Atmospheric health burden across the century and the accelerating impact of temperature compared to pollution

Atmospheric conditions substantially affect human health: exposure to extreme temperatures and air pollution are both linked with increased mortality. The Global Burden of Disease (GBD) estimates indicate that these risk factors together cause roughly 6.5 (95% CI: 5.3–7.5) million deaths per year, with ambient air pollution responsible for about 60% and PM2.5 accounting for ~90% of air-pollution-attributable deaths. PM2.5 penetrates deep into the respiratory and circulatory systems, driving cardiovascular and other diseases. While numerous studies have evaluated each risk factor separately, simultaneous investigation of projected mortality from both long-term exposure to non-optimal temperature (including both heat and cold) and air pollution at global scale has been limited. Human activities influence both risk factors via emissions that alter atmospheric composition and climate, necessitating projections of future air quality and surface temperature. Using CMIP6 Earth System Models and Shared Socioeconomic Pathways (SSPs), this study aims to estimate current (around 2000) and future (to 2090) global mortality attributable to long-term exposure to PM2.5 and non-optimal temperature, quantify regional variations, and disentangle the roles of changing exposure, population growth, and aging.

Prior work links elevated air pollution and non-optimal temperatures to increased mortality and has quantified global burdens (e.g., GBD 2019; Gasparini et al. 2015; Zhao et al. 2021). PM2.5 is identified as the dominant pollutant driving air-pollution mortality. Studies have projected future air-pollution-related mortality under climate change and socioeconomic scenarios (e.g., Fang et al. 2013; Chen et al. 2020; Turnock et al. 2023), and assessed temperature-related mortality patterns across regions and climates (e.g., Burkart et al. 2021). However, comprehensive projections jointly assessing long-term mortality attributable to both non-optimal temperature and PM2.5 globally, and their evolution under multiple SSP scenarios with explicit treatment of demographics, remain scarce. This study addresses that gap by integrating CMIP6 climate-chemistry outputs, downscaled exposures, GBD exposure–response functions, and SSP-consistent population and age structure.

Overview: Mortality attributable to long-term exposure to non-optimal temperature and to PM2.5 is computed on a global grid and aggregated regionally and globally for 20-year time slices centered on decades from 2000 to 2090 under SSP1-2.6, SSP2-4.5, and SSP5-8.5. Computation of attributable mortality: For each grid cell, disease, age group, and risk factor X (temperature or PM2.5), annual attributable mortality Mort is computed as Mort = BMR × Pop × AF, where BMR is the cause-specific baseline mortality rate, Pop is population, and AF is the attributable fraction derived from exposure–response functions (ERFs). For temperature, daily AFs are computed and converted to annual mortality using multi-annual daily means for each 20-year period. Exposure fields:

