Outdoor air pollution in India is not only an urban problem

Significance

The pollution in major cities in India is immense, and they are in the news often. In general, air pollution is viewed as an urban problem. Here, we show that outdoor air pollution due to particulate matter is just as harmful in nonurban (rural) areas as in urban areas in India based on satellite data and calculated health impacts. This finding has a far-reaching impact on how pollution is viewed in India and other developing countries. It is also very timely given that India, and countries like India, are turning attention to addressing air pollution, and they need to be aware of this vast issue.

Abstract

Urban outdoor air pollution in the developing world, mostly due to particulate matter with diameters smaller than 2.5 µm (PM_2.5), has been highlighted in recent years. It leads to millions of premature deaths. Outdoor air pollution has also been viewed mostly as an urban problem. We use satellite-derived demarcations to parse India’s population into urban and nonurban regions, which agrees with the census data. We also use the satellite-derived surface PM_2.5 levels to calculate the health impacts in the urban and nonurban regions. We show that outdoor air pollution is just as severe in nonurban regions as in the urban regions of India, with implications to monitoring, regulations, health, and policy.

Air pollution is one of the leading causes of ill health. Indeed, it has been estimated to be the fourth leading cause of premature mortality in the world (1). Air pollution has been mostly associated with urban pollution (2). It originated in big cities such as London and Los Angeles in the previous century (3, 4). Routinely, we see lists of the most polluted cities globally, again emphasizing the connection between air pollution and urban areas, with PM_2.5 being the dominant pollutant in the last few decades (5). Beijing, Delhi, Mexico City, and Jakarta are associated with high outdoor pollution in the 21st century. The PM_2.5 pollution levels in these regions are, or were, so overwhelmingly large that it attracted a great deal of attention. The rapid industrialization of Asia has led to a vast number of urban areas with PM pollution that is well beyond the World Health Organization’s (WHO) guideline for healthy air (annual mean of 10 µg/m³) (6).

Is Air Pollution in the 21st Century Really Only an Urban Problem?

There have been many health impact studies focusing on PM_2.5 effects in India (7–11). Karambelas et al. (12) have examined the question of urban versus nonurban PM_2.5 impacts over northern India via modeling calculations for 4 mo. They showed that the impacts of air pollution on nonurban residents were comparable to those on urbanites, thereby making a key point about the role of air pollution on people who are not usually considered to be impacted. They focused on the highly polluted Indo-Gangetic Plain (IGP) using 4-mo annual averages based on the results of a model that uses average emissions over large areas and times. In this article, we examine this question using observed (satellite-derived) values. We show that particulate (PM_2.5) outdoor air pollution over the entire country of India is not merely urban, but it also affects the nonurban areas just as much as the urban areas, although there are quantitative differences between regions. For each region, the urban and nonurban impacts are roughly the same, and this is true for the entire country.

Results and Discussion

Demarcation between Urban and Nonurban Areas.

The demarcation of urban from nonurban regions itself is different over India compared to the western world. In the western world, the large cities and their boundaries are reasonably well defined; this is true even with the urban sprawls that are almost contiguous. The population density in nonurban regions of the developed and new world are very low compared to those in the urban regions. Also, the populations in the urban areas far exceed those in rural areas in the western world. In contrast, nonurban areas in India have large population densities, often with dense clusters of people surrounded by fields.

To parse out populations into urban and nonurban regions, one could look at the census data. Here, we have used the nightlight satellite radiances, which will likely mirror large areas of activities, transportation, and industry. The Suomi National Polar-orbiting Partnership Visible Infrared Imaging Radiometer Suite (VIIRS) provides average radiance composite images using nighttime light data that were used to define the urban land areas (13–15). The VIIRS nighttime light data were available at 15 arc-seconds (∼0.463 km at the equator), and we regridded the data to 2.5 arc-minutes (∼4.63 km at the equator). Fig. 1A shows the nighttime light data over India. Fig. 1B shows the population density at the same 2.5 arc-minute resolution across India from the National Aeronautics and Space Administration Socioeconomic Data and Application Center for 2011 Census data (16). The most densely populated regions are the IGP and the southwest coast of India. The major urban areas are clearly visible in the nightlight data. Associated with these regions (see SI Appendix, Fig. S1 for the definition of the regions), the population density is clearly high but not all high population areas have significant night lights. This is particularly so in the eastern IGP (∼75–85°E longitudes), where there is a large population density but not large cities. What is starkly evident in comparing Fig. 1 A and B is that there is a vast population not associated with urban areas. This conclusion is not new, and it is well understood and documented (Census of India, 2011). Essentially, India is a country with a vast nonurban population.

