Air quality impacts of crop residue burning in India and mitigation alternatives

Long-term exposure to ambient PM2.5 is associated with elevated health risks, causing more than four million premature deaths globally each year, 10–25% of which are estimated to occur in India. Crop residue burning is a key source of direct PM2.5 emissions. India generates ~500 million metric tonnes of crop residue annually, about 100 MT of which is burned, primarily after wheat harvest (April–May) and rice harvest (October–November) in northwestern India. Downwind, densely populated areas in the Indo-Gangetic Plain experience high annual mean PM2.5 (50–200 µg m³) with episodic spikes (200–1200 µg m³) during burning seasons, far exceeding WHO guidelines. Despite regulations—crop residue management programs, burning bans, and fines—burning remains prevalent due to agronomic and economic constraints, including the low utility of rice and wheat residues for alternative uses. This study aims to quantify India-wide air quality and health impacts attributable to agricultural residue burning, attribute impacts to locations and times, and evaluate targeted mitigation strategies including district-level actions and within-day shifts in burning time using an adjoint modeling approach.

Prior work has documented severe air pollution episodes in cities such as Delhi during crop residue burning seasons and demonstrated regional-scale transport across South Asia. Epidemiological studies indicate increased mortality risk with any additional PM2.5 exposure, supporting consideration of widespread small exposure increments as well as localized peaks. Existing studies have attributed a portion of India’s PM2.5-related premature mortality to agricultural burning (e.g., GBD MAPS estimated ~66,200 deaths in 2015). However, many studies focus on local/urban scales and single seasons. This work extends the literature by using adjoint modeling to equitably attribute impacts across India to specific locations and times of burning, and by proposing and quantifying a new mitigation lever: shifting burning within the day.

Emissions: Agricultural residue burning emissions are from GFEDv4.1s. Daily emissions (2003–2019) inform adjoint analyses; monthly emissions (1997–2019) inform forward simulations. Burning-attributable PM2.5 is defined as BC + OC. Because GFED’s standard diurnal cycle derives from the Americas and may not reflect Indian practices, an alternative triangular diurnal profile is applied for India with 95% of emissions between 06:30–19:30 local time, peaking at 14:30 LT, consistent with MODIS Terra/Aqua fire count ratios. Adjoint modeling: The GEOS-Chem adjoint (v35) at 0.5°×0.667° over extended Asia quantifies sensitivities of India-wide population-weighted PM2.5 exposure (cost function J) to emissions in space and time. Three annual adjoint simulations represent meteorological regimes: flood (2007), drought (2009), and normal (2012), based on monsoon rainfall. For each year 2003–2019, one sensitivity dataset is assigned according to rainfall regime. Two additional adjoint simulations modify J to represent urban (population density >400 km−2) and highly populated areas (>1,000 km−2) for a normal year, to separate impacts by population density. Meteorology uses GEOS-5; non-fire emissions from EDGAR v4.3.2. Forward modeling: GEOS-Chem Classic (v13.0.2) forward simulations at 0.5°×0.625° over extended Asia are run for October–November (post-monsoon) for each year 1997–2019, with and without Indian agricultural burning, using MERRA-2 meteorology and monthly GFEDv4.1s. These quantify long-term trends, spatial distribution, and transboundary impacts. Exposure and health impacts: Population-weighted exposure is computed over India’s states/UTs and districts. Premature mortality is estimated using the Integrated Exposure Response (IER) function for five causes (COPD, IHD, LRI, lung cancer, cerebrovascular disease), with India-specific population age structure and baseline mortality (GBD/UN). Monte Carlo (1,000 draws) yields mean and 95% CI. Alternative concentration-response functions (log-linear, GEMM, MR-BRT) are evaluated in sensitivity analyses. Monetized impacts use a value of statistical life (VSL) transferred from the US to India via benefit transfer with income elasticities (0.7 intra-US, 1.5 US–India), GDP deflators, and PPP ratios; uncertainty in VSL is propagated via Monte Carlo. Intervention analyses: Spatial targeting efficacy is assessed as percent reduction in India-wide impacts per 1% emission reduction by state/district in pre- and post-monsoon seasons. Temporal interventions shift the diurnal peak of burning by ±1–6 hours, recomputing impacts under adjoint sensitivities. Model evaluation: Forward simulations are evaluated against satellite-derived PM2.5, MODIS AOD, AERONET AOD, and CPCB/US Embassy ground PM2.5. Forward/adjoint consistency checks show <10% discrepancy in exposure estimates across tests.

