Response of the ozone-related health burden in Europe to changes in local anthropogenic emissions of ozone precursors

We simulate O₃ and its sensitivity to various precursor emissions in a European domain ('Europe', 32.75–61.25° N, −15–40° E) using a nested-grid capability of the Goddard Earth Observing System (GEOS)-Chem chemical transport model (v9–02, www.geos-chem.org, last access: 11 October 2022) and its adjoint (v35n) [22]. The model is driven by assimilated meteorology from the GEOS-FP of the NASA Goddard Earth Observing System at a resolution of 0.25°× 0.3125°. Model configurations and emissions are described in text S1. We perform six 2 month simulations to generate the BASE results during April to September in 2015. Each simulation includes one month of adjoint forcing and two months of adjoint integration (in order to e.g. capture the influence of emissions in April on O₃ in May), and can be executed in parallel for computational efficiency. Concentrations from the forward model, which is the standard GEOS-Chem simulation, are used as the restart files so that consistent and reasonable initial conditions can be generated. The restart files for each of these simulations come from a continuous 7 month simulation from March to September. We consider the receptor function as the total health burden attributable to O₃ exposure in all European countries listed in the Global Health Data Exchange included in our nested domain. The adjoint simulation calculates the response, or sensitivity, of the health impact receptor function ($J$) to changes in emissions of NO_x , VOCs, CO, sulfur dioxide (SO₂), and ammonia (NH₃) everywhere in our nested domain:

$$\begin{equation}{\lambda _{i,p,t}} = {\nabla _{{E_{i,p,t}}}}J = \frac{{\partial J}}{{\partial {E_{i,p,t}}}}.\end{equation} \tag{ 1 }$$

where ${\lambda _{i,p,t}}$ is the adjoint sensitivity of the receptor function to emissions of species $p$ in model grid cell $i$ in month $t.$

The O₃-related health burden is calculated according to the estimated relative risk for premature deaths due to respiratory diseases in association with the 6 month average (April to September) of the maximum daily 1 h average O₃ concentration (${X_i}$) based on the CPS-II cohort study of Jerrett et al [5]. The receptor function ($J$) can be expressed as:

$$\begin{align}J &= \mathop \sum \limits_a \mathop \sum \limits_k \mathop \sum \limits_{i \in k} {M_{i,a,\,k}}\left( {1 - HR_i^{ - 1}} \right){P_{i,a}}\end{align} \tag{ 2 }$$

$$\begin{align}H{R_i} &= {\text{ex}}{{\text{p}}^{\beta \Delta {X_i}}}\end{align} \tag{ 3 }$$

where ${M_{i,a,k}}$ and ${P_{i,a}}$ are the baseline mortality rate and the population, respectively, for age group $a$ in grid cell $i$ in country $k$. Age composition and baseline mortality rate are obtained from the Global Burden of Disease (GBD) Results Tool (https://vizhub.healthdata.org/gbd-results/, last access: 1 February 2023). Following [27], mortality causes considered here match the International Classification of Disease tenth revision (ICD-10) codes for respiratory disease in the population $\unicode{x2A7E}$30 years of age. Total population is obtained from the fine resolution (∼1 km) population estimate of the Center for International Earth Science Information Network [35]. The hazard ratio $H{R_i}$ in grid box $i$ is calculated according to the exposure response relationship reported in Jerrett et al [5], which suggested a HR of 1.040 (CI: 1.013, 1.067) per 10 ppb increment in ${X_i}$. $\Delta {X_i}$ is the O₃ increment above the theoretical minimum risk exposure level (TMREL) of 33.3 ppb [5], and $\beta$ is the exposure-response factor capturing the log-linear relationship between the health risk from respiratory disease and O₃ exposure.

Given the non-linear O₃-precusor relationship as well as the non-linearity in the exposure response function (section 2.2), we apply a first-order approximation to quantify the response of O₃-related premature deaths in Europe to a 20% change in anthropogenic emissions from each species and sector:

$$\begin{equation}\Delta {J_{i,p,s,t}} = {\lambda _{i,p,t}} \times 20\% {E_{i,p,s,t}}\end{equation} \tag{ 4 }$$

