Skip to content
Frontiers
Search

Methods

Frontiers in Climate ·

Volume 8 - 2026 | https://doi.org/10.3389/fclim.2026.1847091

An open-access workflow combining climate model projections with epidemiological frameworks to assess air quality mortality

  • Alice F. Wells1*corresponding author
  • James W. Hurrell1
  • Eric Gilleland2
  • G. Brooke Anderson3
  1. 1Department of Atmospheric Science, Colorado State University, Fort Collins, CO, United States
  2. 2Department of Statistics, Colorado State University, Fort Collins, CO, United States
  3. 3Department of Environmental and Radiological Health Sciences, Colorado State University, Fort Collins, CO, United States
Article metrics0 viewsView details
PDFCiteMetrics

Abstract

Surface-level air pollution is a major source of human mortality worldwide, and future emissions projections and resulting climate change are expected to alter concentrations of pollutants such as ozone and particulate matter. Estimating the health burden of these changes requires combining climate model projections with epidemiological frameworks, such as those developed in the Global Burden of Disease (GBD). However, significant barriers hinder this integration. Climate model outputs differ from health metrics in spatial resolution, temporal aggregation, and pollutant metrics. For example, GBD quantifies ozone exposure as the highest seasonal average of 8-h daily maximum concentrations, while most climate models provide hourly or monthly mean data. Furthermore, climate model outputs often require bias correction and spatial downscaling to ensure exposure estimates are as accurate as possible, so that applying exposure-response functions derived from observational data yields more trustworthy mortality projections. We present an open-access workflow explicitly designed to bridge this gap. Our goal is to enable researchers to process climate model data for health impact assessments of air quality. The workflow processes climate model pollutant data to align with GBD metrics, applies bias correction and downscaling methods, and calculates mortality using established GBD exposure-response functions and baseline demographic data. We demonstrate the workflow through a single illustrative worked example, rather than a comprehensive multi-model projection of future air-quality mortality. The underlying pipeline, however, is designed to support consistent, reproducible estimation across different scenarios and models. By making the workflow publicly available, we aim to lower barriers for interdisciplinary research and support collaboration between climate scientists and epidemiologists. Our work provides a foundation for quantifying the health implications of changing air quality under plausible future air pollutant scenarios and climate conditions, improving decision-making around mitigation and adaptation strategies.

Keywords

1 Introduction

Climate change affects human health through shifting exposure to environmental hazards, including increasing extreme weather events (Akhtar, 2020), infectious disease spread (Baker et al., 2022), and changes in air quality (Orru et al., 2017). Air pollution is one of the most significant environmental risks to human health worldwide and the second leading risk factor for pre-mature mortality (Health Effects Institute, 2025). In 2023 an estimated 7.9 million deaths were attributed to exposure to air pollution, approximately one in eight deaths globally (Health Effects Institute, 2025). Of these, over 60% (4.9 million deaths) were linked to ambient fine particulate matter (PM2.5; particles that are 2.5 microns or less in diameter) exposure and 470,000 deaths were attributed to surface ozone exposure (Health Effects Institute, 2025). The primary damage pathways of surface ozone and PM2.5 exposure are respiratory and cardiovascular conditions. Ozone has limited solubility so it can penetrate the upper respiratory tract reaching the lungs, primarily affecting the respiratory system (Alvim-Ferraz et al., 2024) and increasing the risk of chronic obstructive pulmonary disease (COPD) (Kim et al., 2020). PM2.5 can also penetrate deep into the lungs causing respiratory and cardiovascular issues (Xing et al., 2016); however it has also been shown to increase the risk of diabetes, adverse birth outcomes and dementia (Lao et al., 2019; Ghosh et al., 2021; Ru et al., 2021).

The Global Burden of Disease (GBD), an international effort coordinated by the Institute for Health Metrics and Evaluation, systematically quantifies health loss from hundreds of diseases, injuries and risk-factors across countries, sexes and age-groups, drawing on the expertise of over 10,000 collaborators worldwide. As part of its comparative risk-assessment framework, the GBD develops exposure-response functions (ERFs) linking ambient exposures (such as surface PM2.5 and ozone) to mortality and disability outcomes, which can then be applied to population exposures and baseline mortality rates to estimate the attributable disease burden (Hay et al., 2025). Here, we focus on ambient PM2.5 only, excluding the impact of household air pollution. Household air pollution arises primarily from solid-fuel combustion for cooking and heating and exposures are typically far higher than ambient concentrations, particularly in low- and middle-income countries, carrying a correspondingly large attributable mortality burden (Liu et al., 2025). Excluding household air pollution means our estimates should not be interpreted as the total air pollution attributable mortality burden.

Sources of ambient PM2.5 include combustion from vehicles, agricultural and industrial activities, dust and wildfires. Due to the implementation of stricter air quality regulations there has been an observed decrease in ambient PM2.5 exposure in many countries (Health Effects Institute, 2025). Increases in wildfire activity, in contrast, have led to worsening air quality in affected regions (Yang et al., 2022). Ozone, while not directly emitted into the atmosphere, is formed through chemical reactions between nitrous oxides and volatile organic compounds in the presence of sunlight (Crutzen, 1988). Typically surface ozone is most highly concentrated in urban areas due to more sources of precursor emissions; however, it can also travel long distances impacting the air quality of rural regions (e.g., Xu et al., 2025). Exposure to future air pollutants depends both on the direct emissions trajectories and also climate feedbacks which can alter the drivers and meteorological conditions which affect these pollutants. Higher temperatures and droughts can worsen wildfires and dust storms (e.g., Richardson et al., 2022), while changes in humidity, precipitation and circulation can alter the photochemical production of ozone and change the frequency of stagnation events, trapping pollutants near the ground (e.g., Horton et al., 2014).

Future concentrations of precursors and direct emissions of these pollutants depend on socio-economic pathways and emission trajectories. Climate projections for the Coupled Model Intercomparison Project Phase 6 (CMIP6) used Shared Socioeconomic Pathways (SSPs) to describe possible future development pathways for human society, particularly in relation to its use of fossil fuels and the social and economic factors which drive fossil fuel usage (O'Neill et al., 2014; Riahi et al., 2017), and these include narratives on future air pollution control (e.g., Rao et al., 2017). Alongside the emission trajectories, the SSP scenarios provide quantitative key scenario drivers, including population projections. The SSP scenarios are combined with Representative Concentration Pathways (RCPs), which define the stabilized radiative forcing at the end of the century and therefore, the level of climate change policy required for a particular SSP to reach said forcing. Applying this framework to future U.S. air quality, Ganji et al. (2026) found that the socioeconomic pathway (SSP) generally explained more of the variance in projected pollutant reductions than the radiative forcing target (RCP), particularly in the near term. The latest generation of scenarios for CMIP7 move away from the SSP-RCP framework and instead define seven illustrative emissions pathways which include air pollution control assumptions (Van Vuuren et al., 2026), and they also have associated population projections (Paprotny, 2026).

Health-impact assessments can be used to understand the impact of projected changes in air pollution on future mortality and morbidity. These assessments combine modeled projections of surface ozone and PM2.5 with ERFs, population distributions, and baseline mortality rates to estimate excess mortality (e.g., McDuffie et al., 2021a; Malashock et al., 2022; Pozzer et al., 2024). Results can vary by scenario and the choice of ERF, making them difficult to directly compare (e.g., Post et al., 2012). Comparability is further complicated as the outcomes attributed to a given pollutant are not fixed. For example, the GBD added diabetes as a PM2.5 attributable mortality outcome from 2019 onwards, and dementia from 2023 onwards, meaning estimates and workflows built on earlier ERFs will exclude a mortality outcome which more recent studies include (Lin et al., 2025). Nonetheless, there is broad agreement that the effects of climate change will increase air quality mortality in the future (e.g., Silva et al., 2017; Akritidis et al., 2024).

