Estimating Mortality Related to O 3 and PM 2.5 under Changing Climate and Emission in Continental Southeast Asia

Air pollution causes adverse effects not only on the environment but also on human health. This study evaluated the excess mortalities in continental Southeast Asia that are related to future O 3 and PM 2.5 ambient concentration changes attributed to future climate change and emission change. The Environmental Benefits Mapping and Analysis Program - Community Edition (BenMAP-CE) was applied as a health impact assessment tool. In BenMAP-CE simulations, baseline scenarios presenting for the present year (2014) were compared against the control scenarios presenting for the future year (2050). The air pollutant concentrations for the simulations were collected from modeled data. The future population data and baseline incidence rates were as same as the 2014 levels. In four calculating countries namely Laos, Cambodia, Thailand, and Vietnam, on average, impacted by climate change alone, the avoided mortalities of –1164 and –3358 under Representative Concentration Pathway (RCP) 4.5 scenario and the additional mortalities of +758 and +2562 under RCP8.5 scenario were calculated for O 3 and PM 2.5 , respectively. Future emission change alone under Evaluating the Climate and Air Quality Impacts of Short-Lived Pollutants current legislation (ECLIPSE CLE) scenario induces +7113 and +11072 additional O 3 and PM 2.5 related mortalities, respectively. Combined change in climate and emission produces additional O 3 and PM 2.5 related mortalities of +6067 and +7830 under RCP4.5 and ECLIPSE CLE combined scenario and +8763 and +14580 under RCP8.5 and ECLIPSE CLE combined scenario, respectively. The results of this study provided meaningful information for understanding the public health attributed to air pollution in the region.


INTRODUCTION
Southeast Asia has been faced with air pollution issues largely caused by rapid population growth and economic development. Its annual rate of population change was 1.1% during the period 2015-2020 (UN, 2019) and its gross domestic product was 5.0% during the period 2013-2017 (OECD, 2019). Regional air quality varies widely across space and time (Nguyen et al., 2019a). Ambient concentrations of ozone (O3) and particle with an aerodynamic diameter smaller than and equal to 2.5 µm (PM2.5) have continued to increase. Open biomass burning and inappropriate land-use practices cause transboundary smoke haze pollution which has become a serious air quality problem in Southeast Asia (UNEP, 2016). As well-known, along with environmental effects, air pollution produces harmful effects on human health. Exposure to air pollution might increase the risk of heart disease, chronic respiratory diseases, lung infections, and cancer (HEI, 2019). Of the pollutants in the atmosphere, O3 and PM2.5 are believed to be the most detrimental to public health (Boldo et al., 2011;EEA, 2012;Stowell et al., 2017;Yang et al., 2019). Studies have shown a link between O3 and PM2.5 exposures and a variety of adverse health outcomes.

Air Quality Modeling Data
The Weather Research and Forecasting (WRF) model version 3.4 (Skamarock et al., 2008) was online coupled with Community Multiscale Air Quality (CMAQ) model version 5.0.2 (CMAS, 2012) to provide the current and future O3 and PM2.5 levels in continental Southeast Asia. The modeling domain (Fig. 1) had a horizontal resolution of 24 km. It had 98 × 98 grid cells covering 4°N to 26°N and 94°E to 116°E. The number of its vertical grid cells was 30, from the surface up to 100 hPa. In the CMAQ model, the gas-phase chemistry and the aerosol process were based on the Carbon Bond mechanism developed in 2005 (CB05) (Whitten et al., 2010) and the sixth-generation CMAQ aerosol module (AERO6) (Appel et al., 2013), respectively. The clear-sky photolysis rates were calculated using the method that calculates the rates online using O3 and aerosol concentrations projected within a simulation (Binkowski et al., 2007). Comprehensive model evaluations were performed by comparing the model results of baseline simulation with the observations at monitoring sites (Fig. 1). The correlation coefficients between the modeled concentrations and observed concentrations were 0.69 and 0.36 for O3 and 0.80 and 0.69 for PM2.5 in Thailand and Vietnam, respectively. The modeled PM2.5 concentrations were found to be underestimated in Thailand but overestimated in Vietnam. The modeled O3 concentrations were overestimated in both countries. The detailed WRF-CMAQ model configuration and evaluation can be found in Nguyen et al. (2019a).
