Global PM_2.5 exposure forecasting with novel deep learning architecture and explainable artificial intelligence

Abstract

Particulate matter (PM) of fine size (≤2.5 μm) remains one of the most significant global environmental risk factors for early mortality and morbidity, and more than 90% of the global population currently lives in areas exceeding the World Health Organization 2021 guideline value of 5 μg/m³. This study introduces a temporally constrained transformer-based forecasting model to anticipate annual population-weighted PM_2.5 exposure across 204 countries and territories between 1990 and 2020, aimed at supporting evidence-based air quality and climate policy development. The framework is based on a filtered dataset from the State of Global Air, comprising 6,323 country-year observations with harmonized exposure estimates and uncertainty bounds, allowing the model to capture long-range temporal variations and enduring heterogeneity among countries in exposure trends. A time-aware expanding-window cross-validation approach was strictly implemented to prevent information leakage and ensure realistic predictive conditions. Five-fold temporal validation demonstrates strong performance across geographical locations, with mean squared error ranging from 0.00043 to 0.00115, root mean squared error from 0.0207 to 0.0339 μg/m³, and mean absolute error from 0.0094 to 0.0193 μg/m³, with Nash-Sutcliffe efficiencies exceeding 0.95 on average. Continental-scale evaluation shows similar high accuracy in Europe and Oceania (root mean squared error <0.01 μg/m³; R² > 0.98), while systematically higher errors are observed in Asia and Africa, which bear a higher pollution burden. The attention-weight inspection offers clear decompositions of temporal trends and country-specific patterns that drive predictions. The proposed framework is, therefore, a methodological and practical addition to transformer-based environmental forecasting and policy-relevant global health-risk assessment.

Keywords

PM2.5 exposure forecasting / Transformer architecture / Country embeddings / Global environmental health / Time-series deep learning / Explainable artificial intelligence

Funding

None.

Conflict of interest

The authors declare they have no competing interests.

References

[1] Inam SA. A review of artificial intelligence for predicting climate driven infectious disease outbreaks to enhance global health resilience. Discover Public Health. 2025; 22(1):738. doi: 10.1186/s12982-025-01167-4

[2] Inam SA, Khan AA, Mazhar T, et al. PR-FCNN: A data-driven hybrid approach for predicting PM2.5 concentration. Discover Artif Intell. 2024; 4(1):75. doi: 10.1007/s44163-024-00184-7

[3] Inam SA, Zaidi SMH, Khan AA, Ullah S. A neural network approach to carbon emission prediction in industrial and power sectors. Discover Appl Sci. 2025; 7(6):640. doi: 10.1007/s42452-025-07257-x

[4] The World Bank. The Global Health Cost of PM2.5 Air Pollution: A Case for Action Beyond 2021. Washington, DC: The World Bank; 2022. doi: 10.1596/978-1-4648-1816-5

[5] Pai SJ, Carter TS, Heald CL, Kroll JH. Updated world health organization air quality guidelines highlight the importance of non-anthropogenic PM2.5. Environ Sci Technol Lett. 2022; 9(6):501-506. doi: 10.1021/acs.estlett.2c00203

[6] Liu J, He C, Si Y, et al. Toward better and healthier air quality: Global PM2.5 and O3 pollution status and risk assessment based on the new WHO air quality guidelines for 2021. Glob Chall. 2024; 8(4):2300258. doi: 10.1002/gch2.202300258

[7] Edlund KK, Killman F, Molnár P, Boman J, Stockfelt L, Wichmann J. Health risk assessment of PM2.5 and PM2.5- bound trace elements in Thohoyandou, South Africa. Int J Environ Res Public Health. 2021; 18(3):1359. doi: 10.3390/ijerph18031359

[8] Elbarbary M, Oganesyan A, Honda T, et al. Systemic inflammation (c-reactive protein) in older Chinese adults is associated with long-term exposure to ambient air pollution. Int J Environ Res Public Health. 2021; 18(6):3258. doi: 10.3390/ijerph18063258

[9] McDuffie EE, Martin RV, Spadaro JV, et al. Source sector and fuel contributions to ambient PM2.5 and attributable mortality across multiple spatial scales. Nat Commun. 2021; 12(1):3594. doi: 10.1038/s41467-021-23853-y

[10] Ghosh R, Causey K, Burkart K, Wozniak S, Cohen A, Brauer M. 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. 2021; 18(9):e1003718. doi: 10.1371/journal.pmed.1003718

[11] Southerland VA, Anenberg SC, Harris M, et al. Assessing the distribution of air pollution health risks within cities: A neighborhood-scale analysis leveraging high-resolution data sets in the bay area, California. Environ Health Perspect. 2021; 129(3):37006. doi: 10.1289/EHP7679

[12] Fang T, Di Y, Xu Y, et al.Temporal trends of particulate matter pollution and its health burden,1990- 2021, with projections to 2036: A systematic analysis for the global burden of disease study 2021. Front Public Health. 2025; 13:1579716. doi: 10.3389/fpubh.2025.1579716

[13] Booth A, James P, McGough S, Solaiman E. Cross-regional deep learning for air quality forecasting: A comparative study of CO, NO2, O3, PM2.5, and PM10. Forecasting. 2025; 7(4):66. doi: 10.3390/forecast7040066

[14] Benali AAE, Cafaro M, Epicoco I, Pulimeno M, Schioppa EJ. Just in Time Transformers. New York: Cornell University; 2024.

