Abstract
Uncontrolled combustion of domestic waste has been observed in many countries, creating concerns for air quality; however, the health implications have not yet been quantified. We incorporate the Wiedinmyer et al (2014 Environ. Sci. Technol. 48 9523–30) emissions inventory into the global chemical-transport model, GEOS-Chem, and provide a first estimate of premature adult mortalities from chronic exposure to ambient PM2.5 from uncontrolled combustion of domestic waste. Using the concentration-response functions (CRFs) of Burnett et al (2014 Environ. Health Perspect. 122 397–403), we estimate that waste-combustion emissions result in 270 000 (5th–95th: 213 000–328 000) premature adult mortalities per year. The confidence interval results only from uncertainty in the CRFs and assumes equal toxicity of waste-combustion PM2.5 to all other PM2.5 sources. We acknowledge that this result is likely sensitive to choice of chemical-transport model, CRFs, and emission inventories. Our central estimate equates to 9% of adult mortalities from exposure to ambient PM2.5 reported in the Global Burden of Disease Study 2010. Exposure to PM2.5 from waste combustion increases the risk of premature mortality by more than 0.5% for greater than 50% of the population. We consider sensitivity simulations to uncertainty in waste-combustion emission mass, the removal of waste-combustion emissions, and model resolution. A factor-of-2 uncertainty in waste-combustion PM2.5 leads to central estimates ranging from 138 000 to 518 000 mortalities per year for factors-of-2 reductions and increases, respectively. Complete removal of waste combustion would only avoid 191 000 (5th–95th: 151 000–224 000) mortalities per year (smaller than the total contributed premature mortalities due to nonlinear CRFs). Decreasing model resolution from 2° × 2.5° to 4° × 5° results in 16% fewer mortalities attributed to waste-combustion PM2.5, and over Asia, decreasing resolution from 0.5° × 0.666° to 2° × 2.5° results in 21% fewer mortalities attributed to waste-combustion PM2.5. Owing to coarse model resolution, our global estimates of premature mortality from waste-combustion PM2.5 are likely a lower bound.
Export citation and abstract BibTeX RIS
Original content from this work may be used under the terms of the Creative Commons Attribution 3.0 licence. Any further distribution of this work must maintain attribution to the author(s) and the title of the work, journal citation and DOI.
1. Introduction
Open, uncontrolled combustion of domestic waste (i.e. trash burning) occurs on a global scale, emitting particulate matter (PM) and toxic gaseous and particulate compounds (Christian et al 2010, Wiedinmyer et al 2014). Domestic waste can include food and agricultural products, containers and packaging, and other sources of residential trash. Combustion occurs both in homes and at community waste sites (i.e. dumps), and both in developing and developed countries, though waste combustion is more widespread in developing countries (Wiedinmyer et al 2014). While combustion of waste is ubiquitous in rural areas, there are also significant emissions in urban areas, potentially leading to pollution exposure of dense populations (Wiedinmyer et al 2014).
Until recently, inventories of pollutant emissions from combustion processes either did not include emissions from uncontrolled domestic-waste combustion (e.g. Bond et al 2007, Janssens-Maenhout et al 2010), or the domestic-waste-combustion emissions were assumed to be low relative to other sources in all locations (Bond et al 2004). However, Wiedinmyer et al (2014) recently created an emissions inventory focused specifically on the emissions from the uncontrolled open combustion of domestic-waste products. The total estimated PM2.5 emissions were 10 Tg yr−1, globally, an order of magnitude higher than a previous estimate in Bond et al (2004), and similar to the emissions from biofuel combustion in Bond et al (2004). The more-recent estimates were higher due to newer assumptions in waste production, fraction burned, and emission factors. The largest driver of the difference in emissions was the estimated amount of annual waste burned: Bond et al (2004) estimated a total of 33 Tg, while Wiedinmyer et al (2014) predict 970 Tg. This discrepancy is in part due to the assumption in Bond et al (2004) that no waste combustion occurs in rural developing countries. Thus, ambient air quality particularly in developing countries may be substantially impacted by domestic-waste combustion.