In order to project the future air pollution levels attributed to the impacts of climate change alone, emission change alone, and the changes in both climate and emission in future year (2050) relative to the present year (2014), WRF-CMAQ simulations were carried out which were different in using the meteorological and emission input data. The first simulation, a baseline simulation, was used the present meteorology condition and present anthropogenic emission. The next two simulations were used future meteorology condition and present anthropogenic emission. The future climate was calculated following the two popular RCP4.5 and RCP8.5 climate scenarios developed by Intergovernmental Panel on Climate Change (IPCC). The fourth simulation was used present meteorology condition and future anthropogenic emission. Future anthropogenic emission was produced based on the projection of ECLIPSE CLE scenario. The last two simulations were used future meteorology condition under RCP4.5 and RCP8.5 scenarios and future anthropogenic emission under ECLIPSE CLE scenario. Description of key input of the simulations is summarized in Table 1. More details about simulation design together with simulation input data were reported in our previous studies (Nguyen et al., 2019b(Nguyen et al., , 2020. The O3 and PM2.5 concentration results of all the above WRF-CMAQ simulations were used for BenMAP-CE simulations which were described in Section 2.3. The detail future changes in O3 and PM2.5 levels due to future climate and emission changes can be found in Nguyen et al. (2019bNguyen et al. ( , 2020.  (Nguyen et al., 2019b) found that climate change has a non-negligible impact on O3 and PM2.5 concentrations since it changes O3 and PM2.5 concentrations in four targeted countries up to 4.62% and 22.52%, respectively. Under RCP4.5 scenario, O3 and PM2.5 concentrations appear to be reduced in the future atmosphere which indicated a potential benefit for air quality. The largest O3 decrease occurs in Vietnam during the dry season and in Laos during the wet season. The largest PM2.5 decrease is in northern Cambodia and northern Vietnam. Under RCP8.5 scenario, climate change causes increases in O3 and PM2.5 concentrations in most areas in the dry season. In the wet season, there are more areas that have O3 and PM2.5 decreases compared to the dry season. The large O3 increase is located in northern Vietnam and the large PM2.5 increase is located in areas surrounding northern Cambodia and southern Laos in the dry season. O3 decrease was found in Laos and Vietnam and PM2.5 decrease was found in Cambodia, Laos, and Vietnam in the wet season. Under these two scenarios, O3 decrease is attributed to water vapor increase, and O3 increase is driven mainly by the stagnant condition, biogenic emission increase, and temperature increase; PM2.5 decrease is mostly due to the precipitation increase, and PM2.5 increase is caused largely by the more stagnant condition in the future climate (Nguyen et al., 2019b). Comparing to the impacts of climate change alone, the impacts of emission change alone on regional air quality are larger (Nguyen et al., 2020). Overall, attributable to emission growth separately under  1) all year round for PM2.5. The O3 decrease in part of northern Vietnam could be explained as follows. The ratio of NMVOCs/NOx in the area was very low compared to those of the other areas. Therefore, the lower this ratio decreases under the effects of projected anthropogenic emission change, the more favorable the titration is in this area than in the other areas, causing a greater O3 loss. Reduction in PM2.5 was observed in areas in and surrounding PP, HNMR, and HCMCMR because there are anthropogenic primary PM2.5 EC and PM2.5 OC emission decreases implying a success of air quality control activities in these areas (Nguyen et al., 2020). In comparison with Cambodia and Vietnam, Laos and Thailand suffer more O3 increases. The annual average O3 increases are +6.19 and +6.51 ppb in Laos and Thailand while they are +4.43 and +4.09 ppb in Cambodia and Vietnam, respectively (Table 2). A significant O3 increase was found in the wet season in northern Laos. For PM2.5, areas in and surrounding Bangkok Metropolitan Region (BMR), Mae Moh power plant (MMPP), and part of northern Vietnam receive significant increases. Climate change and emission change in combination mostly produce the rises in O3 and PM2.5 concentrations. O3 and PM2.5 rises are larger in RCP8.5 and ECLIPSE CLE combined scenario than that in RCP4.5 and ECLIPSE CLE combined scenario. The spatial patterns of concentration changes in these two scenarios were quite similar to those of ECLIPSE CLE scenario.