Performing these health impact assessments requires the integration of multiple technical components with steps typically split across disciplines. Climate model outputs need to be processed into the variables required for exposure assessments, including conversion into relevant concentration metrics, bias correction and spatial downscaling. At the same time, identifying and correctly applying ERFs, population and baseline mortality rate data can pose additional challenges for researchers less familiar with public-health methodologies. These barriers make it difficult to generate consistent and comparable estimates of air-pollution mortality. Despite a growing body of literature in this space, including global mortality assessments, regional projections, and demographic exposure studies, each of these analyses has, to date, been built as a single-purpose pipeline suited to its own study design, scenario, and region. No open-source workflow currently exists to allow researchers to perform this type of health impact assessment for a region, scenario, or dataset of their choosing. Instead, researchers are required to spend substantial time building the analytical pipeline from scratch.

Our workflow addresses these challenges by integrating the key technical components into a documented, version-controlled pipeline. Existing studies such as McDuffie et al. (2021a), Malashock et al. (2022), and Pozzer et al. (2024) demonstrate that these components can be combined to produce robust mortality estimates, but each was developed to answer a specific research question. Our contribution is therefore not a new epidemiological method, exposure-response function, or mortality estimate, but an openly available, modular, and documented workflow that encompasses established methods into a form which allows others to inspect, adapt, and run for their own regions, scenarios, or datasets. We document datasets, processing steps and their methodological choices, epidemiological inputs, and we supply runnable code so that users can reproduce country- or region-level results. Our goal is to lower the technical barrier to interdisciplinary research and facilitate climate scientists and public-health researchers to perform future mortality assessments. Recent guidance on climate-health research has highlighted the importance of transparent, reproducible, and transdisciplinary approaches to linking climate information with health outcomes (Ebi et al., 2025). Our workflow addresses several practical challenges highlighted in that guidance, particularly those related to data accessibility, methodological consistency, and transparency in analytic choices. The workflow is openly available and designed to be adaptable, allowing users to substitute their own pollutant datasets, baseline mortality information, or population projections as needed.

We also see this workflow as a flexible research tool and a learning resource. Users can apply the pipeline directly to address specific research questions, modify individual components (for example, exposure metrics, epidemiological inputs, or scenarios) to suit their study design, or extend it to related applications. The accompanying example analyses (see Section 5) are designed to function as a guided tutorial, illustrating each step of the health impact assessment process and the links between climate modeling and epidemiological methods. In this way, the workflow can also serve as a shared reference within interdisciplinary teams, supporting a common understanding of assumptions, data flows, and methodological choices across climate science and public health.

The paper is structured as follows: Section 2 outlines the mortality calculation and the required inputs. Section 3 then describes the processing of climate model output in more detail, including the derivation of the air quality variables required for the exposure assessment. Section 4 presents a verification of the workflow and in Section 5, we present a worked example demonstrating one application of the workflow and the types of results it can produce.

2 Mortality calculation

This section introduces the overarching structure of the workflow and the sequence of pre-processing steps required to obtain the inputs to calculate excess mortality.

Excess mortality is defined as the number of deaths that would not have occurred in the absence of exposure. It can be calculated as:

M=BMR × POP × AF(1)

where BMR is the baseline mortality rate, POP is the exposed population, and AF is the attributable fraction, representing the proportion of deaths attributable to the exposure (e.g., Anenberg et al. 2010).

As illustrated in Figure 1, the workflow comprises three parallel preprocessing tracks that produce the variables needed to calculate excess mortality: population, attributable fraction, and baseline mortality rates. The subsections below describe these components, the assumptions made in the workflow, and outlines the transformations required to progress from raw inputs to the point at which these three quantities can be combined in Equation 1. Additionally, each track of the workflow can be adjusted to the users requirements, allowing the workflow to be a flexible modular tool. The following sections highlight where changes can be made to align with the users desired outcome.

Illustrative workflow to calculate excess mortality. Processing steps provided within the workflow are denoted in yellow. The three variables needed to calculate mortality are shown in orange. Circular dashed inputs show where uncertainty, either model and scenario, or epidemiological, contributes to the final result.
Figure 1. Illustrative workflow to calculate excess mortality. Processing steps provided within the workflow are denoted in yellow. The three variables needed to calculate mortality are shown in orange. Circular dashed inputs show where uncertainty, either model and scenario, or epidemiological, contributes to the final result.

The accompanying GitHub repository (https://doi.org/10.5281/zenodo.18436806) is designed to be used alongside the paper as a tutorial, allowing users to follow the steps shown in Figure 1 using the same scenario presented in Section 5.

2.1 Population projections

Projections of population are important for mortality calculations as they define the number of people exposed to an environmental hazard. Even if exposure levels remained constant, changes in population size and geographic distribution could substantially alter the total number of attributable deaths. To quantify the relative contributions of changes in exposure and changes in population size or distribution to total mortality projection, sensitivity analyses can be performed by holding either population or exposure constant at a relevant baseline.

Furthermore, age plays an important role in calculating the total mortality of a country or region. Age groups have different baseline mortality rates, and the size of each age category is projected to change at different rates over time, so both the age structure of a population and its evolution can shift the total mortality projection independently of overall population growth. By using all-age population projections, the base version of this workflow assumes a static age distribution over the projection period, which may not accurately capture mortality trends in regions undergoing demographic transitions such as population aging or shifts in fertility and life expectancy. This could lead to an under- or over-estimate of attributable mortality in age groups whose relative size within the population changes over time. Incorporating age-stratified population projections is therefore an encouraged extension of this workflow for users with appropriate data. The workflow's modular structure allows age-stratified population projections to be substituted alongside corresponding age-stratified baseline mortality rates (Section 2.3). The GBD has not produced age-stratified exposure response functions from 2021 onwards; however, users can either substitute their own age-stratified ERFs or apply the GBD all-age ERFs across all age groups. The base version presented here relies on all-age population projections, offering a more general and immediately accessible starting point.

In general, the choice of which population projections to use depends on the underlying climate scenario that is being analyzed. For example, under the CMIP6 structure and in line with our example (Section 5), a climate scenario labeled SSP2-4.5 requires population projections from SSP2. In this case, users could apply country level age-stratified population projections from IIASA (International Institute for Applied Systems Analysis). However, any population projections can be used as long as they are consistent with the underlying climate scenario.

In our example, we use Global 1-km Downscaled SSP Population projections which can be obtained from Gao (2020). These data are available on a 0.0083° x 0.0083° resolution for each decade from 2000–2100. To align with the resolution of the GBD exposure estimates (Section 3.2), the population data need to be aggregated to a 0.1° x 0.1° spatial resolution. When the GBD calculates population-weighted exposures for historical years they used population data from the Gridded Population of the World database which provides data at the same resolution as these SSP projections. Following the GBD method we aggregate the population data to 0.1° x 0.1° spatial resolution by summing the central 12 x 12 cells (see Supplementary appendix 1 from Brauer et al., 2024) and perform a temporal interpolation to derive annual estimates from the decadal estimates provided.

Alternatively, one could adjust the 1/8th degree SSP population projections (https://www.earthdata.nasa.gov/) to align with the required spatial resolution. These data, however, were not available to use at the time of writing, but could be a resource to consider in future uses of this workflow.

2.2 Attributable fraction

The attributable fraction (AF) can be defined as the fraction of all cases of a disease in the population that can be attributed to the exposure (Equation 2). In other words, the fraction of cases that would not have occurred if exposure had been eliminated. This can be calculated using relative risks (RR).

AF(x, y)= RR(x, y)-1RR(x, y)(2)

The RR for a binary exposure is the ratio between the risk of disease among the exposed and the risk among the non-exposed (Steenland and Ben, 2006). A RR greater than one indicates that the exposure increases the likelihood of disease, whereas a RR less than one suggests a protective effect. For continuous exposures, this concept is extended by defining the RR as the ratio of risk at a given exposure level relative to a specified reference exposure.