where the contribution ($\Delta {J_{i,p,s,t}}$) resulted from a 20% reduction in emissions (${E_{i,p,s,t}}$) from species $p$, sector $s$, and grid cell ($i$) in month $t$ is calculated according to the adjoint sensitivity (${\lambda _{i,p,t}}$). The accuracy of this first-order estimate in describing the O₃-NO_x -VOCs relationship has been demonstrated by Nawaz et al [20], who compared effectiveness of the standard first-order calculation to that of the second-order approach and finite difference test by examining $\lambda$ changes for different emission levels. Their results suggest the response of O₃ to NO_x emission changes can be well characterized by the first-order calculation. The bias and correlation relative to the response obtained by the forward model perturbation are comparable to those calculated by the second-order approach with ±50% NO_x emission changes. As the O₃-VOCs relationship is less non-linear compared to that of O₃-NO_x [36], we assume the response of O₃-NO_x -VOCs relationships to a 20% change in emissions can be well characterized by the first-order approximation in this study. The conducted sensitivity experiments are displayed in table 1.

In 2015, the total O₃-related health burden in Europe is estimated to be 25 432 premature deaths. This estimate lies in the range of previous estimates (16 000–55 000) for Europe, even with some variability across these studies in terms of the year considered and countries included [26–31]. The uncertainty of our estimate is discussed in more detail in section 3.4. As figure 1(a) shows, large amounts of O₃-related premature deaths occur in coastal and central regions, among which Germany (15.4%), Italy (12.6%), Spain (9.5%), and the United Kingdom (UK, 8.3%) are the countries where the highest number of deaths occur, making up 45.8% of the total O₃-related health burden in Europe. The number of O₃-related deaths is closely associated with the exposure level as well as the respiratory mortality rate in each country (figure 1(b)). For example, the high incidence in Italy is driven by high O₃ exposure near the Mediterranean (figure 1(c)) whereas in the UK it is due to the high baseline of respiratory mortality among the local population. In addition, the health burden scales with population as can be seen in densely populated areas like Benelux (figure 1(d)).

Zoom In Zoom Out Reset image size

Figure 2(a) displays the number of avoided O₃-related premature deaths everywhere in Europe resulting from a 20% decrease in anthropogenic emissions in each grid cell in 2015. We refer to this as the marginal benefit as it approximates the contribution of the last unit of pollution emitted on premature deaths (as distinct from the 'total contribution' corresponding to a 100% reduction in emissions). Note that the marginal benefit is negative when O₃ increases. A 20% reduction in emissions prevents 1576 premature deaths in Europe. Decomposed by species, we estimate marginal benefits of anthropogenic emissions as: NO_x (1,105), VOCs (381), CO (99), SO₂ (18), and NH₃ (−28). The small negative contribution from NH₃ emissions is owing to its role in the chemical sink of NO_x via formation of ammonium nitrate aerosol. As NO_x , VOCs, and CO account for 99% of the positive contributions from anthropogenic emission changes, we focus on these species in the remainder of our analysis. The sign of the contributions from NO_x emission changes largely determines whether reducing anthropogenic emissions has net positive or negative impacts on the O₃-related health burden. In most European countries, NO_x emissions make positive contributions to the O₃-related health burden. O₃ formation in these areas is NO_x -limited, and NO_x emission control would be an efficient approach to reducing O₃-related health risk. The negative NO_x contributions in areas like southern UK, Benelux, and Germany suggest that the O₃ formation regime is NO_x -saturated (VOC-limited), and hence NO_x reductions—while providing health benefits via PM_2.5 decreases [21]—aggravate health risks attributable to O₃ exposure. The boundary between NO_x -limited and VOC-limited conditions shifts in space over time. Consistent with satellite-based estimates of patterns of NO_x vs VOC limited regimes [37], our results (figure S1) show that the area of the negative-NO_x -contribution regions expands as temperature decreases. In southern UK and Benelux, O₃ production is NO_x -saturated all year. In Germany, the O₃ formation regime shifts from NO_x -saturated to NO_x -limited during June to August, indicating that regulating NO_x in summer and VOCs in other seasons would maximize O₃-associated health benefits.

