A complete transition to clean household energy can save one–quarter of the healthy life lost to particulate matter pollution exposure in India

Simulations were conducted using the Weather Research and Forecasting model online–coupled with Chemistry (WRF–Chem) version 3.7.1 (Grell et al 2005), incorporating various model improvements, including updated anthropogenic emissions, aqueous chemistry, and a more complex secondary organic aerosol scheme. Detailed information on the model setup is provided in supplementary table 1 (stacks.iop.org/ERL/15/094096/mmedia) and the supplementary methods. Simulations were for the year of 2016 with one month spin–up. The model domain covered South Asia at 30 km (0.3°) horizontal resolution.

Anthropogenic emissions of black carbon (BC), organic matter (OM), non–methane volatile organic compounds (NMVOC), nitrogen oxides (NO_X), other PM_2.5, and sulphur dioxide (SO₂) for residential biomass (wood, dung, and crop residues for cooking, space heating, and water heating), residential LPG (including biogas), residential kerosene, and residential lighting are for 2010 from a new residential inventory for India (Lam and Bond 2020). These emissions were produced at 1 km spatial resolution based on village surveys of energy services required, then aggregated to 0.25° × 0.25° horizontal resolution. Emission factors for the residential sector were completely reassessed to include field–measured emission factors that have recently become available. Residential kerosene use is diminishing in India without the need for further incentives.

Anthropogenic emissions of BC, organic carbon (OC), NMVOC, NO_X, other PM_2.5, and SO₂ for open burning, power plant coal (thermal), industrial coal (heavy and light), brick production, transportation (on–road gasoline/compressed natural gas, on–road diesel, and railways), distributed diesel (agricultural tractors, agricultural pumps, and diesel generator sets), and other sources (informal industry, trash burning, and urban fugitive dust) were taken from (Venkataraman et al 2018) as used by the Global Burden of Disease from Major Air Pollution Sources study for 2015 at 0.25° × 0.25° horizontal resolution (GBD MAPS Working Group 2018, Venkataraman et al 2018). Anthropogenic emissions of carbon monoxide (CO), ammonia (NH₃), acetylene (C₂H₂), and methane (CH₄) were from the Emission Database for Global Atmospheric Research with Task Force on Hemispheric Transport of Air Pollution version 2.2 for 2010 at 0.1° × 0.1° horizontal resolution (Janssens-Maenhout et al 2015).

To understand the impacts of interventions to replace solid fuel use with LPG, we completed annual simulations using five different emission scenarios (table 1). Household energy scenarios were split by access and usage issues, relative to a complete transition.

Figure 1 shows the annual anthropogenic emission totals from these scenarios. In the BASELINE scenario, residential biomass makes a substantial contribution to anthropogenic OC and NMVOC emissions. Power plant and industrial coal use dominate anthropogenic emissions of other chemical components of PM_2.5, SO₂, and NO_X. Anthropogenic dust, open burning, and trash burning contribute strongly to anthropogenic PM_2.5 emissions. Transportation and distributed diesel contribute heavily to anthropogenic NO_X and NMVOC emissions. Under the ALLLPG scenario, total BC emissions are reduced by 47%, and OC emissions by 77%, contributing to a 44% reduction in total primary PM_2.5 emissions relative to the BASELINE. There are also substantial reductions in NMVOC (28%) emissions, while SO₂ and NO_X emissions are reduced by less than 2%. Applying the spatial constraint to the intervention as in URB15, resulted in half the emission reduction that was achieved in ALLLPG. The stove stacking scenario in EMIS50 resulted in similar total emission reductions to URB15, but with different spatial patterns.

Zoom In Zoom Out Reset image size

Model evaluation was conducted using measurements obtained from OpenAQ (OpenAQ 2019). Following Manning et al (2018), sites were accepted for evaluation when there was more than 16 h of data per day, more than 50 d of data, and when hourly–mean PM_2.5 concentrations were greater than 5 μg m^–3 (93% acceptance). Measurements were aggregated to annual–means. There were 34 OpenAQ measurement sites in India for 2016 that passed this criteria. We compared the same days in the model as there are data in the measurements. The normalised mean bias factor (NMBF) and the normalised mean absolute error factor (NMAEF) were used to evaluate the model (Yu et al 2006). The measurements from OpenAQ were primarily collected from the Central Pollution Control Board (Ministry of Environment and Forests 2018). To increase the sample size of measurement sites for evaluation, these were combined with the World Health Organization Global Ambient Air Quality Database for measured annual–mean PM_2.5 concentrations in 2016 (World Health Organization 2018).

