Inequality in air pollution mortality from power generation in India

India’s coal-heavy electricity system is the world’s third largest and a major emitter of air pollution and greenhouse gas emissions. Consequently, it remains a focus of decarbonization and air pollution control policy. Considerable heterogeneity exists between states in India in terms of electricity demand, generation fuel mix, and emissions. However, no analysis has disentangled the expected, state-level spatial differences and interactions in air pollution mortality under current and future power sector policies in India. We use a reduced-complexity air quality model to evaluate annual PM2.5 mortalities associated with electricity production and consumption in each state in India. Furthermore, we test emissions control, carbon tax, and market integration policies to understand how changes in power sector operations affect ambient PM2.5 concentrations and associated mortality. We find poorer, coal-dependent states in eastern India disproportionately face the burden of PM2.5 mortality from electricity in India by importing deaths. Wealthier, high renewable energy states in western and southern India meanwhile face a lower burden by exporting deaths. This suggests that as these states have adopted more renewable generation, they have shifted their coal generation and associated PM2.5 mortality to eastern areas. We also find widespread sulfur emissions control decreases mortality by about 50%. Likewise, increasing carbon taxes in the short term reduces annual mortality by up to 9%. Market reform where generators between states pool to meet demand reduces annual mortality by up to 8%. As India looks to increase renewable energy, implement emissions control regulations, establish a carbon trading market, and move towards further power market integration, our results provide greater spatial detail for a federally structured Indian electricity system.


Introduction
India is the world's third largest emitter of greenhouse gases (GHGs) (Carbon Brief 2019, World Resources Institute 2019, BP 2020), with 40% of its emissions coming from coal-dominated power generation (Mohan et al 2019), which formed approximately 50% of generating capacity and 70% of generation in 2020 (Central of Electricity Authority 2021).
Likewise, largely uncontrolled emissions of conventional air pollutants (CAPs) sulfur dioxide (SO 2 ) and nitrogen oxides (NO X ) from Indian power generation contribute to poor ambient air quality in the country. Between 7% and 21% of the estimated 1.1 million premature deaths in Indian associated with PM 2.5 , solid or liquid particles suspended in the atmosphere, come from power generation (Guttikunda and Jawahar 2014, Lelieveld et al 2015, GBD MAPS Working Group 2018, Conibear et al 2018a, 2018b, Gao et al 2018, Guo et al 2018, Apte and Pant 2019, Reddington et al 2019, Cropper et al 2021. With increasing economic growth, power generation will remain a source of GHGs and CAPs and consequently control policies (Venkataraman et al 2018, Peng et al 2020. The Government of India has targets to increase zero-emission generation to 50% of all power capacity by 2030 and net-zero emissions by 2070 (Government of India 2015, Vaidyanathan 2021). It has also announced stricter limits on SO 2 and NO X emissions in 2015 (Ministry of Environment Forest and Climate Change 2015) which could reduce respective emissions by up to 80%-90% (Srinivasan et al 2018, Sengupta et al 2022a. Consequently, climate and air pollution control policies for the Indian power sector remain an active area of research due to a sector in flux from growing power demand and policy targets. Previous efforts to quantify the premature mortality from Indian power generation under current and future policies largely pair simulations of air quality models with exposure-response functions and mortality rates. Cropper et al (2021) estimate approximately 78 000 deaths (∼9.1% of all PM 2.5 premature mortality) associated with Indian coal power plants in 2018 with most deaths in the most populous states of Uttar Pradesh and Maharashtra. They estimate annual mortality increases to 112 000 deaths from planned construction of new coal power stations. Both (Reddington et al 2019) and Gao et al (2018) likewise estimate 210 000 (∼21% of all PM 2.5 premature mortality) and 270 000 (∼33% of all PM 2.5 premature mortality) annual deaths, respectively, associated with Indian power generation, with most deaths from PM 2.5 exposure in states in the Indo-Gangetic Plain where exposure levels are the highest. McDuffie et al (2021) estimate approximately 74 000 deaths due to PM 2.5 (∼6.5% of all PM 2.5 premature mortality) from coal in energy production in South Asia (India, Pakistan, Bangladesh, Nepal, and Bhutan) with the coal belt in eastern India having large contributions of PM 2.5 from coal mining and burning. Peng et al (2020) quantify tradeoffs to find failing to enforce emission control regulations at Indian power stations results in 2.5× more deaths in 2040 than failing to enforce renewable energy targets. Overall previous analyses quantifying the impact of power generation have found the sector to be a large contributing sector that will remain important as India meets multiple objectives of development, industrialization, and decarbonization (International Energy Agency 2021).
