Introduction

Ambient ozone (O₃) posed a significant threat to human health and the ecological environment worldwide, particularly during warmer days1,2,3,4,5. Several countries, including China, the United States (US), Italy, and Japan, have established standards for ambient O₃ concentrations to improve air quality6. The World Health Organization’s (WHO) latest air quality guideline (AQG), published in 2021, recommended that the maximum daily 8-hour averaged (MDA8) ambient O₃ concentrations should not exceed 100 μg m−3 or 50.11 parts per billion (ppb) for short-term exposure. Additionally, the AQG of warm-season MDA8 ambient O₃ concentrations (the maximum consecutive 6-month average) was suggested as 60 μg m-3 or 30.07 ppb for long-term exposure. In the last few years, global ambient O₃ concentrations have shown an upward trend7,8,9,10. It was estimated that O₃-attributable mortality due to chronic respiratory disease (CRD) increased by 46% from 2000 to 2019 worldwide11. Furthermore, global exposure to ambient O₃ resulted in 0.470 million (95% confidence interval [CI]: 0.100, 0.818) and 0.423 million (95% CI: 0.223, 0.659) deaths from chronic obstructive pulmonary disease (COPD)12 and CRD13 in 2019, respectively. Therefore, it is crucial to accurately and continuously monitor global ambient O₃ distribution and assess O₃-related health risks.

In-situ measurements can provide accurate and reliable ambient O₃ concentrations for estimating exposure levels and associated health impacts14. However, in situ stations were typically located in cities15,16, which limited their representativeness for peri-urban and rural areas. In contrast, the chemical transport model offered an alternative method for simulating ambient O₃ concentrations over large areas17,18. Unfortunately, it required extensive computational resources and relied on temporally lagging emission inventories with unexpected biases over local regions. These factors can significantly increase the time consumed for simulations and result in high uncertainties into the model output.

To date, data fusion algorithms based on statistical or machine learning (ML) techniques7,8,9,10,11,12,13,15,19,20,21,22,23,24,25 have been emerging as preferred methods to acquire ambient O₃ distribution. For example, in national/regional studies, a geographically weighted regression model was proposed by Zhang et al. 19 to acquire monthly 0.25-degree ambient O₃ concentrations over Eastern China. Li et al. 20 developed an enhanced geographically and temporally weighted neural network for obtaining high-resolution (0.05°) MDA8 ambient O₃ concentrations across the Greater Bay Area in China. A daily ambient O₃ dataset at spatial resolution of 0.1° over China was established by Wei et al. 21 based on a space-time extremely randomized trees model. Wang et al. 22 employed the random forest model to generate 0.1-degree MDA8 ambient O₃ concentrations over California in the US. The light gradient boosting machine was adopted by Chen et al. 23 for estimating MDA8 ambient O₃ concentrations (0.1°) in Europe, which also analysed O₃ exposure with other air pollutants.

Regarding globe-scale works7,8,9,10,11,12,13,24,25, a M3Fusion model and its improved version with Bayesian maximum entropy (M3Fusion-BME) were devised by Chang et al. 7 and DeLang et al. 10, respectively, to establish warm-season 0.1-degree ambient O₃ datasets over the globe. Liu et al. 8 mapped globally distributed monthly ambient O₃ exposure at spatial resolution of 0.5° by developing a cluster-enhanced ML model. The random forest model was employed in Xu et al. 24 for estimating daily 0.25-degree ambient O₃ concentrations from landscape fire, which then systematically analysed relevant population exposure worldwide. Sun et al. 9 proposed a spatiotemporal Bayesian neural network to generate global monthly ambient O₃ dataset (0.1°) and assessed O₃-attributable mortality burden for respiratory diseases in 2010. Based on existing modelling algorithms, the global burden of disease (GBD) 202112 discussed the health risks of ambient O₃ for COPD across 204 countries and regions through a M3Fusion-BME model10. The M3Fusion-BME model10 was also introduced by Malashock et al. (2022)11,13 to assess O₃-attributable mortality burden for CRD in cities and peri-urban/rural areas globally. Xue et al. 25 investigated the exposure-response function between O₃ exposure and children mortality in 55 low-income and mid-income countries using a cluster-enhanced ML model8.

Data fusion algorithms based on statistical or ML techniques7,8,9,10,11,12,13,15,19,20,21,22,23,24,25 have significantly advanced the estimation of population exposure to ambient O₃ and its associated health risks. For national/regional studies, localized methods with individual O₃ modelling strategies15,19,20,21 generally outperformed the holistic models that utilized all in situ stations simultaneously. The superior performance was attributed to the pronounced local heterogeneity of ambient O₃ distribution, driven by spatiotemporally varying surface emissions of O₃ precursors5,26 and meteorological conditions27,28. Furthermore, the established ambient O₃ datasets mostly achieved high spatial (e.g., 0.05°) and temporal (e.g., daily) resolutions in national/regional studies. Conversely, the globe-scale works typically utilized the holistic models to acquire ambient O₃ concentrations, often with coarse spatial (e.g., 0.25°) or temporal (e.g., monthly) resolutions. As for health impacts, they mainly aimed at assessing long-term O₃ exposure referencing part of WHO standards. Meanwhile, the globe-scale works usually concentrated on investigating O₃-related health risks for specific diseases, selected countries, or single source.

Previous studies7,8,9,10,11,12,13,15,19,20,21,22,23,24,25 have yielded robust results and significantly contributed to society. However, the severity of current global O₃-related health risks still remained underappreciated in the globe-scale works7,8,9,10,11,12,13,24,25, which exhibited considerable shortcomings. To be specific, the geospatially local apriority of ambient O₃ distribution has not been adequately incorporated into global modelling, unlike in national/regional studies, potentially introducing greater uncertainties in the modelled results. In the meantime, previous globe-scale works generally adopted statistical or ML models with coarser spatial (e.g., 0.25°) or temporal (e.g., monthly) resolutions compared to national/regional studies. This likely led to cumulative errors or defects in the estimation of O₃ exposure levels and mortality burden29,30. For instance, triple coarser spatial resolution (e.g., 36 versus 12 km) might cause biased (> 10 %) O₃-attributable mortality burden in national regions30. The coarse temporal resolution (e.g., monthly) also cannot support the assessment of short-term population exposure to ambient O₃. Moreover, these globe-scale works have not sufficiently explored the connection between short- and long-term O₃-related health risks with multiple standards from WHO. Both short- and long-term health impacts should be considered to evaluate the acute and chronic risks due to O₃ exposure over the globe. Importantly, previous globe-scale works simply focused on limited conditions (e.g., diseases, countries, or O₃ sources), which might substantially underestimate current worldwide O₃-attributable mortality burden.

