Predicting fine-scale daily NO2 over Mexico city using an ensemble modeling approach

• DOI: 10.1016/j.apr.2023.101763
• Publication Date: 2023-04-17T12:31:24Z
• Authors: Mike Z. He; Maayan Yitshak-Sade; Allan C. Just; Iván Gutiérrez-Avila; Michael Dorman; Kees de Hoogh; Bas Mijling; Robert O. Wright; Itai Kloog
• Keywords: Generalized additive model; Gradient boosting

In recent years, there has been growing interest in developing air pollution prediction models to reduce exposure measurement error epidemiologic studies. However, efforts for localized, fine-scale have predominantly focused the United States and Europe. Furthermore, availability of new satellite instruments such as TROPOsopheric Monitoring Instrument (TROPOMI) provides novel opportunities modeling efforts. We estimated daily ground-level nitrogen dioxide (NO2) concentrations Mexico City Metropolitan Area at 1-km2 grids from 2005 2019 using a four-stage approach. stage 1 (imputation stage), we imputed missing NO2 column measurements Ozone (OMI) TROPOMI random forest (RF) 2 (calibration calibrated association ground monitors meteorological features RF extreme gradient boosting (XGBoost) models. 3 (prediction predicted model over each grid our study area, then ensembled results generalized additive (GAM). 4 (residual used XGBoost local component 200-m2 scale. The cross-validated R2 were 0.75 0.86 respectively, 0.87 GAM. Cross-validated root-mean-squared (RMSE) GAM was 3.95 μg/m3. Using approaches newly available remote sensing data, multi-stage presented high fits reconstructs estimates further studies City.