Comparison of Three Functional Regression Methods on Air Pollution Throughout the First Two COVID-19 Lockdown Phases Across 31 Iranian Provinces
Published in Journal of Air Pollution and Health, 11(2), 217–230, 2026
Title
Comparison of Three Functional Regression Methods on Air Pollution Throughout the First Two COVID-19 Lockdown Phases Across 31 Iranian Provinces
Author
Mohammad Fayaz
Journal Information
Journal: Journal of Air Pollution and Health (JAPH)
Volume: 11
Issue: 2 (Spring 2026)
Pages: 217–230
Publication Year: 2026
Keywords
- Air Pollution
- COVID-19 Restrictions
- Functional Data Analysis (FDA)
- Functional Principal Component Analysis (FPCA)
- Function-on-Function (FOF) Regression
- Functional Regression
- Meteorological Covariates
- Sentinel-5P
- Google Earth Engine (GEE)
- Iran
Abstract
Introduction
Exposure to air pollution increases respiratory vulnerability, particularly during infectious disease outbreaks. The COVID-19 lockdowns in Iran created a natural experiment for evaluating the impact of reduced anthropogenic activities on air quality across all 31 provinces. This study compared three functional regression approaches for modeling temporal changes in air pollution during two nationwide lockdown periods.
Materials and Methods
Satellite-derived air quality and meteorological data were obtained from Sentinel-5P, the NASA GLDAS-2 dataset, and Google Earth Engine (GEE). Seven atmospheric indicators (CO, H₂O, NO₂, O₃, SO₂, AER, and HCHO) together with meteorological variables (temperature, pressure, precipitation, and wind speed) were analyzed. Sparse temporal observations were reconstructed using Functional Data Analysis (FDA) and Functional Principal Component Analysis (FPCA). Three Function-on-Function (FOF) regression models—standard, smooth, and principal component-based—were fitted with and without meteorological adjustments. Model performance was evaluated using R², Akaike Information Criterion (AIC), and Bayesian Information Criterion (BIC).
Results
Air pollutant concentrations decreased significantly during both COVID-19 lockdown phases compared with the corresponding pre-pandemic periods, although regional differences reflected industrial activity and meteorological conditions. Including meteorological covariates substantially improved predictive performance, particularly for nitrogen dioxide (NO₂) and carbon monoxide (CO). Among the evaluated approaches, the principal component-based FOF regression model demonstrated the best overall performance, explaining more than 80% of the variability in the major pollutants.
Conclusion
COVID-19 lockdowns resulted in measurable improvements in air quality across Iran, although the magnitude of these improvements varied geographically. Functional regression models combined with meteorological adjustments provide an effective framework for environmental monitoring and can support evidence-based air pollution management during future public health emergencies.
Full Text
- PDF File
- Juornal Website
- Appendix : Extra Analysis
- Appendix : R Codes
- Appendix : Dataset
Citation
Fayaz M. Comparison of Three Functional Regression Methods on Air Pollution Throughout the First Two COVID-19 Lockdown Phases Across 31 Iranian Provinces. Journal of Air Pollution and Health. 2026;11(2):217–230.
Recommended citation: Fayaz M. Comparison of Three Functional Regression Methods on Air Pollution Throughout the First Two COVID-19 Lockdown Phases Across 31 Iranian Provinces. Journal of Air Pollution and Health. 2026;11(2):217-230.
Download Paper