In our study, the objectives are threefold: to propose a geospatially dynamic ensemble ML model (global-local coupled ensemble forest, GL-CEF) (1) for the global modelling of daily seamless high-resolution (0.05°) ambient O₃ concentrations; to provide an in-depth assessment of global O₃ pollution, with prominent focuses on short-/long-term population exposure (2) and all-cause mortality burden (3). It is crucial to incorporate geospatially local apriority into the model due to the pronounced local heterogeneity in ambient O₃ distribution. Nevertheless, the spatial locations of in situ stations were highly nonuniform across the globe, with sparse or even nonexistent stations in many regions. To fix this issue, we devise three modules (global, local, and global-local) in the GL-CEF model for the estimation over station-sparse (or no-station) and station-dense regions, exhibiting superior performance to popular holistic ML models. The GL-CEF model can generate consistent global ambient O₃ distribution with previous globe-scale works7,8,9,10,11,12,13,24,25, but performed at higher resolutions (daily and 0.05°). The high resolutions result in data volumes 25 to 1,460 times larger per year (yr-1) compared to them, offering richer spatiotemporal details. By combining all three objectives, our study investigates current O₃-related health risks worldwide, considering the latest WHO standards and land use disparities. The ambient O₃ standards from WHO include the AQG, interim target 2 (IT2), and interim target 1 (IT1), with the values of 50.11/30.07, 60.13/35.08, and 80.18/50.11 ppb for short-/long-term O₃ exposure, respectively. Our ambient O₃ dataset supports for fine-scale assessment of health risks from both short- and long-term O₃ exposure, potentially reducing cumulative errors in global analyses. Our findings reveal that O₃-related health impacts might have been underappreciated and identify key polluted regions over the globe. Therefore, our study may have implications for investigating global O₃ pollution and benefiting the O₃-exposed population in the future.

Results

Multiscale model performance

Figure 1a, b depict the performance of GL-CEF model based on two cross validation schemes, which include the space-informed cross validation (SICV) and temporally extrapolated SICV (TESICV)15,16,20,21,29. The collocated samples are sufficient, 3.67 million for SICV and 1.15 million for TESICV, providing reliable validation metrics. All metrics are computed under the significance levels of p < 0.01. At daily scale, the GL-CEF model demonstrates satisfactory performance, with the coefficient of determination (R2) of 0.87 and 0.73 for SICV and TESICV, respectively. At monthly (warm-season) scale (Supplementary Fig. 1a–d), the model errors decrease further, with the root mean square error (RMSE) improving by 2.200 (2.624) ppb for SICV and 4.537 (5.192) ppb for TESICV compared to daily scale. Additionally, the GL-CEF model generally yields favourable accuracy on collocated grids that involve in situ stations (Fig. 1c–f), with 82.59%/50.54% and 87.97%/81.26% of them showing the R2 of > 0.7 and mean bias of < ± 5 ppb for SICV/TESICV, respectively. Meanwhile, the collocated grids with high R2 ( > 0.8) and near-zero mean bias are mostly located in densely populated regions, such as China, Europe, and the US. This indicates that the global MDA8 ambient O₃ concentrations established by the GL-CEF model are of both significant quality and practical utility.

Fig. 1: Overall model performance of the GL-CEF model.
figure 1

a, b Density scatter plots of cross-validation results and c–f spatial distribution of metrics on collocated grids worldwide at daily scale. Black dashed and red solid lines stand for 1:1 and fitted lines in a, b, respectively. Colour bars denote the normalized densities of data pairs in a, b and the values of metrics in c–f. Unit for RMSE and mean bias: ppb. Definitions of acronyms: GL-CEF (global-local coupled ensemble forest), N (number of collocated samples), SICV (space-informed cross-validation), TESICV (temporally extrapolated space-informed cross-validation), and RPE (relative percentage error).

Full size image

Global population exposure to ambient O₃

The spatial patterns of annual MDA8 ambient O₃ are generally similar to those observed during the warm season (Fig. 2a, b). However, significant regional differences still exist. For example, there is much higher MDA8 ambient O₃ during the warm season than the annual levels over Africa and southern Asia, likely due to massive surface emissions of O₃ precursors from biomass burning31,32,33. Meanwhile, the high values of MDA8 ambient O₃ are also observed in southern Europe and the western US during the warm season, exceeding annual levels. At country scale (Fig. 2c, d), the large population-weighted MDA8 ambient O₃ (> 48 ppb) is predominantly found in Asia and Africa, particularly in northern regions of middle latitudes. By contrast, the population-weighted MDA8 ambient O₃ is smaller (<32 ppb) across the countries of South America and Oceania.

Fig. 2: Maps and statistics of current global ambient O₃ distribution.
figure 2

a, b Annual and warm-season globally gridded, c, d country-scale population-weighted, and e–j population-weighted statistics for typical countries of MDA8 ambient O₃ concentrations. Colour bar stands for the values of ambient O₃ concentrations (unit: ppb) in a–d. Light grey denotes the unavailable regions with snow, ice, and lake for the GL-CEF model in a–d. The black and red dashed lines represent population-weighted MDA8 ambient O₃ concentrations worldwide for annual (42.39 ppb) and warm season (49.76 ppb) in e–j, respectively. The vertical lines indicate the ranges of one standard deviation in e–j. The countries that present significantly larger (difference > 10 ppb) and smaller (difference < −10 ppb) population-weighted MDA8 ambient O₃ concentrations than the global average are marked with red up arrow and blue down arrow in e–j, respectively. Definitions of acronyms: PW (population-weighted) and SA (South Africa).

Full size image

Globally, the population-weighted MDA8 ambient O₃ are 42.39 ± 11.29 and 49.76 ± 13.75 ppb for annual and warm-season levels, respectively (Fig. 2e–j). In some Asian countries, annual and warm-season population-weighted MDA8 ambient O₃ are distinctly higher than the global average (difference > 10 ppb), likely due to sufficient solar radiation intensity combined with favourable meteorological conditions27,28 and surface emissions of O₃ precursors5,26. By contrast, a few countries in South America and Oceania exhibit much lower population-weighted MDA8 ambient O₃ compared to the global average (difference < −10 ppb), indicating milder O₃ pollution. Notably, severe O₃ pollution can occur at country scale for brief periods (Supplementary Fig. 7). For instance, the population-weighted MDA8 ambient O₃ is observed extremely high over China and South Korea from March to May (MAM), which surpasses the IT1 standard (i.e., > 80.18 ppb).