Health Impact Assessment
The BenMAP-CE health impact assessment tool version 1.5 released in May 2021 was employed to estimate excess mortalities related to O3 and PM2.5 changes in the future period due to changes in climate and emission. BenMAP-CE is an open source software that is powerful in estimating health impacts resulting from changes in air pollution concentrations (U.S. EPA, 2021). To quantify health impacts associated with air pollution, BenMAP-CE first estimates the air pollution change using user-specified air quality data. The change in air pollutant concentration is the difference between the starting pollutant concentration in a baseline scenario and the pollutant concentration after some change in a control scenario. Next, this pollutant concentration change is related to certain health effects. The relationship between concentration change and health effect is referred to as the health impact function (HIF). These functions are derived from epidemiology studies relating pollutant concentrations with health outcomes. BenMAP-CE utilizes HIFs to the population experiencing the air pollution change to calculate health impacts. The expression for HIF is defined as shown in Eq. (1) (U.S. EPA, 2021).
where ∆Y represents the rate of adverse health effects due to the change of pollutant concentration. The statistical coefficient β, which is unitless, is determined from the epidemiological analysis measuring the response of a health effect to one-unit change in pollutant concentration. ∆AQ represents the concentration difference between the control scenario and the baseline scenario. Y0 is the baseline incidence rate for the health endpoint per unit population, and POP represents the population exposed to the air pollution.
In BenMAP-CE, there is a library of HIFs which is used for different health endpoints including work loss days, morbidity, and mortality. In this study, the selected health effect was mortality from all-cause. Table 3 summarizes the HIFs embedded in BenMAP-CE that were selected to apply in this study. The HIF for O3 (Bell et al., 2005) was selected since it represents the entire age groups and is the only choice for the health endpoint of all-cause mortality related to daily 24 (D24) hour mean O3 within in datasets of BENMAP-CE. The selected all-cause mortality HIF for PM2.5 (Turner et al., 2016) is the most recent HIF which covers larger numbers of locations and age groups compared to the others in the datasets.
To calculate O3 and PM2.5 related excess mortality, the key input data was obtained as follows. Air quality changes, ∆AQ, were calculated by using two sets of pollutant concentrations: future pollutant concentrations minus present concentrations. The future and present pollutant concentrations were simulated as documented in Section 2.1. The exposed population, POP, was estimated as shown in Section 2.2. All-cause baseline mortality rates, Y0, in 2014 were calculated from crude death rates collected from Work Bank Open Data (https://data.worldbank.org/). All the data inputs were adjusted according to formats that BenMAP-CE recognizes. All other options and parameters were set as default within the tool.
In order to estimate the health impacts produced by the change of O3 and PM2.5 concentrations caused by climate change alone, emission change alone, and combined climate change and emission change, five simulations were performed. The present and future air quality (AQ) data were referred to as the baseline scenario and the control scenario in each simulation, respectively. In all simulations, the future exposed population and baseline mortality rates were kept constant as in the present levels so that the estimated mortalities were affected only by pollutant concentration changes between present and future air quality. In the first and second simulations, the O3 and PM2.5 levels caused by climate change alone under RCP4.5 and RCP8.5 scenarios, respectively, were used as the future air quality data, therefore, these mortality assessments can quantify the air pollution-related health impacts due to climate change alone. Similarly, in the third simulation, the O3 and PM2.5 levels caused by emission change alone under ECLIPSE CLE scenario were used as the future air quality data, so that, this assessment estimated mortality attributed to air quality change due to future emission change alone. In the fourth and fifth simulations, the O3 and PM2.5 levels caused by both climate change under RCP4.5 and RCP8.5 scenarios, respectively, and emission change under ECLIPSE CLE scenario were used as the future air quality data, and these assessments provided mortality attributed to climate change along with emission change. This type of simulation design has been carried out in several previous works (Knowlton et al., 2004;Sun et al., 2015;Yang et al., 2019). Table 4 summarizes the simulation design of this study. Table 5 summarizes