Since the Clean Air Act of 1970, atmospheric ozone (O₃) concentrations have declined in the US. Nevertheless, the American Lung Association reported that, as of 2013, over 138 million people in the US (~44%) continue to live in areas where O₃ levels exceed regulatory standards (ALA, 2015). Among common air pollutants that impact public health, O₃ is one of the most detrimental. Risk of O₃-related adverse outcomes is a public health concern due to widespread O₃ exposure, which is ubiquitous in industrialized regions. Research has consistently linked O₃ exposure to a variety of adverse health outcomes including increased emergency room (ER) visits and hospitalizations, asthma exacerbation, cardiovascular stress, impaired lung function, and premature death (Bell et al., 2005; Bell et al., 2007; Bell et al., 2004; Bernard et al., 2001; Jackson et al., 2010; Post et al., 2012; Tagaris et al., 2009; Levy et al., 2005; Gryparis et al., 2004; Stieb et al., 2009; Jerrett et al., 2009). Multiple studies have demonstrated the connections between climate change to O₃ concentrations and these potential health outcomes. For example, Tagaris et al. found the highest climate-induced O₃ increases coincided with the most densely populated areas in the US and increases in national premature mortality of approximately 300 additional deaths annually (Tagaris et al., 2009). Bell et al. also showed that climate change-induced O₃ increases are associated with significant increases in premature mortality and ER/hospital admissions (Bell et al., 2007; Bell et al., 2004). Additionally, by comparing future O₃ concentrations and associated adverse health outcomes from seven published studies, Post et al. showed substantial heterogeneity in the projections when different models and methods were considered (Post et al., 2012). One such example found in this comparison of studies demonstrated a large discrepancy in O₃-related excess mortality due to climate change among the studies examined (ranging from – 600 deaths to over 2500 deaths annually).

The primary drivers of ground-level O₃ generation are precursor emissions (nitrogen oxides (NO_x) and volatile organic compounds (VOCs)), presence of methane, and favorable meteorological conditions (Dawson et al., 2007; Jacob and Winner, 2009; Nolte et al., 2008). Because both emissions and meteorology vary in space, O₃ concentrations can be spatially heterogeneous at the scale of a few kilometers to tens of kilometers (Diem, 2003). Therefore, spatially-resolved estimates of O₃ levels are important when evaluating its potential impact on air quality and human health as well as developing applicable mitigation and adaptation policies. However, as Post et al. reported, the coarse spatial resolution of global climate models (GCMs) cannot resolve the fine-scale features in future O₃ levels (Post et al., 2012).

Both dynamical and statistical downscaling approaches have been developed to address this resolution incongruence. Dynamical downscaling involves executing high-resolution regional climate models (RCMs) and air quality models using GCM outputs as boundary conditions. This method integrates atmospheric chemistry composition, allowing for extrapolation of future atmospheric conditions (Nolte et al., 2008). However, the high computational demand (due to high-resolution, full-chemistry simulations) limits the application to multiple GCM outputs and reduces the availability of these methods (Gao et al., 2013; Gao et al., 2012; Murphy, 2000). Previous studies have used dynamical downscaling methods to study the impact of climate change on future O₃ and air quality. At 36 km resolution, Nolte et al. used dynamical downscaling methods to show significant increases in summer O₃ and a lengthening of the O₃ season under a high emissions scenario as well as substantial decreases during the summer season under a lower emissions scenario (Nolte et al., 2008).

Statistical downscaling methods use efficient statistical methods based on historical atmospheric patterns to relate coarse-resolution GCM simulations to finer grid results, which is much less computationally demanding (Gao et al., 2013). Previous studies have investigated the relationship between O₃ and changes in meteorological conditions using statistical models. For example, Cox and Chu examined 100 meteorological variables for potential effects on ambient O₃, and found that maximum surface temperature, wind speed, relative humidity, mixing layer, and cloud cover were significant. Both Dawson et al. and Camalier et al. found similar statistically significant results showing that daily maximum temperature, relative and absolute humidity, wind speeds, and mixing height greatly affect O₃ concentration (Dawson et al., 2007; Cox and Chu, 1996; Camalier et al., 2007). The limitations of statistical downscaling are mainly due to the assumption that the statistical association between O₃ levels and meteorological conditions will remain the same in the future, which may not be realistic given potential future variations in atmospheric chemistry and emissions (Mahmud et al., 2008).