Four land use classes (Natural Vegetation, Cropland, Urban, and Bareland) are adopted to assess the population exposure to ambient O₃ (Fig. 3). Natural Vegetation includes forests, shrublands, savannas, and grasslands (see Supplementary Table 4 for details). The total (three) O₃ interval (intervals) represents (represent) > AQG (AQG-IT2, IT2-IT1, and > IT1). Regarding short-term O₃ exposure (Fig. 3a–f), 66.2%/54.9%/47.5% of the global population is exposed to excess ambient O₃ for more than 30/60/90 days yr-1. The global O₃ exposure patterns are mainly influenced by those of Asia (large population), with similar land use fractions on various O₃ exposure intervals (e.g., large contribution from Cropland). In North America (Africa), the population exposed to O₃ pollution primarily lives in Urban (Natural Vegetation and Cropland) areas, especially for the MDA8 ambient O₃ exceeding IT1. Europe and South America both exhibit relatively low O₃ exposure levels, predominantly within the AQG-IT2 interval. The O₃ pollution is very mild in Oceania and consequently not analysed in detail. As for long-term O₃ exposure (Fig. 3g–m), 94.3% of the global population experiences excess ambient O₃, with nearly half (45.1%) exposed to severe O₃ pollution (> IT1). Asia again dominates the global O₃ exposure patterns, generally presenting consistent land use fractions across different O₃ exposure intervals. Specifically, the fraction of Cropland increases with higher O₃ pollution levels (Fig. 3g, h), which likely suggests that more people suffer from high MDA8 ambient O₃ during the warm season11,13,34. Europe, North America, and Africa exhibit similar O₃ exposure patterns, though varying by land use disparities. Severe O₃ pollution (> IT1) is not observed over South America and Oceania, indicating low O₃ exposure levels there.

Fig. 3: Current population exposure to ambient O₃ worldwide.
figure 3

a–f Population fractions divided by land use classes falling on total (> AQG) and three short-term O₃ exposure intervals (AQG-IT2, IT2-IT1, and > IT1). g–m Population fractions divided by land use classes falling on total (> AQG) and three long-term O₃ exposure intervals (AQG-IT2, IT2-IT1, and > IT1). T1, T2, and T3 represent the thresholds of more than 30, 60, and 90 days yr−1 at the right Y-axis in (a-f), respectively. Definitions of acronyms: AQG (air quality guideline, 50.11 ppb for short term and 30.07 ppb for long term), IT2 (interim target 2, 60.13 ppb for short term and 35.08 ppb for long term), IT1 (interim target 1, 80.18 ppb for short term and 50.11 ppb for long term), and NV (Natural Vegetation).

Full size image

Overall, the population amount exposed to short-term ambient O₃ gradually declines with the O₃ pollution increasing (Fig. 3a–f). However, the O₃ exposure interval of IT2-IT1 involves the largest global population (20.1%) for the duration of > 90 days yr-1 (Fig. 3a). This implies that more people worldwide are exposed to the MDA8 ambient O₃ within the IT2-IT1 interval for a long period, likely leading to higher short-term mortality risks. On the other hand, the areas of polluted regions for long-term O₃ exposure on IT2-IT1 (near 70%) distinctly exceed that on > IT1 (less than 20%) globally (Supplementary Fig. 5g, h). Nevertheless, more people (45.1%) are exposed to the warm-season MDA8 ambient O₃ of > IT1 (Fig. 3g), which also potentially results in higher long-term mortality risks.

Global O₃-attributable mortality burden

Our study adopts a log-linear exposure-response function to obtain the global O₃-attributable mortality burden, based on the pooled relative risks of short- and long-term O₃ exposure for all-cause deaths from two worldwide epidemiological researches35,36. We estimate that the short- and long-term population attributable fraction (PAF) are 3.42 × 10−3 (95% CI: 2.71 × 10−3, 4.12 × 10−3) and 0.0272 (95% CI: 0.176, 0.365) globally during 2019–2021 (Supplementary Fig. 5i, j), respectively. This indicates that 0.34% (95% CI: 0.27%, 0.41%) and 2.72% (95% CI: 1.76%, 3.65%) of the global total deaths are attributed to O₃ pollution. The long-term PAF substantially exceeds that of the short-term (eight times), suggesting its much higher mortality risks. This is potentially attributed to that short-term O₃ exposure may stimulate the production of antioxidant enzymes, which is deemed as a compensatory mechanism37. The upregulation of antioxidant enzymes can decrease oxidative stress and is likely related to the suppression of proinflammatory responses. Conversely, long-term O₃ exposure probably engenders the upregulation of internal redox homoeostasis, which will result in an increment of systematic inflammation37.

The total (three) exposure interval (intervals) stands (stand) for > AQG (AQG-IT2, IT2-IT1, and > IT1) in Fig. 4. For short-term O₃-attributable mortality (Fig. 4a–d), the high values of PAF ( > 3 × 10-3) are predominantly observed in Asian countries of middle latitudes (30°N), while other countries generally present relatively smaller PAF ( < 2 × 10-3). Within different O₃ exposure intervals, the high values of country-scale PAF ( > 3 × 10-3) typically emerge on IT2-IT1. Notably, China also demonstrates large PAF (3.46 × 10-3 [95% CI: 2.75 × 10-3, 4.17 × 10-3]) for the > IT1 interval, indicating more frequent occurrences of extremely high MDA8 ambient O₃ (i.e., > 80.18 ppb) than other countries. Furthermore, the four land use classes (Natural Vegetation, Cropland, Urban, and Bareland) are applied to assess the O₃-attributable mortality burden (Supplementary Fig. 5i). Cropland shows the largest total PAF (4.68 × 10−3 [95% CI: 3.71 × 10−3, 5.64 × 10−3]), reflecting significant mortality risks due to heavy O₃ pollution. This is likely because surface emissions of biogenic volatile organic compounds (O₃ precursor) from vegetation and carbon monoxide (which contributes to radical formation) from incomplete biomass burning are likely pronounced over Cropland areas34. Additionally, nitrogen oxides (another O₃ precursor) can also transport from Urban to Cropland areas34. Favourable solar radiation intensity further accelerates photochemical reactions, leading to the accumulation of ambient O₃ concentrations5,26,27,28. Among three O₃ exposure intervals, the distribution of PAF based on land use classes is diverse. Generally, the PAF on IT2-IT1 is larger compared to other O₃ exposure intervals, which results from its higher short-term O₃ exposure levels over the globe (Fig. 3a).

Fig. 4: Current O₃-attributable mortality burden worldwide.
figure 4

a–d Country-scale PAF for total (> AQG) and three short-term O₃ exposure intervals (AQG-IT2, IT2-IT1, and > IT1). e–h Country-scale PAF for total (> AQG) and three long-term O₃ exposure intervals (AQG-IT2, IT2-IT1, and > IT1). The short- and long-term gridded PAF with more details at a high spatial resolution can be referred to in Supplementary Fig. 8. Colour bars stand for the values of PAF. Definitions of acronyms: AQG (air quality guideline, 50.11 ppb for short term and 30.07 ppb for long term), IT2 (interim target 2, 60.13 ppb for short term and 35.08 ppb for long term), and IT1 (interim target 1, 80.18 ppb for short term and 50.11 ppb for long term).