In cases where a risk-free exposure level (i.e., RR = 1) theoretically exists but has not been well established, the reference exposure can instead be defined by the theoretical minimum risk exposure level (TMREL). The TMREL represents a counterfactual exposure (or exposure range) associated with the lowest health risk supported by available epidemiological evidence and that is theoretically possible (Hay et al., 2025). In the GBD framework, uncertainty in the TMREL for surface ozone (U[29.1, 35.7] ppb) and PM2.5 (U[2.4, 5.9] μg/m3) is represented using uniform distributions, due to the lack of evidence for a single, well-defined exposure level below which no adverse health effects occur (Hay et al., 2025). This is broadly consistent with WHO guidance, which recommends peak seasonal ozone exposure and annual mean PM2.5 concentrations do not exceed 30 ppb and 5 μg/m3, respectively.

Exposure-response functions (ERFs) provide a mathematical relationship between the magnitude of exposure and the corresponding RR, allowing risk to be estimated continuously across different exposure levels. The GBD provides estimates of ERFs for a variety of risk factors, including ambient surface ozone and PM2.5. The GBD only provides ERFs for mortality outcomes where there is sufficient literature to support a relationship between exposure and mortality due to that outcome (risk-outcome pairs). Previously, the GBD would include risk-outcome pairs that met the World Cancer Research Fund grades of convincing or probable evidence (World Cancer Research Fund, 2007). Since 2021, a new Burden of Proof methodology has been introduced to provide a conservative, standardized measure of the strength of evidence for associations between risk factors and health outcomes (see Brauer et al., 2024; Supplementary appendix 1).

We have chosen to use the GBD exposure-response functions because they offer a robust, peer-reviewed, globally comparable set of relationships for each mortality cause. They have been derived through extensive statistical analysis performed by the GBD to incorporate information from a range of studies from different locations and across different demographics (see Brauer et al., 2024; Supplementary appendix 1 for a more detailed review of the statistical methods). In our workflow, the use of ERFs from the GBD is a design choice to enable users to utilize the GBD's well-established aggregation of epidemiological evidence. However, users may choose to replace these ERFs with specific studies' values to best represent the location or research question they are investigating, or in cases where a risk-outcome pair does not have an ERF available through the GBD.

For exposure to ambient surface ozone pollution the only included mortality outcome is COPD. In this case, the ERF follows a log-linear relationship, defined by a relative risk of 1.074 (95% CI: 1.014–1.137) per 10 ppb increase in ozone. To derive this ERF the GBD performed a literature review of studies examining long-term ozone exposure and COPD mortality. They used the meta regression-Bayesian, regularized, trimmed (MR-BRT) meta-regression tool to conduct a meta-analysis on those studies (Hay et al., 2025). The relationship is expressed as

RR={ eβ(X−TMREL)1if X>TMRELif X≤TMREL (3)

Where β= ln(1.07410), X is ozone concentration (see Section 3.2) and TMREL is the theoretical minimum risk exposure level which the GBD defines for surface ozone as a uniform distribution of 32.4 (95 % CI: 29.1–35.7) ppb (Hay et al., 2025).

For exposure to ambient PM2.5 pollution the GBD includes seven mortality outcomes: COPD, Lower Respiratory Infection, Type II Diabetes, Ischemic Heart Disease, Stroke, Lung Cancer, and Dementia. Unlike surface ozone, these relationships are more complex and non-linear, so the GBD provides tabulated concentration-response functions (i.e., arrays of relative risk values across an exposure range) for each outcome (Institute for Health Metrics Evaluation, 2024). The GBD performed a similar meta-analysis using the MR-BRT tool to calculate the respective ERFs from the literature and they define the TMREL as a uniform distribution of 4.15 (95% CI: 2.4–5.9) μg/m3 (Hay et al., 2025). The RR curves provided by the GBD do not account for the TMREL, so in our workflow, we shift the RR curves for each mortality outcome to ensure the RR at the TMREL is 1 and the RR at other exposure levels are relative to the TMREL. To do this we take the log of the RR, and shift the log(RR) curve to be relative to the TMREL baseline by subtracting the log(RR) of the TMREL. We then set all values of the log(RR) below the TMREL as 0 and exponentiate the curve to generate estimates of RR so that all values of RR below the TMREL are 1 and the RR at other exposures are estimated relative to the TMREL baseline.

2.3 Baseline mortality rate

The baseline mortality rate (BMR) represents the underlying rate of deaths within a given population in a specified time period. It describes expected mortality prior to the application of exposure-related excess risk and provides the basis for estimating exposure-attributable mortality. Country specific BMRs are calculated by taking the average mortality rate (per 100,000) for a given mortality outcome (i.e., COPD, Stroke, Lung Cancer, etc.) over a specified baseline period. For the example we share in Section 5, the baseline period is 1990–2009.

Calculating country specific BMRs is essential because mortality patterns vary substantially across populations. Differences in demographic structure, prevalence of pre-existing health conditions, healthcare access and quality, and socioeconomic status all influence the underlying susceptibility to disease and death. For example, regions with higher rates of cardiovascular or respiratory illness, or with limited healthcare infrastructure, will exhibit higher baseline mortality rates for the same cause. Accounting for these differences through country specific rates allows for a more accurate assessment of the additional deaths attributable to a given exposure. Further granularity can be included by calculating country-level BMRs for different age categories. Age-stratifying the BMRs allows for more accurate mortality estimates, as described in Section 2.1, and can be calculated by choosing the same age categories as those found in the age-stratified population data.

In this workflow, similar to previous health impact assessments (e.g., Pozzer et al., 2024; Akritidis et al., 2024), we hold the BMR constant throughout our projections of future mortality. This is a design choice in order to minimize uncertainty, however it could be seen as a limitation of our mortality projections. Choosing to keep BMRs constant assumes that healthcare, social and physical environments, including pollution levels, remain the same as our chosen baseline level which is unlikely to be the case. In particular, the historical BMR reflects the mortality burden attributable to historical pollution exposure, not a “pollution-free” baseline. There may also be an expectation that BMR will decrease over time if healthcare, for instance, improves. However, across the 7 health outcomes we are investigating, the global mortality rate has both increased (COPD, Lung Cancer, Type II Diabetes and Dementia) and decreased (Lower Respiratory Infection, Ischemic Heart Disease and Stroke) across the period 1980–2023 (Supplementary Figure S1). Additionally, for each mortality outcome, every country demonstrates a varying trend, which may or may not continue. With that being said, should the user wish to, projections of BMR could be applied here (e.g., Hughes et al., 2014). This would likely include additional uncertainty to the projected mortality calculations.

To calculate country specific BMRs, we use the GBD Results Tool (Global Burden of Disease Collaborative Network, 2024) to extract the mortality rate for each country and cause for the baseline years we use in our example, 1990–2009. Users can adjust the baseline years to match their requirements. The seven mortality outcomes for which we calculate the BMR are: COPD, Lower Respiratory Infection, Type II Diabetes, Ischemic Heart Disease, Stroke, Lung Cancer, and Dementia. For each BMR an upper and lower bound of a 95% confidence interval (CI) are provided alongside the mean value. Section 2.4 describes how these confidence intervals are incorporated into the mortality calculation. To match the resolution of the GBD exposure-response function, country specific BMRs are then mapped onto a 0.1° x 0.1° spatial resolution using country masks from McDuffie et al. (2021b).

2.4 Parametric bootstrapping

Epidemiology methods are primarily statistical and therefore include a range of uncertainty to be considered. To characterize the uncertainty in the BMR, TMREL, exposure-response parameter (β, for ozone and RR for PM2.5) we apply a parametric bootstrap approach.

For each quantity, we use the reported summary statistics from the GBD (e.g., mean and 95% confidence interval) to define an appropriate probability distribution. For each mortality outcome, the BMR is sampled from a normal distribution derived from the mean and 95% confidence interval, while the TMREL is sampled from a uniform distribution bounded by the lower and upper bounds. For ozone the log-linear slope parameter (β) is sampled from a normal distribution derived from the RR (Equation 3) and its 95% uncertainty interval. For PM2.5, uncertainty in the RR is generated by assuming log-normal uncertainty; the 95% confidence intervals for each exposure level, provided by the GBD, are transformed to the log scale and used to derive a normal distribution for log(RR), from which samples are drawn and exponentiated to obtain RR values. We generate n random samples from each distribution and repeat the mortality calculation across all samples to capture the full range of plausible outcomes.