However, no study has analyzed policies within Indian power sector operations accounting for the exchange of electricity between states. Due to historical, constitutional, and geographical reasons (Dubash and Rajan 2001, Ramanathan 2001, Tongia 2007, the electricity sector varies by state in India. Both central government and state governments have overlapping jurisdiction over the Indian power sector according to the Indian constitution (Tongia 2007). Each state individually contracts and schedules generating capacity to meet demand within their borders. Consequently, not only is there state-to-state heterogeneity in power generation impacts on ambient air quality as previous analyses quantify, but also fuel mixes and emissions from the electricity consumed by each state (Safiullah et al 2017, Power System Operation Corporation Limited 2020. For example, while renewable energy is growing in India, its deployment is uneven. Despite the world's fourthlargest generation of solar and wind electricity in 2019 (107 GWh) (International Renewable Energy Agency 2022), just six states in India generated 85% of this renewable electricity mostly in western and southern India (Sengupta et al 2022b). Moreover, power plants in southern, western and northern India furthest away from coal mining areas in eastern India have higher electricity costs due to transport costs (Kamboj and Tongia 2018), inducing further interstate differences. To analyze premature mortality of policies such as carbon taxes or further integration of wholesale power markets between states, a representative accounting of power sector operations and associated emissions is needed, especially in light of growing zero-emission generation. Likewise quantifying the PM 2.5 premature mortality embedded in power consumed by each state in addition to power produced by each state can inform emissions reductions policies at the state-level. States choose which plants where will generate electricity to meet demand.
Consequently here, we feed emissions estimates from reduced-form model of Indian power generation as inputs into a reduced-form air quality model to (a) quantify the annual PM 2.5 premature mortality associated with power produced and consumed by each Indian state in 2017-2018 and (b) analyze the impact on PM 2.5 premature mortality of future emission control, market integration, and carbon tax policies in the Indian power sector. Our approach improves upon previous work by accounting for electricity exchanged between states to understand the import and export of air pollution between states as well as power sector operations.

Reduced-form power generation model
We use a power generation model because of the lack of readily available continuous emissions monitoring data and granular power generation data associating power consumption with specific plants in India. We use a reduced-form dispatch (productioncost) model that estimates the hourly dispatch and generation of power generators in each Indian state based on electricity demand (Sengupta et al 2021(Sengupta et al , 2022a. Previous analyses have used the model to quantify marginal emission factors for Indian power generation (Sengupta et al 2022b). We direct readers to these previous analyses for a full description of the model. What follows is a short summary. States are primarily responsible for procuring and dispatching power based on demand in India, with some regional coordination (Safiullah et al 2017

Reduced-form air quality model
We feed our estimates of SO 2 , NO X , and primary PM 2.5 emissions into a reduced-form air quality model, the intervention model for air pollution (InMAP) (Tessum et al 2015, 2017, Gilmore et al 2019. Previous analyses have used this model to evaluate mortality associated with power sector emissions in the United States (Thind et al 2019). We use a reduced-form model instead of a full-form chemical transport model to reduce computational resources. Moreover, these models may be better suited for initial screening and policy analyses over regions in the Global South with data and resource constraints. We use the global version of the model which uses outputs of the Goddard Earth Observing System-Chem (GEOS-Chem) chemical transport model to approximate steady-state ambient PM 2.5 concentrations (Thakrar et al 2022). The reduced-form model uses simplified parameterizations of advection, diffusion, deposition, and chemical reactions based on underlying GEOS-Chem simulations (Hammer et al 2020). Grid definitions and emissions and meteorology inputs for the underlying GEOS-Chem simulations follow (Hammer et al 2020) who largely use Emissions Database for Global Atmospheric Research (Crippa et al 2016) and modern-era retrospective analysis for research and applications (Gelaro et al 2017) emissions inputs. However, Global InMAP deviates from this setup by simulating with variable grid cell size based on 0.01 • gridded 2020 global population estimates, increasing spatial resolution (smaller cells) in areas of higher population and population density. Grid cell size varies between 5 • × 4 • (∼500 km at equator) in remote areas to 0.04 • × 0.03 • (∼4 km at equator) in urban areas. Global InMAP takes annual emissions estimates of PM 2.5 precursors gases to directly estimate annual-average ambient PM 2.5 concentrations chemical resolved by sulfate, nitrate, ammonium, elemental carbon, and secondary organic components. We feed our emissions estimates as elevated emissions with uniform height of 200 m all plants, stack diameter 5 m, exit velocity 23 m s −1 , and temperature 416.5 K (U.S. Environmental Protection Agency 2017).