PM air pollution is a major contributor to the Global Burden Of Disease (GBD), and the most important environmental contributor to morbidity and mortality in the world (Lim et al 2012). Thus, exposure to ambient pollution from uncontrolled domestic-waste combustion may have serious health implications, yet these health effects have not been quantified. A number of past studies have linked exposure to ambient PM with aerodynamic diameters less than 2.5 μm (PM2.5) to increased risk of mortality for adults from ischaemic heart disease (IHD), cerebrovascular disease (CeVD, or stroke), chronic obstructive pulmonary disease (COPD), and lung cancer (LC) (Pope et al 2002, Laden et al 2006, Krewski et al 2009, Burnett et al 2014). Using a combination of satellite and ground observations, and chemical-transport models to estimate PM2.5 exposure from all sources, the GBD Study 2010 estimated 3.1 million premature mortalities, globally, in 2010 (Lim et al 2012).
For global/regional scale studies, chemical-transport models have been used to estimate surface-level ambient PM2.5 exposure for use in health-impact assessments (e.g. Anenberg et al 2010, Lelieveld et al 2013, 2015, Butt et al 2016). Despite uncertainties in emissions and atmospheric processes, models have an advantage over ground-based measurements in that they provide exposure estimates in regions where few surface measurements are available. In addition, models can provide source-specific PM2.5 estimates, thereby allowing for policy-relevant mortality estimates from specific emission sectors (e.g. Corbett et al 2007, Butt et al 2016). Recently, Lelieveld et al (2015) used the ECHAM model to estimate source-specific mortalities, finding dominant sources in Asia are from residential solid-fuel combustion and biomass burning; however, this study does not include emissions from open waste combustion.
One potential drawback of using chemical-transport models on a global scale is that computational efficiency often limits model spatial resolution. Lower-resolution global models average PM2.5 concentrations over larger areas, limiting the ability to predict very-polluted regions and sharp gradients outside these regions. As emissions are often co-located with population, this may lead to a dependence of mortality estimates on model resolution. Punger and West (2013) estimated ambient PM2.5 mortalities in the United States using different model resolutions, and find substantially (∼30%) lower estimates when scaling PM2.5 to spatial scales representative of global models (>250 km) relative to 12 km. One method to correct for sub-grid scale gradients involves using satellite aerosol optical depth (AOD) at resolutions higher than the model (e.g. ∼10 km versus ∼200 km) to interpolate PM2.5 exposure fields at the resolution of the satellite AOD. However, this method relies on the ratio of modeled AOD to surface PM2.5 (van Donkelaar et al 2015). Ford and Heald (2016) estimated 20% uncertainty in mortality due to methodology in estimating surface PM2.5 concentrations from satellite AOD. Additionally, while this method was used in several studies to estimate all-source PM2.5 mortality (Lim et al 2012, Evans et al 2013), satellite observations cannot be used to isolate individual sources of PM2.5; one would need to assume that spatial distribution of AOD from each individual source follows the spatial distribution of the total AOD.
In this study, we provide the first estimates of mortality due to chronic exposure to PM2.5 from domestic-waste combustion. We include the recent Wiedinmyer et al (2014) waste-combustion inventory into the global/regional chemical-transport model, GEOS-Chem. In section 2, we discuss the model setup and methodology for estimating mortality. In section 3.1, we present model results for waste-combustion PM2.5 and in section 3.2 we present the global and country-level attributed mortality. In section 3.3, we estimate the deaths averted due to complete removal of emissions. In section 3.4, we test the sensitivity of mortality rates to uncertainties in emission mass and model resolution. We share our conclusions in section 4. In a separate study, Kodros et al (2016), we estimate the aerosol radiative impacts from domestic-waste combustion.
2. Methods
2.1. Chemical-transport model overview
To estimate PM2.5 concentrations, we use the Goddard Earth Observing System chemical-transport model (GEOS-Chem) version 10.01. GEOS-Chem is driven by GEOS-5 assimilated meteorology fields (http://gmao.gsfc.nasa.gov), and includes PM tracers for black carbon, organic aerosol, dust, sea salt, sulfate, nitrate, and ammonium in addition to 52 gas-phase species. Globally, we use the EDGAR emissions inventory with regional overwrites as described in the supplemental material. Waste-combustion emissions are not included in these base GEOS-Chem emissions inventories. We incorporate to GEOS-Chem the waste-combustion inventory of Wiedinmyer et al (2014), which estimates the mass of waste burned in urban and rural areas for developing countries, and rural areas for developed countries. The inventory primarily uses emission factors compiled in Akagi et al (2011). Estimated waste-combustion emissions include 0.6 Tg yr−1 of black carbon and 5.1 Tg yr−1 of primary organic carbon, as well as more-minor contributions from gas-phase species including sulphur dioxide, ammonia, and nitrous oxides, which contribute to additional PM2.5 (<10% of the total addition) through chemical reactions. Evaluation of the modeled aerosol burden with and without waste-combustion emissions is included in the supplemental material.