Sensitivity analyses are performed to determine the number of bootstrap samples required to effectively capture the parameter uncertainty. Ultimately selecting the number of samples is a balance between computational cost, given the high spatial resolution and associated memory requirements, and the need for a sufficiently large sample size to robustly span the uncertainty space. In our example (Section 5), we choose 300 bootstrap replications which is sufficient to characterize the distribution of outcomes while limiting computational expense; however, this choice can be adjusted based on the users requirements and computational resources. A testing script is available in the workflow to help users choose how many bootstrap replications to use.

3 Data processing

3.1 Climate model output and variable calculation

The first step in processing the climate model data into the format needed to calculate mortality is to extract the correct variables. One way to do this is to use the Earth System Grid Federation, an open source platform which provides access to a selection of Coupled Model Intercomparison Project (CMIP) data (Cinquini et al., 2014; https://esgf.github.io/index.html). The CMIP variables required for the analysis are provided in Table 1 (see https://clipc-services.ceda.ac.uk/dreq/mipVars.html for variable descriptions).

Table 1. Variables required for the analysis outline in the workflow.

VariableStandard nameTemporal frequencyDimensions
sfo3O3 volume mixing ratio in lowest model layer1* or 3 hTime lat lon
mmrbcElemental carbon mass mixing ratioMonthly or annual+Time lev lat lon
mmroaTotal organic aerosol mass mixing ratioMonthly or annual+Time lev lat lon
mmrso4Aerosol sulfate mass mixing ratioMonthly or annual+Time lev lat lon
mmrssSea-salt aerosol mass mixing ratioMonthly or annual+Time lev lat lon
mmrdustDust aerosol mass mixing ratioMonthly or annual+Time lev lat lon
taAir temperatureMonthly or annual+Time lev lat lon

Temporal frequencies with an asterisk (*****) are preferred and those with a cross (+) require some adjustment to the code to integrate with the workflow but do not change the output.

For ambient ozone exposure, 1-hourly surface ozone concentrations are preferred for the calculation of the required variable, however not all models provide this. If 3-hourly resolution is available, these can be interpolated to 1-hourly resolution, however this should be interpreted with caution as linear interpolation between 3-hourly points may smooth sharp concentration peaks, possibly underestimating MDA8 on high ozone days. To integrate with the framework the data needs to be converted to ppb. In our example we convert surface ozone concentrations from mol/mol to ppb, however we also provide an example script using another climate model converting surface ozone from kg/kg to ppb.

Not all climate models provide a singular PM2.5 variable so an approximate PM2.5 is calculated to ensure intermodal comparisons are consistent. The calculation uses a combination of the monthly aerosol mass mixing ratio of black carbon (BC), total organic aerosol (OA–both primary and secondary sources), sulfate aerosol (SO4), sea salt (SS), and dust (DU) using the formula from Turnock et al. (2020).

PM2.5=BC+OA+ SO4+(0.25 × SS)+(0.1 × DU)(4)

These data are then converted from a mass mixing ratio in kg/kg to a concentration in μg/m3 by multiplying by the air density, calculated at each grid point from the local pressure and temperature using the ideal gas law.

While this approach provides a consistent method for approximating PM2.5 across models that do not output a unified PM2.5 diagnostic, it carries several sources of uncertainty that should be considered when interpreting the results. The formula omits nitrate and ammonium aerosols, a design choice to allow model intercomparison since many CMIP6 models did not include the simulation of ammonium nitrate in their aerosol scheme (Turnock et al., 2020). Nitrate and ammonium can be substantial contributors to PM2.5 in regions with high ammonia and NOx emissions, which suggests an underestimate of total PM2.5 in those regions and correspondingly, a conservative estimate of associated mortality. Furthermore, the scaling factors applied to sea salt (0.25) and dust (0.1), intended to approximate the fine-mode fraction of these pre-dominantly coarse-mode species, can vary substantially by model and are likely to depend on the individual aerosol scheme and the simulated aerosol size distribution within a particular model (Turnock et al., 2020). For example, Turnock et al. (2020) found that for one CMIP6 model a sea salt factor of 0.01 was more appropriate, owing to differences in the model's treatment of sea salt hygroscopicity and assumed size range. Finally, the underlying representation of primary and secondary organic aerosol, black carbon, and sulfate can differ across models in number of aerosol modes, assumed size distribution, and mixing state. Therefore, the use of Equation 4 is likely to introduce some errors (Turnock et al., 2020), but it does provide an estimate that is consistent across models. The benefit of the modular nature of this framework means that users can replace this calculation of PM2.5 with data that meets their use case. For example, more detailed output from a chemistry transport model can be used as input here, instead of following this formula.

To leverage the GBD framework, climate model data must match the GBD metrics in both space and time. For mortality attributable to ambient surface ozone the GBD uses the relevant peak-season ozone health metric, OSDMA8, in ppb. This is defined as the highest seasonal average of 8-h daily maximum surface ozone concentrations. OSDMA8 is calculated in three steps. First, the daily maximum 8-h average (MDA8) ozone concentration is calculated from 8-h rolling means. Second, a monthly mean MDA8 is calculated for each month, and a 6-month rolling mean is calculated across a 15 month window from January of year y to March of year y+1. Using a 15-month window rather than a calendar year allows the metric to capture the peak exposure season in the Southern Hemisphere as well as the Northern Hemisphere. Third, OSDMA8, the seasonal maximum exposure, for year y is the maximum of the 6-month rolling means.

Surface ozone has a strong diurnal and seasonal cycle as it is a secondary pollutant produced by photochemical reactions between sunlight and precursor gases such as nitrogen oxides and volatile organic compounds. OSDMA8 is used to capture the seasonal variation and assess the long-term risk of exposure to surface ozone on human health. It is also consistent with the current World Health Organization air quality guidelines (https://iris.who.int/handle/10665/345329).

For mortality attributable to ambient PM2.5 the GBD uses the annual mean concentration of PM2.5 in μg/m3. The workflow calculates this as a straightforward annual average of the monthly data.

3.2 Bias correction and downscaling

Global climate models exhibit systematic errors in simulating surface ozone and PM2.5 concentrations (Turnock et al., 2020; Griffiths et al., 2021). These biases are often considered negligible when analyzing changes between time periods, under the assumption that the bias remains constant. However, this assumption may not hold when applying model output to health impact assessments, especially in cases when the ERF is non-linear. Therefore, a bias correction is needed to align the simulated concentrations with observation-based datasets before further analysis.

The bias correction in this workflow is performed using the delta method (Räisänen, 2007), a widely used and simple approach applied to climate change simulations. This method assumes that the relative bias between modeled and observed climate is constant over time, such that the ratio of future to baseline modeled values can be applied to the observed baseline to produce bias-corrected future estimates. The delta method has been used in several other health impact assessments for air quality changes (e.g., Akritidis et al., 2024; Tarín-Carrasco et al., 2022; Silva et al., 2017). A delta value, or climate change signal, is calculated as the ratio between the average modeled parameter (e.g., OSDMA8 or annual PM2.5) for a future period (m¯f) to the average modeled parameter for the baseline period (m¯base) (Equation 5).

Δf = m¯fm¯base(5)

This delta is then multiplied by the average observed parameter for the baseline period (o¯base) resulting in a bias corrected average modeled parameter for the future period.

m¯bc f= Δf × o¯base(6)

To align with the GBD exposure-response functions the data must also be at the same spatial resolution. The observations used within the GBD framework for both surface ozone and PM2.5 are at a 0.1° x 0.1° spatial resolution, much finer than typical climate model resolutions (Lee et al., 2021). To downscale the bias corrected data a bilinear interpolation is applied to the delta value (Δ*f*) before the final step (Equation 6), resulting in a 0.1° x 0.1° resolution bias corrected average modeled parameter. Similarly to the BMR, we choose our baseline period as the years 1990–2009.