Estimating premature mortality from PM 2.5
We use a similar risk-exposure relationship and input data used in previous analyses (Apte et al 2015, GBD MAPS Working Group 2018, McDuffie et al 2021 to quantify annual PM 2.5 premature mortality, including specifically for India (Balakrishnan et al 2019, Pandey et al 2021. We focus on long-term health impacts represented by annual-average concentrations which satisfy the assumptions of the riskexposure models we use. Short-term or sub-annual, seasonal exposures are outside the scope of this analysis. This includes PM 2.5 mortality attributable to the power sector (m i,j,power , annual deaths from power sector PM 2.5 ) and total PM 2.5 premature mortality (m i,j , annual deaths from all source PM 2.5 ) for each disease endpoint j (e.g. heart disease, lung cancer, etc) in each grid cell i simulated by InMAP located in states: (1) Here I s,j is the mortality rate for each disease endpoint (annual deaths per capita from each PM 2.5related disease) in each Indian state, s, P i refers to the population in each grid cell i, and RR i,j refers to the relative risk of mortality from each disease in each grid cell as a function of annual average concentrations of PM 2.5,i. We omit subscripts for sex and age for clarity, but all variables except for PM 2.5,i vary by sex and age as well. We scale the mortality rates I s,j by a population-weighted average relative risk, RR s, j because mortality rates are reported by state and not grid cell. However, reported mortality rates for each state do vary by location within each state because mortality rates account for deaths associated with and without PM 2.5 exposure. Consequently, to account for this variability, we scale mortality rates by a weighted average relative risk of PM 2.5 exposure ( which scale mortality rates by relative risk in each grid cell and not a weighted average of grid cells in each state. Furthermore, we direct readers to previous analyses that derive this fractional attribution approach that quantifies the PM 2.5 mortality attributable to power generation by scaling total PM 2.5 mortality by the portion of PM 2.5 exposure attributable to power generation ( We use InMAP estimates for the absolute contribution of PM 2.5 from the power sector, PM 2.5,i,power and population, P i . Because global InMAP takes emissions estimates as perturbations on top of the emissions in its underlying GEOS-Chem simulations, we assume the model's estimates of annual average ambient PM 2.5 concentrations represent the absolute contribution of PM 2.5 from the power sector. To account for other sources in PM 2.5 exposure we use 2018 global gridded PM 2.5 estimates at 0.01 • resolution (Hammer et al 2020) aggregated to the InMAP modeling grid as inputs for PM 2.5,i . These estimates represent PM 2.5 from all sources harmonized from satellite measurements, model estimates, and ground measurements. We note the underlying GEOS-Chem simulations are the same in both InMAP and global gridded PM 2.5 estimates (Thakrar et al 2022). We obtain mortality rates, I i , and relative risk curves for outdoor PM 2.5 , RR(PM 2.5,i ) from the 2019 Global Burden of Disease (GBD) (Global Burden of Disease Collaborative Network 2021) for six disease endpoints, j: ischemic heart disease, stroke, chronic obstructive pulmonary disease (COPD), lower respiratory infections, lung cancer, and type 2 diabetes. Efforts to establish causal relationships between PM 2.5 exposure and additional disease endpoints are active areas of research (U.S. Environmental Protection Agency 2022). Mortality rates are specific to Indian states split by age (5-year intervals from 0 to 95+ years), sex (male and female), and disease endpoint (Indian Council of Medical Research et al 2017). GBD 2019 risk curves are MR-BRT spline fits of various pollution exposure cohort studies across global ranges of PM 2.5 exposure assuming a theoretical minimum-risk exposure level (TMREL) which is the level of PM 2.5 exposure at which and below relative risk is equal to one. GBD 2019 curves are based off statistical draws with mean, 2.5th and 97.5th percentile values. For this analysis we construct risk curves by scaling the summary GBD 2019 risk curves (mean, upper