We perform GEOS-Chem simulations for year 2010 with a pair of simulations: one with emissions from all sources (including waste burning; 'WASTE_ON'), and another, otherwise-identical simulation but without waste-combustion emissions ('WASTE_OFF'). Comparison of these pairs of simulations allows us to isolate the impacts of waste combustion on PM2.5 concentrations and mortality. In order to test the sensitivity of mortality rates to uncertainties in waste-combustion emission mass, we assume waste-combustion PM2.5 scales linearly with PM emission mass and simply double ('HIGHMASS') and half ('LOWMASS') modeled waste-combustion PM2.5 concentrations. The factor-of-2 uncertainty in emission mass is similar to uncertainties in waste-combustion emission factors reported in Akagi et al (2011). In order to test the sensitivity of mortality rates to model resolution we repeat the WASTE_ON and WASTE_OFF simulations at three resolutions: 2° × 2.5° (∼200 km) and 4° × 5° (∼400 km) resolution globally, and a 0.5° × 0.666° (∼50 km) resolution over Asia.
2.2. Calculation of premature mortality
Our methodology to estimate mortalities from waste-combustion PM2.5 is based on the methods used in the GBD 2010 to estimate ambient PM2.5 mortalities. We discuss a number of uncertainties and limitations of this method in section 4. We use year 2010 gridded population (from the NASA Socioeconomic Data and Application Center (SEDAC, http://sedac.ciesin.columbia.edu/) and baseline mortality rates compiled for the GBD 2013 for IHD, CeVD, COPD, and LC for adults (ages greater than 25) (Naghavi et al 2015). Baseline mortalities are reported at the country level, and we aggregate them to the model resolution assuming no sub-national gradients. Data on baseline mortality variability at a sub-national scale were not available.
The fraction of all premature mortalities due only to PM2.5 is based on concentration-response functions (CRFs) that relate exposure to ambient PM2.5 to increased risk of premature mortality from specific diseases. We calculate the relative risk (RR) from all PM2.5 sources (RRtotal) using the integrated CRFs of Burnett et al (2014). We choose the Burnett et al (2014) study as it uses smoking and household air pollution to constrain the risk at high PM2.5 concentrations (such as ambient PM2.5 found in India and China). The Burnett et al (2014) study makes several assumptions relevant to this study: the toxicity of PM2.5 is not dependent on composition, increased mortality risk is a result of long-term exposure, CRFs apply globally, there exists a minimum PM2.5 threshold (modeled as a uniform distribution ranging from 5.8 to 8.8 μg m−3) below which no further negative health impacts are assumed to occur, and the relationship between relative risk of mortality and PM2.5 exposure is nonlinear. We note that CRFs are an active area of research, and our results are likely sensitive to our choice of CRF. Burnett et al (2014) fits coefficients using a Monte Carlo approach, reporting 1000 sets of coefficients for each cause (and each age group for CeVD and IHD). We use these sets of coefficients to determine median, 5th, and 95th percentiles of mortality. The attributable fraction of premature mortalities by PM2.5 is then calculated from the RR by: (RRtotal − 1)/RRtotal.
We estimate total premature adult mortality from PM2.5 as the product of population, cause and age specific baseline mortality rates, and attributable fraction from all PM2.5 sources in each model gridcell. We estimate 2.7 million premature adult mortalities from ambient PM2.5 in the 2° × 2.5°-resolution simulation (figure S1). This number is lower than estimates of 3.1 million from Lelieveld et al (2015) and Lim et al (2012); however, it is within the reported uncertainty ranges (2.5 to 3.7 million for Lim et al 2012 and 1.5 to 4.6 million for Lelieveld et al 2015). The coarser resolution of our global model likely contributes to the lower estimate.
