We use results from a global atmospheric modeling exercise performed using a coupled general circulation model with interactive chemistry, the LMDz-INCA chemistry–climate model [8,10,11]. Modeled ozone concentrations for present conditions using this model are shown to agree reasonably well with surface ozone measurements [8,10].

The global simulations used in this study are described fully by Szopa et al. [24] and Szopa and Hauglustaine [25], and these simulations were shown previously to be within the range of several other models in the Photocomp experiment [7,23]. While Szopa et al. [24] and Szopa and Hauglustaine [25] also present results for a regional model imbedded over Europe, and consider climate change scenarios in the future, we consider here only the results of the global model using present-day meteorology (for 2000 from the ECMWF ERA40 reanalysis).

Four scenarios are considered in this study: a simulation for 2000, and projected 2030 emissions under the SRES A2 scenario, the CLE (current legislation) scenario, and the MFR (maximum feasible reduction) scenario. The SRES A2 scenario [17] is a high-growth scenario with rapid increases in emissions of air pollutants. The CLE scenario takes into consideration recently-enacted legislation to improve air quality in nations around the world, and the MFR scenario assumes that currently available emission control technologies are aggressively employed globally [6]. LMDz-INCA is run with a horizontal resolution of 3.75° in longitude and 2.5° in latitude, and mortality effects are calculated on this grid also. Using hourly surface ozone concentrations, we calculate the daily maximum 8-h average ozone concentration on each day and at each grid cell, and use these 8-h maxima to drive the ozone mortality estimates.

We estimate effects of changes in ozone on premature human mortality for several scenarios, following the methods used by West et al. [27]. We use the results of an epidemiological daily time-series study that relates ozone with daily mortalities, using a distributed lag method and a large database for 95 cities in the United States [2]. The equation for a change in human mortalities (ΔMort) due to a change in ozone concentration (ΔO₃) used in the epidemiological study and in this study is:

The β that we have selected from Bell et al. [2] is consistent with, but generally smaller than, meta-analyses of ozone mortality that have been published since 2001 [1,3,12,14,15,22,26]. We selected this β because it is not subject to publication bias (the possible tendency to selectively publish positive results), which may bias meta-analyses high. While this value of β was estimated in the US, we assume that this ozone–mortality relationship is valid globally, as similar results have been demonstrated in Europe [9] and in some locations in the less industrialized world [5,13,19]. We use a value of β that was derived for the daily maximum 8-h ozone. While some studies have tried to distinguish the ozone–mortality relationship for different populations, such as by age or by the proximate cause of death, these attempts generally do not find significant differences from the whole population [2]. Consequently, we use a value of β derived for the whole population, and apply it to the whole population. Since long-term effects of ozone on mortality have not been demonstrated [20], we do not consider possible chronic effects of ozone on mortality, nor the years of life lost due to premature mortality.

The growth in global population and its spatial distribution is modeled in four world regions, according to the SRES A2 scenario [17], totaling 6.17 billion in 2000 and 9.17 billion in 2030. The spatial distribution of population within each region is based on the 2003 population from the LandScan database [18], which provides population data at very fine resolution, and which we then map onto the atmospheric modeling grid.

Equation (1) is applied in each grid cell and on each day, using the change in the daily maximum 8-h ozone concentration on that day, the population of that grid cell, and the value of y₀ corresponding to the appropriate world region. Since it is applied on each day, β is divided by 365.25 days per year. Equation (1) is prescribed for a change in ozone concentration, and epidemiological studies can effectively evaluate β over the range of ozone concentrations for which there are observations. While most ozone measurements in these studies are in the range of modest ozone, some studies show evidence that a similar ozone–mortality relationship holds at low concentrations, well below current national standards, and question whether a low-concentration threshold exists [4,9]. Estimating the total mortality burden due to ozone would require one to specify a reference ozone concentration or field against which current levels would be compared. Because the ozone–mortality relationship has not been firmly established at low concentration, and because specifying a reference case is beyond the scope of this study, we do not estimate the total mortality burden of ozone. Rather, we estimate ozone mortality due to differences in ozone concentration between the different scenarios considered.

We evaluate ozone mortality assuming a low-concentration threshold of 25 ppbv, below which changes in ozone are assumed to have no effect on human mortality. Using such a threshold replaces estimating mortality in particular seasons; epidemiological studies in temperate regions have often segregated results by season, but these seasons do not have the same relevance in tropical nations [27]. We choose a threshold at roughly the current background concentration, as the effect of ozone on mortality is uncertain at low concentrations. Since this choice of a threshold is fairly arbitrary, we test the sensitivity of the results to this threshold level.

