CAN THE WEATHER TELL WHICH DAYS IN ASTANA WILL BE POLLUTED? CLASSIFYING PM2.5 EXCEEDANCE DAYS WITH MACHINE LEARNING AND CHECKING THE DATA AGAINST A GROUND MONITOR

Abstract

Astana has long cold winters, and part of the city is still heated with coal. In this study I asked whether weather data alone can show in advance on which days the daily mean concentration of fine particles (PM2.5) in Astana will be above the World Health Organization guideline of 15 µg/m³. I used free hourly PM2.5 estimates from the Copernicus Atmosphere Monitoring Service (CAMS) and ERA5 weather data for 1,490 days between 4 August 2022 and 1 September 2026. Only 141 days (9.5%) were above the guideline, so a model that always answers "normal" is right 90.5% of the time and still finds no polluted day. I trained logistic regression, random forest and gradient boosting models on 2022-2024 and tested them once on 2025-2026. With the weather of the same day, the random forest found 64% of the exceedance days with 55% of its alarms correct (F1 = 0.59, 95% CI 0.47-0.68), better than the rule "tomorrow will be like today" (F1 = 0.42). When the model could use only the weather of the previous day, F1 fell to 0.32, below that simple rule. The model missed all ten summer exceedances and most days with a lot of dust in the air. Finally, I compared CAMS with the US Embassy monitor in Astana on 613 days. The monitor showed about twice as much PM2.5 (17.4 against 8.4 µg/m³ on average) and exceedances on 39% of days instead of 9%, and only the monitor showed higher pollution in the heating season. Weather predicted the monitor's exceedance days at least as well as the CAMS ones. My conclusion is that weather contains real information about polluted days in Astana, but CAMS shows a much cleaner city than the ground monitor, so models trained only on CAMS should be used with care.

Keywords: PM2.5, air quality, Astana, machine learning, imbalanced classification, CAMS, ground-based monitoring