Full size image

As for long-term O₃-attributable mortality (Fig. 4e–h), the high values of country-scale PAF ( > 0.03) also primarily concentrate in Asia of middle latitudes (30°N), which signifies the substantial mortality risks from both short- and long-term O₃ exposure over these regions. Other countries generally show moderate PAF of > 0.01. Among three O₃ exposure intervals, the high values of country-scale PAF ( > 0.03) are predominantly observed for > IT1. By contrast, most countries exhibit significantly small PAF on AQG-IT2 ( < 1 × 10-3), suggesting that the ambient O₃ standard of long-term IT2 (35.08 ppb) from WHO may need to be revised upward, which has shown a similar effect with AQG (30.07 ppb) across various countries. Moreover, Cropland and Bareland both present large total PAF (0.0332 [95% CI: 0.0216, 0.0446] and 0.0355 [95% CI: 0.0230, 0.0477]) (Supplementary Fig. 5j). However, excess ambient O₃ is more hazardous for Cropland worldwide due to its larger population amount (2.885 billion) compared to Bareland (0.114 billion). Concerning various O₃ exposure intervals, the PAF gradually increases with O₃ pollution rising, since more people are exposed to higher warm-season MDA8 ambient O₃ over the globe (Fig. 3g).

We estimate that a total of 0.177 million yr-1 (95% CI: 0.139, 0.214) and 1.407 million yr-1 (95% CI: 0.909, 1.896) deaths globally stem from short- and long-term O₃ exposure, respectively (Supplementary Table 5). The mortality burden attributed to long-term O₃ exposure distinctly exceeds that from short term. Meanwhile, most O₃-attributable deaths due to short- and long-term exposure occur in the IT2-IT1 (53.01%) and > IT1 (73.04%) intervals, respectively. This aligns with the above analyses of O₃ exposure. Notably, the polluted regions on IT2-IT1 for short-term exposure present a similar distribution with that on > IT1 for long-term exposure (Supplementary Fig. 5c, h). Considering the top-10 countries simultaneously (Supplementary Table 2, 3), we find that Asia of middle latitudes (30°N) and the western US experience high mortality burden attributed to short- and long-term O₃ exposure.

Discussion

This study provides an assessment of current global O₃ pollution. We propose a geo-ensemble ML model (GL-CEF) to acquire MDA8 ambient O₃ concentrations worldwide. The establishment of global ambient O₃ dataset integrates remote sensing observations, chemical apriori data, meteorological fields, geographic elements, and in-situ measurements from > 7000 stations across more than 100 countries. The GLC-CEF model is rigorously validated and stably displays global daily seamless high-resolution (0.05°) patterns of O₃ pollution during 2019–2021, achieving the R2 of 0.87 and 0.73 for SICV and TESICV, respectively. Based on the modelled dataset, we characterize short- and long-term O₃ population exposure and their associated all-cause mortality burden globally, with consideration of the latest ambient O₃ standards from WHO and land use disparities. The assessment of population exposure and mortality burden is conducted at various scales (the globe, continent, country, and grid), with results that corroborate across different levels.

Our study improves the modelling algorithm to derive global ambient O₃ concentrations. The spatial distribution of ambient O₃ generally presented strong local heterogeneity, making it essential to incorporate geospatially local apriority into the model. Nevertheless, there were sparse or even no in situ stations established in plenty of regions worldwide, which rendered previous localized methods15,19,20,21 challenging for global modelling tasks. To address this issue, the developed GL-CEF model comprises three components: global module (employing global features), local module (incorporating geospatially local apriority), and global-local module (connecting global and local modules). Notably, the local module consisted of two strategies: sliding block strategy (to accelerate model convergence) and variable local sub-model (to improve model robustness). The GL-CEF model can adopt global and local knowledge over station-sparse (or no-station) and station-dense regions, respectively. Additionally, we introduce a spatiotemporal-based imputation method for recovering missing information in the model inputs based on their self-correlation. Validation results demonstrate that the GL-CEF model achieves favourable accuracy, with the mean bias of −0.03 and −1.41 ppb for SICV and TESICV, respectively. The performance of the GL-CEF model also surpasses those of widely-used holistic ML models (Supplementary Fig. 2), such as random forest and light gradient boosting machine.

Using the GL-CEF model, we generate a daily spatiotemporally continuous 0.05-degree dataset for assessing short- and long-term O₃-related health risks. This dataset exhibits improved performance (an increment of 0.39 in R2) and spatial resolution five times higher than the GEOS-CF product from the National Aeronautics and Space Administration (NASA) (Supplementary Fig. 3). The modelled dataset captures the seasonality of current ambient O₃ worldwide, showing high values during the warm season over Asia, Africa, and North America, which is consistent with previous reports7,8,9,10,16. In addition, the ambient O₃ concentrations from previous globe-scale works7,8,9,10,11,12,13,24,25 had coarse spatial (e.g., 0.25°) or temporal (e.g., monthly) resolutions (Supplementary Table 1), potentially introducing spatially or temporally cumulative errors in final analyses29,30. Conversely, our study considers O₃ precursors3,5,26 and employed high-resolution (~ 5 km) remote sensing observations as key inputs8,19,20,21,22 of the model, which include nitrogen dioxide (NO₂) and formaldehyde (HCHO) from the tropospheric monitoring instrument (TROPOMI). By integrating various multisource datasets, the modelled dataset achieves spatial and temporal resolutions of 0.05° and daily, respectively. The high resolutions lead to massive data volumes, ranging from 25 to 1460 times yr-1 compared to previous globe-scale works7,8,9,10,11,12,13,24,25, which provide richer spatiotemporal details and information.

Our study refines the land use disparities in global O₃-related health risks assessment. Land use disparities potentially signified different sources of ambient O₃38,39, which affected O₃ exposure levels and mortality burden worldwide. Previous reports9,11,13,34 implied that O₃-related health risks were generally higher in peri-urban or rural areas compared to those in cities. We further divide global land use disparities into four classes, including Natural Vegetation, Cropland, Urban, and Bareland. Although more people experience short- and long-term O₃ exposure in Cropland areas over Asia, the population exposed to O₃ pollution primarily reside in Urban areas over North and South America. In addition, a large amount of population suffers from short- and long-term O₃ exposure in Natural Vegetation areas over Africa. We reveal the geographically diverse impacts of land use disparities on O₃-related health risks globally.

Our study increases the understanding of ambient O₃ standards in the context of global O₃-related health risks assessment. The ambient O₃ standards from WHO were applied in some previous globe-scale works8,11,24,25. Liu et al. 8 reported that 37.13% of global population lives in the regions beyond the long-term IT1 suggested by WHO during the warm season. The short-term WHO AQG was adopted in Xu et al. 24 to define the events of substantial fire-sourced air pollution. Malashock et al. 11 implied that the number of cities with warm-season MDA8 ambient O₃ of > long-term AQG from WHO increased by 865 from 2000 to 2019. A threshold value near the long-term WHO IT1 was designed by Xue et al. 25 for obtaining the non-linear exposure-response function of O₃ and under-5 mortality. However, these globe-scale works only considered partial WHO standards and did not account for the different O₃ exposure levels and mortality burden defined by AQG, IT2, and IT1. Additionally, they focused solely on either short- or long-term WHO standards, without exploring the connection between short- and long-term O₃-related health risks.