Szopa et al. [24] and Szopa and Hauglustaine [25] present global results for the set of scenarios that we consider; here we present population-weighted concentrations, as indicators which are relevant for human health. Table 1 shows the population-weighted annual average 8-h daily maximum surface ozone concentration, globally and in each of ten world regions. Definitions of these ten regions are shown in Fig. 1.

The results in Table 1 indicate a very substantial increase in ozone globally in 2030 under the A2 scenario, relative to the 2000 simulation, with the global population-weighted 8-h ozone increasing by 9.4 ppbv. The projected increase in ozone in the A2 scenario is greatest over South Asia (nearly 15 ppbv), with large increases also in the Middle East, Africa, Southeast Asia, Latin America, and East Asia. This substantial growth in ozone in the A2 scenario, particularly over South Asia, has been simulated previously using other models [7,21].

These regional changes in ozone are the result of a combination of changed emissions within each region, and background influences due to increases in emissions in other regions and growth in the global background ozone concentration itself. Fig. 2 shows the regional changes in emissions in the 2030 scenarios, relative to the 2000 base simulation. In the A2 scenario, emissions of CO, NO_x, and NMVOCs increase in all world regions. While the absolute growth in emissions in South Asia is not larger than in other regions, it has the highest growth in population-weighted ozone. The Middle East, Africa, and Southeast Asia also have large increases in ozone, despite the fact that emissions increases are comparable to those in East Asia and North America. This suggests that the production of ozone per unit change in emissions is greater in tropical regions than in temperate regions, which is consistent with other model results [16].

The population-weighted ozone is likewise projected to increase in all world regions in the CLE scenario, again with very large increases in South Asia (9 ppbv). Ozone increases in the CLE scenario despite decreases in NO_x emissions in Europe, the former Soviet Union, Japan, and Australia, due to the increase in global ozone concentrations and long-range transport [25]. Ozone is estimated to decrease in all regions in the MFR scenario, relative to 2000, with the largest decreases in East Asia, the Middle East, and South Asia. The increase in all regions under CLE, and decrease in MFR is consistent with results from several models [7]; our reported ozone changes are greater than Dentener et al. [7] reported, because of our use of population-weighted 8-h ozone indicators, rather than area-weighted annual averages.

We focus on changes in human mortality in the three 2030 scenarios. Table 2 shows the avoided mortalities in 2030 under the CLE scenario relative to A2, using the low-concentration threshold of 25 ppbv. We estimate a total of 191,000 avoided mortalities globally, which is roughly 0.2% of the projected total number of annual deaths in 2030. The largest numbers of avoided mortalities occur in Africa, South Asia, and East Asia, as these three regions have the highest projected total population. We also show the avoided mortalities per million people, showing that Africa, the former Soviet Union, and South Asia are the highest. Africa has the highest total mortalities and mortalities per million people mainly because it has the highest baseline mortality rate in the world (mainly due to poor access to health care, food, and sanitation), followed by the former Soviet Union (because of an aging population). The avoided mortalities per million people are least in Southeast Asia, Europe, and North America, due to small changes in ozone and/or low baseline mortality rates.

In our estimates of the total avoided premature mortalities, we apply globally the ozone–mortality relationship derived from a study in the US [2]. While few ozone–mortality studies have been conducted in the less industrialized world, the general causes of death differ among world regions. Because of this, we also show the avoided cardiovascular and respiratory (C&R) mortalities. C&R causes are the most likely means by which ozone affects mortality, although Bell et al. [2] found that the β for C&R mortalities was not significantly different from the β for total mortalities. We use the β for C&R mortalities [2] and use the baseline mortality rates for only C&R mortalities. We estimate that 103,000 avoided premature C&R mortalities result under CLE relative to A2, and Table 2 shows that Africa does not have the highest number of avoided C&R mortalities nor C&R mortalities per million people. Rather, the former Soviet Union has the most C&R mortalities per million people, as it has an aging population for whom most deaths are due to C&R causes. The estimated avoided C&R mortalities can be thought of as a lower bound on the total mortalities, but it likely omits other means by which ozone may affect mortality.

Table 3 shows the avoided mortalities in 2030 for the MFR scenario relative to A2, with an estimated 458,000 avoided mortalities in 2030 (∼0.5% of total 2030 mortalities), and 259,000 avoided C&R mortalities. The MFR scenario causes ozone in South Asia to decrease substantially, by more than 20 ppbv for the population-weighted 8-h ozone. Consequently, South Asia has the largest number of avoided mortalities per million people in this scenario.