Fig. 1.

Annual VIIRS nighttime light data for 2015 (A) and population density (based on Census 2011) (B).

We define the grids with nighttime light exceeding 1.15 nW/cm²/sr as urban land. Based on this criterion, we estimate the urban and nonurban population to be 400 million (31%) and 890 million (69%), respectively. These numbers are consistent with the separation between urban (31.16%) and nonurban (68.84%) populations given by the Census of India 2011. SI Appendix, Fig. S2 shows good correspondence between the nightlight and the census data for the classification of urban and nonurban regions. It also shows that a large fraction of India’s population lives outside of the urban areas.

Surface PM_2.5 Across Urban and Nonurban Regions.

We calculated the annually averaged aerosol optical depth (AOD) from three satellite instruments (Methods) that were converted to surface PM_2.5 abundances using PM_2.5:AOD ratios from the GEOS-Chem chemical transport model following van Donkelaar et al. (17). (The obtained AOD data separated into urban and nonurban regions is shown in SI Appendix, Fig. S3.) We acknowledge that satellite-derived data has an element of modeling since the measured AOD is converted to surface PM_2.5 abundances using a model. We compared the satellite-derived daily and annual PM_2.5 with the surface PM_2.5 measured by Central Pollution Control Board (18) at 20 monitoring sites (mostly in the urban area) (SI Appendix, Fig. S4). The slope and correlation coefficient for the annually averaged and daily data comparisons are, respectively, 0.845 and 0.757 and 0.772 and 0.632; the mean bias for both are less than ±30% of the mean. Such comparisons are usually better at higher spatial resolutions. We have used footprints of 4.63 km × 4.63 km for surface PM_2.5. Hammer et al. (19) have produced data at 0.01° × 0.01° (∼1 × 1 km) resolution. The annual mean PM_2.5 values from this work, GEOS-Chem at 0.25° × 0.3125° resolution, and Hammer et al. (19) are shown in SI Appendix, Fig. S5.

The satellite-derived annual mean surface PM_2.5 is shown in Fig. 2A. Comparison of Fig. 2A with Fig. 1 A and B clearly shows the surface PM_2.5 pollution levels are not always significantly different between urban and nonurban areas within the six regions in India. Indeed, Fig. 2B shows the population-weighted PM_2.5 levels separated between urban and nonurban areas for six regions of India, which have different sources, population densities, and meteorologies (20) and shown in SI Appendix, Fig. S1. The IGP stands out as the area of highest population-weighted PM_2.5 (>100 µg/m³ throughout) and has the largest population density. In this region, the pollution is almost evenly spread across urban and nonurban regions. Most of India outside of the IGP has PM_2.5 levels between 55 and 90 µg/m³ (Fig. 2A). These pollution levels (on average) are only about a factor of two different from the highest pollution areas that stand out (red-yellow color). Yet, even in the other five regions, there is little difference between population-weighted PM_2.5 levels between urban and nonurban regions. The annual mean PM_2.5 levels are much larger than India’s national ambient air quality standard (40 µg/m³, annual mean) in all of the six regions. The mountainous areas in the very northern parts and the lower part of the peninsula have noticeably smaller pollution levels (<40 µg/m³) than the rest of India. Available in situ PM_2.5 observations in nonurban areas for the period 2010–2016 are shown in SI Appendix, Table T1). As evident in Fig. 2A and SI Appendix, Table T1, high (>90 μg/m³) PM_2.5 concentrations are seen in nonurban areas in the IGP. This is to be compared with southern India, where the mean PM_2.5 concentrations in nonurban locations are in the range of 30–50 μg/m³.

Fig. 2.

Satellite-derived annual mean surface PM_2.5 (A) and variation in population-weighted PM_2.5 over urban and nonurban locations in six regions (SI Appendix, Fig. S1) of India for 2015 (B). The black dashed and solid horizontal lines correspond to World Health Organization’s (10 µg/m³) and India’s (40 µg/m³) annual mean PM_2.5 air quality standards, respectively.