By contrast, our study provides an assessment of both short- and long-term health risks associated with current O₃ pollution globally, with consideration of all ambient O₃ standards from WHO. We estimate that a large amount of global population (20.1%) is exposed to the MDA8 ambient O₃ of between IT2 and IT1 standards for more than 90 days yr-1, likely leading to higher mortality risks from short-term exposure. Meanwhile, nearly half of the people worldwide (45.1%) are exposed to the warm-season MDA8 ambient O₃ exceeding the IT1 standard, which also potentially results in elevated mortality risks due to long-term exposure. As a consequence, the majority of O₃-attributable deaths from short- and long-term exposure are associated with the IT2-IT1 (53.01%) and > IT1 (73.04%) intervals, respectively. Importantly, this analysis reveals that mid-latitude Asia (30°N) and the western US experience high mortality burden due to short- and long-term O₃ exposure, likely driven by intense solar radiation, favourable meteorological conditions27,28, surface emissions of O₃ precursors5,26. Our study identifies the similar key polluted regions at grid level for global short- and long-term O₃-related health risks.

A highlight of our study is the assessment of both short- and long-term exposure levels and all-cause mortality burden due to current all-source O₃ pollution across > 200 countries or regions, rather than focusing on specific diseases, limited countries, or single-source O₃. Especially for specific diseases, previously reported global O₃-attributable mortality burden primarily involved respiratory diseases, such as CRD and COPD9,11,12,13,17,18,40. The O₃ exposure could stimulate the respiratory system to generate a large amount of inflammatory cell hormones and then accumulate toxic lipid oxidation products, which finally engendered respiratory diseases. Nevertheless, a recent review41 has provided the epidemiological evidence and biological mechanisms linking O₃ exposure to cardiovascular diseases. Specifically, the O₃ exposure could trigger chain reactions including respiratory and systemic inflammation, oxidative stress, disruption of autonomic nervous and neuroendocrine systems, impairment of coagulation function, glucose, and lipid metabolism. These reactions can ultimately lead to vascular dysfunction and the progression of cardiovascular diseases. Furthermore, a number of existing meta-analyses have indicated that the mortality of all-cause diseases was positively associated with short-35,42,43 and long-term36,44,45 O₃ exposure globally. Therefore, focusing solely on respiratory diseases might significantly underestimate the broader mortality burden associated with ambient O₃ worldwide, which ignored potential influences from other diseases.

Specifically, we estimate that 1.407 million yr-1 (95% CI: 0.909, 1.896) all-cause deaths are attributed to long-term O₃ exposure globally, which substantially exceeds estimates from the GBD 202112 (COPD, 0.470 million [95% CI: 0.100, 0.818]) and Malashock et al. 11,13 (CRD, 0.423 million [95% CI: 0.223, 0.659]). Nevertheless, Malley et al. 17 and Chowdhury et al. 18 claimed that 1.04 million (95% CI: 0.72, 1.37) and 1.30 million (95% CI: 0.93, 1.68) respiratory deaths were attributed to long-term O₃ exposure worldwide in 2010 and 2015, respectively. The O₃-attributable mortality burden from respiratory diseases might be overestimated in these two globe-scale reports17,18. This was mainly because they adopted chemical transport models (e.g., GEOS-Chem) with large errors against in-situ measurements worldwide (e.g., a mean bias of 10.8 ppb in Malley et al. 17), which also had coarse spatial resolutions (e.g., 2.5°). Conversely, the global ambient O₃ dataset established via our model is performed at high spatial resolution of 0.05°, achieving the mean bias of less than -1.5 ppb, suggesting more reasonable results of O₃-attributable mortality burden. In addition, we find that global short-term O₃ exposure results in 0.177 million yr-1 (95% CI: 0.139, 0.214) all-cause deaths. By adopting a comprehensive perspective, our assessment reveals that the severity of current worldwide O₃-related health risks is fairly greater than previously recognized from specific diseases.

However, there are some limitations in this study. The global module in the GL-CEF model is trained with all collocated samples over station-sparse regions, where only a few in situ stations can be applied in the validation. This probably leads to insufficient representativeness of the validation results compared to other regions. Meanwhile, our study is restricted by the availability of satellite observations from TROPOMI (operating since 2018), preventing the comparison with the historical sequence (e.g., 2000-2018). Additionally, our study assumes even personal O₃ exposure on each grid, likely resulting in misclassification bias due to the population migration and diverse time patterns of anthropogenic activities (e.g., time spent outdoors versus indoors). Moreover, the calculation of premature deaths uses a log-linear exposure-response function between ambient O₃ concentrations and premature deaths from epidemiological researches35,36. Limited by existing meta-analyses, it is challenging to account for all conditions (e.g., ages and socioeconomic differences) of worldwide population in the computation of pooled relative risks due to O₃ exposure. The calculation of O₃-attributable premature deaths also ignores other related influencing factors on mortality burden, such as heat waves. Heat waves are favourable meteorological conditions for photochemical reactions, which can exacerbate O₃ pollution46,47. The synergistic effects between heat waves and ambient O₃ may lead to greater health risks. Finally, the impacts of O₃ exposure on the mortality burden attributed to the coronavirus disease 2019 (COVID-19) are not evaluated in our study. There was literature claiming that higher mortality risks of COVID-19 were associated with short-/long-term O₃ exposure48,49. Nevertheless, the exact mechanisms of chronic and acute exposure to ambient O₃ on COVID-19 mortality are unclear and required to be elucidated, which need further exploration in the future. These limitations potentially introduce some uncertainties in the estimation of O₃-attributable health risks.

In conclusion, we conduct a globe-scale study on current O₃ pollution, utilizing a daily seamless high-resolution (0.05°) perspective. This study features a geospatially dynamic modelling algorithm and assesses short- and long-term O₃ exposure levels as well as their associated all-cause mortality burden. Our ambient O₃ dataset offers insights into a fine-scale assessment of both short- and long-term O₃-related risks worldwide. Notably, our study reveals significant global O₃-related health impacts that might have been underappreciated in recent years. Effective mitigation of global O₃ pollution, particularly over the regions identified in this study, may lead to a reduction of significant O₃-related health risks and benefit the population exposed to O₃ pollution in the future.

Methods

In this study, all variates are first imputed for missing values and re-sampled before being considered as inputs. Next, the processed data and ground truths (output) should be spatiotemporally collocated and fed into the GL-CEF model for training. A total of two validation schemes are then exploited to verify the performance of modelled results. Eventually, the global exposure levels and all-cause mortality burden due to ambient O₃ are carefully assessed and discussed.

Datasets

Previous works widely considered O₃ precursors as key inputs of the model8,19,20,21,22, including NO₂ and HCHO. In our study, the high-resolution (~ 5 km) NO₂ and HCHO tropospheric vertical column density (TroVCD) from TROPOMI50,51 are adopted as the primary variates. The O₃ profile from CAMS52 is also introduced as a primary variate to provide the apriori information of ambient O₃. Meanwhile, multiple frequently used factors are selected as the auxiliary variates to improve the performance of the model, which consist of solar radiation intensity (necessary conditions for photochemical reactions53,54,55), meteorological fields, and geographic elements. The in-situ measurements of MDA8 O₃ from the Open Air Quality (OpenAQ), China National Environmental Monitoring Center, US Environmental Protection Agency, and European Environment Agency are used as the ground truths (output). At last, the latest replay ambient O₃ product from GEOS-CF56 is applied for comparison to the modelled results. More specific details are given as follows.