To attribute the number of mortalities owing to waste-combustion emissions, we scale total mortalities from all PM2.5 by the fraction of PM2.5 from waste combustion (attribution method). This is represented by equation (1):
where Mwaste is the premature mortalities due to waste combustion, PM2.5, WASTE_ON and PM2.5,WASTE_OFF are the modeled PM2.5 concentrations in the WASTE_ON and WASTE_OFF simulations, respectively, and Mall (WASTE_ON) is the premature mortalities due to all sources in the WASTE_ON simulation. Additionally, we estimate the number of mortalities averted due to removal of waste-combustion emissions by subtracting the total mortalities in the WASTE_OFF simulation from the total mortalities in the WASTE_ON simulation (subtraction method). This is represented by equation (2):
where Mall (WASTE_OFF) is the premature mortalities from all sources in the WASTE_OFF simulation. The subtraction method yields different results than the attribution method as the CRF has strong nonlinearities (discussed below).
3. Results
3.1. Model increases in PM2.5 due to waste-combustion emissions
Figure 1 shows the absolute increase (panel a) and percent increase (panel b) in PM2.5 from waste-combustion emissions at 2° × 2.5° model resolution. Waste-combustion emissions lead to over 10% increases in ambient PM2.5 concentrations in South and South-East Asia, Eastern Europe, Central America and Coastal South America. Notably, PM2.5 increases by more than 40% in Sri Lanka and central Mexico (e.g. Mexico City). The absolute increases in PM2.5 are greatest in eastern China and northern India, but there are anomalously large fractional increases in grid cells corresponding to large urban areas such as Johannesburg, Cairo, Moscow, and Mexico City.
Figure 1. The (a) absolute difference and (b) percent difference in modeled boundary layer PM2.5 from including emissions from combustion of domestic waste in our simulations at 2° × 2.5° resolution.
Download figure:
Standard image High-resolution image3.2. Global and country-level mortality rates due to waste-combustion PM2.5
The mortality rates per area associated with waste-combustion PM2.5 calculated through the attribution method are shown in figure 2(a), and global annual mortalities listed by cause of death are in table 1. We attribute waste-combustion PM2.5 to 270 000 (5th–95th:213 000–328 000) adult mortalities per year (calculated as the sum of IHD, CeVD, COPD, and LC). The majority of these mortalities are caused by IHD (120 000, 5th–95th:106 000–136 000) and CeVD (108 000, 5th–95th:86 000–130 000). Spatially, the highest concentration of deaths occurs in eastern China and northern India where mortality exceeds 900 deaths per 104 km2 (a 100 × 100 km box) per year. There is also a substantial number of mortalities in Eastern Europe. Significant increases in PM2.5 co-located with dense populations lead to greater than 300 deaths per 104 km2 in densely populated cities. Figure 2(c) shows the waste combustion mortality rate per 106 people (in each grid cell). A combination of large relative PM2.5 increases and high baseline mortality rates in Eastern Europe and Russia lead to a similar mortality rate as in Asia. Figure 2(c) also shows more widespread health impacts in Africa and the Middle East, where lower population densities limit the total number of annual mortalities.
Figure 2. Annual mortalities from chronic exposure to ambient PM2.5 from waste combustion per 104 km2 in the (a) 2° × 2.5° global domain and (b) in the 0.5° × 0.666° Asian domain as well as per 106 people in the (c) 2° × 2.5° global domain and (d) in the 0.5° × 0.666° Asian domain.
Download figure:
Standard image High-resolution imageTable 1. Annual mortalities (thousands) by cause and model resolution for the global and Asian domains due to exposure to PM2.5 from combustion of domestic waste.
| Resolution | IHD | CeVD | COPD | LC | Sum | |
|---|---|---|---|---|---|---|
| Global Domain | 4° × 5° | 103a (91–116) | 89 (71–107) | 23 (12–34) | 12 (6–34) | 227 (180–275) |
| 2° × 2.5° | 120 (106–136) | 108 (86–130) | 27 (14–40) | 15 (7–21) | 270 (213–328) | |
| 2° × 2.5°-HIGHMASS | 229 (203–258) | 208 (165–249) | 52 (28–77) | 30 (14–41) | 518 (180–275) | |
| 2° × 2.5°-LOWMASS | 62 (55–70) | 55 (44–66) | 13 (7–20) | 7 (3–11) | 138 (109–167) | |
| 2° × 2.5°-SUBTRACT | 70 (63–80) | 84 (67–98) | 23 (14–30) | 13 (8–17) | 191 (151–224) | |
| Asia Domain | 4° × 5° | 76 (68–84) | 76 (61–91) | 21 (11–31) | 11 (6–15) | 184 (146–222) |
| 2° × 2.5° | 90 (81–101) | 93 (74–110) | 24 (13–36) | 13 (6–19) | 221 (174–266) | |
| 0.5° × 0.666° | 113 (102–126) | 120 (96–143) | 29 (15–44) | 16 (8–23) | 279 (221–336) | |
Abbreviations: IHD, ischaemic heart disease; CeVD, cerebrovascular disease; COPD, chronic obstructive pulmonary disease; LC, lung cancer. aThe first number is the median estimate, and the 5th and 95th percentile are in parentheses.
