As Indian epidemiological evidence grows and concentration response function models are being developed and improved, future work may benefit from the adoption of a new method. Third, we assume a single diurnal cycle for burning emissions based on satellite information due to limited data of hourly burning activities from local sources . Lastly, since this study focuses on the broader air quality impacts over a large dispersion population, we do not specifically look at individual pollution hot-spots such as Delhi. We do however provide additional assessments for densely populated areas, where Delhi is a main recipient of pollution from agricultural fires .Our approach allows any proposed emissions change to be related to the eventual air quality impacts for the Indian population and sets the stage for future research into crop residue burning. Since we have focused most of our analysis on a single intervention, it would be a natural next step to examine the effects of such interventions in downwind locations using conventional forward modeling techniques. Online modeling considering aerosol-meteorology interactions is also needed to better understand whether these feed backs would suppress or enhance reductions in exposure. Furthermore, since our assumed diurnal pattern of burning may not reflect true fire activities, focused observational work on burning practices is needed to verify that these benefits are realizable. In addition, a deep assessment of any single alternative is needed to determine how plausible such an intervention would be in practice. Our study estimates the total annual premature deaths and the value of mortality risk reduction attributable to PM2.5 exposure from crop residue burning in India over 2003–2019. We also estimate the efficacy of marginal changes to reduce these impacts at the district level,stacking flower pot tower finding that a small number of administrative regions could be prioritized to provide the maximum air quality improvement. We find that six districts in Punjab are responsible for 40% of the nationwide air quality impacts as a result of meteorological factors, the size of the downwind population, and the use of residue-intensive crops.
Our work provides additional insights into potentially low-cost interventions that may significantly reduce the air quality impacts, such as shifting to burning in the morning rather than afternoon and promoting less residue-intensive crops . These findings provide a quantitative basis for the design and optimization of mitigation strategies for crop residue burning on a broad scale, as well as providing new opportunities for future regional and local studies on agricultural fires in India.Consistent with GBD 2018 India Special Report, we calculate emissions from agricultural residue burning using the Global Fire Emissions Database v4.1s from 2003 to 2019 and from 1997 to 2019. GFEDv4.1s is a hybrid emissions inventory that incorporates satellite and ground-based measurements to estimate fire emissions of various types . In particular, it includes a small fire boost based on active fire detections outside the burned area extent, which improves estimation of emissions from frequent and/or short-lived burning events. A comparison using alternative fire emissions inventories is provided in the Supplemental Information. Similar to Koplitz et al. 2016, we define burning-attributable PM2.5 as the sum of black carbon and organic carbon , the primary components of fire smoke-related PM2.5. The diurnal pattern of fire activity in the standard GFEDv4.1s product is estimated using an emissions redistribution approach. The diurnal cycles of burning are estimated based on observational data from geostationary satellites over the Americas, which are then applied to other parts of the world by matching three broad fuel types. While appropriate for many applications over North and South America, this method is not likely to accurately reflect agricultural residue burning in India because the crops grown, crop cycles, field size, and crop practices are different. We therefore apply an alternative diurnal cycle for agricultural residue burning in India based on satellite information from prior literature. The fire activity in sub-tropical areas is typically more intense in the early- to late-afternoon. Over India, the fire counts from MODIS Aqua are three to four times greater than those from Terra during periods of crop residue burning.
Based on this information, and in the absence of more reliable and/or accurate observational data specific to agricultural burning, we assume that agricultural burning emissions have a triangular profile , where 95% of emissions occur between 06:30 LT and 19:30 LT, with a peak at 14:30 LT. Sensitivity to this assumption is explored in Supplementary Discussion.We use the adjoint of the GEOS-Chem atmospheric chemistry and transport model to quantify the sensitivity of annual mean population exposure to PM2.5 in India with respect to emissions sources in the extended Asia domain. Adjoint simulations are performed at a resolution of 0.5° × 0.667° , with 47 uneven vertical layers from the surface up to 80 km altitude. Boundary conditions are saved from global runs at a resolution of 2° × 2.5°. The adjoint model quantifies the effect of changes in any emissions species at any time and any grid cell in India on a scalar quantity J. In our case, the cost function is the India-wide population weighted exposure to PM2.5. The adjoint approach has been widely applied in inverse problems such as air quality impact attribution, which suits the need of this study. We use GEOS-5 meteorological fields from the Goddard Earth Observing System of the NASA Global Modeling Assimilation Office and non- fire anthropogenic emissions from the Emissions Database for Global Atmospheric Research v4.3.2. Each adjoint simulation first requires a conventional, forward simulation to be performed; the data from these forward simulations is compared against observational data in our model validation . Two sets of simulations are run with the GEOS-Chem adjoint model. First, we perform three sets of simulations for three full years which respectively represent a typical rainfall condition for a “flood”, “drought” and “normal” year, based on 20-year monsoon rainfall data in India . Each set includes an adjoint run and a forward run . For each year we calculate the sensitivity of annual population-weighted exposure to PM2.5 across all of India with respect to emissions from December 1st the previous year to January 31st the year after. The first and the last month are discarded due to model spin-up and down, such that data for the whole year are used in the analysis. We then classify 2003–2019, where daily fire emissions are available, into three categories by meteorology type . By applying adjoint sensitivities with gridded agricultural fire emissions corresponding to their rainfall condition, we estimate the total change in population-weighted exposure for the entire Indian population due to emissions from crop residue burning for each year. Second, we perform two other full-year adjoint simulations, where the cost function is modified to annual population-weighted PM2.5 exposure for population in urban areas and highly populated areas, for a typical “normal” year . We define urban and densely populated areas as locations in which the population density exceeds 400 and 1,000 people per km2 , respectively. Besides estimating impacts on India as a whole, this allows us to separately quantify the impact of residue burning on different population groups, as people living in densely populated areas may be exposed to different exposure levels than those living in rural areas .