Another way to look at this PM_2.5 distribution is to plot the population exposed to specific levels of PM_2.5 pollution, as done by Apte et al. (21). Fig. 3A shows the distribution of population in urban and nonurban areas over India, and Fig. 3 B and C shows the normalized population in urban and nonurban areas in the six regions over India with respect to PM_2.5 concentrations. Some key findings are evident: 1) Most people in India (∼84% of the population) are well above the limit of India’s standard (40 μg/m³), and almost the entire country is exposed to levels higher than the WHO standard (10 μg/m³); 2) there is a long tail of very high concentrations (>160 μg/m³) in the urban regions in the IGP and parts of nonurban areas in eastern and western India (Fig. 3B), and these are the levels that are often publicized; 3) since the health impacts vary logarithmically with the PM_2.5 pollution levels (for the levels of PM_2.5 over India), the high concentrations do not proportionally enhance the health burden; 4) there are clean parts of the country where there could be larger relative differences between urban and nonurban regions, but they are relatively small; and 5) the large hump seen in Fig. 3A at high PM_2.5 concentrations is primarily due to the IGP. This was the area analyzed by Karambelas et al. (12) to show that the urban and nonurban areas have similar PM_2.5 loading and health impacts. Our analyses for the other regions show that the same trend is true although the PM_2.5 levels are lower than in the IGP.

Fig. 3.

Distribution of population in urban and nonurban areas (A) and normalized population in urban and nonurban areas in the six regions over India with respect to surface PM_2.5 concentrations (B and C). The solid and dashed lines, respectively, represent urban and nonurban regions. The black dashed and solid vertical lines correspond to WHO’s (10 µg/m³) and India’s (40 µg/m³) annual mean PM_2.5 air quality standards, respectively.

Health Impacts of the Observed Surface PM_2.5 Abundances.

We have calculated the premature mortality attributable to PM_2.5 in the urban and nonurban areas for six causes of death (ischemic heart disease [IHD], stroke, lower respiratory infections [LRI], chronic obstructive pulmonary disease [COPD], lung cancer [LC], and diabetes mellitus type 2). The annual premature mortality attributable to PM_2.5 for India (urban and nonurban) is 1.05 (5–95 percentiles: 0.687–1.37) million and is comparable to those reported by other studies (1, 10, 22). These premature mortalities attributable to the PM_2.5 parsed between urban and nonurban areas are shown in Fig. 4A. The larger mortality in the nonurban region is due to its larger population. Fig. 4B shows the fraction of the premature mortalities that are attributable to PM_2.5 that occur in urban versus nonurban areas. This distribution is very similar to the population distribution of 31% urban and 69% nonurban. Fig. 4C shows premature mortality per million people. Clearly, there is little difference between urban and nonurban regions. The PM_2.5 levels are comparably high in both urban and nonurban regions. Furthermore, the relative risk changes roughly logarithmically with PM_2.5 at the levels seen over India. Therefore, the relative risks for the population are almost the same in the two areas. PM_2.5 air pollution, the major pollution problem in India, is just as much a contributor to the ill health of the nonurban population as the urban population. These mortality numbers will not change significantly if we include the impact of ozone pollution since they are much less than those due to PM pollution (12).

Fig. 4.

(A) Total annual premature deaths attributed to PM_2.5 over urban and nonurban regions in India. The lower and upper limit of the error bars correspond to 5th and 95th percentile; fraction of mortality attributed to urban and nonurban regions in India (B); and total annual premature deaths attributed to PM_2.5 per million people over the urban and nonurban regions in India (C).

Our calculated premature deaths do have large uncertainties. We include only the uncertainties due to the response functions and not those due to PM_2.5, baseline mortality rates, model resolution, as these uncertainties dominate in most regions (23). There are significant uncertainties in the PM_2.5 levels that are not included in the current mortality uncertainties. However, these uncertainties should not affect our conclusion that the nonurban population is at almost the same risk as to the urban population. This is because we are calculating the fraction of PM_2.5 between urban to nonurban areas using the same satellite and model data. Simply put, the differences in PM_2.5 levels breathed by urban and nonurban populations are not sufficiently dissimilar to make a significant difference.

The Health Effects Institute report (24) on India’s PM_2.5 air pollution clearly shows the comparable levels of the PM between urban and rural areas. A recent study has shown that the PM_2.5 levels are very high surrounding the Delhi region (25). Our conclusions, based on the satellite observations, are consistent with these findings. Our study extends the findings of Karambelas et al. (12) that urban and nonurban impacts of PM_2.5 are the same in the highly polluted IGP. Our study further emphasizes that even in less polluted (relative to the IGP) regions of India, PM_2.5 affects the urban and nonurban populations similarly. There is 16% of India’s population that is not affected by this pollution (below 40 µg/m³), and that is mostly confined to the very northwestern parts of India, the Western Ghats, and a few regions within India (Fig. 2A). The fraction that is below the WHO standards is very small (<0.001%).