At the country level, the estimated premature mortality rates due to waste-burning PM2.5 (normalized either by population or area) varies by several orders of magnitude between regions and economic level (see supplemental table 1). This mortality range is due to (1) differing amounts of waste generation (and hence emissions) and pollution transport leading to differing PM2.5 concentrations, (2) large variations in total number and density of the exposed population, and (3) differences in baseline mortality rates. In addition, non-waste-combustion PM2.5 plays an important role. In both WASTE_ON and WASTE_OFF, some countries are uniformly below the minimum concentration threshold set as the counterfactual in Burnett et al (2014) for PM2.5 health effects (though, we note that the actual minimum concentration threshold has not been clearly identified in the literature). By aggregating mortalities to the country level, we find the countries with the largest total mortalities due to waste combustion are China, India, Pakistan, and Russia. These four countries amount to 78% of the global mortality. The countries with the highest mortalities per capita are Montenegro, Bulgaria, Moldova, and Ukraine. Normalizing by mass of waste generated, we find that the countries with the highest mortalities per mass of waste generated are Nepal, Montenegro, Uruguay, and Bulgaria (supplemental table 1). Supplemental figure S4 shows box-and-whisker plots of country-level generated waste per capita per year (a) (from Wiedinmyer et al, 2014), mortality rates per capita (b), and mortality per mass of waste generated (c) split between 4 economic strata: high income, upper-middle income, lower-middle income, and low income. While on average high-income countries generate two times more waste, only 10 people die for every Tg of waste generated compared to 82 deaths for every Tg of waste generated in low- to upper-middle-income countries. This difference is driven by a higher fraction of waste generated being burned in low- to upper-middle income countries compared to high-income countries.
3.3. Deaths averted due to removal of waste-combustion emissions
We estimate 191 000 (5th–95th:51 000–224 000) premature mortalities per year may be saved by eliminating waste combustion (subtraction method; see explanation in section 2.2) (table 1 and figure S6). Thus, removing waste combustion would reduce premature mortality rates by a smaller number than the mortality rate attributed to waste combustion. This is due to the nonlinear CRFs where the mortality response to PM2.5 saturates with increasing PM2.5 concentrations. This saturation effect is strongest in the heavily polluted regions of India and China (which lie on the sub-linear portion of the CRF). This saturation effect is partly balanced out by a larger number of mortalities avoided in cleaner regions where waste-combustion emissions elevates PM2.5 concentrations from below to above minimum-PM2.5-threshold values.
3.4. Sensitivity of mortality rates to emission mass and model resolution
Wiedinmyer et al (2014) acknowledges large uncertainties in PM2.5 emission mass. To test the sensitivity of mortality rates to uncertainties in emission mass, we introduce a factor-of-2 uncertainty in waste-combustion PM2.5 mass. Halving waste-combustion PM2.5 (LOWMASS) results in 138 000 (5th–95th:109 000–167 000) mortalities per year and doubling PM2.5 (HIGHMASS) results in 518 000 (5th–95th:410 000–626 000) through the attribution method (table 1 and figure S5). The relationship between waste-combustion PM2.5 and mortality is sub-linear (additional PM2.5 impacts health less when PM2.5 concentrations are already high) because many of the waste-combustion source regions occur in already heavily polluted areas; thus, the waste-combustion emissions in these regions lead to fewer mortalities than if PM2.5 concentrations were lower.