In addition to air pollution levels estimated at fine spatial scales, the impacts on future O₃ levels due to climate change and future emissions need to be assessed separately for effective mitigation measures. Above all, the impact of air pollution emissions control can have a more immediate effect on air quality and subsequent human health than the effects from slowing down climate change (Fiore et al., 2015). Previously utilized emission scenarios, however, do not allow for such separation of O₃ levels due to climate change and emissions. The latest Representative Concentration Pathways (RCPs) differ from previous emission scenarios such as the Special Report on Emissions Scenarios (SRES) by integrating current and planned environmental policies (IIASA, 2013; Moss et al., 2010; van Vuuren et al., 2011). As a result, RCP-based climate model simulations reflect the combined impact of both climate change and planned emission control on air pollutant levels (IIASA, 2013). This integrated combination provides a platform to develop methods to examine the separate contributions of climate change and emissions. There are multiple RCP scenarios with underlying population growth, economic, and emissions assumptions. RCP2.6, 4.5 and 6.0 all represent some form of improvement upon our current trajectory of growth and environmental policy. RCP8.5, however, represents a “business-as-usual” scenario in which nations choose to retain current economic, environmental, and social tracks. For example, RCP4.5 represents a future scenario with medium to low greenhouse gas emissions, medium-level air pollution, less crop land, and low population growth. RCP8.5, on the other hand, is characterized by high population growth, low to medium crop land use, increasing trends for methane and nitrous oxide, and higher concentrations of almost all air pollutants (van Vuuren et al., 2011).

The objective of this study is to estimate the contribution of climate change and emissions control to future O₃ levels separately at high spatial resolution in the Continental US. We present a hybrid dynamical-statistical downscaling approach to project and separate the impacts of climate change and air pollution emissions control on future O₃ levels under both RCP4.5 and 8.5. Additionally, we expand our analysis and estimate county-level excess mortality due to projected O₃ exposure in the 2050s and evaluate the spatial and temporal patterns of associated estimated health risks. The 2050s were selected for the future projected years based on the IPCC common use of 2050 as a threshold for major global temperature divergence (i.e. potential to rise above 2 °C) (IPCC, 2013).

Our four-step hybrid health impact projection approach is shown in Fig. 1. Step 1 involves a dynamical downscaling framework following two RCPs respectively. This framework is composed of a GCM, a RCM, and an atmospheric chemistry model, which estimates county-level O₃ concentrations in the 2050s due to the combined effects of climate change and environmental policies as described in RCPs. Step 2 develops a statistical downscaling model to estimate future changes in O₃ concentrations from climate change, which uses both real-world historical climate conditions and high-resolution future climate conditions simulated by the RCM in Step 1. Step 3 estimates the future change in O₃ concentrations due to emissions only by subtracting the statistical downscaling results (Step 2) from the dynamical results (Step 1). Finally, in Step 4, the results from Steps 1–3 are placed in a human health context by estimating the future excess mortality due to projected changes in O₃ concentrations.

Our results point to significant differences in the contributions of climate and emissions mitigation to future O₃. Under both RCP scenarios, future emissions control policy will likely have a substantial impact on O₃ levels and its associated health effects. Our hybrid downscaling approach suggests that changes in emissions may be the source of the main incongruities between RCP4.5 and RCP8.5. Thus, while climate change alone may cause some adverse health effects due to poorer air quality, substantial and more immediate health benefits may be achieved by emission mitigation of O₃ precursors regardless of changing climate conditions especially under RCP8.5.

In all RCPs, most emissions of O₃ precursors are expected to decrease in the US due to nearly worldwide implementation of stricter environmental policies. However, the results under RCP8.5 suggest a rise in future O₃ concentrations. It is important to note that, with the climate change effect removed, the predictions still include background O₃ conditions. As is explained in Gao et al. (2013), the O₃ changes in RCP8.5 primarily occurs in spring and winter, especially over the western US, due to increases in global methane emissions (60% by the end of 2050s). In summer and fall, particularly over the eastern US, the increase of methane emissions is offset by the large reduction in anthropogenic VOC and NOx emissions, leading to decrease of O₃ concentrations (Gao et al., 2013). Increases of O₃ in RCP 8.5 were also found across a majority of the troposphere in Young et al. (2013), which attributes the higher ozone concentration to large increase of methane and greater stratospheric influx (Young et al., 2013). The tropospheric ozone increase in RCP 8.5 was shown to enhance background ozone, further impacting regional modeling results through the boundary conditions. This enhanced background ozone can lead to elevated spring ozone in the western US and has also been documented by Lin et al. (2012).

Several recent studies have demonstrated findings comparable to our results regarding the ozone changes in RCP 4.5 and RCP 8.5 (Clifton et al., 2014; Rieder et al., 2015; Yahya et al., 2017). In particular, as is shown in Gao et al. (2013; Fig. 5c), the ozone increases largely disappear when the boundary conditions were cleaner and methane emission increases were removed in RCP 8.5. The effect of global background ozone increase on regional downscaling results through boundary conditions was further shown by a more recent sensitivity study (Yahya et al., 2017, Fig. 7i vs. Figure 7iii), showing clearly that the high ozone boundary conditions inherited from global models under RCP 8.5 contributed to majority of the ozone increases in US in regional model results (Yahya et al., 2017). Thus, the combination of global background ozone increases and methane emission increases may be the main contributing factor for increases in O₃ and O₃-related EM under RCP8.5. Researchers have evaluated and proposed that the control of methane emissions may be an efficient way to reduce both tropospheric O₃ and radiative forcing (Nolte et al., 2008; West and Fiore, 2005; West et al., 2013).