OpenAQ can provide globally distributed ambient concentrations of major air pollutants, which came from various sources over more than 100 countries. The air quality records from OpenAQ have been broadly utilized in worldwide studies of recent years24,56,57,58,59. In the present study, the global in-situ O₃ measurements during 2019–2021 are collected from OpenAQ and calculated to MDA8 ambient O₃ concentrations (regarded as the ground truths). It’s worth noting that the duration of in-situ measurements per day should exceed 20 h (from the first to the last). Furthermore, only the source of “governments” is employed to guarantee the data quality. The units of all values are transformed to ppb based on corresponding references. Supplementary Fig. 9a illustrates the spatial locations of in situ stations in the globe, using the symbols of red circles. A total of > 7,000 in situ stations (by 2021) are considered in this study, which densely cover China, Europe, the US, etc. Since the data from OpenAQ could be irregularly missing, the in-situ O₃ measurements from the China National Environmental Monitoring Center, US Environmental Protection Agency, and European Environment Agency are selected for supplement.

TROPOMI devised a self-appropriate atmospheric NO₂ retrieval algorithm based on that from the Ozone Monitoring Instrument, which can generate NO₂ TroVCD worldwide51. The differential optical absorption spectroscopy method, a chemical transport model (or data reanalysis system), and an air-mass factor lookup table were all introduced in the TROPOMI NO₂ retrieval algorithm. Related steps are as below: (1) Retrieve NO₂ total slant column density via the differential optical absorption spectroscopy method. (2) Separate the stratospheric and tropospheric parts using the chemical transport model (or data reanalysis system). (3) Transform the tropospheric slant column density to TroVCD according to the air-mass factor lookup table. In our study, the record of “nitrogendioxide_tropospheric_column” is applied as a primary variate for modelling global ambient O₃ concentrations from 2019 to 2021. Supplementary Table 6 lists more information about the TROPOMI NO₂ TroVCD.

TROPOMI applied a method based on differential optical absorption spectroscopy and combined ultraviolet spectral bands to produce HCHO TroVCD globally50. The detailed procedures of TROPOMI HCHO retrieval algorithm were similar to those of NO₂. In the present study, the global record of “formaldehyde_tropospheric_vertical_column” during 2019–2021 is adopted as a primary variate. Supplementary Table 6 provides more details about the TROPOMI HCHO TroVCD.

CAMS was the fourth generation of globally gridded atmospheric reanalysis product from the European Centre for Medium-Range Weather Forecasts52. Based on the mechanisms of physics and chemistry, CAMS can provide multiple chemical components by integrating simulations of chemical transport model with worldwide measured data. The general spatial and temporal resolutions for CAMS were 0.75° and 3-hour, respectively. CAMS reanalysis product has been extensively exploited for previous atmospheric works over the globe52, which suggested its reliable data quality. In this study, the global record of “ozone_mass_mixing_ratio” (O₃ profile) from 2019 to 2021 is selected as a primary variate to introduce the apriori information of ambient O₃ into the model. Considering that ground truths were MDA8 ambient O₃ concentrations, the daily maximum 9-hour averaged O₃ profile is employed. Supplementary Table 6 shows the specific information of the CAMS O₃ profile.

Similar to CAMS, ERA5 reanalysis product was also devised by the European Centre for Medium-Range Weather Forecasts60, which involved simulations of chemical transport model and actual measured data. In general, ERA5 can generate atmospheric/surficial parameters with spatial and temporal resolutions of 0.25° and hourly, respectively. In our study, the records of solar radiation intensity (i.e., “surface_solar_radiation_downwards”) and several meteorological fields are regarded as auxiliary variates for modelling global ambient O₃ concentrations during 2019–2021. Attributed to that ground truths are MDA8 ambient O₃ concentrations, the MDA8 solar radiation intensity, air temperature, and dew point temperature are utilized. As for other meteorological fields, the hourly values for each day are averaged in this study. Detailed information can be referred to in Supplementary Table 6.

Previous works broadly adopted the geographic elements as auxiliary inputs of the model8,9,20,21,61 due to their significant association with the spatial distribution of ambient O₃. In this study, the normalized differential vegetation index (NDVI)62 and land use classes63 from the moderate resolution imaging spectroradiometer (MODIS) with LandScan population density64 worldwide are deemed as auxiliary variates to increase the robustness of the model. Supplementary Table 6 lists the specific details of geographic elements.

GEOS-CF was developed by NASA in 2021, including two versions: forecast and replay (improved through reanalysed meteorological fields)56. It can produce various global atmospheric chemical components based on simulations of chemical transport model, with spatial and temporal resolutions of 0.25° and hourly, respectively. In the present study, the replay record of “surface_ozone” is exploited in the validation for comparison with the modelled MDA8 ambient O₃ concentrations globally. More information about GEOS-CF can be found in Keller et al. 56.

Data preprocessing

The surface emissions of O₃ precursors had continuous and complicated impact on ambient O₃ distribution65,66,67. Therefore, the NO₂ and HCHO TroVCD are averaged to acquire monthly data, which reflect the conditions for surface emissions of O₃ precursors in our study. Meanwhile, the average by month can reduce data noise (especially for HCHO TroVCD68) and improve the coverage of available values.

Afterward, the data interpolating empirical orthogonal functions (DINEOF) method69 is employed to recover the missing information in monthly NO₂ TroVCD, HCHO TroVCD, and NDVI, relying on their spatiotemporal self-correlation. The brief procedures of the DINEOF method are as follows: 1) Initialize missing values and unfold the 3-dimensional origin data to a 2-dimensional matrix (M) along the spatial dimension; 2) Decompose M using the singular value decomposition; 3) Reconstruct a matrix Mr with top-k singular values and replace the missing values of M with those of Mr; and 4) Repeat the procedures of 2) and 3) until the errors reach a pre-determined threshold and reshape M to final results according to the origin dimensions. More details of the DINEOF method can be referred to in Alvera-Azcárate et al. 69. Supplementary Table 7 shows the simulated experiment results of the DINEOF method. The missing masks for validation in the simulation are acquired from the real scenes of origin products. As listed, the imputed results in monthly NO₂ TroVCD, HCHO TroVCD, and NDVI present expected accuracy via the DINEOF method, with the correlation coefficient (CC) of 0.85, 0.8, and 0.91, respectively.