In order to explore the dependence on estimated attributed mortality rates to model resolution, we use a coarser 4° × 5° and finer 2° × 2.5° global simulation and a 0.5° × 0.666° simulation over Asia. The total number of annual mortalities in the global 4° × 5° resolution simulation is 16% lower than the 2° × 2.5° resolution simulation (table 1 and figure S7). The decrease in mortality rates with coarser resolution is caused by two main factors. First, as grid-box area increases, ambient PM2.5 is averaged over a larger area and may reduce concentrations below the minimum threshold for mortality (between 5.8 and 8.8 μg m−3) in some locations. Second, higher-resolution simulations are better able to co-locate dense urban populations with PM2.5 increases from waste combustion. Figure 2(b) shows the annual mortality per 104 km2 and figure 2(d) shows annual mortality per 106 people over Asia at 0.5° × 0.666° resolution. In this domain, total mortalities increase from 184 000 (5th–95th:146 000–222 000) in the 4° × 5°, to 221 000 (5th–95th:174 000–266 000) in the 2° × 2.5°, to 279 000 (5th–95th:221 000–336 000) in the 0.5° × 0.666° resolution simulation. The higher-resolution simulation predicts more mortalities just in the Asia domain than the 2° × 2.5° simulation predicts globally, which highlights the importance of model resolution when using simulated PM2.5 concentrations to estimate mortality rates.
Figure 3 shows the fractions of the population impacted by waste-combustion PM2.5 and mortality risk. Figure 3(a) shows the complementary cumulative distribution functions (CCDF) of the percent of the global population exposed to different levels of PM2.5 from domestic-waste combustion. In the global 2° × 2.5° domain, 50% of the total population is exposed to a greater than 1.3 μg m−3 increase in PM2.5 from waste combustion, while 10% is exposed to a greater than 5.3 μg m−3 increase. Increases in exposure due to waste-combustion PM2.5 are generally smaller in the coarser 4° × 5° domain due to dilution of emissions into larger grid cells. Figure 3(b) shows the corresponding CCDFs for the relative increase in PM2.5-mortality risk in the WASTE_ON compared to the WASTE_OFF simulation. The all-cause relative risk is the mean value of the relative risk of the four causes weighted by the proportion of each cause to the total baseline mortality. For the two global resolutions, more than 50% of the population has greater than a 0.5% increased risk of mortality by any cause due to waste-combustion emissions, while 10% of the population has a greater than 1.5% increased risk of mortality. While in figure 3(a) 91% of the population is exposed to at least a 0.1 μg m−3 increase in PM2.5, only 75% of the population has a greater than 0.02% increased risk of mortality in figure 3(b) (due to minimum PM2.5 thresholds in the CRF). Figures 3(c) and (d) show increased PM2.5 and increased risk of mortality in the Asian domain for the three model resolutions. The majority of global mortalities occur in Asia (table 1), where a larger fraction of the population is exposed to waste-combustion PM2.5 and thus increased risk of mortality. All three resolutions show that nearly 99% of the population in Asia is exposed to at least a 0.1 μg m−3 increase in PM2.5 and nearly 83% of the population have an greater than 0.02% increased risk of mortality from waste combustion. The 0.5° × 0.666° resolution simulation predicts that a higher fraction of the population is exposed to larger increases in PM2.5 and mortality risk than the coarser resolutions due to the reasons discussed above.
Figure 3. Complementary cumulative distribution functions weighted by percent of the population for (a) the change in PM2.5 due to combustion of domestic waste for the global domain, (b) the increased risk for mortality due to combustion of domestic waste for the global domain, (c) the change in PM2.5 for the Asia domain, and (d) the increased risk of mortality for the Asia domain.
Download figure:
Standard image High-resolution image
4. Discussion and conclusions
As stated in the results, we estimate 270 000 (5th–95th:213 000–328 000) annual adult mortalities from waste combustion. Our estimate of mortality attributed to waste combustion is similar to recent estimates of mortalities due to PM2.5 from solid-fuel cookstoves (300 000–400 000 mortalities per year), another emission source associated with the developing world (Chafe et al 2014, Butt et al 2016). Waste-combustion emissions are approximately 10% of our total ambient PM2.5 mortalities and 9% of the total ambient PM2.5 mortalities estimated in the GBD 2010. We note that our estimate of global mortalities from waste-combustion emissions are of a similar magnitude to the Lelieveld et al (2015) estimate of mortalities from PM2.5 from biomass burning (∼165 000), land traffic (∼165 000), and industrial emissions (∼231 000); however, we note that our total mortality estimate is 10% lower than Lelieveld et al (2015).
Waste combustion has high mortality rates due to proximity of emissions to population. In developing countries, substantial amounts of waste combustion can occur in urban areas. We find high mortality densities (>300 deaths per 104 km2 yr−1) in cities such as Mexico City, Moscow, Johannesburg, and Rio de Janeiro. Four countries (China, India, Pakistan, and Russia) account for slightly more than 2/3 of the global mortality burden due to waste combustion. Globally, more than half of the world's population is exposed to an additional 1.3 μg m−3 of PM2.5 due to waste combustion leading to a 0.5% increased risk of mortality each year just from waste-combustion emissions.