Zoom In Zoom Out Reset image size

Figure 2(b) displays the marginal benefit aggregated to the 27 European Union (EU) member states and UK. Reducing emissions by 20% in France, Italy, and Spain yields the largest health benefits by avoiding, respectively, 250, 211, and 171, O₃-related premature deaths in Europe. In contrast, emission reductions in Belgium, Netherland, and UK make negative contributions to the O₃-related health burden due to the NO_x -saturated conditions (figures 2(a) and S1). The road transport (ROAD) sector constitutes the dominant source of NO_x emissions in most of Europe, contributing 26.9%–80.4% of the total premature deaths that could be avoided by a 20% decrease in NO_x emissions in each country. The marginal benefit of reducing VOC emissions is mostly contributed by the solvent and agriculture livestock (AGLI) sectors, while the industry (INDU), residential (RESI), and ROAD sectors are the leading contributors to the marginal benefit of CO emission reductions. Our results also indicate that NO_x emissions from different sectors within the same country might have opposite impacts on O₃-related health impacts. For instance, in Germany, a 20% reduction in NO_x emissions generally helps to avoid deaths, in particular for emissions emanating from transportation activities. In contrast, NO_x control at German power plants has a negative effect on public health in Europe. As figure S2 shows, hotspots of NO_x emissions from the energy (ENER) sector exhibit a strong spatial overlap with NO_x -saturated areas in Germany, as identified by high NO_x /VOC emissions ratios and relatively high ambient ratios of NO_x to formaldehyde. While transportation generates more NO_x emissions overall, those emissions are concentrated in urban areas where the presence of high VOC emissions prevents NO_x -saturated O₃ regimes. In contrast, high NO_x /VOC ratios in both emissions and ambient concentrations, together with the magnitude of emissions and corresponding negative contributions, suggest that energy emissions are mainly responsible for NO_x -saturated O₃ regimes in Germany. In the other NO_x -saturated regions identified by figure S2(c), especially where the ENER emissions are low (e.g. the Netherlands), shipping and aviation emissions should also be important sources that contributed to O₃ depletion, given the large NO_x /VOC emissions ratios.

According to calculations for the BASE and EMI2005 scenarios, the number of O₃-related premature deaths in Europe (a.k.a. receptor function) falls by about 10% between 2005 and 2015, from 28 312 to 25 432. Despite precursor emissions falling by 19%–32% (table S1), however, the marginal benefit of a further 20% decrease in anthropogenic emissions increases from 1304 avoided premature deaths in 2005 to 1576 in 2015. Subdividing by pollutant and normalizing by absolute emissions reductions yield that the marginal benefit of abating one unit of NO_x more than doubles (+128%) between 2005 and 2015. Following the economic logic of equalizing marginal benefits and marginal abatement costs, this finding provides strong support for tighter regulations of NO_x emissions in Europe. In contrast, marginal benefits per unit of abatement for VOCs and CO decline only slightly over this period (−11% and −9%, respectively), owing to the non-linear nature of O₃ formation.

Figure 3(a) shows the spatial extent of those changes by plotting the response of O₃-related premature deaths to 1 kg km⁻² yr⁻¹ change in NO_x emissions in 2005 and 2015. Negative O₃ response to NO_x emission changes generally weakens in NO_x -saturated areas such as southern UK, Benelux, and Germany, while the positive O₃ response becomes stronger in NO_x -limited areas. The results point to a change in the predominant O₃ formation regime in Europe, in that O₃ formation becomes less sensitive to VOC emissions but more limited by NO_x emissions due to the emission changes during 2005–2015. The anthropogenic emission changes during the studied period are shown in figure S3. The changes in marginal contributions in most NO_x -saturated areas are related to stronger reductions in NO_x emissions than in VOC emissions. In the Netherlands, the lack of VOC emission reductions further weakens the negative O₃ response to NO_x , since the VOC emissions exhibit distinct increases during the studied period. Such VOC increases are related to the intensified agriculture, with the VOC emissions from the AGLI sector increasing over 60% during April to September according to the HTAPv3 emission inventory [34]. In NO_x -limited areas, O₃ formation is not sensitive to VOC emission changes. The increased marginal contribution is thus mostly caused by the large NO_x emission reductions, especially those from the ROAD sector.

Zoom In Zoom Out Reset image size

As figure 3(b) shows, most increases in the total marginal benefit are due to the contribution from NO_x emissions, which increases by over 85% (∼509 deaths) from 2005–2015. This indicates that per-unit NO_x emission reduction yields higher health benefits at lower overall emissions levels. The share of contributions from ground transportation emissions (ROAD and other ground transport (OTTR)) decreases while that of contributions from international aviation emissions increases in each species contribution. The results are consistent with those reported by prior studies that ground transportation emissions are effectively reduced while aviation emissions increase in Europe during the studied period [38, 39]. In tables S1 and S2, we present more details of the marginal benefit and the total anthropogenic emissions in 2005 and 2015. Considering the magnitude and relative changes, the marginal benefits associated with the ROAD, INDU, RESI, and OTTR sectors exhibit the largest increases due to the O₃ chemistry changes, despite the decreased NO_x emissions. The negative-to-positive shifts of the marginal benefit of NO_x emission reductions from the ENER sector indicate that the adverse impacts of the energy-related NO_x reductions on public health decrease owing to the emission changes. As the energy emissions are important sources affecting NO_x -saturated regions in Europe, further emission reductions from energy sources could help to mitigate the negative impacts of emission reductions in these areas and bring increasing health benefits as emissions are further regulated.