annual and seasonal all-cause mortality findings caused by future changes in O3 attributable to climate change and emission change. The additional mortalities are presented by positive values, the avoided mortalities are presented by negative values. The mean values were given together with median values, and 90% confidence intervals (CI). National mortality was gained by adding up all the estimates of the nation's cells. In response to the decrease in future O3, in Cambodia, Laos, Thailand, and Vietnam, BenMAP-CE estimated -53, -30, -242, -419 avoided mortalities in the dry season, -29, -28, -140, -226 avoided mortalities in the wet season, and -82, -58, -382, -642 avoided mortalities all year round in average under RCP4.5 scenario, respectively. On the other hand, the predicted increase in future O3 under RCP8.5 scenario results in additional mortalities except in Laos in the wet season. Cambodia has +23, + 4, +27 additional mortalities, Thailand has +197, +136, +339 additional mortalities, Vietnam has +349, +38, +383 additional mortalities in the dry season, the wet season, and all year round, respectively. Lao has +17 additional deaths in the dry season but -7 avoided mortalities in the wet season. All year round, it has +9 additional mortalities. Under the impact of climate change alone, larger health effects occur in the dry season compared to the wet season that was consistent with the higher projected O3 changes found in this season (Fig. 2). Future emission under ECLIPSE CLE scenario mostly increases (Nguyen et al., 2020) causing high additional mortalities significantly compared to climate change impact under RCP8.5 scenario. Additional mortalities were estimated to be +292, +188, +2177, +304 in the dry season, +292, +223, +1722, +1919 in the wet season, and +582, +410, +3888, +2233 all year round in Cambodia, Laos, Thailand, Vietnam, respectively. The combined impacts of climate change under RCP4.5 scenario and emission change under ECLIPSE CLE scenario produce +240, +162, +2006 additional mortalities in the dry season, +260, +193, +1567 additional mortalities in the wet season, and +498, +355, +3560 additional mortalities all year round at Cambodia, Laos, Thailand. Different from these countries, Vietnam has -90 avoided mortalities in the dry season but +1731 additional mortalities in the dry season. For the entire year, it suffers +1654 additional mortalities. RCP8.5 scenario together with ECLIPSE CLE scenario brings out more additional mortalities compared to RCP4.5 scenario in combination with ECLIPSE CLE scenario as RCP8.5 scenario causes O3 concentration increase while RCP4.5 scenario causes the decrease in O3 concentration (Nguyen et al., 2019b). The additional mortalities are +336, +234, +2701, +937 in the dry season, +298, +227, +1867, +2168 in the wet season, and +631, +460, +4563, +3109 all year round in Cambodia, Laos, Thailand, Vietnam, respectively. In general, among countries, it turned out that Vietnam and Thailand have higher excess mortality numbers than Cambodia and Laos because of their higher population. The national mortality increases or decreases corresponding to the increase or decrease of national average O3 concentration. However, in Vietnam, under RCP8.5 scenario, the national average O3 concentration was projected to decrease -0.05 ppb (Table 2) while mortality was estimated to increase in the wet season. As shown in Fig. 2 and Fig. 4, parts of Vietnam where have O3 increase also have high population density, causing large additional mortality compared to the avoided mortality from the parts of Vietnam where have O3 decrease and lower population density. The additional mortality is +91 deaths and the avoided mortality is -54 deaths. It results in net mortality of +38 deaths. Similarly, under RCP4.5 and ECLIPSE CLE combined scenario, there is the O3 increase of +1.43 ppb over Vietnam in the dry season (Table 2). However, the mortality decreases. It was because the large O3 decrease, that is up to -8.03 ppb, in part of northern Vietnam (Fig. 2) where has a high population density (Fig. 4) causes high mortality reduction. The avoided mortality is -747 deaths. In the areas where have O3 increase, the additional mortality is +657 deaths. Thus, the net result is -90 avoided deaths. Fig. 5 shows the annual and seasonal spatial distributions of the all-cause mortalities. Overall, the mortality patterns are relevant to the corresponding patterns of O3 concentration changes, the mortality increases or decreases along with the rising or reducing O3 concentration. The spatial distribution of O3 related mortality showed high mortality in areas where have dense populations. It indicated the dominant influence of the population on mortality results that was also found in Sun et al. (2015). Significant additional mortality was observed in areas in and surrounding PP, HCMCMR, MMPP, and BMR all year round and in areas in and