (9204), Mumbai (8817), Bangalore (4528) Hyderabad (4157) and Chennai (3818). Together these cities account for 7.86% of the total premature mortalities in India and 42.47% of total premature mortalities in UAs.

Fig. 2(b) and Table S8 show the premature mortality per capita, i.e. the mortality per population due to PM_2.5 exposure in each UA. This number is directly related to the PM_2.5 concentrations in each UA [see Eq. (3), [39]]; the higher the PM_2.5 concentrations (Fig. 4) the higher will be the premature mortality per capita. The most vulnerable UAs are Delhi, Kolkata, Lucknow, Kanpur, Ghaziabad, Patna, Agra, Ludhiana, Faridabad, Meerut, Jamshedpur, Asansol, Allahabad, Amritsar and Dhanbad, with 0.065% premature mortality per capita. All these are higher than the country average of 0.047% premature mortality per capita. That is, premature mortality due to PM_2.5 exposure was 2% to 59% higher in the regions containing UAs than the country average. Across all the cities, the causes of mortalities were Stroke ≈ IHD > COPD > LC (Fig. S6).

The premature mortality with respect to population in the rural regions ranged from 0.02% to 0.069% with the average of 0.052% (Section S3, Table S9–S11, Fig. S7–S8; rural regions obtained from the Ministry of Jal Shakti, Government of India, website, https://ejalshakti.gov.in/IMISReports/Reports/BasicInformation/rpt_RWS_RuralPopulation_S.aspx?Rep=0&RP=Y, last accessed 05 April, 2021; [40, 41]). The annual average PM_2.5 concentrations in these rural sites was 50.1 μg m⁻³ (12.5 μg m⁻³ at Pangi region of Chamba district of Himachal Pradesh to 105.33 μg m⁻³ at Jakhal region of Fatehabad of Haryana state). As can be seen, some values were lower than the country average of 0.047% and some were higher (1.2–50.6%) than the country average. We then plotted the PM_2.5 concentrations in UAs and rural regions by five administrative zones of India (northern, southern, eastern, western and central India as shown in Fig. S8). It is seen that PM_2.5 concentrations in rural regions in southern and eastern India do not statistically differ from those in UAs (Fig. S9). Whereas, in the northern, western and central India the PM_2.5 concentrations in rural regions were lower than UAs at p < 0.05. When observed together with the spatial annual mean concentrations of NO₃⁻, NH₄⁺, SO₄²⁻, BC and OC, and the overall seasonal PM_2.5 map (Fig. 1), it appears that the pollution from PM_2.5 is largely zonal. That is, in most cases, the percent BoD would be similar in rural, semi-urban and urban locations that belong to one particular region, and the absolute premature mortality then essentially dependent on the absolute population residing in that region. Therefore, pollution reduction measures taken in certain high emission regions of a particular zone may be beneficial to the entire zone.

Fig. 3 and Table S12 present the premature mortality contributed by different chemical components of PM_2.5 in India for the simulated year 2010. Cities like Delhi, Kolkata, Surat, Mumbai, Hyderabad, Chennai, Bangalore and Ahmedabad had the premature mortalities ranging from 337 to 1,136 by OC, 215 to 996 by NO₃⁻, 210 to 828 by SO₄²⁻, 150 to 625 by NH₄⁺ and 51 to 567 by BC respectively. The contribution of differnet chemical components was in the order: OC > NO₃⁻ > SO₄²⁻ > NH₄⁺ > BC. Pemature mortality due to NO₃⁻, NH₄⁺ and OC are proportionately much higer in the Indo-Gangetic Plains (Fig. 3), which could be attributable to vehicular, agrilcutural and biomass burning emissions (e.g. [34]). Premature mortality due to SO₄²⁻ coincides relatively well with the locations of coal-fired power plants (e.g., in Jharkhand, greater Delhi region, Uttar Pradesh and Maharashtra).