and lower percentiles) assuming a TMREL of 4.15 µg m −3 based on an average of the uniform distribution between 2.4 µg m −3 and 5.9 µg m −3 . We only consider exposure to outdoor ambient air pollution and not household air pollution, i.e. pollution from burning kerosene, dung, wood, and charcoal (Abbafati et al 2020, Global Burden of Disease Collaborative Network 2021, Pandey et al 2021). As a sensitivity case, we also estimate mortality using relative risk curves, RR(PM 2.5,i ), from the global exposure mortality model (GEMM) (Burnett et al 2018), keeping the same underlying mortality rates, I i for five disease endpoints: ischemic heart disease, stroke, COPD, lung cancer, and lower respiratory infections.

Policy scenarios
We run the model in a business as usual (BAU), 2017-2018 case where equation (1) represents the PM 2.5 mortality attributable to power generation. We also modify inputs to simulate policy scenarios of greater pollution control, further market integration, and carbon taxes. We describe these methods in the supplementary material with an overview of all scenarios in table S1.
We run InMAP to estimate the mortality associated with each Indian state by running the model for each state with only the state's BAU emissions. We take both a (a) a production-oriented view with emissions from plants located within the borders of a state and a (b) consumption-oriented view with emissions from plants supplying power to meet a state's demand. In the latter case, these plants can be within a state or outside a state. We denote the mortality associated with an individual state (production or consumption) as: Here PM 2.5,i,power,state represents the InMAP estimate of PM 2.5 in a grid cell from the emissions associated with each individual state, while s is state with location of grid cell i. Consequently, using equation (4) these simulations quantify the burden of deaths associated either production of electricity within a state or consumption of electricity within a state.

Results
We show the results of our state-wise attribution of PM 2.5 premature mortality associated with power generation in India in figures 1 and (2. See figure S17 for a map of India labeled with state names and regions. We discuss mean estimates using GBD relative risk curves (Global Burden of Disease Collaborative Network 2021) here, and present similar estimates using GEMM risk curves (Burnett et al 2018) in the figures S9 and S10. We adopt a similar approach used by Thind et al (2019) (Thind et al 2019) for attributing electricity generation deaths in the United States. We derive these estimates from our BAU InMAP simulation and simulations with only BAU emissions associated with power production or consumption in each state. We find that in each grid cell, the sum of the PM 2.5 estimates of all the state-bystate simulations equals the PM 2.5 estimate from the BAU simulation, achieving closure with equation (1). Across India, we find an annual population-weighted ambient PM 2.5 concentration of 76 µg m −3 and an annual population-weighted attributable to power generation PM 2.5 concentration of 3.8 µg m −3 . In total for 2017-2018, across all grid cells and disease endpoints ( ∑ i ∑ j m i, j ), we estimate that electricity generation is associated with ∼62 000 (51 000-72 000) of the annual ∼1.1 million (900 000-1.3 million) premature deaths from ambient PM 2.5 in India (figures S4-S7).
Data underlying figure 1(a) are shown in figure S10 and table S2. We find Tamil Nadu in south India to have the highest burden of electricity PM 2.5 deaths: 15 000 deaths due high sulfur-emitting lignite power plants in the state. This is higher than the next two highest states combined: West Bengal in eastern India (6500 deaths), and Uttar Pradesh in northern India (6100 deaths). Other large states facing high burdens include Maharashtra (5200 deaths), Gujarat (4300), and Andhra Pradesh (3700 deaths), Small northeastern states (Sikkim, Mizoram, Nagaland, Tripura, Assam, Arunachal Pradesh, Manipur, Meghalaya) and predominantly Himalayan states in north India (Jammu and Kashmir, Himachal Pradesh, Uttarakhand, Delhi, Punjab, and Haryana), face little or no burden from electricity generation deaths (0-600 deaths).
