Comparison of three functional regression methods on air pollution throughout the first two COVID-19 lockdown phases across 31 Iranian province

Abstract

Introduction: Exposure to air pollution heightens respiratory vulnerability, particularly during pandemics. The COVID-19 lockdowns in Iran provided a natural experiment to investigate how reduced human activity influenced air quality across 31 provinces. Understanding these environmental responses is vital for informing sustainable public health and pollution mitigation policies.

Materials and methods: Satellite-derived data on air pollutants and air quality and meteorological variables were obtained for all 31 provinces of Iran from Sentinel-5P, the GLDAS-2 dataset developed by National Aeronautics and Space Administration (NASA), and Google Earth Engine (GEE). The study covered two COVID-19 lockdown periods and their corresponding pre-pandemic periods from the previous year. The evaluated air quality indices consisted of Carbon monoxide (CO), Water Vapor (H₂O,) Nitrogen dioxide (NO₂), Ozone (O₃), Sulfur dioxide (SO₂), Absorbing Aerosol Index (AER), and Atmospheric Formaldehyde (HCHO). Meteorological covariates comprised temperature, pressure, precipitation, and wind speed. Sparse temporal data were reconstructed using FDA and FPCA, representing Functional Data Analysis and Functional Principal Component Analysis, respectively. Three Function-on-Function (FOF) regression models— standard, smooth, and principal component-based—were developed, with and without meteorological adjustments. Model performance was assessed using R², AIC, and BIC, representing the coefficient of determination, Akaike Information Criterion, and Bayesian Information Criterion, respectively.

Results: Air pollutant levels significantly declined during both lockdowns compared with the corresponding pre-pandemic periods, with spatial variations influenced by meteorological and industrial factors. Incorporating meteorological covariates markedly improved model accuracy, particularly for NO₂ and CO. The principal component-based FOF model provided the best fit, explaining over 80% of variance in major pollutants.

Conclusion: COVID-19 lockdowns produced measurable, regionally heterogeneous improvements in air quality across Iran. Integrating meteorological adjustments and advanced functional regression approaches enhances environmental modeling and supports evidence-based air pollution control strategies during health emergencies.