Environmental Modeling and Methods for Estimation of the Global Health Impacts of Air Pollution

Abstract

Air pollution is increasingly recognized as a significant contributor to global health outcomes. A methodological framework for evaluating the global health-related outcomes of outdoor and indoor (household) air pollution is presented and validated for the year 2005. Ambient concentrations of PM2.5 are estimated with a combination of energy and atmospheric models, with detailed representation of urban and rural spatial exposures. Populations dependent on solid fuels are established with household survey data. Health impacts for outdoor and household air pollution are independently calculated using the fractions of disease that can be attributed to ambient air pollution exposure and solid fuel use. Estimated ambient pollution concentrations indicate that more than 80% of the population exceeds the WHO Air Quality Guidelines in 2005. In addition, 3.26 billion people were found to use solid fuel for cooking in three regions of Sub Saharan Africa, South Asia and Pacific Asia in 2005. Outdoor air pollution results in 2.7 million deaths or 23 million disability adjusted life years (DALYs) while household air pollution from solid fuel use and related indoor smoke results in 2.1 million deaths or 41.6 million DALYs. The higher morbidity from household air pollution can be attributed to children below the age of 5 in Sub Saharan Africa and South Asia. The burden of disease from air pollution is found to be significant, thus indicating the importance of policy interventions.

Appendices

Appendix I: Representing Urban/Rural Fractions of PM2.5 in TM5

TM5 model simulations were performed at a spatial resolution of 1° × 1° longitude–latitude, corresponding to a nominal longitudinal resolution of ca. 111 km at 0° latitude (tropics), 79 km at 45° latitude, and 56 km at 60° latitude (latitudinal resolution is always 111 km). Ambient concentrations of some air pollutants may show strong variability at a much finer scales (e.g., in urban areas, at hot spots close to industrial point sources of emission, etc.), and could thus result in variable impacts on populations. We also separately estimate for all regions, an urban increment at the grid cell resulting from anthropogenic primary aerosol emissions, assuming that the model calculations are sufficient to cover aerosols from natural and secondary sources. The sub-grid increment parameterization attributes calculated primary aerosol concentrations according to population density and the area over which they are emitted. Population density is derived from the high (0.1° × 0.1°) resolution CIESIN population dataset provided by Columbia University (http://www.ciesin.org/). The urban increment of primary aerosol concentration at the 1°×1° grid cell is calculated according to population density and the area over which they are emitted.

Assuming that the concentration of Primary PM in each 1° × 1° grid cell of the model is given by

$$ \text{CTM5} = \frac{E}{\lambda } $$

(1)

With E = in-cell emission intensity of BC + PPOM (primary emissions of black carbon and particulate organic matter), λ = in-cell mixing rate, including dilution.

If we distinguish rural from urban emissions, we can define the rural concentration as

$$ {C_{\text{RUR}}} = \frac{{{E_{\text{RUR}}}}}{\lambda } = \frac{{1 - {f_{\text{up}}}}}{{1 - {f_{\text{ua}}}}}\frac{E}{\lambda } $$

(2)

With f _up = urban population fraction in the 1° × 1° grid cell derived from 0.1° × 0.1° population statistics, f _ua = urban area fraction in the grid cell.

The urban and rural population fractions are estimated by setting a threshold on the population density in high-resolution sub-grids. To conserve the grid-average concentration, after the calculation of C _RUR, the urban concentration must fulfill the requirement that:

$$ {f_{\text{ua}}}{C_{\text{URB}}} + \left( {1 - {f_{\text{ua}}}} \right){C_{\text{RUR}}} = {\text{CTM5}} $$

(3)

where according to Equations 1 and 2,

$$ {C_{{{\text{RUR}}}}} = \frac{{1 - {f_{\text{up}}}}}{{1 - {f_{\text{ua}}}}}{\text{CTM5}} $$

(4)

C _URB follows immediately from Eq. (3)

Equation 4 basically rescales the sub-grid concentration of primary emitted components according to population density and the area over which they are emitted.