Fig. 4(a) and Fig. 4(b) present the concentrations and the contribution of SO₄²⁻_, NO₃⁻, NH₄⁺, BC and OC on PM_2.5 for the UAs of India for the year 2010. OC concentration contributed the most (18–30%) to the total PM_2.5 concentration, followed by NO₃⁻ (10–28%), SO₄²⁻ (7–21%), NH₄⁺ (7–12%) and BC (2.5–4%), respectively for the different UAs. In general, strategies to reduce OC and NOx pollution may be expected to yield health benefits across all the cities. Furthermore, measures that target the residential or biomass burning [42], and transportation sector should lead to benefits in most UAs.

However, measures to reduce the impact of NO₃⁻ should be applied across all major categories of transportation (buses, taxis, cars, private, public). For example, cities like Delhi have implemented short-duration odd-even schemes in the past whereby all private four-wheelers with an odd registration numbers are allowed to be on-road one day and even registration numbers on the next day. And the most recent scheme applied between November 5 and 15, 2019, had several concessions, with restrictions being applicable only between 8 am and 8 pm, with exemptions for female drivers with children up to age 12, vehicles with physically disabled person and elderly people and on Sundays. These schemes have been applied to obtain short-term improvement in air quality and their long-term effectiveness has been judged to be marginal only [43, 44].

If we assume that such a scheme is applied consistently throughout the year, considering 48% of NO₂ contribution from passenger vehicles [45] and 33% from cars and jeep [43], an overall NO₂ reduction of the order of 8% can be expected if half of the cars and jeep were removed from the roads. NO₃⁻ contributed approximately 16% to total PM_2.5 concentrations in Delhi region in our model results and an overall 1.3% reduction of PM_2.5 concentration can be expected. Model predicted PM_2.5 and NO₂ concentrations for the Delhi grid will be 66.87 μg m⁻³ and 7.72 μg m⁻³, respectively, which are comparable to the measured PM_2.5 and NO2 concentrations by CPCB, 79–507 μg m⁻³ and 9–159 μg m⁻³, respectively during the 1^st to 15^th January 2016 (http://cpcbenvis.nic.in/pdf/CPCB%20Report%20on%20Odd-Even%20Scheme.pdf). Thus, even though our 1.3% reduction calculations are derived using assumptions, they suggest that such odd-even schemes applied to a small section of overall traffic will not lead to high reductions; instead, they may be more useful in reducing traffic congestion and travel times [43]. Further considering the natural variability in measurements, such reductions may not be observed at a statistically significant level. In contrast, switching to tighter emissions norms [such as under the Bharat Stage (BS)-VI], introduced on 1^st April 2020, will reduce NOx emissions from petrol-driven vehicles by 25% and from diesel-driven vehicles by 68%. Therefore, here, more than 10% reduction in PM_2.5 can be expected due to local NOx emissions reductions alone.

SO₂ emissions are largely driven by the industrial and power sectors, and their contribution to urban PM_2.5 load is variable (10–30%). However, technologies such as flue gas desulfurization (FGD) installed in all coal-fired power plants [46] can remove as much as 90% of this contribution.

Contributions of NH₄⁺, BC and OC to PM_2.5 are consistent across all analyzed UAs (total 35–50%) and are primarily influenced by the residential sector (like emissions from traditional stove, biofuels, cow dung and agricultural waste used for cooking). This is consistent with previous country-wide model results that the residential sector could account for 52% of population-weighted annual mean PM_2.5 concentrations [37]. The percent premature mortality is much higher in urban areas (2–59%) than the country-wide estimate. Therefore, a move away from low efficiency combustion of domestic fuels [47], limiting biomass burning during winters and alternative power sources such as diesel generators would substantially reduce PM_2.5 pollution, and the associated exposure and premature mortality in urban regions as well as the entire country.

As such, our analysis shows policies targeting multiple sectors would lead to larger reductions in PM_2.5 pollution in urban regions (reductions of more than 19–30%) compared to policies in a single sector alone. Overall, a multi-pronged approach will be required to improve the overall air quality of urban regions.