• Climate/air quality models: CMIP6 outputs providing daily near-surface temperature and monthly PM2.5 for historical (1980–2014) and future (2015–2099) were used from three global models meeting data and resolution requirements: CESM2-WACCM, GFDL-ESM4, and MIROC-ESM2-0. Models are equally weighted.
• Downscaling and bias correction (temperature): Daily temperature is bias-adjusted and downscaled using the Pacific Climate Impacts Consortium’s ClimDown package: calibration against WFDE5 (0.5°) for 1990–2014 via a climate impacts algorithm, followed by climate-signal-preserving Quantile Delta Mapping using calibrated/observed data (1980–2014) and future projections (2015–2041). Final fields are statistically disaggregated per scenario/model.
• Downscaling and bias correction (PM2.5): Monthly PM2.5 is downscaled via bilinear interpolation and bias-corrected using the delta method relative to an observationally constrained reference (satellite–model fused product calibrated with ground observations via Geographically Weighted Regression). Bias correction is applied to 20-year climatological means (e.g., 1990–2009 baseline; subsequent 20-year time slices stepped every 10 years). Theoretical minimum-risk exposure level is uniformly distributed between 2.4–5.9 μg/m³.
• Temporal aggregation: Temperature (daily) and PM2.5 (annual means) are averaged into 20-year periods stepped every 10 years to align with mortality calculations.
• Target grid: All fields are regridded to the population grid at 7.5 arc-min (0.125° × 0.125°); analysis domain spans −54.95° to 67.9° N to match observational PM2.5 coverage. Population and baseline mortality:
• Population: Gridded population for 2000 and projections (2000–2090, 10-year steps) at 0.125° are from SEDAC and consistent with SSPs.
• Age structure: Country-level age distributions (2000–2100, 5-year steps) from IIASA are used to apportion gridded population by age (10-year steps assumed; 2010 used to represent 2000 where needed).
• Baseline mortality rates (BMR): Cause-specific BMRs (1990–2009) by age and disease from IHME/GBD are averaged over 20 years and held constant for future estimates. Exposure–response functions and attributable fractions:
• Temperature ERFs: From GBD 2019 (MR-BRT), defined for the whole population (no age split) for 17 causes (e.g., CVD, IHD, stroke, COPD, CKD, diabetes). ERFs are stratified by climate zone (based on multi-annual mean daily temperatures; 21–23 zones spanning ~6–28 °C). Daily AFs are computed per grid cell using zone-specific ERFs and multi-annual daily mean temperatures.
• PM2.5 ERFs: From GBD 2019 (MR-BRT splines), age-specific for IHD, stroke, COPD, lung cancer, type 2 diabetes (25+ years), and for lower respiratory infections in children 0–5 years. ERFs are global (not climate-zone-specific) and use annual mean concentrations. AFs are computed per grid cell using cause-specific ERFs and annual mean PM2.5. Uncertainty: Confidence intervals for attributable mortality are derived from the RR/ERF confidence intervals provided by GBD2019 and propagated through equations. Model ensemble is equally weighted across the three CMIP6 models. Outputs: Mortality is estimated for each disease and age group; totals for non-optimal temperature and PM2.5 are the sums across causes and ages. Results are presented globally and for GBD super-regions for year-2000 and end-century time slices and across the 2000–2090 period.

• Baseline (circa 2000): Global long-term mortality attributable to non-optimal temperature is 5.71 (0.34–15.88) million yr⁻¹ and to PM2.5 is 7.69 (0.44–20.15) million yr⁻¹. Population-weighted mortality rate is 96 (52–153) deaths per 100,000.
• End-of-century totals: Under SSP2-4.5, overall attributable mortality rises to 30.3 (13.6–53.1) million yr⁻¹; under SSP1-2.6 to 36.9 (15.9–66.1) million yr⁻¹; under SSP5-8.5 to 43.4 (17.6–79.7) million yr⁻¹. Population-weighted mortality rates increase to 337 (51–590), 509 (220–913), and 581 (231–1049) deaths per 100,000 under SSP2-4.5, SSP1-2.6, and SSP5-8.5, respectively.
• Relative growth by risk factor: Globally, PM2.5-attributable mortality increases by factors of ~4.8–7.5 (SSP1-2.6 to SSP2-4.5) and ~6.6 (SSP5-8.5), while non-optimal temperature-attributable mortality rises by factors of ~6.7 (SSP2-4.5), ~9.1 (SSP1-2.6), and ~10.9 (SSP5-8.5). Warm-temperature mortality shows the largest relative growth (order-of-magnitude increases; e.g., ~21–52× depending on scenario), whereas cold-related mortality grows more modestly (~6–7.5×), with reductions from warming countered by population growth and aging.
• Regional patterns: South and East Asia bear the largest absolute burdens (reflecting large populations and historically high pollution). By end-century, temperature-related mortality is projected to surpass PM2.5-attributable mortality in several regions (e.g., Central/Eastern Europe, parts of Latin America) across scenarios. In High-Income regions, temperature-related mortality is projected to be 3–7 times that from air pollution by end-century, despite successful air quality controls.
• Exposure trends: Global population-weighted PM2.5 exposure declines from ~43 μg/m³ (early century) to ~31 μg/m³ (SSP2-4.5 and SSP5-8.5) and ~25 μg/m³ (SSP1-2.6) by end-century. Nevertheless, total mortality attributable to PM2.5 rises due to population growth to ~9.3–9.8 billion mid-century and substantial population aging (average age increasing from ~32 to ~46–56 years across scenarios by end-century).
• Overall implication: Temperature-related mortality accelerates faster than pollution-related mortality, becoming a more important health risk than air pollution for at least 20% of the world’s population by end-century, and potentially offsetting gains from air quality improvements.