The delta method assumes that the historical bias between modeled data and observations remains approximately stationary over the projection period. This is a reasonable assumption where the primary driver of future change is a broadly-scaled shift in emissions or meteorology that the model captures proportionally correctly, even if its absolute concentrations are biased. However, this assumption may not hold when future concentrations depend on non-linear atmospheric chemistry, changes in precursor emission composition, and contributions of PM2.5 from wildfire activity. Where such processes are expected to substantially change over the projected period, more advanced bias-correction methods that preserve changes in the shape of the concentration distribution (e.g., quantile mapping), or direct use of chemical transport model output at appropriate resolution, may be preferable to the delta method used here. The model we use in our example (Section 5) does not include interactive wildfires, for example, so changes in future wildfires are not included. Therefore we consider the delta method appropriate in this instance and it has benefit for model comparisons of CMIP6 projections. However, its assumptions should be re-evaluated for applications involving significant changes to emission projections, those with interactive wildfires and for studies examining changes in future atmospheric chemistry, chemistry transport models should be considered. The modular nature of this workflow allows users to test and compare other bias correction methods at this stage to align best with their use case.

To bias-correct the climate model data in our example, we use the 1990–2009 average of yearly global surface ozone (OSDMA8) estimates produced by DeLang et al. (2021). These estimates are provided at 0.1° x 0.1° spatial resolution globally and were generated for 1990–2017 using a combination of observations and models to support the Global Burden of Disease.

For PM2.5, we use exposure estimates from the Global Burden of Disease Collaborative Network (2026) provided at 0.1° x 0.1° spatial resolution globally. From 2006 onwards, these estimates are derived from the satellite-based V5.GL.04 product, which combines aerosol optical depth with the GEOS-Chem chemical transport model, calibrated against ground based observations (Van Donkelaar et al., 2021). This represents a change from GBD 2021, which used the Data Integration Model for Air Quality (DIMAQ) and estimates from Shaddick et al. (2018). For years prior to 2006 (1990, 1995, and 2000–2005), GBD 2023 provides GBD 2021 exposure estimates rescaled by location using overlapping years between the two products, with a 3-year rolling average applied across 2006–2008 to smooth the transition (Hay et al., 2025). Our baseline window is 1990–2009, so we used the average across these years as our reference period for bias correction.

4 Validation

To assess the workflow's ability to reproduce established estimates, we compared the workflow-derived mortality with the GBD 2023 cause-specific estimates for the year 2000. We used the observational datasets (Section 3.2) applied for bias correction in our workflow as the exposure inputs which is consistent with the exposure data underlying the GBD's own calculations for this period. We apply the BMR and population data for the year 2000 to calculate cause specific mortality.

We calculate the mean country-level mortality, using central estimates for the BMR and epidemiological parameters, and mean global mortality with a 95% confidence interval using the parametric bootstrapping approach explained in Section 2.4.

For ozone, our workflow's country-level and global mortality estimates closely reproduce the GBD 2023 values (Figures 2a, b). For PM2.5, our mean global estimate is 78% higher than the mean GBD 2023 value (Figures 2c, d; ~5 vs. ~2.8 million). The discrepancy between these estimates is concentrated in a small number of countries (< 10% of all countries account for the discrepancy). The largest single contributor is China, where our estimate exceeds the GBD value by ~883,000 deaths, accounting for almost 40% of the total discrepancy between the estimates. We attribute this to differences in the resolution of the BMR data. The GBD incorporates subnational (e.g., state or province level) BMR data in their mortality estimates where available; however, for some countries, including China, they are unable to make these subnational data available. In our workflow we use country-level BMRs for all locations, and we believe this is likely driving the total discrepancy between our estimates and those of the GBD, rather than the mortality calculation methodology. This is supported by the close agreement for ozone, which uses broadly the same calculation framework, and the close agreement across many other countries. PM2.5 mortality is likely more sensitive to this BMR resolution mismatch than ozone because it is calculated across seven mortality causes compared to the single cause attributed to ozone.

Verification of ozone-attributable **(a, b)** and PM<sub>2.5</sub>-attributable **(b, c)** mortality for the year 2000. **(a, c)** Country and global level mortality as calculated using the workflow for the year 2000. **(b, d)** Country and global level mortality estimates for year 2000 from GBD 2023.
Figure 2. Verification of ozone-attributable (a, b) and PM2.5-attributable (b, c) mortality for the year 2000. (a, c) Country and global level mortality as calculated using the workflow for the year 2000. (b, d) Country and global level mortality estimates for year 2000 from GBD 2023.

5 Example of applying workflow

In this section, we present the output from a worked example demonstrating the application of the workflow and illustrating the types of results it can produce. The accompanying GitHub repository (https://doi.org/10.5281/zenodo.18436806) is designed to be used alongside this section as a tutorial and includes the functions, processing, and plotting scripts required to reproduce the results shown here, together with the input data (https://doi.org/10.5281/zenodo.18436834). Each directory contains a README.md file describing the required inputs, the pre-processing and processing steps, and the expected outputs. As the full workflow described in the preceding sections is implemented and documented in the repository, this section focuses on illustrating the outputs of the workflow and presenting selected results from the example application, although these should be interpreted as examples of results not robust projections as only one ensemble member from one model is used.

The following example shows results from one ensemble member of the SSP2-4.5 scenario using the Community Earth System Model version 2 (CESM2) (Danabasoglu et al., 2020). The SSP2-4.5 scenario describes a “middle of the road” pathway where future social, economic and technological trends do not substantially change from historical patterns. Global mean temperature under this scenario increases to approximately 2.5 °C above pre-industrial by the end of the century. Global population increases to around 9.5 billion in the later half of the century, leveling off to ~9 billion by 2100. Regarding air pollution, the SSP2 scenario projects ambitious pollution concentration targets over the following century as more value is placed on health and the environment, however it is not the most ambitious scenario (Rao et al., 2017).

The results of the workflow provide annual estimates of excess mortality due to surface ozone and PM2.5 on a 0.1° x 0.1° resolution. These results can be aggregated globally, by country or by region. Figures 3, 4 show examples of how these can be visualized and the workflow provides code to calculate these aggregated values. In this example we looked at the years 2020–2084. However, for surface ozone, results for the year 2084 are not included since calculating the peak-season ozone metric for all regions requires the months January of the first year to March of the following year (Section 3.2). The global and regional results include uncertainty ranges from the BMRs and epidemiological parameters (Section 2.4) and show results from only 1 year, not a difference between two time periods. The global and regional figures show the 95% confidence interval from the epidemiological parameter uncertainty. Whereas, the country level results are based on the central estimates only.

COPD mortality per 100,000 due to surface ozone exposure **(a)** GBD regions for the year 2080. The orange circle represents the mean, the white diamond the median and the whiskers are the 95% confidence interval. **(b)** Global time series from 2020–2083 including the mean and 95% confidence interval. **(c)** Country level for the year 2080 using central estimates of the BMR and epidemiological parameters.
Figure 3. COPD mortality per 100,000 due to surface ozone exposure (a) GBD regions for the year 2080. The orange circle represents the mean, the white diamond the median and the whiskers are the 95% confidence interval. (b) Global time series from 2020–2083 including the mean and 95% confidence interval. (c) Country level for the year 2080 using central estimates of the BMR and epidemiological parameters.
Mortality per 100,000 due to PM<sub>2.5</sub> exposure for seven outcomes (COPD, IHD, LRI, Type 2 Diabetes, Lung Cancer, Stroke, and Dementia) **(a)** GBD regions for the year 2080. The orange circle represents the mean, the white diamond the median and the whiskers are the 95% confidence interval. **(b)** Global time series from 2020–2084. The black line shows the total mortality from all seven causes and the gray hatching shows the 95% confidence interval. The mean for each mortality outcome is stacked under the total mortality to show the relative contribution of each cause. **(c**) Country level for the year 2080 using central estimates of the BMR and epidemiological parameters.
Figure 4. Mortality per 100,000 due to PM2.5 exposure for seven outcomes (COPD, IHD, LRI, Type 2 Diabetes, Lung Cancer, Stroke, and Dementia) (a) GBD regions for the year 2080. The orange circle represents the mean, the white diamond the median and the whiskers are the 95% confidence interval. (b) Global time series from 2020–2084. The black line shows the total mortality from all seven causes and the gray hatching shows the 95% confidence interval. The mean for each mortality outcome is stacked under the total mortality to show the relative contribution of each cause. (c) Country level for the year 2080 using central estimates of the BMR and epidemiological parameters.