The origin of this PM_2.5 is not addressed here but would be of great interest for mitigation efforts. Based on the current emission inventories in the model (10, 20), the total anthropogenic PM and precursor emissions in the urban and nonurban regions are 25.1 Tg/yr and 53.4 Tg/yr, respectively (SI Appendix, Fig. S6). The residential energy use is a significant emissions source, primarily due to household cooking with solid fuels in the nonurban areas. Other sources of air pollution in nonurban areas include stubble burning, brick kilns, coal-fired factories, agricultural processing, power generation, cement factories, and cottage industries. The urban areas are affected by the transport of pollution from the nonurban regions and vice versa. The nonurban emissions in the IGP is a factor of 2–14 higher compared to other regions in India. We have shown previously that eastern India is greatly affected by pollution transport from the IGP (10). It appears that nonurban regions create a significant amount of their own pollution. The contribution of urban vs. nonurban emissions to premature deaths in nonurban areas needs to be investigated in future work.

Our work, along with those of Karambelas et al. (12), highlights the critical need for monitoring PM_2.5 levels in nonurban areas in India. Enhancing measurements in these regions, which are virtually nonexistent, could help better assess the risks. Further, it highlights the need for significant reductions in PM_2.5 levels across India. A pan-India approach to reduction would be beneficial. Of course, the outdoor air quality degradation in the nonurban regions is related to the larger indoor PM_2.5 levels in those regions. Currently, the nonurban population has a lesser ability to reduce their risks because of economic reasons and, thereby, raising equity issues. The nonurban regions have the further confounding influence of indoor air quality due to the use of solid fuels for cooking in inefficient cookstoves. The inclusion of this impact would only tend to enhance the air pollution impacts in nonurban areas relative to urban areas. If one were to add indoor pollution and synergistic effects of nutritional deficiencies in the rural areas (26), not the topics of this paper, there could be even larger impacts on the nonurban populations, especially to women and children. Clearly, the nonurban regions in India are being affected by air pollution as much as the more visible urban regions.

Methods

Definition of Urban and Nonurban Population in India.

Satellite data for calculating urban and nonurban populations.

The Earth Observation Group in the National Oceanic and Atmospheric Administration’s National Centers for Environmental Information provide the nighttime light data from the VIIRS Day/Night Band (DNB). The VIIRS can detect nighttime lights within the radiation range from 3 × 10⁻⁹ W/cm²/sr to 0.02 W/cm²/sr, with an extensive spatial coverage from 75°N latitude to 65°S and a swath of 3,000 km (13). We used the annual VIIRS nighttime light data, which contains cloud-free average radiance values (in nW/cm²/sr) that have removed outliers to filter out fires, other ephemeral lights (e.g., lights on fishing boats and fires), and background (nonlights). The products are produced in 15 arc-seconds that were regridded to 2.5 arc-minutes to match the population grid. The grids with nightlight satellite radiance exceeding 1.15 nW/cm²/sr were defined as urban land.

Census data.

The Indian census of 2011 deemed urban areas as 1) all places with a municipality, corporation, cantonment board (civic administration body in India), or notified town area committee; and 2) all other places which satisfied the following criteria of a minimum population of 5,000 with at least 75% of the male working population engaged in nonagricultural pursuits and a population density of at least 400 persons/km².

Comparison of our nighttime light data determined separation between urban and nonurban land compares very well with census data and the established consensus that a large fraction of India’s population lives outside of the urban areas (SI Appendix, Fig. S2). It should be noted that some of the most densely populated regions in India are the nonurban areas in the eastern IGP (Fig. 1B). Many of these nonurban regions have a population density comparable to the US suburbs or many cities.

PM_2.5 Surface Abundance.

India is data-sparse for surface observations. Even the few stations that are available are located in urban regions and provide long-term averaged data. Therefore, we cannot rely on surface data for calculating PM_2.5 across India, especially in nonurban regions where such data are virtually absent. Consequently, we have used satellite-retrieved AOD and used that information with the PM_2.5:AOD ratios from a chemical transport model to derive surface PM_2.5.

Satellite data.

We used the AOD from three different satellite instruments: 1) the Moderate Resolution Imaging Spectroradiometer (MODIS) instrument aboard Terra satellite with an equator crossing time at ∼10:30 AM local time (LT); 2) the MODIS instrument aboard the Aqua satellite with an equator crossing time at ∼1:30 PM LT; and 3) the Multiangle Imaging SpectroRadiometer (MISR) instrument aboard the Terra satellite. MODIS measures visible and infrared radiation from 36 spectral bands (0.4–14.5 µm with spatial resolutions of 250 m, 500 m, and 1 km) at nadir. MISR features nine push broom cameras observing the Earth from different angles in four spectral bands (0.443–0.865 m) with resolutions of 275 m to 1.1 km. Both Terra and Aqua MODIS instruments view the entire Earth in 1–2 d, and it takes 9 d for MISR to view the entire globe.