In figure 1(b), shows annual electricity generation deaths occurring within each state's borders from annual electricity production within that state. Here the highest deaths occur in Tamil Nadu (13 500 deaths) followed by West Bengal (4500 deaths), Gujarat (3700), Maharashtra (2600 deaths), Uttar Pradesh (2500 deaths), Andhra Pradesh (1700 deaths), and Rajasthan (1300 deaths). Northeastern states remain relatively unaffected because of few emitting power plants (coal and gas) located within those states. Likewise, north Indian states located in in the Himalayas show low in-state death burdens.
In figure 1(c) shows annual electricity generation deaths occurring outside each state's borders from annual electricity production within that state. See figure S18 for a map of which states outside the state's borders are affected by production within the state. Here the relative burden of out-of-state deaths differs: Gujarat (3800 deaths) and Uttar Pradesh (3100 deaths) both lead in terms of out-of-state deaths from electricity generated within the state. Maharashtra (3100 deaths), Madhya Pradesh (3000 deaths), Rajasthan (2700 deaths), Tamil Nadu (2100 deaths), Chhattisgarh (1900 deaths), Andhra Pradesh (1500 deaths), Telangana (1200 deaths), Haryana (1000 deaths), Karnataka (1000 deaths), West Bengal (800 deaths), and Punjab (800 deaths) follow them. Once again, northeastern and northern, Himalayan states remain relatively unaffected.
Lastly, figure 1(d) shows the unequal mortality burden from electricity production between states in India. To derive these estimates, we first take the mortality burden in each state from all electricity generation in India (figure 1(a)) and subtract from it the sum of in-state deaths associated with electricity production in each state ( figure 1(b)). This gives the burden of deaths in each state from plants located outside the state. Then we compare this value to the burden of deaths outside the state from plants within a state ( figure 1(c)). The difference between these two values gives the relative mortality burden of power production in each state. If all states shared equal mortality burdens tied to their electricity production, values in figure 1(d) would be approximately zero, i.e. the deaths outside a state associated with plants in the state would equal the deaths in the state associated with plants located outside the state. However here, distinct spatial patterns emerge, and estimates are net negative for states predominately in western, southern, and central India ('production death exporters'). That is, the power plants in these states are disproportionately responsible for electricity-associated deaths. For example, power generation is responsible for approximately ∼4300 deaths in Gujarat ( figure 1(a)), ∼3700 of which are from power plants in Gujarat ( figure 1(b)). The remainder (a) and (b) of the Gujarat's mortality burden (∼600) comes from plants outside Gujarat. However, plants in Gujarat meanwhile contribute to approximately ∼3800 deaths outside Gujarat ( figure 1(c)). Consequently, deaths in Gujarat from plants outside Gujarat (∼600) are fewer than deaths outside Gujarat from plants in Gujarat (∼3800) by ∼3200 deaths (figure 1(d)). Likewise, other states who produce electricity associated with higher deaths include Rajasthan (−1500 deaths), Chhattisgarh (−700 deaths), Haryana (−700 deaths), Maharashtra (−600 deaths), Tamil Nadu (−600 deaths), Madhya Pradesh (−600 deaths), and Punjab (−400 deaths). Estimates are net positive for states predominately in northern and eastern India, except for one state in the south (Karnataka). This means these states face a disproportionate burden of electricity-production deaths. They have a higher burden of deaths within their borders from power plants located outside their borders than deaths outside their borders from plants located within their borders ('production death importers'). States include Bihar (1500 deaths), West Bengal (1200 deaths), Karnataka (1200 deaths), Jharkhand (700 deaths), Odisha (700 deaths), and Uttar Pradesh (600 deaths).
