Comprehensive 24-hour ground-level ozone monitoring: Leveraging machine learning for full-coverage estimation in East Asia

Abstract

Cloud cover often hinders satellite-derived ozone (O3) concentration estimation, leading to incomplete spatial coverage. To address this limitation and obtain gap-free hourly ground-level O3 estimates, this study developed a novel all-sky O3 estimation model based on a light gradient boosting machine, combining clear- (cLearGBM) and cloudy-sky (cLoudGBM) models. Unlike earlier studies focusing mainly on daytime, this study provides comprehensive O3 variations over a full 24-h cycle at an hourly 2 km resolution. The all-sky O3 estimation model was developed using Himawari-8 brightness temperature (BT) as a key input, alongside other satellite-derived variables, meteorological variables, and auxiliary parameters, with ground-based O3 observations serving as the dependent variable. The models were evaluated using three 10-fold cross-validation methods (random, spatial, and temporal), showing high estimation accuracy (cLearGBM: coefficient of determination (R2) = 0.90, root mean square error (RMSE) = 8.77 ppb; cLoudGBM: R2 = 0.87, RMSE = 9.44 ppb). Notably, BT data improved the accuracy and spatial resolution of the O3 estimates. The estimated ground-level O3 distribution followed a typical diurnal pattern across the study area, with urban regions showing higher O3 concentrations during the day and rural areas exhibiting higher concentrations at night. Compared to the Copernicus Atmosphere Monitoring Service reanalysis data, the proposed model offered a representation of East Asia that was 40 times better spatial resolution and 2.26 times more accurate when evaluated against in-situ observations. The 24 h ground-level O3 data for East Asia provided by this study is expected to serve as a valuable foundation for applied research and to support effective O3 pollution management.