To inform estimates of long-term trends in exposure and the spatial distribution of impacts, we use the “forward” model GEOS-Chem Classic and perform 23 sets of conventional, forward-running simulations for September 1st to December 31st for each year between 1997 and 2019, where monthly fire emissions are available. September is discarded due to model spin-up, and only October to November are considered for the “post-monsoon season”. Each simulation is performed over the extended Asia domain at a resolution of 0.5° × 0.625° , with 73 uneven vertical layers from the surface up to 80 km altitude. Similar to adjoint simulations, boundary conditions are saved from global runs at a resolution of 2° × 2.5°. Each set includes two simulations with and without Indian agricultural residue burning emissions,ebb and flow which provides information on the long term impact of Indian post-monsoon crop residue burning on population living in neighboring countries including Bangladesh, Nepal and Pakistan. We use meteorological data from the ModernEra Retrospective analysis for Research and Applications, Version 2 and monthly agricultural residue burning emissions data from GFEDv4.1s. This data is also used in our model validation.Here the cost function for the adjoint simulation, J, is the annual mean population-weighted exposure to PM2.5 within India, including 29 states and seven union territories which are further divided into 666administrative districts. The adjoint method quantifies a linearized relationship between emissions and PM2.5 exposure. This makes it well suited for computing the impact of marginal emissions changes of a particular type at a particular location or time. Although there may be non-linearities that are not captured by this approach, the atmospheric processes relevant to PM2.5 ––wet deposition and advection––are accurately represented as linear operations in the GEOS-Chem model. As such, the error due to atmospheric non-linearities is expected to be small. Atmospheric chemistry-transport models depend on emissions inventories to compute air quality impacts, and our estimate and attribution of population exposure is subject to the specific choice of emissions inventory. While various global fire emissions inventories have been developed, differences across inventories such as satellite image interpretation and adjustment for small fires can result in large regional differences in emissions estimates. We select six global emissions inventories, including five commonly used and one newly-developed for Indian agricultural residue burning, and make an inter-comparison by calculating PM2.5 exposure due to post-monsoon crop residue burning using each of the emissions inventories . We find that exposure estimates vary by up to a factor of seven due to uncertainty in emissions inventories . However, we find that this does not significantly affect our conclusions, which are focused on the relative reduction in harm which could be achieved through targeted interventions. Detailed comparison and discussion can be found in Supplementary Discussion.Do political leaders benefit from anti-poverty programs? There is a large and growing literature on the targeting of government expenditures, but less is known about the political effects of distributive programs, particularly large-scale poverty-reduction efforts that target substantial portions of the population. Across Africa, governments have increasingly adopted agriculture subsidy programs in recent years to combat rural poverty and food insecurity, embracing a strategy common in the 1960s and 1970s before structural adjustments programs reduced such market interventions in the 1990s . While the political appeal of agricultural subsidies in countries where the majority of the population is engaged in smallholder agriculture are obvious, there has been little quantitative research on their effects.In part this lacuna stems from the difficulty of quantifying the political effects of subsidy programs. Because subsidy programs may be targeted, often for political reasons , researchers must confront the thorny challenge of teasing apart selection effects from potential treatment effects. This paper contributes to studies of distributive politics by examining Malawi’s Agricultural Input Subsidy Programme , one of the largest and most expensive programs implemented to date. To examine whether the incumbent party, the Democratic Progressive Party headed by president Bingu wa Mutharika, benefited from Malawi’s subsidy program, we draw on panel data from a survey of 1,846 respondents interviewed in 2008 and again in 2010. We proceed in two steps. We first investigate whether the program was targeted at the local level. We propose that because of informational constraints and the weakness of party institutions at the grassroots level, the subsidy is likely to be untargeted with respect to party support and the main determinant of party allegiances – ethnicity – at the village level. Consistent with these expectations, we find no evidence of partisan or ethnic targeting in our sample area. This finding is interesting in its own right, especially given dominant theories of distributive politics that argue whether politicians benefit by targeting material transfers to core supporters or swing voters . The second step in the analysis is to test for potential effects on preferences. While we find no evidence of political targeting at the individual level, we do not claim that distribution was random. Accordingly, testing for political effects requires accounting for potential confounding factors. We employ two alternative methods for addressing possible omitted variables.