- Average (2003–2019) annual India-wide population-weighted PM2.5 exposure due to burning: 6.7 µg m³; pre-monsoon contributes 28%, post-monsoon 64%. - Geographic attribution: >90% of exposure increase is due to northwest states; Punjab 64%, Haryana 11%, Uttar Pradesh 5.7% on average. Urban and densely populated regions experience 2.0 µg m³ (31%) and 4.8 µg m³ (73%) higher annual PM2.5 than the national average due to burning. - Health burden: Mean 69,000 (95% CI: 57,000–80,000) premature deaths per year attributable to burning over 2003–2019; across years the range is 44,000–98,000, with 2016 the highest at 98,000 (95% CI: 82,000–110,000). Punjab, Haryana, and Uttar Pradesh consistently contribute 67–90% of impacts; by state shares: Punjab 48–75%, Haryana 7.8–14%, Uttar Pradesh 3.7–9.5%. - Monetized cost: Average annual cost 23 (95% CI: 3.5–53) billion USD (≈38% of total health expenditure; 7.8% of agricultural gross value added on average), increasing sixfold from 2003 to 2019 (to 44 [6.7–100] billion USD in 2019). - District contributions: Six Punjab districts account for ~40% of national impacts; Patiala and Sangrur together contribute ~20%. District categorization shows that combinations of higher emissions per unit production, higher mortalities per unit emission, and larger production drive outsized impacts. - Efficacy of spatial reductions (post-monsoon): 1% emission reduction yields 0.57% decrease in national impacts from Punjab and 0.065% from Haryana, averting ~380 (95% CI: 320–450) and 45 (37–52) deaths annually, valued at ~130 (20–300) and 15 (2.3–34) million USD, respectively. Within Punjab, Sangrur, Patiala, and Ludhiana dominate potential benefits. - Temporal shifting: Shifting burning earlier by 1–4 hours reduces annual impacts by 0.5–19% on average; burning 2–3 hours earlier in November reduces annual India-wide burning-related impacts by 15–23%. Restricting to Punjab, burning 2 hours earlier in November yields a 14% reduction on average, averting 9,600 (95% CI: 8,000–11,000) premature deaths valued at 3.2 (0.49–7.3) billion USD. - Interannual variability: Relative to the 17-year mean, impacts are ~2.4% lower in drought years and ~4.8% higher in flood years. Delhi’s post-monsoon period average carbonaceous PM2.5 consistently exceeds 120 µg m³ (1997–2019). - Policy context 2018–2019: Despite mechanization subsidies, nationwide impacts in 2018–2019 (~86,000 deaths each year) remained similar due to increased pre-monsoon burning offsetting post-monsoon reductions; with fixed meteorology, pre-monsoon share in the three key states rose to 38% in 2019 (from 9% in 2015). - Transboundary impacts: Indian burning increases population exposure in Pakistan, Bangladesh, and Nepal by about 0.24–12% of India’s increase; 88–95% of exposure is borne within India.

The study addresses how agricultural residue burning contributes to India-wide PM2.5 exposure and associated mortality, and how targeted interventions can most effectively reduce harm. Using adjoint sensitivities, the work links incremental emissions changes by district, season, and hour to national exposure, revealing highly concentrated contributions from a few Punjab districts and substantial benefits from modest within-day timing shifts toward earlier burning when dispersion is more favorable. These findings imply that spatially and temporally targeted measures—complementing broader bans and mechanization—could deliver large health and economic benefits at potentially lower cost and disruption. The results also show that seasonal shifts in emissions can offset gains, underscoring the need to address both pre- and post-monsoon burning. Transboundary effects exist but most exposure remains within India due to prevailing winds and population distribution. The analysis supports prioritizing districts cultivating residue-intensive rice varieties and implementing measures that align burning with meteorology to reduce downwind exposure.

This work quantifies and disaggregates India-wide air quality and health impacts from crop residue burning over 2003–2019, identifies outsized contributors at the district level, and proposes a novel mitigation lever—shifting burning to earlier hours—to reduce exposure. Six districts in Punjab account for ~40% of national impacts, suggesting substantial benefits from geographically targeted actions, crop diversification toward less residue-intensive varieties (e.g., basmati rice, pulses, oilseeds), and improved residue management. Timing interventions (morning vs afternoon) could avert up to ~14% of impacts in key regions. Future research should assess feasibility and cost-effectiveness, evaluate local downwind effects via forward modeling with online aerosol–meteorology interactions, and collect in-field observations of burning practices and diurnal cycles to validate and refine intervention design.

- Health impact estimates consider primary carbonaceous PM2.5 (BC+OC) from burning and exclude other species/secondary formation and ozone. - The IER assumes equal toxicity across PM2.5 species and is not India-specific; alternative concentration-response functions yield different magnitudes (−56% with log-linear; up to +45% with GEMM/MR-BRT). - Use of a single diurnal emissions cycle based on satellite proxies may not capture local hourly burning behaviors; district-level diurnal patterns are not available. - Spatial model biases exist in some regions; aerosol–PBL and aerosol–wind interactions are not fully resolved in offline simulations. - Baseline mortality rates are applied at the national level; subnational socioeconomic heterogeneity is not explicitly represented. - Uncertainties in satellite observations and emissions inventories are explored via inter-comparisons but not fully quantified in central estimates.