The study demonstrates that, even with anticipated declines in PM2.5 exposure under multiple SSPs, global mortality attributable to both PM2.5 and non-optimal temperature increases markedly through 2090, driven primarily by population growth and aging. The relative rise in temperature-related mortality outpaces that from PM2.5, with warm-temperature exposure becoming the dominant subcomponent in many regions. These results align with GBD-based estimates for 2000–2019 and extend them by jointly projecting both risk factors under unified climate–demographic scenarios with downscaled CMIP6 exposures. Regionally, South and East Asia carry the largest absolute burdens, while in High-Income and several temperate regions, temperature-related mortality overtakes PM2.5-attributable mortality as air quality improves and climate warming intensifies heat exposure. The findings imply that air pollution controls alone will not be sufficient to reduce the atmospheric health burden; ambitious climate mitigation and effective heat adaptation strategies are essential to curtail the growing temperature-related risks. By the end of the century, at least one-fifth of the global population is expected to face temperature as a more significant atmospheric health risk than pollution, underscoring the need for integrated health, climate, and air quality policies.

This work provides a unified global projection of mortality attributable to long-term exposure to non-optimal temperature and PM2.5 across SSP1-2.6, SSP2-4.5, and SSP5-8.5 to 2090, using bias-corrected/downscaled CMIP6 simulations and GBD2019 exposure–response functions. We find that total attributable mortality could quadruple to roughly 30 million deaths per year under a medium scenario by end-century, with temperature-related mortality growing faster than PM2.5-related mortality and becoming the dominant atmospheric risk factor for a substantial share of the global population. The results highlight that climate change mitigation and heat adaptation policies are urgently needed alongside continued air quality improvements to prevent large future health losses. Future research should expand the model ensemble and spatial coverage, incorporate additional pollutants (e.g., ozone), evaluate adaptation and behavioral/physiological acclimatization, explore alternative baseline mortality trajectories and healthcare improvements, and refine ERFs with age- and region-specific temperature sensitivities.

• Model ensemble: Only three CMIP6 models (CESM2-WACCM, GFDL-ESM4, MIROC-ESM2-0) met data/resolution criteria; equal weighting may not capture full model uncertainty.
• Downscaling/bias correction: Temperature and PM2.5 bias adjustments rely on specific algorithms (ClimDown/QDM; delta method) and reference datasets; using 20-year climatological means may smooth variability and extremes.
• Spatial/temporal coverage: Analysis limited to −54.95° to 67.9° N due to PM2.5 observational constraints; exposures aggregated to 20-year periods stepped every 10 years may underrepresent interannual extremes.
• Baseline mortality: Cause-specific BMRs averaged over 1990–2009 and held constant into the future, not reflecting potential healthcare or socio-demographic transitions beyond population/age structure.
• ERFs and assumptions: Temperature ERFs are not age-specific; PM2.5 ERFs assume global validity; theoretical minimum-risk exposure for PM2.5 set to 2.4–5.9 μg/m³; adaptation to heat/cold and behavioral changes are not explicitly modeled.
• Data inconsistencies: Scenario labeling and some reported numeric details contain minor inconsistencies/typos in the text; uncertainties are represented via GBD-derived CIs but full propagation of structural/model uncertainties is limited.