In an analogous figure to figures 1 and 2 shows annual electricity generation deaths occurring within each state's borders from each respective state's annual power consumption. In figure 2(a), we show the same values as figure 1(a). We present results for each state by consumption because states are responsible for choosing and contracting the plants from which they procure power for their consumers, showing a potential lever for emissions control policies. States can procure power from plants within their borders or outside their borders, so these values incorporate deaths embedded in electricity imports and exports. Northern, Himalayan states and northeastern states show again the lowest instate deaths (0-200 deaths). Similar patterns emerge as figure 2(b), with the highest in-state deaths tied to consumption occurring in Tamil Nadu (11 100 deaths), West Bengal (4000 deaths), Gujarat (3500 deaths), and Maharashtra (2500 deaths).
Figure 2(c) shows annual electricity generation deaths occurring outside each state's borders from each respective state's annual power consumption. See figure S19 for a map of which states outside the state's borders affected by consumption within the state. Here clearer geographic patterns emerge where the highest out-of-state deaths occur in western, southern, and northern India, but the lowest out-of-state deaths occur in coal-mining, eastern states. Northeastern areas remain unchanged. Here, the highest out-of-state deaths occur from electricity consumption in Gujarat (4400 deaths) followed by Maharashtra (3900 deaths), Rajasthan (3300 deaths), Uttar Pradesh (2700 deaths), Tamil Nadu (2600 deaths), Karnataka (2300 deaths), and Haryana (1700 deaths). States in eastern Indian meanwhile see fewer out-of-state deaths: West Bengal (600 deaths), Chhattisgarh (600 deaths), Bihar (600 deaths), Odisha (600 deaths), and Jharkhand (200 deaths).
Lastly, figure 2(d) shows the unequal mortality burden from electricity consumption between states in India. We derive these estimates in a similar manner as figure 1(d), where we first take difference between the mortality burden for each state from all electricity generation in India (figures 1(a) and 2(a)) and subtract from it the number of in-state deaths from plants that supply the state's electricity consumption ( figure 2(b)). This gives the number of deaths in each state from plants supplying other states' electricity consumptions. If all states shared equal mortality burdens tied to their electricity consumption, values in figure 2(d) would be approximately zero, i.e. the deaths outside a state from the state's electricity consumption would equal deaths in the state associated with other states' electricity consumptions. However, here like production, distinct spatial patterns emerge, estimates are net negative for states predominately in western, southern, and northern India ('consumption death exporters'). States in western and southern India generate and consume a disproportionate amount of India's renewable energy generation (Sengupta et al 2022b). Using Gujarat as an example again, ∼4300 deaths (figure 2(a)) occur from all electricity consumption in India, and ∼3500 of those deaths ( figure 2(b)) occur in the state's from its own electricity consumption. The rest, ∼800 occur from other states' consumptions. Meanwhile, Gujarat's electricity consumption is associated with ∼4400 deaths outside Gujarat ( figure 2(c)). Gujarat's electricity consumption results in ∼3500 more deaths (figure 2(d)) outside the state than deaths in Gujarat from other states' consumptions. Numbers are approximate due to rounding error. Likewise, other states who consume electricity associated with higher deaths include Rajasthan (−2100 deaths), Haryana (−1400 deaths), Maharashtra (−1300 deaths), Kerala, (−1300 deaths), Punjab (−1000 deaths) and Delhi (−1000 deaths). Estimates are net positive for states predominately in eastern and central India which are more coal dependent. This means these states face a disproportionate burden of electricityconsumption deaths. They have a higher burden of deaths within their borders from electricity consumption located outside their borders than deaths outside their borders from their own electricity consumption ('consumption death importers'). States include West Bengal (1900 deaths), Chhattisgarh (1700 deaths), Bihar (1600 deaths), Uttar Pradesh (1600 deaths), Tamil Nadu (1200 deaths), Madhya Pradesh (1200 deaths), Jharkhand (900 deaths), and Andhra Pradesh (700 deaths).