It is important to note that the estimated EM attributable to O₃ changes vary significantly, showing both negative and positive results by region, as seen in Table 1 and Supplemental Tables 2 and 3. Additionally, relatively high standard deviations are captured for most of the regional predictions. These reflect the varying EM from county-to-county within the regions. Further uncertainty is introduced with the CRF values which have been derived from short-term O₃ exposure (robust long-term estimates remain unavailable in the literature) with the assumption that county-specific CRF coefficients are normally distributed. However, despite these limitations, the high-resolution hybrid downscaling system presented here allows the examination of EM at the county level, which is a major strength of our current study. As expected, county-level O₃-related EM is high in counties with higher populations. However, US counties, in general, stand to benefit from emission changes under RCP4.5 scenarios.

Since climate change can have important ramifications in California such as more severe and frequent wildfires and air pollution episodes, it has been the focus of extensive air pollution and climate research (Mahmud et al., 2008; Fujita et al., 2016; He et al., 2016; Fujita et al., 2013). For example, Mahmud et al. performed statistical downscaling methods in the state of California using temperature data from the National Center for Atmospheric Research (NCAR) Reanalysis and 1-hour maximum O₃ values from two ground monitors (Mahmud et al., 2008). The air temperature data have a coarse spatial resolution (2.5° × 4°), which makes it difficult to be directly associated with daily O₃ levels. Instead, linear regression models were developed between 850-hPa air temperature and quartiles of O₃ concentrations, posing an obstacle when linking O₃ exposure with population-specific concentration-response functions. Fujita et al., while using a high resolution (5 km) chose to utilize a chemical box model in which only one parameter was allowed to change at a time. While the methods vary significantly from those used in our study, the general trend of increasing O₃ and emissions is still evident (Fujita et al., 2013). Additionally, He et al. designed a study at a relatively low resolution (30 km) using multiple scenarios: climate change only, emissions only, and a boundary effect. Using CMAQ and the Sparse Matrix Operator Kernel Emissions Model (SMOKE), these scenarios were based on SRES A1B and A1FI. Generally, A1B is similar to the RCP4.5 scenario used in our study in terms of greenhouse gas emission increases and anthropogenic emission decrease, however, A1FI is very different from RCP8.5. In A1fi, all anthropogenic emissions are projected to increase. In RCP8.5, VOC/NOx is actually projected to decrease with large increases in methane leading to consequent rises in ozone levels (He et al., 2016). Taken together, overall conclusions of these case studies are consistent with our results, but the use of hybrid downscaling and the updated RCP scenarios strengthens our current study and lends more insight into the nuances of future ozone changes.

While we have sought to improve on previous methods, some limitations remain. Uncertainties may lie in the estimation of the future mortality rate, CRF, population projection, the potential effect of interactions between temperature and O₃, and O₃ concentration predictions (Brown et al., 2014; Chang et al., 2014; Deser et al., 2012; Henneman et al., 2017; Meehl and Stocker, 2007). We attempt to account for some uncertainty by evaluating EM estimates using a robust Monte Carlo method. We acknowledge that other modeling frameworks have been proposed, however, we deemed our approach reasonable due to sufficient high model performance (Chang et al., 2014; Henneman et al., 2017). Exploration of additional techniques, though a future direction of study, was beyond the scope of this analysis. Another drawback lies in the cubic splines of time used in the model as they may underestimate the contribution of climate change to O₃ concentrations due to the removal of long-term trends. Additionally, the use of county-level resolution, while more precise than previous studies, may still cause some loss in detail of future O₃ predictions. However, it is necessary to keep the resolution of this data consistent with the resolution of the health data for analysis purposes (i.e., health data kept at county level for privacy protection). Future efforts to improve on this analysis could include enhancements in pollutant data collection locations, the addition of co-pollutant effects, and the effects of pollutants on human morbidity.

The results of this study demonstrate that potential increases in premature death and in adverse health effects of climate change-induced O₃ increases in the US may be substantially offset by the effect of emission reductions planned under RCP4.5. However, even with the reduction of O₃ precursors, O₃-related excess mortality may still increase in the US, due to increases in methane emissions under RCP8.5. Thus, with responsible emissions policy, the effects of emission reduction of O₃ precursors is poised to significantly offset the adverse health effects of O₃ due to climate change. To prevent adverse health effects of this potential driver, it is important to continue to intensify mitigation efforts towards both GHGs and O₃ precursor emissions. These efforts are likely to avoid great cost to human health and quality of life.