Furthermore, the spatial resolutions of various variates need to be consistent in our study. Considering the spatial resolution of TROPOMI products (~ 5 km), a global grid of 3600 × 7200 (0.05°) is adopted. To be specific, the monthly NO₂ and HCHO TroVCD are re-sampled to 0.05° through the nearest neighbouring interpolation70. The re-sampling methods for other variates are inverse distance weighted interpolation71 and area-weighted aggregation72 (Supplementary Table 6).

Remote-sensing and reanalysis products were gridded data with various spatial and temporal resolutions, while the applicable ranges of in-situ measurements only focused on small regions. Therefore, it is required to unify the spatial and temporal dimensions between gridded data and ground truths. Initially, the variates with multiple temporal resolutions, such as monthly NO₂ TroVCD and daily meteorological fields, are mutually aligned. Next, all the ground truths falling on the same grid are averaged to collocate with the aligned gridded data.

Model description

The GL-CEF model includes global, local, and global-local modules, which can adopt global and local knowledge over station-sparse (or no-station) and station-dense regions, respectively. Specific information about the three modules is shown in the following parts.

Global module: as depicted in Supplementary Fig. 9b, a certain number of collocated grids are firstly given (blue circles). Next, a transition zone is set in the regions that are [lmin, lmax] from the collocated grids. If the distance between the target grid i (yellow square) and its h-th nearest collocated grid is greater than lmax, i is deemed as station-sparse, and the global module should be exploited for the estimation (G-value). In our study, the global module utilizes all collocated samples in the modelling, which adopts the deep forest73 (see the Supplementary Methods for details) as the global sub-model. The general expression of the global sub-model is defined in Eq. (1).

$${{VG}}_{O3}={F}_{G}({V}_{{THCHO}},{V}_{{TNO}2},{V}_{{CAMS}},{V}_{{ESRI}},{V}_{{EMF}},{V}_{{GE}},{V}_{{TC}})$$
(1)

where \({{VG}}_{O3}\) represents the modelled MDA8 ambient O₃ concentrations through the global sub-model (G-value). \({F}_{G}\) stands for the global sub-model based on the deep forest. \({V}_{{THCHO}}\), \({V}_{{TNO}2}\), \({V}_{{CAMS}}\), \({V}_{{ESRI}}\), and \({V}_{{EMF}}\) indicate the TROPOMI HCHO TroVCD, TROPOMI NO₂ TroVCD, CAMS O₃ profile, ERA5 solar radiation intensity, and EAR5 meteorological fields. \({V}_{{GE}}\) denotes geographic elements, including NDVI, land use classes, and population density. \({V}_{{TC}}\) signifies the temporal encoding74. Supplementary Table 8 lists the parameters of the global sub-model designed in this study.

Local module: as illustrated in Supplementary Fig. 9c, if the distance between the target grid i and its h-th nearest collocated grid is less than lmin, i is regarded as station-dense, and the local module ought to be adopted for the estimation. The local module involves two highlights: sliding block strategy and variable local sub-model.

Sliding block strategy. Previous localized methods15,19,20 normally built independent sub-model for each target grid, which yielded commendable performance but required large time consumptions. These methods were likely unfit for the global modelling task. As a result, the sliding block strategy is proposed for the fast training of local sub-models in our study. Related procedures are as follows. A sliding block with the radius of r is first selected and then traverses all the target grids with the step size of s. To ensure that collocated samples can smoothly change in adjacent sliding blocks, a buffer distance of db is introduced, which is larger than r. If the collocated samples in some sliding blocks are too few (<cm), they should be discarded. Next, the collocated samples in each sliding block are utilized for the independent modelling, which can generate several well-trained local sub-models. The local sub-models trained by the nearest N sliding blocks from the target grids are adopted for their intermediate estimation (Ls-values). Finally, the Ls-values are aggregated to acquire the modelled result (L-value). In addition, if the collocated samples included in a few adjacent sliding blocks remain unchanged, they need to be discarded. This can avoid multiple identical local sub-models that likely lead to discontinuous modelled results. Supplementary Table 8 shows the parameters of the sliding block strategy devised in this study.

Variable local sub-models. Generally, the complexity of the model is supposed to be positively correlated with the number of collocated samples. In our study, the variable local sub-models are developed deriving from the light gradient boosting machine75 (see the Supplementary Methods for details). The parameters of variable local sub-models can automatically vary depending on the number of collocated samples, whose key parameters (see bold fonts in Supplementary Table 8) are determined as provided in Eq. (2).

$$p=\left\{\begin{array}{c}{p}_{\min },{cou} \, < \, {q}_{\min }\\ {Rou}\left(\frac{{{cou}-q}_{\min }}{{q}_{\max }-{q}_{\min }}\times \left({p}_{\max }-{p}_{\min }\right)+{p}_{\min }\right),{q}_{\min } \, < \, {cou} \, < \, {q}_{\max }\\ {p}_{\max },{cou}\ge {q}_{\max }\end{array}\right.$$
(2)

where \({p}_{\min }\) and \({p}_{\max }\) stand for the minimum and maximum of key parameters, respectively. \({q}_{\min }\) and \({q}_{\max }\) represent the minimum and maximum of number thresholds, respectively. \({cou}\) is the number of collocated samples. \({Rou}(\bullet )\) indicates the rounding function. The general expression of the variable local sub-models is defined in Eq. (3).

$${{VL}}_{O3}=\frac{1}{N}\mathop{\sum }\limits_{n=1}^{N}{F}_{{VLn}}({V}_{{THCHO}},{V}_{{TNO}2},{V}_{{CAMS}},{V}_{{ESRI}},{V}_{{EMF}},{V}_{{TC}})$$
(3)

where \(N\) denotes the number of variable local sub-models. \({{VL}}_{O3}\) signifies the modelled MDA8 ambient O₃ concentrations (L-value) aggregated from the intermediate estimation through the variable local sub-models (Ls-values). \({F}_{{VLn}}\) indicates the n-th VLSM deriving from the light gradient boosting machine. It is worth noting that the inputs of variable local sub-models discard geographic elements, which ensures their collocated samples can present sufficient temporal differences.

Global-local module: as displayed in Supplementary Fig. 9d, a transition zone is set to smooth the modelled results in the connections of global and local modules. Regarding the target grid i in the transition zone, the modelled results of global (G-value) and local (L-value) modules are merged using a geospatial weighting method. Related weights are defined in Eqs. (4)–(6).

$${{VW}}_{O3}={{VG}}_{O3}\times {w}_{1}+{{VL}}_{O3}\times {w}_{2}$$
(4)
$${w}_{1}=\frac{{l}_{1}^{2}}{{l}_{1}^{2}+{l}_{2}^{2}}$$
(5)
$${w}_{2}=\frac{{l}_{2}^{2}}{{l}_{1}^{2}+{l}_{2}^{2}}$$
(6)

where \({{VW}}_{O3}\) indicates the MDA8 ambient O₃ concentrations after geospatial weighting (W-value). \({l}_{1}\) and \({l}_{2}\) stand for the distances between i and two boundaries (lmin and lmax) of the transition zone, respectively.