Due to a factor-of-2 uncertainty in PM2.5 emission mass, we estimate a range of mortalities per year of 138 000 to 518 000. We note that even on the low end of this range, waste combustion contributes to a substantial number (greater than 100 000) of mortalities per year. Additionally, we estimate that a complete removal of waste-combustion would avoid 191 000 mortalities per year (see explanation in section 2.2). In these sensitivity simulations, the calculated decrease of premature mortality is less than the proportional decrease in PM2.5. The sub-linear nature of these sensitivity tests demonstrates that most of the mortalities from waste combustion occur in heavily polluted areas.
Our estimates of premature mortality from waste-combustion PM2.5 are likely a lower bound due to the coarse resolution of the global model. Combustion of domestic waste often occurs near where people live and work, sometimes in streets or in front of homes, and none of our model simulations resolve this near-field pollution exposure. The 2° × 2.5° resolution is unable to capture high PM2.5 gradients, which are often co-located with dense populations. We include an additional simulation at 4° × 5° and calculate 227 000 mortalities (16% lower than the estimate at 2° × 2.5°). This suggests increasing mortalities at progressively finer model resolutions. Due to the high mortality rates in Asia, we include a higher resolution 0.5° × 0.666° simulation in this domain. We estimate 279 000 mortalities in Asia at this model resolution. The sensitivity of health-response estimates to model resolution has been explored in past studies (e.g. Thompson and Selin 2012, Punger and West 2013).
The methods and data sources used here are similar to the GBD 2010 and several recent health-impact studies (e.g. Lim et al 2012, Apte et al 2015, Lelieveld et al 2015); however, there are substantial uncertainties and limitations inherent in these methods. First, the Burnett et al (2014) CRF is largely based on epidemiologic studies that occur in the United States, and while there are age modification factors for CeVD and IHD, there are no modifications for other sociodemographc factors. We assume here that this function applies globally; however, as demographics and pollution sources vary regionally, this may not be a valid assumption. Second, we consider all species of PM2.5 to be equally toxic; however, some studies have suggested that combustion particles may be more toxic (Krzyzanowski et al 2005). Third, due to lack of data, baseline mortality rates used here are reported at the national or regional level; however, demographics often vary sub-nationally, and national and regional level mortality rates may introduce uncertainty. Chowdhury and Dey (2016) estimate 15% fewer premature mortalities from ambient PM2.5 exposure in India when assuming a uniform country-level baseline mortality rate as opposed to varying baseline mortality at the state level using gross domestic product as a proxy. The uncertainties listed here are in addition to uncertainties in emissions (Akagi et al 2011, Wiedinmyer et al 2014) and model processes.
In this study, we focus on mortality from chronic exposure to ambient PM2.5, but we note that waste combustion also emits a number of gas- and particulate-phase toxins that may have additional significant health implications. While generally accepted that these toxins have negative health impacts, the CRFs are less well-developed making it difficult to estimate a global health impact. Thus, because PM2.5 response functions exist, it is common to use PM2.5 estimates as a proxy for exposure to a broad array of toxic species. Therefore, the net health impacts from waste combustion could be greater than estimates for only PM2.5.
Uncontrolled domestic-waste-combustion emissions are a potentially significant emission source that has been largely overlooked in past studies. Waste combustion creates health-effect disparities between high-income countries and lower-income countries, where high-income countries generate more waste but combust less, resulting in a factor of 8 fewer deaths per mass waste of generated than lower-income countries (figure S4). Due to the coarse model resolution and exclusion of gaseous toxic species, our estimate of mortalities from waste-combustion emissions is likely a lower bound. Despite this, we find a substantial burden of disease (9% of mortalities from all PM2.5) that are on the same order of magnitude as modeling estimates of industry, land traffic, and biomass burning. Waste combustion also affects aerosol number concentrations and radiative impacts, and we explore this in a separate study (Kodros et al 2016).
Acknowledgments
This research has been supported by a grant from the US Environmental Protection Agency's Science to Achieve Results (STAR) program through grant no. 83543801. Although the research described in the article has been funded by the US Environmental Protection Agency's STAR program, it has not been subjected to any EPA review and therefore does not necessarily reflect the views of the Agency, and no official endorsement should be inferred.