surrounding HNMR in the wet season under ECLIPSE CLE scenario, RCP4.5 and ECLIPSE CLE combined scenario, RCP8.5 and ECLIPSE CLE combined scenario (Figs. 5(c), 5(d), and 5(e)). It was due to the significant effect of the population as mentioned since the O3 concentration changes in these areas were not so large in comparison with those of the other areas in most scenarios. Large avoided mortality from future O3 reduction occurs in areas in and surrounding HNMR in the dry season in all scenarios except RCP8.5 scenario ( Fig. 5(b)). The largest avoided mortality was found under RCP4.5 and ECLIPSE CLE combined scenario. Table 6 summarizes the annual and seasonal PM2.5 related all-cause mortality attributed to climate and emission changes. Resulting from the future PM2.5 decrease under RCP4.5 scenario, the avoided mortalities are -116, -58, -778, -782 in the dry season, -19, -46, -628, -943 in the wet season, and -132, -104, -1407, -1715 all year round at Cambodia, Laos, Thailand, and Vietnam, respectively. Under RCP8.5 scenario, results show +248, +95, +848 additional mortalities in the dry season, -29, -4, -820 avoided mortalities in the wet season, and +221, +90, +32 additional mortalities all year round in Cambodia, Laos, Vietnam, respectively. Thailand has additional mortalities in both seasons due to PM2.5 concentration increase. The mortalities are +1984, +231, and +2219 in the dry season, the wet season, and all year round, respectively. For ECLIPSE CLE scenario, a high increase in mortality was observed in the four target countries compared to RCP8.5 scenario. It was found +192, +6111 additional mortalities in the dry season, +151, +4276 additional mortalities in the wet season, and +341, +10369 additional mortalities all year round in Laos, Thailand, respectively. In Cambodia and Vietnam, mortalities increase by +70, +391 in the dry season and decrease by -45, -28 in the wet season, resulting in the additional mortalities of +22, +340 all year round. Under RCP4.5 and ECLIPSE CLE combined scenario, results showed mortality decrease in Cambodia and Vietnam but mortality increase in Laos and Thailand. The avoided mortalities are -52, -529 in the dry season, -71, -858 in the wet season, and -124, -1406  all year round in Cambodia, Vietnam, respectively. The additional mortalities are +130, +5483 in the dry season, +101, +3666 in the wet season, and +229, +9131 all year round in Laos, Thailand, respectively. RCP8.5 and ECLIPSE CLE scenarios in combination lead to the largest health impact associated with PM2.5. Laos and Thailand have +312 and +8838 additional mortalities in the dry season, +156 and +4677 additional mortalities in the wet season, and +465 and +13519 additional mortalities all year round. Cambodia and Vietnam have +328 and +1146 additional mortalities in the dry season, and -78 and -786 avoided mortalities in the wet season, and +248, +348 additional mortalities all year round. The mortality increases or decreases in accordance with the increase or decrease of the PM2.5 concentration as for O3 except in the following cases. Even the average national PM2.5 concentrations in Cambodia and Vietnam increase by +0.88 and +3.08 µg m -3 in the wet season under ECLIPSE CLE scenario, +0.09 and +2.52 µg m -3 in the dry season and +0.61 and + 2.11 µg m -3 in the wet season under RCP4.5 and ECLIPSE CLE combined scenario, and +0.68 and +2.69 µg m -3 in the wet season under RCP8.5 and ECLIPSE CLE combined scenario (Table 2), respectively, the corresponding mortalities decrease across these two countries. The large PM2.5 decrease together with high population density in areas in and surrounding PP, HNMR, and HCMCMR (Figs. 3 and 4), results in largely avoided mortalities. The avoided mortalities in Cambodia, Vietnam are -119, -1814 in the wet season under ECLIPSE CLE scenario, -96, -2110 in the dry season and -127, -2314 in the wet season under RCP4.5 and ECLIPSE CLE combined scenario, and -138, -2358 in the wet season under RCP8.5 and ECLIPSE CLE combined scenario, respectively. On the other hand, the increase of PM2.5 concentration in the other areas of the countries where had lower population density produces fewer additional mortalities compared to avoided mortalities. The additional mortalities are +74, +1787 in the wet season under ECLIPSE CLE scenario, +45, +1581 in the dry season and +56, +1455 in the wet season under RCP4.5 and ECLIPSE CLE combined scenario, and +60, +1573 in the wet season under RCP8.5 and ECLIPSE CLE combined scenario in Cambodia, Vietnam, respectively. Thus, the avoided mortalities dominated the estimated net results of -45, -28 deaths in the wet season under ECLIPSE CLE scenario, -52, -529 deaths in the dry season, and -71, -858 deaths in the wet season under RCP4.5 and ECLIPSE CLE combined scenario, and -78, -786 deaths in the wet season under RCP8.5 and ECLIPSE CLE combined scenario in Cambodia and Vietnam, respectively.