Validation scheme

The spatial performance of the GL-CEF model need to be emphatically validated. Hence, the SICV and TESICV schemes of 5 folds are utilized in the present study, which focus on spatial accuracy and spatiotemporally predictive ability, respectively. As depicted in Supplementary Fig. 6a, all the collocated samples are first divided into 5 folds based on spatial locations in the SICV scheme. Next, the GL-CEF model will be trained and validated using 80% (4 folds) and 20% (1 fold) of the collocated samples, respectively. Finally, the above step should be repeated 4 times until each fold has been adopted. As for the TESICV scheme, the only difference is that the training and validation sets came from 2020–2021 and 2019, respectively. In our study, the global imputed and modelled results are verified with the help of 5 metrics: R2, as defined in Eq. (7); CC, as defined in Eq. (8); RMSE, as defined in Eq. (9); relative percentage error (RPE), as defined in Eq. (10); and mean bias, as defined in Eq. (11).

$${{\rm{R}}}^{2}=1-\frac{\sum (\hat{v}-{v})^{2}}{\sum (\bar{v}-{v})^{2}}$$
(7)
$${\rm{CC}}=\frac{k\sum v\hat{v}-\sum v\sum \hat{v}}{\sqrt{k\sum {v}^{2}-{\left(\sum v\right)}^{2}}\sqrt{k\sum {\hat{v}}^{2}-{\left(\sum \hat{v}\right)}^{2}}}$$
(8)
$${\rm{RMSE}}=\sqrt{\frac{1}{k}\sum {(\hat{v}-v)}^{2}}$$
(9)
$${\rm{RPE}}=\frac{{\rm{RMSE}}}{\bar{v}}$$
(10)
$${\rm{mean\; bias}}=\frac{1}{k}\sum (\hat{v}-v)$$
(11)

where \(k\) represents the number of collocated samples. \(\hat{v}\), \(v\), and \(\bar{v}\) stand for the modelled, benchmark, and mean benchmark values, respectively.

Population exposure estimation

In this study, the population-weighted MDA8 ambient O₃ concentrations are acquired using Eq. (12)13,76.

$${v}_{{pw}}=\frac{\sum {{pop}}_{i}\times {\hat{v}}_{i}}{\sum {{pop}}_{i}}$$
(12)

where \({\hat{v}}_{i}\) indicates the modelled MDA8 ambient O₃ concentrations on grid i. \({{pop}}_{i}\) stands for the population density on grid i. \({v}_{{pw}}\) represents the population-weighted MDA8 ambient O₃ concentrations over target region. In the meantime, the standard deviation of \({v}_{{pw}}\) (i.e., \({{std}}_{{pw}}\)) is computed with Eq. (13)57,77.

$${{std}}_{{pw}}=\sqrt{\frac{\sum {{pop}}_{i}\times ({\hat{v}}_{i}-{v}_{{pw}})}{\sum {{pop}}_{i}\times (e-1)/e}}$$
(13)

where \(e\) signifies the grid count of the population density exceeding 0. Moreover, the exposure levels during a period are calculated by Eq. (14)77,78, which temporally cumulates the MDA8 ambient O₃ concentrations on each grid.

$${{EL}}_{U,D}\left({U}_{x},{D}_{y}\right)=\frac{\sum {{pop}}_{i}\times {I}_{U,D}\left({U}_{i} \, > \, {U}_{x},{{D}_{i} \; > \; D}_{y}\right)}{\sum {{pop}}_{i}}\times 100\%$$
(14)

where \({{EL}}_{U,D}({U}_{x},{D}_{y})\) reflects the fraction of the population exposed to MDA8 ambient O₃ concentrations > \({U}_{x}\) ppb for more than \({D}_{y}\) days. \({I}_{U,D}({U}_{i} \, > \, {U}_{x},{{D}_{i} \, > \, D}_{y})\) denotes positive as the day count of MDA8 ambient O₃ concentrations > \({U}_{x}\) ppb surpasses \({D}_{y}\) days. Otherwise, it is configured to negative.

Mortality burden estimation

In the present study, the long-term all-cause mortality burden due to O₃ exposure is computed according to a log-linear exposure-response function between ambient O₃ and premature deaths from epidemiological researches. The all-cause diseases are defined by A00 to R99 in the International Classification of Diseases 10. The full procedures can be expressed as Eqs. (15)–(18)2,11,34,79.

$${{RR}}_{i}={e}^{\gamma ({v}_{{mi}}-{v}_{s})}$$
(15)

where \({{RR}}_{i}\) represents the pooled relative risk on grid i. \(\gamma\) indicates the pooled effect value of long-term ambient O₃ exposure for all-cause deaths from a worldwide meta-analysis involving 226 million participants36, with a value of 1.39 × 10−3 (95% CI: 8.9597 × 10−4, 1.8822 × 10−3) per ppb. This suggests that an increment of 10 ppb in warm-season MDA8 ambient O₃ concentrations is associated with a pooled relative risk of 1.014 (95% CI: 1.009, 1.019) for long-term exposure. \({v}_{{mi}}\) is the warm-season modelled MDA8 ambient O₃ concentrations on grid i. \({v}_{s}\) stands for the long-term threshold of counterfactual concentrations advised by WHO in 202180, which denotes its AQG (30.07 ppb) as the starting point.

$${{paf}}_{i}=1-\frac{1}{{{RR}}_{i}}$$
(16)

where \({{paf}}_{i}\) reflects the PAF on grid i, indicating the proportion of mortality burden that will be eliminated when ambient O₃ is decreased to the threshold of counterfactual concentrations (30.07 ppb for long term).

$${tpaf}=\frac{\sum {{pop}}_{i}\times {{paf}}_{i}}{\sum {{pop}}_{i}}$$
(17)
$${dea}=\sum {{pop}}_{i}\times {tpaf}\times {mor}$$
(18)

where \({tpaf}\) signifies the PAF over target region. \({dea}\) denotes the annual long-term O₃-attributable deaths over target region. \({mor}\) indicates the annual baseline mortality from the GBD 202112 over target region. For different land use classes, \({tpaf}\) is acquired using the population of a single class.

As for short-term O₃ exposure, the associated all-cause deaths are also calculated through Eqs. (15)–(18). By comparison, \(\gamma\) can be obtained from another worldwide meta-analysis for short-term all-cause deaths, with a value of 8.5627 × 10−4 (95% CI: 6.7736 × 10−4, 1.035 × 10−3)35 per ppb. This demonstrates that an increment of 10 ppb in MDA8 ambient O₃ concentrations is associated with a pooled relative risk of 1.0086 (95% CI: 1.0068, 1.0104) for short-term exposure. \({v}_{{mi}}\) stands for the modelled MDA8 ambient O₃ concentrations on grid i. \({v}_{s}\) indicates the short-term threshold of counterfactual concentrations provided by WHO in 202180, which deems its AQG (50.11 ppb) as the starting point. The daily PAF is first accumulated and then averaged to acquire the annual value for the short-term O₃ exposure.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.