[15] Muthukumar P, Cocom E, Nagrecha K, et al. Predicting PM2.5 atmospheric air pollution using deep learning with meteorological data and ground-based observations and remote-sensing satellite big data. Air Qual Atmos Health. 2022; 15(7):1221-1234. doi: 10.1007/s11869-021-01126-3

[16] Nedungadi V, Munir MA, Rußwurm M, et al. AirCast: Improving Air Pollution Forecasting through Multi-Variable Data Alignment. New York: Cornell University; 2025.

[17] Oldenburg V, Cardenas-Cartagena J, Valdenegro-Toro M. Forecasting Smog Clouds with Deep Learning. New York: Cornell University; 2024.

[18] Boogaard H, Pant P, Künzli N. Editorial: Science to foster the WHO air quality guideline values. Int J Public Health. 2025; 69:1608249. doi: 10.3389/ijph.2024.1608249

[19] Tello-Leal E, Ramirez-Alcocer UM, Macías-Hernández BA, Hernandez-Resendiz JD. Evaluation of deep learning models for predicting the concentration of air pollutants in urban environments. Sustainability. 2024; 16(16):7062. doi: 10.3390/su16167062

[20] Huang S, Xiao X, Guo H. A novel method for carbon emission forecasting based on EKC hypothesis and nonlinear multivariate grey model: Evidence from transportation sector. Environ Sci Pollut Res Int. 2022; 29(40):60687-60711. doi: 10.1007/s11356-022-20120-5

[21] Ran Q, Bu F, Razzaq A, et al. When will China’s industrial carbon emissions peak? Evidence from machine learning. Environ Sci Pollut Res Int. 2023; 30(20):57960-57974. doi: 10.1007/s11356-023-26333-6

[22] Costantini L, Laio F, Mariani MS, Ridolfi L, Sciarra C. Forecasting national CO 2 emissions worldwide. Sci Rep. 2024; 14(1):22438. doi: 10.1038/s41598-024-73060-0

[23] Lee JB, Lee JB, Koo YS, et al. Development of a Deep Neural Network for Predicting 6 Hour Average PM_2.5Concentrations up to Two Subsequent Days Using Various Training Data. [Preprint]; 2021. doi: 10.5194/gmd-2021-356

[24] Damkliang K, Chumnaul J. Deep learning and statistical approaches for area-based PM_2.5forecasting in Hat Yai, Thailand. J Big Data. 2025; 12(1):36. doi: 10.1186/s40537-025-01079-9

[25] Utku A, Can U, Alpsülün M, et al. Advancing air quality monitoring: Deep learning-based CNN-RNN hybrid model for PM2.5 forecasting. Atmosphere. 2025; 16(9):1003. doi: 10.3390/atmos16091003

[26] Zhang Y, Zhang W, Wu B, Yi W. Regional PM_2.5pollution forecasting using a hybrid model based on multi-scales feature fusion and deep learning algorithms. PLoS One. 2025; 20(10):e0333489. doi: 10.1371/journal.pone.0333489

[27] Yeo I, Choi Y, Lops Y, Sayeed A. Efficient PM_2.5forecasting using geographical correlation based on integrated deep learning algorithms. Neural Comput Appl. 2021; 33(22):15073-15089. doi: 10.1007/s00521-021-06082-8

[28] Samal KKR, Babu KS, Das SK. Multi-directional temporal convolutional artificial neural network for PM2.5 forecasting with missing values: A deep learning approach. Urban Clim. 2021; 36:100800. doi: 10.1016/j.uclim.2021.100800

[29] Potavijit C, Pattarapanitchai P, Nontapa C. A hybrid deep learning model for forecasting PM2.5 concentrations in Northern Thailand from satellite images. Glovento J Integr Stud. 2025; 1(1):6. doi: 10.63665/gjis.v1.6

[30] Wu Y, Wang X, Wang M, Liu X, Zhu S. Time-series forecasting of PM2.5 and PM10 concentrations based on the integration of surveillance images. Sensors (Basel). 2024; 25(1):95. doi: 10.3390/s25010095

[31] Dairi A, Harrou F, Sun Y.Efficient Deep Learning-Driven Approach for PM2.5 Forecasting at Different Locations in Spain. In: 2021 IEEE 3^rdEurasia Conference on Biomedical Engineering, Healthcare and Sustainability (ECBIOS). IEEE; 2021. p. 173-178. doi: 10.1109/ECBIOS51820.2021.9510462

[32] Vignesh PP, Jiang JH, Kishore P. Predicting PM2.5 concentrations across USA using machine learning. Earth Space Sci. 2023; 10(10):1-15. doi: 10.1029/2023ea002911