The model captures the spatial variation and magnitude of annual–mean APM_2.5 concentrations across India, with greatest concentrations over the Indo–Gangetic Plain (figure 2). The model slightly underestimates observed APM_2.5 concentrations (NMBF = −0.12 and NMAEF = 0.39). Our previous work used a similar model configuration and similarly underestimated measured APM_2.5 concentrations (Conibear et al 2018a). Overall, the simulated APM_2.5 concentrations show adequate skill to address questions of relative change in long–term APM_2.5 concentrations over India.

Zoom In Zoom Out Reset image size

All the health impact assessments were for the same year (2015) to remove confounding influences of changing population size, population age, and baseline mortality rates. The health impact assessment estimated the disease burden attributable to PM_2.5 exposure using population attributable fractions (PAF). Intervention–driven variations in exposure were used to predict associated variations in outcome. This study followed the approach of the Global Burden of Diseases, Injuries, and Risk Factors Study (GBD) 2017 (GBD 2017 Risk Factor Collaborators 2018). The GBD2017 apportioned the disease burden attributable to TPM_2.5 between APM_2.5 and HPM_2.5 for solid fuel users and non–solid fuel users following relationships in supplementary table 2. The proportional PAF allowed for the individual disease burdens to be additive, due to the joint effects of APM_2.5 and HPM_2.5 being considered on single integrated exposure–response (IER) function per disease. The primary metrics used in the health impact assessment are the annual number of premature mortalities (MORT) and the rate of DALYs per 100 000 population, i.e. the total loss of healthy life. Detailed information on the methodology for the health impact assessment for APM_2.5 and HPM_2.5 are in the supplementary methods, where supplementary figure 1 shows the HPM_2.5 concentrations and supplementary figure 2 shows the IER. The epidemiological data underlying health impact assessments are rapidly developing and a range of uncertainties remain (see supplementary methods). Recent developments to the IER for GBD2017 allow a better understanding of the combined impacts of household and ambient PM_2.5 exposure. As the main results of this paper are comparative, future epidemiological developments that influence the absolute disease burdens will not impact the comparative lessons drawn from this paper.

We calculated annual–mean population–weighted APM_2.5 concentrations of 75.4 μg m^–3. This estimate is lower than the annual–mean measured APM_2.5 concentrations (90 μg m^–3, figure 2(b)) and the latest GBD (91 μg m^–3, GBD 2017 Risk Factor Collaborators 2018), but higher than Chowdhury et al (2019b) (55 μg m^–3) (supplementary figure 3). We estimated annual–mean population–weighted HPM_2.5, based on data from Shupler et al (2018), of 248.6 μg m^–3, 178.9 μg m^–3, and 216.2 μg m^–3 for females, males, and children, respectively.

Figure 3 shows the disease burden associated with TPM_2.5 exposure under the BASELINE scenario. We estimated 1 190 000 (95UI: 764 000–1 601 000) premature mortalities per year associated with TPM_2.5 exposure, with 44% from APM_2.5 exposure and 56% from HPM_2.5 exposure. The DALYs rate associated with TPM_2.5 exposure was 2900 (95UI: 1900–3900) per 100 000 population, with 45% from APM_2.5 exposure and 55% from HPM_2.5 exposure. The individual disease burdens associated with APM_2.5 and HPM_2.5 exposure are shown in supplementary figures 4 and 5, respectively.

Zoom In Zoom Out Reset image size

Our estimates for the disease burden associated with TPM_2.5 exposure are slightly larger (+5% for MORT) than those from the GBD2017 (Institute for Health Metrics and Evaluation 2019). This is the result of our larger estimated disease burdens associated with HPM_2.5 exposure (+31% for MORT), and smaller estimates of disease burdens associated with APM_2.5 (–17% for MORT). The larger disease burden estimates associated with HPM_2.5 exposure are due to our use of higher HPM_2.5 exposures and the state–specific solid fuel use, both with higher estimates in the densely populated Indo–Gangetic Plain. Overall, our disease burden estimates associated with PM_2.5 exposure in India are in general agreement with those from the GBD2017.

The health impact assessment was also estimated for ambient ozone (O₃) exposure, following the methodology of the GBD2017 (supplementary figure 6 and supplementary methods). The model underestimated O₃ concentrations (NMBF = −0.40 and NMAEF = 0.49), though the magnitude of the bias is similar to many regional modelling studies over India (supplementary figure 7 and supplementary methods). We found no difference between the disease burden estimates across the five scenarios. Hence, we focus on the public health impacts associated with TPM_2.5 exposure.

A complete transition to clean household energy (i.e. ALLLPG relative to BASELINE) reduced APM_2.5 concentrations by 25% (population–weighted from 75.4 μg m^–3 to 56.4 μg m^–3, annual–mean) as shown in figure 4(a). Improvements in air quality occur across India, with the largest reductions in pollution across the Indo–Gangetic Plain. The 25% reduction in APM_2.5 concentrations found here is similar to the 24% reduction estimated by the GBD MAPS Working Group (2018) with similar emissions. Previous studies estimated that a complete removal of residential emissions, without any replacement fuel, would lead to a 22–56% reduction in APM_2.5 concentrations (Chafe et al 2014, Lelieveld et al 2015, Butt et al 2016, Silva et al 2016, Karagulian et al 2017, Conibear et al 2018, Guo et al 2018, GBD MAPS Working Group 2018, Gao et al 2018, Upadhyay et al 2018, Reddington et al 2019, Chowdhury et al 2019a).