Figure 3 shows the excess COPD mortality due to surface ozone. In this example we have normalized the excess mortality by population and present mortality as a rate per 100,000 people. Figure 3b shows that for this one ensemble member, in CESM2 under SSP2-4.5, the global ozone-related mortality is projected to decrease from 4.8 per 100k [95% CI: 1.3–8.4 per 100k] in 2020 to 3.4 per 100k [95% CI: 0.9–5.8 per 100k] in 2083. This is likely due to a decrease in ozone precursor emissions as a result of air pollution targets in SSP2; however, for a robust projection of future global mortality in this scenario, several ensemble members and/or models should be considered.

Figure 4 shows the total excess mortality due to PM2.5, including the 7 mortality outcomes the GBD has linked to PM2.5 exposure. Figure 4b demonstrates the contribution of each mortality outcome to the overall global mortality rate per 100,000 (black line), and includes the 95% confidence interval around the total mortality rate from the epidemiological parameters uncertainty. By breaking the global mortality rate down by mortality outcome, it becomes clear which health outcomes contribute most to overall PM2.5-related mortality and if any are decreasing faster or slower than others.

Further examples of ways to visualize results from this workflow can be found in the supplement and accompanying GitHub, including the total mortality from both PM2.5 and surface ozone exposure (Supplementary Figure S2). In this example, the contribution from PM2.5 related mortality dominates the total air quality mortality assessed here.

6 Discussion

The main objective of this paper is to provide a transparent, standardized framework that allows climate model projections to be applied to epidemiological methods to estimate mortality due to surface ozone and PM2.5. The Global Burden of Disease (GBD) provides statistical relationships to estimate mortality due to these air pollutants; however, the inputs required for these exposure-response functions are rarely available in the format produced by Earth system models. This workflow clearly outlines the steps required to convert climate model output into the health-relevant exposure metrics, providing code that others can easily apply or adapt. This includes calculating pollution metrics that match GBD definitions, bias correction and spatial downscaling.

Many health impact assessments use study specific exposure-response functions to estimate the relationship between exposure to air pollutants and mortality (e.g., Anenberg et al., 2010; Silva et al., 2017). We chose to implement those provided by the GBD as they are a well-established aggregation of epidemiological evidence. By incorporating these standardized relationships within the workflow, we hope it can be utilized to make consistent comparisons between scenarios and studies. Since the GBD publishes regular updates to their relationships based on new studies, this framework can be easily updated if and when new exposure-response functions are implemented.

Our workflow enables a multitude of analyses, including how future air quality changes may influence mortality at global, regional or national levels. It can be used to compare mortality across different future climate scenarios and quantify the potential benefits or risks of different mitigation strategies. The workflow also provides an opportunity to explore the uncertainties that arise from internal variability: the natural, chaotic fluctuations of the climate system that occur even in the absence of an external forcing. Internal climate variability can substantially affect the timing, magnitude and spatial distribution of climate-driven air quality changes, particularly at regional scales and over shorter time scales (e.g., Callahan and Justin, 2020). Typically, climate modeling studies will provide multiple ensemble members, the same scenario run several times with slightly different initial conditions, to evaluate the range and impact of internal climate variability. Whilst this is regularly examined in climate modeling studies, it is often not considered when examining downstream climate impacts (e.g., Schwarzwald and Nathan, 2022; Hart et al., 2025). It is worth noting that the workflow directly incorporates uncertainty only in the epidemiological parameters, through the confidence intervals provided by the GBD ERFs. Other sources of uncertainty relevant to the overall mortality projection, including those arising from the choice of climate model, emissions scenario, bias correction method, and population projection, are not directly propagated through the workflow. Users wishing to explore these additional sources of uncertainty can do so by running the workflow using different combinations of inputs and comparing the resulting range of mortality estimates.

Beyond conventional climate change scenarios, our workflow could also be used to estimate the health impacts of different plausible climate futures, for instance those resulting from proposed solar climate intervention approaches. Solar climate intervention is the deliberate large-scale intervention into the climate system, intended to reduce the impacts of climate change by reflecting a small portion of sunlight back to space (National Academies of Sciences, 2021). One could evaluate the health implications of stratospheric aerosol injection (SAI), a type of solar climate intervention, which may alter the distribution and magnitude of surface ozone and concentrations of PM2.5 (e.g., Xia et al., 2017; Visioni et al., 2020). Our workflow allows for a consistent comparison of a climate future with and without the intervention and provides an opportunity to examine the wider potential risks and benefits of these proposed interventions.

Beyond the assumptions noted elsewhere in the paper, some additional points are worth stating explicitly. Firstly, the GBD ERFs used here assume that PM2.5 carries equal toxicity per unit mass, regardless of its chemical composition or emission source. In reality, PM2.5 toxicity is thought to vary with composition and source (e.g., Franklin et al., 2008). Since this workflow is modular, users with access to source or composition specific ERFs could substitute these in place of the GBD ERFs. Secondly, wildfire activity is expected to increase under climate change in many regions and can substantially affect PM2.5 concentrations; users interested in assessing the mortality implications of changing fire activity should therefore ensure their chosen model configuration includes interactive, rather than fixed or prescribed, fire emissions. The example we share in Section 5 does not include the model configuration which interactively models future changes in fires. Finally, this workflow currently estimates mortality only, not morbidity. Mortality refers to deaths attributable to pollutant exposure, whereas morbidity captures non-fatal outcomes such as disease incidence, hospitalization, or years lived with disability. The GBD publishes relative risks for many non-fatal outcomes, and future work could incorporate these into the workflow using the same structure.

Overall this work aims to provide a reliable and adaptable tool which lowers barriers to interdisciplinary research on climate and health. By bringing together the key steps, datasets, and providing reproducible code to process model data and estimate mortality, it creates an accessible starting point for inter- and cross-disciplinary analyses. We hope it encourages further collaboration between climate scientists and epidemiologists and facilitates a wide range of analysis.

Generative AI statement

The author(s) declared that Generative AI was used in the creation of this manuscript. ChatGPT was utilized to help develop the code and plots.

Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us.

Publisher's note

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.

Supplementary material

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fclim.2026.1847091/full#supplementary-material

Supplementary material Supplementary material at the source

Footnotes

Edited by: Warren Dodd, University of Waterloo, Canada

Reviewed by: Fatemeh Ganji, Iowa State University, United States

Chris Wells, University of Leeds, United Kingdom

Statements

Conflict of interest

The author(s) declared that this work was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Funding

The author(s) declared that financial support was received for this work and/or its publication. This work is funded by a grant from Reflective. JWH is supported by the Walter Scott Jr. College of Engineering, Colorado State University.

Data availability

The workflow presented here, and the data used for the example, are available in the Zenodo repository via 10.5281/zenodo.18436806 and 10.5281/zenodo.18436834 respectively.

Author contributions

AW: Conceptualization, Data curation, Formal analysis, Investigation, Methodology, Software, Visualization, Writing – original draft, Writing – review & editing. JH: Funding acquisition, Project administration, Resources, Supervision, Writing – review & editing. EG: Methodology, Writing – review & editing. GA: Conceptualization, Funding acquisition, Methodology, Project administration, Supervision, Writing – review & editing.