Uncertainties in our results arise from two major areas: the CTM-related calculations and the health impact assessment. As covariance between these two aspects is not well known, we treat them independently to estimate total uncertainty.

CTM-related uncertainties arise from uncertainty in estimating O₃ concentrations using the forward model and uncertainty in calculating O₃ source-receptor sensitivities using the adjoint model. These both are related to uncertainties in emissions, meteorology, and the chemical and physical processes represented by GEOS-Chem. A measure of these uncertainties is the accuracy of the model estimated O₃ compared to in-situ measurements provided by the European Environment Agency Air quality e-reporting database. As figure S4 shows, the overall mean bias in the model estimated O₃ concentration (${X_i}$) is approximately +0.27 ppb over the 1794 European monitoring sites, which translates into a slight overestimation (∼1%) in O₃ levels and related health impacts over the domain. Though simulated O₃ concentrations exhibit relatively larger bias in low O₃ areas (observed O₃ concentrations <30 ppb), they show good agreement with observations in most monitoring sites, and the normalized mean bias is within ±10% at over 73% of the sites. Additional uncertainty arises when estimating the marginal benefit changes during 2005–2015, since we only account for impacts of emission changes. This neglects uncertainty owing to interannual variability in meteorology, which also influences the accuracy of O₃ simulation [40–43]. To bound this, our sensitivity results show that the year-to-year meteorology variability causes 2.5%–22.5% changes in the absolute species contributions (figure 4) and −12.0%–+5.3% changes in the proportional species contributions (figure S5). Compared to those (−51.6% to +148.1%) induced by emission changes, the differences are small. The change in O₃ chemistry associate with emission changes is thus the dominant factor influencing the marginal benefit changes. Our adjoint calculations can be numerically verified by comparison to results from the forward model perturbation (FWD_TEST), where a 20% decrease in anthropogenic emissions avoids 1779 and 1415 premature deaths in 2015 and 2005, respectively. As the adjoint results depend largely on emissions, we discuss possible impacts of the uncertainty in the emission inventory in text S2.

Zoom In Zoom Out Reset image size

Health assessment-related uncertainties arise from uncertainty in estimates of population, baseline mortality rate, and the exposure response relationship. We adopt the health assessment approach from Jerrett et al [5] rather than a later method from Turner et al [6] since the latter is based on annual average O₃ exposures. As we consider the marginal benefit over a large European region, the annual total contribution would be mitigated by enhanced negative ones in winter due to a shift in the spatial extent of different O₃ formation regimes (figure S1). There is also greater certainty in all-cause mortality impacts of peak O₃ exposure compared to annual [44]. We thus only focus on the O₃ pollution season from April to September. The uncertainty bounds of the population and mortality rate have been discussed in Gu et al [21], where the former is 1.9%–11.3% (population-weighted) in total European population in 2015 and the latter are provided explicitly by the GBD results. Using those bounds and ranges for the hazard ratio (1.013–1.067) [5], we estimate the total number of O₃-related premature deaths from respiratory diseases ($\unicode{x2A7E}$30 years of age) to be 25 432 (7356–53 160) over Europe in 2015. The uncertainty (−71%–+109%) associated with the health assessment is still the major source of uncertainty, and thus we report these ranges in the summaries of our findings. To quantify the impacts of population and baseline mortality rate, we find that demographic changes alone under conditions of constant exposure and meteorology lead to a decrease in NO_x and VOC contributions by −4.3% and −11.8%, respectively in 2005, while the source attribution results are only slightly affected (figure S6). We consider the lower bound of TMREL (33.3 ppb) in Jerrett et al [5]. Given there's limited evidence that a 'safe' O₃ threshold exists below which there's no health risk, our estimates might still underestimate the absolute health burden in Europe. However, as the 6mDM1h O₃ concentrations are all above the TMREL over our receptor region, the health assessment-related uncertainties would only greatly affect the absolute values of the estimated premature deaths. The relative contributions of different species and sectors are largely determined by the CTM calculations. Thus, our relative results of the source attribution should be generally representative, with uncertainties close to those of the CTM calculations.