For MODIS, we used the level 2 (collection 6.1) at 3-km spatial resolution “Optical depth land and ocean” product (merged Dark Target and Deep Blue) with the highest quality at 550 nm. For MISR, we used level 2 (version 23) AOD with a spatial resolution of 4.4 km. Although the optical depths derived from these three instruments differ, our calculated differences between urban and nonurban regions are robust and essentially the same (SI Appendix, Fig. S3). Both MODIS and MISR show similar seasonal variations except for the monsoon months. The lower AOD during the monsoon season seen by MISR has been attributed to the sampling bias due to cloudy conditions during the monsoon (June–September) months.

GEOS-Chem simulation.

We used the GEOS-Chem global three-dimensional model (version 12.0.3) driven by assimilated meteorological data (acmg.seas.harvard.edu/geos). The model has fully coupled tropospheric NO_x-O_x-hydrocarbon-aerosol chemistry. The global GEOS-Chem simulations were made at 2° × 2.5° resolution using Goddard Earth Observing System-Forward Processing (GEOS-FP) meteorology to generate temporally varying boundary conditions for all species for higher-resolution nested simulations at 0.25° × 0.3125° over India. The details of the emissions used are given elsewhere (20). The nested simulations were carried out to determine AOD and surface PM_2.5. The simulated AOD was evaluated against observations, as described in David et al. (20). The simulated PM_2.5 was evaluated against observations from Central Pollution Control Board. The correlation coefficient and bias were 0.793 and 17.8 ± 50.8 μg/m³, respectively.

Estimation of Surface PM_2.5.

The satellite-derived surface PM_2.5 was derived using satellite-observed AOD using the following relationship (27):

$$\mathrm{PM}2\mathrm{.}{5}_{\mathrm{surface}}\mathrm{=\eta \times AO}{\mathrm{D}}_{\mathrm{satellite}},$$

where η is the ratio of daily simulated surface PM_2.5 to AOD. The satellite-observed AOD within a grid (at 0.25° × 0.3125° resolution) was subjected to the same η value. The daily AOD from each instrument was converted to PM_2.5 separately. After that, the satellite-derived PM_2.5 values from each instrument that lies within a 2.5 arc-minute grid were averaged to get the daily PM_2.5. We averaged the daily PM_2.5 to calculate the annual mean PM_2.5.

Calculation of Premature Mortality.

To estimate the ambient PM_2.5 exposure in India, the estimated daily surface PM_2.5 from MODIS (Terra/Aqua) and MISR were averaged and regridded to the population grid (at 2.5 arc-minute resolution). Since the density of MODIS data are larger than that of MISR, the calculated premature mortality is weighted toward the MODIS data. However, the ratio of the urban to nonurban surface PM_2.5 level does not change with the choice of the MODIS vs. MISR data. The premature mortality due to ambient PM pollution was calculated using the relative risks at different PM_2.5 concentrations for six causes of death from Global Burden of Disease 2017 (28). The theoretical minimum risk exposure level for ambient PM was defined as a population-weighted annual mean PM_2.5 between 2.4 and 5.9 μg/m³. We have used the state-wise baseline mortality rates (22, 29). The relative risks and baseline mortality rates for LRI, COPD, LC, and diabetes are for all ages, and for IHD and stroke it is age-dependent (above 25 y). The number of mortalities within a grid cell (∼4.63 km) for each cause of death was calculated. After that, the total mortality was calculated for the urban and nonurban regions.

Data Availability.

The MODIS aerosol products (Collection 6.1 at 3 km) MOD04_3K from Terra and MYD04_3K from Aqua are available from the Atmosphere Archive and Distribution System Distributed Active Archive Center, https://ladsweb.modaps.eosdis.nasa.gov/. The MISR aerosol product (version 23 at 4.4 km) MIL2ASAE.003 is available from the ftp site ftp://l5ftl01.larc.nasa.gov/MISR/. The satellite-derived PM_2.5 data used in this study are available at https://mountainscholar.org/handle/10217/210896 (30). The gridded population data for India is available from the NASA Socioeconomic Data and Application Center (https://sedac.ciesin.columbia.edu/), which is based on the 2011 census. The average radiance composite images using nighttime light data from the VIIRS DNB is available from https://payneinstitute.mines.edu/eog-2/viirs/.