Mortality Related to PM2.5
Comparing among target countries, Thailand was estimated to suffer the highest additional mortality resulting from both high population size and PM2.5 increase. Between the two seasons, the excess mortality related to PM2.5 is mostly higher in the dry season than that in the wet season because of a larger change in the future PM2.5 concentration in the dry season (Fig. 3). Comparing to O3, PM2.5 has larger health impacts. One of the potential reasons was that the β coefficient of the PM2.5 related all-cause mortality is higher than that of the O3 related one (Table 3) (Sun et al., 2015). Fig. 6 shows the spatial distributions of annual and seasonal all-cause mortalities related PM2.5. The spatial distributions of PM2.5 related mortality outcomes correspond to those of trends of future PM2.5 concentration changes (Fig. 2). The major health impacts occur in areas where have large future PM2.5 change and dense populations. Having a large increase in PM2.5 concentration together with the dense population, areas in and surrounding BMR showed high additional mortalities except under RCP4.5 scenario. RCP8.5 and ECLIPSE CLE combined scenario produces the largest additional mortality in these areas. The major health benefits would be observed in areas in and surrounding HNMR and HCMCMR except under RCP8.5 scenario in the dry season, resulting from the large reductions in future PM2.5 and the great number of inhabitants that congregate in these areas as mentioned. The highest PM2.5 related avoided death occurs in areas in and surrounding HNMR under RCP4.5 and ECLIPSE CLE combined scenario. In addition, in northern Cambodia and southern Laos with a large increase of PM2.5 in the dry season under RCP8.5 scenario, the health impact is not high. It was because of the low population distribution within the area.

CONCLUSIONS
In this study, BenMAP-CE health impact assessment tool was utilized to quantify O3 and PM2.5 related health impacts due to the changes in climate and emission. The results indicated that health benefit in term of avoided deaths was calculated for RCP4.5 scenario. By contrast, under other scenarios, additional mortalities related to O3 and PM2.5 were mostly found. Change in emission causes higher impacts on mortalities in comparison with change in climate. Between the two seasons, the dry season has more excess mortalities than the wet season. Among four countries, Thailand and Vietnam undergo higher excess mortalities than Laos and Cambodia mainly contributed by the high population size. PM2.5 change contributes greater mortality than O3 change which was consistent with that of Yang et al. (2019).
The BenMAP-CE estimate reliability depends on the precise and relevance of data used in calculations. The input data includes air pollutant concentrations, baseline incidence rates, and estimated populations. The BenMAP-CE results are also affected by used HIFs. Each of them may be uncertain that leads to having impacts on the final estimates of mortality changes. Uncertainties may exist in modeled air pollutant concentrations coming from the uncertainties in the air quality model configurations, formulations, and inputs (Yang et al., 2019). As stated in the previous study (Nguyen et al., 2019a), WRF-CMAQ modeling system that produced the air quality data for this study was evaluated comprehensively. The model performance showed that the model system could reasonably reproduce the observations. The estimated population data and baseline incidence rates in this study were calculated using data from reliable sources. The HIFs used in this study may produce several uncertainties since they were gained from studies that were carried out in a limited number of locations (Yang et al., 2019). The assessment of this study relied on the results from these epidemiological studies because of the absence of robust Asia HIFs for exposure to ambient PM2.5 and O3. Despite these limitations, the results of this study could provide meaningful information for air quality control and public health protection.
The limitation of the health impact analyses of this study was that O3 and PM2.5 related causespecific mortalities were not included because of the lack of health data in the region. The existing HIFs tend to be generic. They are applied commonly to both sexes, and do not differentiate well across age groups, although the effects on these subgroups may not be the same (Sun et al., 2015). In addition, this study just focused on the impacts of climate and emission changes on O3 and PM2.5 related mortalities in the future period. Hence, it did not consider the influences of population growth and future baseline incidence rate on the changes of mortalities related O3 and PM2.5. Therefore, combining these facts in further studies can give more information for understanding the public health impacts and developing proper policies in the region.