Zoom In Zoom Out Reset image size

We estimate that a complete transition to clean household energy would prevent 29% of the present–day disease burden associated with PM_2.5 exposure, preventing 348 000 (95UI: 284 000–373 000) premature mortalities each year (figure 4). A complete transition to LPG reduces DALYs by 800 (95UI: 600–900) per 100 000 population.

Chowdhury et al (2019b) found a complete transition to household LPG in India could reduce the number of premature mortalities associated with APM_2.5 exposure by 13%, where the population–weighted APM_2.5 concentrations reduced by 31% to 38 μg m^–3. We find that a complete transition to clean household energy can reduce the disease burden associated with TPM_2.5 by 29%, when the population–weighted APM_2.5 concentrations reduce by 25% to 56.4 μg m^–3. A key finding of Chowdhury et al (2019b) is that a transition to clean household energy would allow India to meet the National Ambient Air Quality Standard of 40 μg m^–3. While we found a similar percentage reduction in APM_2.5 concentrations, annual–mean APM_2.5 concentrations in India remained above the National Ambient Air Quality Standard due to higher baseline population–weighted APM_2.5 concentrations (75.4 μg m^–3 relative to 55 μg m^–3). The key novelties of our study are the consideration of TPM_2.5 exposure to account for the joint risks between APM_2.5 and HPM_2.5 exposure, and the use of a high spatial resolution (1 km) residential emission inventory based on village surveys of energy services required. As we account for TPM_2.5 exposure, the disease burden estimates associated with APM_2.5 exposure are not directly comparable to those from Chowdhury et al (2019b). Despite these differences, our study confirms the major findings of Chowdhury et al (2019b), namely that a complete transition to clean household energy can substantially improve public health in India.

Table 2 and figure 5 summarise the impacts of different household energy scenarios on air quality and public health at the national scale in India. Detailed data per state are provided in the Supplementary Data.

Zoom In Zoom Out Reset image size

An incomplete transition to clean household energy might be limited by spatial access to LPG distribution (i.e. URB15 relative to ALLLPG). This transition, which reaches 80% of the population, nevertheless reduces the potential health benefits of a clean fuel transition by about 50%. Under this scenario, APM_2.5 concentrations are reduced by 12%, in contrast to the 25% reduction in the complete transition. This incomplete transition to clean household fuels results in a 13% reduction in total premature mortality, compared to a 29% reduction in the complete transition to clean fuels. Spatial constraints on access to LPG, falling under the category of 'distribution potential' (Lam and Bond 2020), therefore reduce the avoided premature mortalities by 197 000 (95UI: 160 000–211 000) per year. This scenario also reduces the avoided DALYs by 600 (95UI: 400–600) per 100 000 population (supplementary figure 9).

If emission reductions are spread evenly among remote rural areas compared with urban centres (i.e. STATE50 relative to URB15), the national–mean APM_2.5 concentrations are also reduced by 12%. The spatial distribution of emission reductions means that APM_2.5 concentrations under STATE50 are larger in urban areas (up to + 15 μg m^–3 in Delhi) and smaller in rural areas (up to −15 μg m^–3 in Uttar Pradesh and Bihar) relative to URB15 (supplementary figure 10(a)). The disease burden from TPM_2.5 exposure under STATE50 increases by 5% relative to URB15, and the number of premature mortalities increases by 55 000 (95UI: 41 000–62 000) per year (supplementary figure 10(b)). This increase is the net of two opposing changes: a 45% increase in the disease burden from HPM_2.5 exposure under STATE50 relative to URB15, because all households using solid fuels under the BASELINE retain some solid fuel use in STATE50, and a 15% decrease in the disease burden from APM_2.5 exposure caused by reduced emissions in high–population urban areas. The dominating role of HPM_2.5 is due to the proportional PAF and the non–linear IER, where large HPM_2.5 exposures under STATE50 drive large disease burdens from TPM_2.5 exposure. Comparing STATE50 to ALLLPG, where remote rural areas have relatively small emission reductions, the health benefits of the clean fuel transition are reduced by approximately 75%. The implication here is that in addition to reaching remote rural areas, the reductions in HPM_2.5 exposure need to be substantial.