We present reductions in annual PM 2.5 mortality from simulating sulfur emissions control, carbon taxes, and market integration figures S1-S3 in the supplementary material. In summary, greater sulfur emissions control would reduce mortality throughout India by approximately 50%. Increasing carbon taxes of up to $100 per ton CO 2 would reduce annual deaths in the short-term by about 9% with most reductions in Tamil Nadu. Lastly, integrating markets by pooling dispatch across states instead of the current siloed approach by states would reduce annual deaths by about 8% likewise mostly in Tamil Nadu.

Discussion and conclusion
In this work we use InMAP, a reduced-complexity air quality model, to evaluate annual PM 2.5 mortalities associated with electricity production and consumption in India. We evaluate electricity associated mortalities by each Indian state because states are largely responsible for scheduling and dispatching power in India. Moreover, under a federal system, states in India along with the central government share jurisdiction over power sector policies. Consequently, considerable heterogeneity exists between Indian states in terms of electricity demand, generation fuel mix, and emissions of GHGs and PM 2.5 precursors (Sengupta et al 2022a). States will continue to be a focus of policy efforts given these features. We evaluate this heterogeneity by further resolving the premature mortality embedded in the production and consumption of electricity in each state. Furthermore, we test several policy scenarios including emissions control, carbon taxes and market integration to understand how changes in power sector operations in the current Indian grid affect ambient PM 2.5 concentrations and associated mortality. Here, we discuss our baseline results and apportionment of premature mortality between states. We contextualize our remaining results for emissions control, carbon taxes, and market integration scenarios in the supplementary material. These scenarios are relevant due to Government of India regulations to control power sector air pollution, establish national carbon markets, and integrate power dispatch between states.
Overall, our baseline 2017-2018 estimates of PM 2.5 mortality attributable to power generation are consistent with previous estimates. We find approximately 62 000 PM 2.5 deaths (∼5.6% of total PM 2.5 mortality) attributable to power generation. Most recent 2018 estimates from Cropper et al (2021) find approximately 113 000 PM 2.5 deaths (∼9.1% of total PM 2.5 mortality) attributable to the power sector in India, assuming exposure to only outdoor, ambient PM 2.5 as we have done here. Previous analyses quantifying PM 2.5 mortality attributable to power generation in India since 2011 find best estimates of 71 000-270 000 deaths with most analyses finding ∼100 000 deaths (Guttikunda and Jawahar 2014, Lelieveld et al 2015, GBD MAPS Working Group 2018, Gao et al 2018, Guo et al 2018, Conibear et al 2018b, Apte and Pant 2019. Estimates differ due to differences in assumptions of emission estimates (from power and non-power sectors), PM 2.5 exposure estimates predicted by chemical transport models at varying spatial resolutions, and risk-exposure relationships. As a sensitivity case, we estimate mortality using GEMM risk curves (see supporting information) and arrive at approximately 83 000 (∼5.7% of total PM 2.5 mortality) assuming the same underlying mortality rates and PM 2.5 estimated exposure from InMAP and Hammer et al (2020).
Likewise our state wise mortality estimates attributable to power generation are fairly consistent with estimates assuming no household air pollution exposure from Cropper et al (2021). Differences arise because we assume higher baseline total PM 2.5 exposure with (Hammer et al 2020) (table S3) and higher total emissions from power generation (table S4), which result in lower PM 2.5 concentrations attributable to power generation when fed into InMAP (table  S3). This on balance decreases the fraction of total PM 2.5 mortality attributable to power generation. Our state-wise breakdowns of total PM 2.5 mortality from all sources (table S5) are in closer agreement to those reported Cropper et al (2021) because despite differences in total PM 2.5 exposure, both our estimates produce similar relative risk values, i.e. both our analyses end up the on flatter part of the GBD riskexposure curve at higher concentrations (Global Burden of Disease Collaborative Network 2021). Likewise, using GEMM risk curves (Burnett et al 2018) which estimate higher levels of risk (and thus mortality) than GBD risk curves at the same PM 2.5 exposure, our total PM 2.5 mortality estimates are consistent with Cropper et al (2021). The largest deviation in state-wise estimates occur in Tamil Nadu, which we estimate to have very high PM 2.5 concentrations due to a cluster of high sulfur-emitting lignite power plants. While  Consequently, our estimates likely represent an upper bound on impacts from this cluster, and we hypothesize that variability in spatial resolution along with variability in emissions estimates likely drive discrepancies between different analyses. Previous nationwide analyses for the United States quantifying air pollution damages from multiple sectors using chemical transport models have shown the impact of spatial resolution on mortality and damage estimate (Paolella et al 2018, Goodkind et al 2019. Consequently this warrants further scrutiny and investigation for India because it is unclear how varying assumptions on source emissions, chemical transport model spatial resolution, and riskexposure affect the range of estimates for PM 2.5 mortality source attribution.