References

  1. Akhtar R. (2020). Extreme Weather Events and Human Health: International Case Studies. Springer International Publishing. doi: doi:10.1007/978-3-030-23773-8
  2. Akritidis D., Sara B., Prodromos Z., Chowdhury S., Horowit L. W., Naik V.. (2024). Strong increase in mortality attributable to ozone pollution under a climate change and demographic scenario. Env. Res. Letters 19: 024041. doi: doi:10.1088/1748-9326/ad2162
  3. Alvim-Ferraz M. C. M., Sofia I. V. S., Fernando G. M., Maria P. F. (2024). Tropospheric and stratospheric ozone: scientific history and shifts in early perspectives regarding the impact on human health. Atmosphere 15: 1504. doi: doi:10.3390/atmos15121504
  4. Anenberg S. C., Larry W., Horowitz D. Q. T., Jason W. (2010). An estimate of the global burden of anthropogenic ozone and fine particulate matter on premature human mortality using atmospheric modeling. Env. Health Perspectiv. 118, 1189 – 1195. doi: doi:10.1289/ehp.0901220
  5. Baker R. E., Ayesha S. M., Ian F. M., Rajeev M., Rasambainarivo F., Rice B. L.. (2022). Infectious disease in an era of global change. Nature. Rev. Microbiol. 20, 193 – 205. doi: doi:10.1038/s41579-021-00639-z
  6. Brauer M., Roth G. A., Aravkin A. Y., Zheng P., Abate K. H., Abate Y. H.. (2024). Global burden and strength of evidence for 88 risk factors in 204 countries and 811 subnational locations, 1990–2021: a systematic analysis for the global burden of disease study 2021. Lancet. 403, 2162 – 2203. doi: doi:10.1016/S0140-6736(24)00933-4
  7. Callahan C. W., Justin S. M. (2020). The influence of internal climate variability on projections of synoptically driven beijing haze. Geophys. Res. Letters 47: e2020GL088548. doi: doi:10.1029/2020GL088548
  8. Cinquini L., Daniel C., Chris M., John H., Galen S., Feiyi W.. (2014). The earth system grid federation: an open infrastructure for access to distributed geospatial data. Future Generation Computer Syst. 36, 400 – 417. doi: doi:10.1016/j.future.2013.07.002
  9. Crutzen P. J. (1988). “ Tropospheric ozone: an overview,” in Tropospheric Ozone, ed Isaksen I. S. A. (Springer Netherlands).
  10. Danabasoglu G. J. -F., Lamarque J. B., Bacmeister J., Bailey D. A., DuVivier A. K., Edwards J.. (2020). The community earth system model version 2 (CESM2). J. Adv. Modeling Earth Syst. 12:e2019MS001916.
  11. DeLang M. N., Jacob S. B., Kai-Lan C., Marc L. S., Owen R. C. Martin G. S.. (2021). Mapping yearly fine resolution global surface ozone through the bayesian maximum entropy data fusion of observations and model output for 1990–2017. Env. Sci. Technol. 55, 4389 – 4398. doi: doi:10.1021/acs.est.0c07742
  12. Ebi K. L. A., Haines R. F. S. Andrade, Åström, C., Barreto M. L., Bonell A.. (2025). The attribution of human health outcomes to climate change: transdisciplinary practical guidance. Climatic Change 178: 143.
  13. Franklin M., Petros K., Joel S. (2008). The role of particle composition on the association between PM2.5 and mortality. Epidemiology 19, 680 – 689. doi: doi:10.1097/EDE.0b013e3181812bb7
  14. Ganji F., Jon S., Lu L., Qin Y., Zhou Y., Graham N.. (2026). Disparities in ambient air pollution exposure among the U.S. population amid climate change. iScience 29: 114603. doi: doi:10.1016/j.isci.2025.114603
  15. Gao J. (2020). Global 1-Km downscaled population grids, SSP-consistent projections and base year, v1.01 (2000-2100). Version 1.0. With Jing G. (Harvard Dataverse).
  16. Ghosh R., Kate C., Katrin B., Sara W., Aaron C., Michael B. (2021). Ambient and household PM2.5 pollution and adverse perinatal outcomes: a meta-regression and analysis of attributable global burden for 204 countries and territories. PLoS Med. 18: e1003718. doi: doi:10.1371/journal.pmed.1003718
  17. Global Burden of Disease Collaborative Network (2024). Global Burden of Disease Study 2023 (GBD 2023) Results. Seattle: Institute for Health Metrics and Evaluation (IHME). Available online at:
  18. Global Burden of Disease Collaborative Network (2026). Global Burden of Disease Study 2023 (GBD 2023) Air Pollution Exposure Estimates and Risk Curves 1990-2023. Seattle, WA: Institute for Health Metrics and Evaluation (IHME).
  19. Griffiths P. T., Lee T. M., Guang Z., Shin Y. M., Abraham N. L., Archibald A. T.. (2021). Tropospheric ozone in CMIP6 simulations. Atmospheric Chemistry and Phys. 21,4187–4218. doi: doi:10.5194/acp-21-4187-2021
  20. Hart W. S., James W. H., Alexander R. K., Meera C., Matt J. K., Robin N. T. (2025). Climate variability amplifies the need for vector-borne disease outbreak preparedness. Proc. Nat. Acad. Sci. 122: e2507311122. doi: doi:10.1073/pnas.2507311122
  21. Hay S. I., Kanyin L. O, Damian, F. S., Aalipour M. A., Aalruz H., Ababneh H. S.. (2025). Burden of 375 diseases and injuries, risk-attributable burden of 88 risk factors, and healthy life expectancy in 204 countries and territories, including 660 subnational locations, 1990–2023: a systematic analysis for the global burden of disease study 2023. Lancet 406, 1873 – 1922. doi: doi:10.1016/S0140-6736(25)01637-X
  22. Health Effects Institute (2025). State of Global Air 2025: A Report on Air Pollution and Its Role in the World's Leading Causes of Death. Boston, MA: Health Effects Institute.
  23. Horton D. E., Christopher B. S., Deepti S., Noah S. D. (2014). Occurrence and persistence of future atmospheric stagnation events. Nature Climate Change 4, 698 – 703. doi: doi:10.1038/nclimate2272
  24. Hughes B., Peterson C., Rothman D., Solorzano J. (2014). IFs Health Model Documentation, Pardee Center for International Futures, Denver. Available online at: (Accessed March 2025).
  25. Institute for Health Metrics and Evaluation (2024). Global burden of disease study 2021 (GBD 2021) air pollution exposure estimates 1990-2021. Institute for Health Metrics and Evaluation.
  26. Kim S. -Y., Esther K., Woo J. K. (2020). Health effects of ozone on respiratory diseases. Tuberculosis and Respiratory Dis. 83, S6 – 11. doi: doi:10.4046/trd.2020.0154
  27. Lao X. Q., Cui G., Ly-yun C, Bo, Y., Zhang Z., Chuang Y. C.. (2019). Long-term exposure to ambient fine particulate matter (PM2.5) and incident type 2 diabetes: a longitudinal cohort study. Diabetologia 62, 759 – 769. doi: doi:10.1007/s00125-019-4825-1
  28. Lee J.-Y., Marotzke J., Bala G., Cao L., Corti S., Dunne J. P.. (2021). ”Future global climate: scenario-based projections and near-term information,” in Climate Change 2021: The Physical Science Basis. Contribution of Working Group I to the Sixth Assessment Report of the Intergovernmental Panel on Climate Change, eds. Masson-Delmotte V., Zhai P., Pirani A., Connors S. L., Péan C., Berger S.. (Cambridge/New York, NY: Cambridge University Press), 553 – 672. doi: doi:10.1017/9781009157896.006
  29. Lin K., Junxian Q., Jiafen L., Yiyue C., Bing Z., Xi X. (2025). Comprehensive Insights into the disease burden linked to air pollution exposure across U.S. states: findings from the 2021 global burden of disease study. Front. Public Health 13: 1636544. doi: doi:10.3389/fpubh.2025.1636544
  30. Liu Q., Dun L., Zhe X., Yaogang W., Jue L. (2025). Disparities in global disease burden attributed to ambient particulate matter pollution and household air pollution from solid fuels. Ecotoxicol. Environ. Safety 291: 117908. doi: doi:10.1016/j.ecoenv.2025.117908