Stove stacking with solid fuels (i.e. EMIS50 relative to ALLLPG) reduces the public health benefits of a clean fuel transition by approximately 75%. Under the stove stacking scenario, APM_2.5 concentrations are reduced by 13%, compared to 25% in the complete transition. Stove stacking reduces the number of avoided premature mortalities by 255 000 (95UI: 201 000–278 000), and reduces the avoided DALYs by 600 (95UI: 500–700) per 100 000 population (supplementary figure 11). The implication of these constraints, coupled to the non–linear IER where risk decreases substantially at the lowest TPM_2.5 concentrations, suggests that large public health benefits are possible, but only if there is a nearly complete and exclusive transition to clean household energy.

A key implication of these results is that it is critical for air pollution studies of residential emissions to consider the disease burden attributed to TPM_2.5 exposure, and not only the portion of TPM_2.5 attributed to either APM_2.5 or HPM_2.5 exposure. This is important because the exposure–outcome associations are non–linear and the joint effects of APM_2.5 and HPM_2.5 are proportional. This means that as the disease burden attributed to HPM_2.5 exposure decreases, the disease burden attributed to APM_2.5 exposure can increase, and vice versa. For example, under the BASELINE scenario 522 000 (95UI: 327 000–716 000) premature mortalities were attributed to APM_2.5 exposure, while under the ALLLPG scenario this attribution increased to 842 000 (95UI: 481 000–1 228 000) premature mortalities, despite a 25% reduction in APM_2.5 exposure. This increased attribution to APM_2.5 exposure when HPM_2.5 exposure is removed is due to the non–linear exposure–outcome association, where both APM_2.5 and HPM_2.5 exposures are individually high enough so that the relative risk is in the flatter section of the response (supplementary figure 2). Hence, a slightly lower total risk from TPM_2.5 exposure is now entirely attributed to APM_2.5 exposure, rather than being attributed approximately evenly between APM_2.5 and HPM_2.5 exposures. The importance of these joint effects are also demonstrated by the variation in frequency distributions between the URB15, STATE50, and EMIS50 scenarios (supplementary figure 8). All three scenarios have similar frequency distributions of APM_2.5 exposure (supplementary figure 8(a)), however, they vary in HPM_2.5 exposures (supplementary figure 8(b)) which drives variations in the overall disease burden associated with TPM_2.5 exposure (supplementary figure 8(f)), and the corresponding attribution between APM_2.5 and HPM_2.5 exposure (supplementary figure 8(d) and 8(e)). The importance of integrating APM_2.5 and HPM_2.5 exposures in India and other areas of high residential solid fuel use has been emphasised in previous studies (Balakrishnan et al 2014, Aunan et al 2018). High residential solid fuel use leads to large HPM_2.5 exposures and substantial source contributions to APM_2.5 exposures, whereby the reduction of residential emissions is an important equity issue in India (Kathuria and Khan 2007, Cowling et al 2014).

We showed that the transition to clean household energy has the potential to reduce the disease burden associated with PM_2.5 exposure by 29%, preventing 348 000 (95UI: 284 000–373 000) premature mortalities every year. These potential public health benefits are dependent on the complete transition to clean fuels. Limited spatial access to LPG reduced health benefits by 50%, and stove stacking with solid fuels reduced health benefits by 75%. This dependency of public health benefits on a complete transition to clean fuels has been seen in India (Pillarisetti et al 2014, 2018, Hanna et al 2016, Smith 2017a, Aung et al 2018).

To complete the transition to clean household energy and provide these substantial public health benefits, remaining access and usage issues need to be overcome (Gould and Urpelainen 2018, Tripathi and Sagar 2019, Kar et al 2019, Pattanayak et al 2019, Harish and Smith 2019). These include extending access to all, especially the most remote, poor, and vulnerable (Harish and Smith 2019), increasing refill rates (Jain et al 2018, Pillarisetti et al 2019), improving continual affordability (Harish and Smith 2019, Tripathi and Sagar 2019), and improving awareness of the need for continual use of clean fuels (Smith 2018, Harish and Smith 2019). The complete transition to clean household energy may also provide multiple benefits to sustainable development, human wellbeing, the climate, ecosystems, and the economy (Smith and Haigler 2008, Wilkinson et al 2009, Venkataraman et al 2010, Smith et al 2014, Bailis et al 2015, World Health Organization 2016, Rosenthal et al 2018).

We explored how hypothetical transitions to clean household energy could change air pollution exposure and the consequent change in associated health outcomes. We did not attempt to evaluate the impact of specific clean energy programmes. The isolated impacts of exposure change were quantified by holding demographics and background mortality rates constant. To directly assess effectiveness of specific air quality policies to improve human health, causal epidemiology study designs will be required once the policies are complete and the appropriate data is available (van Erp et al 2012, Zigler and Dominici 2014, Zigler et al 2016, Boogaard et al 2017, Burns et al 2019). Exposure–outcome differentials and effect modification among equity groups are important future research topics for air pollution studies in India.