Список литературы

1. Kerimray A., Bakdolotov A., Sarbassov Y., Inglezakis V., Poulopoulos S. Air pollution in Astana: analysis of recent trends and air quality monitoring system // Materials Today: Proceedings. 2018. Vol. 5, No. 11. P. 22749-22758. DOI: 10.1016/j.matpr.2018.07.086.
2. Biloshchytskyi A., Neftissov A., Kuchanskyi O., Andrashko Y., Biloshchytska S., Mukhatayev A., Kazambayev I. Fractal analysis of air pollution time series in urban areas in Astana, Republic of Kazakhstan // Urban Science. 2024. Vol. 8, No. 3. Art. 131. DOI: 10.3390/urbansci8030131.
3. Agibayeva A., Kumisbek A., Nauyryzbay A., Avcu E., et al. Towards sustainable air quality in coal-heated cities: a case study from Astana, Kazakhstan // Sustainability. 2025. Vol. 17, No. 22. Art. 10214. DOI: 10.3390/su172210214.
4. World Health Organization. WHO global air quality guidelines: particulate matter (PM2.5 and PM10), ozone, nitrogen dioxide, sulfur dioxide and carbon monoxide. Geneva: World Health Organization, 2021. 273 p. ISBN 978-92-4-003422-8. URL: https://www.who.int/publications/i/item/9789240034228 (accessed: 02.10.2026).
5. Mohammadi F., Teiri H., Hajizadeh Y., Abdolahnejad A., Ebrahimi A. Prediction of atmospheric PM2.5 level by machine learning techniques in Isfahan, Iran // Scientific Reports. 2024. Vol. 14. Art. 2109. DOI: 10.1038/s41598-024-52617-z.
6. Makhdoomi A., Sarkhosh M., Ziaei S. PM2.5 concentration prediction using machine learning algorithms: an approach to virtual monitoring stations // Scientific Reports. 2025. Vol. 15. Art. 8076. DOI: 10.1038/s41598-025-92019-3.
7. Saminathan S., Malathy C. Ensemble-based classification approach for PM2.5 concentration forecasting using meteorological data // Frontiers in Big Data. 2023. Vol. 6. Art. 1175259. DOI: 10.3389/fdata.2023.1175259.
8. Ravindiran G., Karthick K., Sivarethinamohan S., Rajamanickam S., et al. A cluster-aware synthetic resampling and machine learning framework for multi-class Air Quality Index classification // Environmental Monitoring and Assessment. 2026. Vol. 198, No. 3. Art. 270. DOI: 10.1007/s10661-026-15121-7.
9. Saito T., Rehmsmeier M. The precision-recall plot is more informative than the ROC plot when evaluating binary classifiers on imbalanced datasets // PLOS ONE. 2015. Vol. 10, No. 3. Art. e0118432. DOI: 10.1371/journal.pone.0118432.
10. Chicco D., Jurman G. The advantages of the Matthews correlation coefficient (MCC) over F1 score and accuracy in binary classification evaluation // BMC Genomics. 2020. Vol. 21. Art. 6. DOI: 10.1186/s12864-019-6413-7.
11. Zippenfenig P. Open-Meteo.com Weather API [software]. Zenodo, 2024. DOI: 10.5281/zenodo.7970649.
12. Open-Meteo. Air Quality API: CAMS global atmospheric composition forecasts, 0.4°, from August 2022. URL: https://open-meteo.com/en/docs/air-quality-api (accessed: 02.10.2026).
13. Rémy S., Kipling Z., Huijnen V., Flemming J., et al. Description and evaluation of the tropospheric aerosol scheme in the Integrated Forecasting System (IFS-AER, cycle 47R1) of ECMWF // Geoscientific Model Development. 2022. Vol. 15, No. 12. P. 4881-4912. DOI: 10.5194/gmd-15-4881-2022.
14. Ali M.A., Bilal M., Wang Y., Nichol J.E., et al. Accuracy assessment of CAMS and MERRA-2 reanalysis PM2.5 and PM10 concentrations over China // Atmospheric Environment. 2022. Vol. 288. Art. 119297. DOI: 10.1016/j.atmosenv.2022.119297.
15. Jin C., Wang Y., Li T., Yuan Q. Global validation and hybrid calibration of CAMS and MERRA-2 PM2.5 reanalysis products based on OpenAQ platform // Atmospheric Environment. 2022. Vol. 274. Art. 118972. DOI: 10.1016/j.atmosenv.2022.118972.
16. Qor-el-aine A., Béres A., Géczi G. Calibration of CAMS PM2.5 data over Hungary: a machine learning approach // Environmental Research Communications. 2024. Vol. 6, No. 7. Art. 075026. DOI: 10.1088/2515-7620/ad6239.
17. Hersbach H., Bell B., Berrisford P., Hirahara S., et al. The ERA5 global reanalysis // Quarterly Journal of the Royal Meteorological Society. 2020. Vol. 146, No. 730. P. 1999-2049. DOI: 10.1002/qj.3803.
18. OpenAQ. Open data archive, location 7094 "US Diplomatic Post: Astana" (AirNow), hourly PM2.5, 2022-2025. URL: https://openaq-data-archive.s3.amazonaws.com (accessed: 02.10.2026).
19. Breiman L. Random forests // Machine Learning. 2001. Vol. 45, No. 1. P. 5-32. DOI: 10.1023/A:1010933404324.
20. Pedregosa F., Varoquaux G., Gramfort A., Michel V., et al. Scikit-learn: machine learning in Python // Journal of Machine Learning Research. 2011. Vol. 12. P. 2825-2830.

Как цитировать

Askarbek N. CAN THE WEATHER TELL WHICH DAYS IN ASTANA WILL BE POLLUTED? CLASSIFYING PM2.5 EXCEEDANCE DAYS WITH MACHINE LEARNING AND CHECKING THE DATA AGAINST A GROUND MONITOR // Central Asian Scientific Journal. – 2026. – No. 4(32), Vol. 1. – Pp. 212–222.