In order to avoid very spiky artifacts associated with a small fraction of the grid occupied by a densely populated sub-area, we introduce empirical limitations to the ratio C _RUR/C _URB and to CTM5/C _RUR:

• 1. Primary BC and POM (C _RUR ) should not be lower than 0.5 times the TM5 grid average. This is based on observations in Europe [51, 52]

• 2. Urban primary BC and POM should not exceed the rural concentration by a factor 5.

Finally, the concentration edges between urban and rural areas are smoothed numerically (linear interpolation over the 0.1° × 0.1° sub-grid cells at the rural–urban border to avoid artificial gradients).

Appendix II: Methodology for Estimation of Health Impacts from Outdoor and Household Air Pollution

We estimate health impacts from ambient air pollution using the PAF approach based on the gradient of risk between the theoretical minimum level of air pollution exposure and the estimated observed exposure [34]. We apply an approach similar to that detailed in [50] which involved: (1) estimating total population exposures to PM2.5; (2) choosing appropriate exposure-response factors for PM2.5 as discussed earlier in the text; (3) determining the current rates of morbidity and mortality in the population of concern using data from [45] and (4) estimating the attributable number of deaths and diseases.

The population-attributable fraction to exposure is calculated based on [53] and is estimated as:

$$ {\text{PAF}} = \frac{{P \times ({\text{RR}} - 1)}}{{[P \times ({\text{RR}} - 1) + 1]}} $$

(5)

where P = exposure expressed in PM2.5 concentrations, and RR = relative risk for exposed versus non-exposed populations. Once the fraction of a disease that is attributed to a risk factor has been established, the attributed mortality or burden is simply the product of the total death or DALY estimates for the disease and the attributed fraction.

We estimate the effects by combining information on the exposed population and the fraction of current disease levels attributable to solid fuel use. The approach utilizes relative risk estimates for health outcomes that have been associated with exposures to household pollution due to indoor smoke from solid fuel use and uses the population dependent on solid fuels as an exposure surrogate. In contrast to the pollutant based approach, which focuses on PM2.5 concentrations from combustion, the fuel-based approach takes advantage of the large number of epidemiological investigations conducted primarily in rural areas of developed countries that treat exposure to household air pollution from SFU as a single category of exposure and appears to be the most reliable method for assessing the environmental burden of diseases from SFU in developing countries [50].

The attributable fraction to SFU, AF_sfu, can be estimated as:

$$ {\text{A}}{{\text{F}}_{\text{sfu}}} = \left[ {\frac{{{P_{\text{e}}}\left( {{r_{\text{r}}} - 1} \right)}}{{{P_{\text{e}}}\left( {{r_{\text{r}}} - 1} \right) + 1}}} \right] $$

(6)

where p _e represents the population exposed to the solid fuels and r _r the relative risk due to SFU.

Similarly, attributable burden due to the solid fuel, AB_sfu use can be estimated as

$$ {\text{A}}{{\text{B}}_{\text{sfu}}} = {\text{A}}{{\text{F}}_{\text{sfu}}}{\text{CDL}} = \left[ {\frac{{{P_{\text{e}}}\left( {{r_{\text{r}}} - 1} \right)}}{{{P_{\text{e}}}\left( {{r_{\text{r}}} - 1} \right) + 1}}} \right]{\text{CDL}} $$

(7)

Appendix III: Comparison of Preliminary and Scaled Values of Average PM2.5 Concentrations (Neglecting the Effects of Dust, Sea Salt and SOA, Without Urban Increment)

Rescaling involved calculating for each grid cell, the ratio of change in concentrations to changes in emissions for each component separately and scaling for the change in emissions. This assumes no regional transfer of emissions but assuming that emission changes are not at the grid level but rather at country/state/province level, the relative change in emissions within the cell is similar to the relative changes of the surrounding cells. Shown above are the comparisons of PM2.5 estimates before and after scaling. The differences were found not to impact the health impacts significantly due to the further truncation of the response above 50 μg/m³.