  31. Malashock D. A., Marissa N. D., Jacob S. B., Serre M. L., West J. J., Chang K.-L.. (2022). Global trends in ozone concentration and attributable mortality for urban, peri-urban, and rural areas between 2000 and 2019: a modelling study. Lancet Planetary Health 6, e958 – e67. doi: doi:10.1016/S2542-5196(22)00260-1
  32. McDuffie E., Michael B., Randall M., Hammer M., van Donkelaar A., Smith S.. (2021b). GBD-MAPS-global: gridded fractional source contribution results. Version 2021_v1. Zenodo.
  33. McDuffie E. E., Martin R. V., Spadaro J. V., Burnett R., Smith S. J., O'Rourke P.. (2021a). Source sector and fuel contributions to ambient PM2.5 and attributable mortality across multiple spatial scales. Nature Communications. 12: 3594. doi: doi:10.1038/s41467-021-23853-y
  34. National Academies of Sciences Engineering, and Medicine. (2021). Reflecting sunlight: recommendations for solar geoengineering research and research governance. National Academies Press.
  35. O'Neill B. C., Elmar K., Keywan R., Ebi K. L., Hallegatte S., Carter T. R.. (2014). A new scenario framework for climate change research: the concept of shared socioeconomic pathways. Climatic Change 122, 387 – 400. doi: doi:10.1007/s10584-013-0905-2
  36. Orru H., Ebi K. L., Forsberg B. (2017). The interplay of climate change and air pollution on health. Current Env. Health Rep. 4, no. 4: 504 – 13. doi: doi:10.1007/s40572-017-0168-6
  37. Paprotny D. (2026). Historical and future population density data for CMIP7 and ScenarioMIP. Zenodo.
  38. Post E. S., Anne G., Chris W. Philip M Jin H., Lai-Yung L.. (2012). Variation in estimated ozone-related health impacts of climate change due to modeling choices and assumptions. Environ. Health Perspectiv. 120, 1559 – 1564. doi: doi:10.1289/ehp.1104271
  39. Pozzer A., Brendan S., Yiannis P., Jean S., Dimitris A., Sourangsu C.. (2024). Atmospheric health burden across the century and the accelerating impact of temperature compared to pollution. Nat. Communications 15: 9379. doi: doi:10.1038/s41467-024-53649-9
  40. Räisänen J. (2007). How reliable are climate models? Tellus A: Dynamic Meteorology and Oceanography 59: 2. doi: doi:10.1111/j.1600-0870.2006.00211.x
  41. Rao S., Zbigniew K., Steven J. S., Rita V. D., Frank D., Lex B.. (2017). Future air pollution in the shared socio-economic pathways. Global Environmental Change 42, 346 – 358. doi: doi:10.1016/j.gloenvcha.2016.05.012
  42. Riahi K., van Vuuren D. P., Kriegler E., Edmonds J., O'Neill B. C., Fujimori S.. (2017). The shared socioeconomic pathways and their energy, land use, and greenhouse gas emissions implications: An overview. Glob. Environ. Change 42, 153 – 168. doi: doi:10.1016/j.gloenvcha.2016.05.009
  43. Richardson D., Amanda S. B., Damien I., Matear R. J., Monselesan D. P., Risbey J. S.. (2022). Global increase in wildfire potential from compound fire weather and drought. NPJ Climate and Atmospheric Sci. 5: 23. doi: doi:10.1038/s41612-022-00248-4
  44. Ru M., Michael B., Jean-François L., Drew S. (2021). Exploration of the global burden of dementia attributable to PM2.5: what do we know based on current evidence? GeoHealth 5: e2020GH000356. doi: doi:10.1029/2020GH000356
  45. Schwarzwald K., Nathan L. (2022). The importance of internal climate variability in climate impact projections. Proc. Nat. Acad. Sci. 119: e2208095119. doi: doi:10.1073/pnas.2208095119
  46. Shaddick G., Matthew L. T., Heresh A., David B., Aaron C., Joseph F.. (2018). Data integration for the assessment of population exposure to ambient air pollution for global burden of disease assessment. Environ. Sci. Technol. 52, 9069 – 78. doi: doi:10.1021/acs.est.8b02864
  47. Silva R. A., Jason W. J., Jean-François L., Drew T. S., William J. C., Greg F.. (2017). Future global mortality from changes in air pollution attributable to climate change. Nature Climate Change 7, 647 – 651. doi: doi:10.1038/nclimate3354
  48. Steenland K., Ben A. (2006). An overview of methods for calculating the burden of disease due to specific risk factors. Epidemiology 17, 512 – 519. doi: doi:10.1097/01.ede.0000229155.05644.43
  49. Tarín-Carrasco P., Ulas I., Camilla G., Laura P. –P., Pedro J. –G. (2022). Reducing future air-pollution-related premature mortality over Europe by mitigating emissions from the energy sector: assessing an 80 % renewable energies scenario. Atmospheric Chemistry and Physics 22, 3945 – 3965. doi: doi:10.5194/acp-22-3945-2022
  50. Turnock S. T., Robert J. A., Martin A. (2020). Historical and future changes in air pollutants from CMIP6 models'. Atmospheric Chemistry and Physics 20, 14547 – 14579. doi: doi:10.5194/acp-20-14547-2020
  51. Van Donkelaar A., Melanie S. H., Liam B., Michael B., Jeffery R. B., Michael J.. (2021). Monthly global estimates of fine particulate matter and their uncertainty. Environ. Sci. Technol.doi: doi:10.1021/acs.est.1c05309
  52. Van Vuuren D. P., Brian C. O., Claudia T., Benjamin M. S., Louise P. C., Pierre F.. (2026). The scenario model intercomparison project for CMIP7 (ScenarioMIP-CMIP7). Geosci. Model Develop. 19, 2627 – 2656. doi: doi:10.5194/gmd-19-2627-2026
  53. Visioni D., Eric S., Douglas G. M., Mahowald N. M., Goodale C. L., Lili X. (2020). What goes up must come down: impacts of deposition in a sulfate geoengineering scenario. Environ. Res. Letters 15: 094063. doi: doi:10.1088/1748-9326/ab94eb
  54. World Cancer Research Fund/American Institute for Cancer Research (2007). Food, Nutrition, Physical Activity, and the Prevention of Cancer: a Global Perspective. Washington, DC: AICR, 2007.
  55. Xia L., Peer J. N., Simone T., Alan R. (2017). Impacts of stratospheric sulfate geoengineering on tropospheric ozone. Atmospheric Chemistry and Physics. 17, 11913 – 11928. doi: doi:10.5194/acp-17-11913-2017
  56. Xing Y. –F., Yue-Hua X., Min-Hua S., Yi-Xin L. (2016). The impact of PM2.5 on the human respiratory system. J. Thoracic Dis. 8, E69 – E74.
  57. Xu L., Bo W., Ying W., Huipeng Z., Danni X., Yibing Z.. (2025). Characterization and source apportionment analysis of PM2.5 and ozone pollution over Fenwei Plain, China: insights from PM2.5 component and VOC observations. Toxics 13: 123. doi: doi:10.3390/toxics13020123
  58. Yang C. –E., Joshua S. F., Yongqiang L., Xinyi D., Yang L. (2022). Projections of future wildfires impacts on air pollutants and air toxics in a changing climate over the Western United States. Environ. Pollution 304: 119213. doi: doi:10.1016/j.envpol.2022.119213

Citation

Wells, A. F., et al. (2026). An open-access workflow combining climate model projections with epidemiological frameworks to assess air quality mortality. Frontiers in Climate, Volume 8. https://doi.org/
Received
Accepted
Published
Volume
8 - 2026

Copyright

Copyright © 2026 Wells, Hurrell, Gilleland and Anderson.

License: https://creativecommons.org/licenses/by/4.0/

*Correspondence: Alice F. Wells

Article metrics

Views
0
Downloads
0
Citations
0
Shares
0

Related articles

Original ResearchRapid intensification of Arabian Sea cyclonic storm and numerical simulation of ESCS Tauktae using the high-resolution ARW modelRohini Ashok, Kuvar Satya Singh, Ambily Thankachan, Reshm M. S.ReviewPolicy and institutional drivers of climate-smart agriculture transition in Sub-Saharan Africa: a review of governance, regulation, and institutional support with implications for South AfricaAdetomiwa Kolapo, Melusi Sibanda, Nolwazi Z. KhumaloPerspectiveMeasuring large-scale mCDR using a novel Atmospheric CO₂ Perimeter Flux (ACPF) methodPeter Fiekowsky, Akhilesh Vijay, Anoohya Jami