Our analysis of net deaths associated with electricity production and consumption in India further show spatial inequalities among states with respect to climate and air pollution impacts. Previous studies (Conibear et  have quantified baseline spatial differences in mortality burdens between states. However we extend this quantification by resolving inter-and intra-state differences in mortality caused by the interactions in electricity generation between states. In the case of both production and consumption, we see states in eastern India predominately coal-mining areas with plants that provide the cheapest power near mines (Kamboj and Tongia 2018) disproportionately face the burden of PM 2.5 mortality from electricity in India. This difference becomes even more prominent in the consumption case where we see the states that are net death exporters are also those with the highest amounts of renewable capacity and generation (e.g. Gujarat, Rajasthan, Maharashtra, and Telangana). These patterns remain consistent assuming both GBD and GEMM risk curves. This suggests that as these states have adopted more renewable generation, they have shifted their emitting coal generation to predominately eastern areas, shifting associated PM 2.5 mortalities as well. As renewable energy continues to grow disproportionately in India in a handful of states in southern and western India and states individually contract for and schedule power, these inequalities may grow in the future. Several analyses of hypothetical future Indian grids have shown these disproportionate capacity addition and generation patterns to persist (Rose et al 2020, Spencer et al 2020, Abhyankar et al 2021. Furthermore, we identify two limitations with this analysis. The first limitation is that we ignore exposure to household air pollution in our mortality quantification unlike previous analyses (Pandey et al 2021, Cropper et al 2021). Our total outdoor ambient PM 2.5 mortality estimates (∼1.1 million) are larger than previous estimates (∼900 000) which indicates that our estimates attribute some PM 2.5 exposure from household air pollution to outdoor air pollution. The states in eastern and central India we identify facing a disproportionate burden of electricity PM 2.5 mortality are also poorer and likely to have higher household air pollution exposures due to lower rates of modern energy access (Mani et al 2021). Evaluating exposure levels to household air pollution were beyond the scope of this work, but future analyses may refine estimates to account for this source. Consequently, the unequal burdens of eastern and central states quantified here are likely more than if they incorporated household air pollution.
The second limitation we identify is our use of InMAP (Thakrar et al 2022) to quantify the changes in concentration from changes in emissions from policy scenarios. Changes in PM 2.5 concentrations due to changes in emissions are non-linear and depend spatially and temporally, so using a full-form air quality (chemical transport) model to quantify concentration will yield more precise estimates. While InMAPs global version overpredicts concentration changes when increasing emissions inputs versus the same estimates from GEOS-Chem, its performance is best over South Asia with similar results as the full-form model (Thakrar et al 2022). Given that in our BAU and sulfur control scenarios, we arrive at similar estimates as previous analyses (Srinivasan et al 2018, Cropper et al 2021 using both InMAP estimates and underlying exposure estimates constrained by ground measurements, satellite observations and chemical transport models (Hammer et al 2020), we believe our results are robust to choice of air quality model. However, as suggested by developers of InMAP, our approach may not yield as consistent results regions outside South Asia.

Data availability statement
The data that support the findings of this study are available upon reasonable request from the authors.

Funding
This material is based on work supported by the National Science Foundation Graduate Research Fellowship Program Grant Nos. DGE-1252522 and DGE-1745016. Any opinions, findings, and conclusions or recommendations expressed in this work are those of the authors and do not necessarily reflect the views of the National Science Foundation. This publication was developed as part of the Center for Air, Climate, and Energy Solutions (CACES), which was supported under assistance Agreement No. R835873 awarded by the US Environmental Protection Agency (EPA). It has not been formally reviewed by the EPA.