As some people consider the CLE scenario the most likely scenario to 2030, we also consider the avoided mortalities in the MFR scenario with respect to the CLE scenario (Table 4). The results show 267,000 avoided premature mortalities globally (∼0.3% of total projected mortalities), and 155,000 C&R mortalities, with the largest number of avoided mortalities in South Asia because of the very large reduction in ozone in this region (about 15 ppbv for the population-weighted 8-h ozone). Note that the results in Tables 2 and 4 do not add simply to give Table 3, due to the nonlinearity in the ozone–mortality function and the presence of the low-concentration threshold.

Because both ozone and population change between 2000 and 2030, it is not possible to estimate directly the increased mortalities between the 2000 base simulation and the 2030 scenarios. We approximate this change by estimating the change in mortality in 2000 with respect to a hypothetical case where ozone is uniformly equal to the 25-ppbv threshold, which one might argue would represent the total mortality burden of ozone. We subtract this estimate of mortality from similar results when the 2030 A2 scenario is estimated relative to the uniform 25 ppbv threshold. The results give 500,000 total excess mortalities annually under A2 (with respect to 2000), and 271,000 C&R mortalities, where these differences are due to both increasing ozone globally and increasing population.

The mortality estimates shown previously are estimated with an ozone–mortality threshold of 25 ppbv. Fig. 3 shows the total avoided mortalities as a function of the threshold assumed, for three cases, corresponding to Tables 2–4. The results show that for thresholds below 25 ppbv, there is very little effect on the total estimated mortalities, with results at 25 ppbv only 1–2% lower than with no threshold. At higher thresholds, the importance of the threshold is substantial. While the threshold has a minor effect at 25 ppbv, a substantial number of grid cell-days fall below the threshold. At a threshold of 25 ppbv, 12% of the populated grid cell-days (that is, for grid cells where at least one person lives) are below the threshold in the A2 case, and 24% in MFR, while mortality only decreases by 2% because of the threshold. At a 50 ppbv threshold, 86% are below the threshold for A2, and 95% for MFR, while mortality decreases by 53%. This difference arises because areas with high ozone also tend to be highly populated.

In addition to uncertainty in the low-concentration threshold, we consider the sensitivity of the estimated global avoided premature mortalities to other uncertain parameters in Fig. 4. The value of β has a large effect on the estimated avoided mortalities, and if we use the range of β from three recent meta-analyses, higher ranges of avoided mortalities result. The response of ozone concentration to the change in emissions is also associated with a large uncertainty and causes a broad range of results. We do not consider uncertainty in the emissions scenarios themselves, since these are considered as alternative scenarios intended to illustrate a range of plausible futures. Because the ozone response and the low-concentration threshold are not independent, an assessment of the total uncertainty would require many simulations. Fig. 4 suggests that the total uncertainty is substantial.

As indicated by these three scenarios, the future of ozone air pollution and its implications for human health in 2030 can vary considerably. We find that in 2030 under the A2 and CLE scenarios, ozone increases in each of 10 world regions, relative to 2000, with substantial increases in mainly the tropical regions. Ozone decreases in all world regions under the MFR scenario.

These ozone changes cause substantial changes in cases of premature human mortality. In 2030, we estimate a reduction in annual mortalities of about 190,000 due to the emission reductions in the CLE scenario relative to A2, with the greatest reductions in Africa due to the high baseline mortality rates. The MFR scenario causes reductions of about 460,000 mortalities relative to A2, and about 270,000 relative to CLE; avoided mortalities are estimated to be greatest in South Asia due to the substantial growth in ozone in South Asia in A2 and CLE, and substantial decrease in MFR. These benefits of avoided mortalities could be combined with other benefits of reduced ozone for human morbidity, agricultural productivity, and ecosystems, and compared against the estimated costs of achieving the emission reductions under these scenarios [27]. Future work should evaluate changes in surface concentrations of PM under these scenarios, and estimate the avoided cases of PM mortality.

These estimates have large uncertainties, and depend significantly on several parameters. We show that the results decrease strongly for low-concentration thresholds greater than 25 ppbv (for the daily maximum 8-h concentration). Substantial uncertainty also lies in the relationships between emissions and ozone concentrations, as well as the ozone–mortality relationship from epidemiologic studies. Our estimated mortalities would have been roughly 35% higher had we used the results of recent meta-analyses [3,12,15]. Future research is needed to improve scientific understanding and ability to model ozone on a global scale, with greater attention to urban scale processes important for population exposure. Future research should also aim to improve understanding of the mechanisms by which ozone affects human health, and to better quantify its effects on diverse populations globally.