Altitude-resolved prediction of roadside air pollution using UAV measurements and machine learning

Abstract Roadside areas are major sources of traffic-related air pollutant emissions. This study identified the spatial distribution of roadside pollutants using Unmanned Aerial Vehicle (UAV) measurements and predicted their concentrations with machine learning models. Pollutant concentrations decreased with increasing altitude and distance from roads, with NO₂ and O₃ showing similar patterns. NO₂ levels peaked at 7 a.m. due to heavy traffic. Seasonally, O₃ concentrations were higher in summer due to higher solar radiation. PM2.5 levels also increased in summer, with long-range transport influences confirmed using the HYSPLIT model. The UAV/Ground concentration ratio was predicted for each pollutant using machine learning models. Among them, the CatBoost model, with the highest R² value (0.65–0.95), was selected as the final model. SHAP-based feature importance analysis indicated that NO2 and PM2.5 UAV/Ground ratio predictions were influenced by ground-level concentrations and meteorology, while O3 was primarily associated with meteorological and seasonal features. CO was influenced by traffic speed, which showed a negative association with the ratio. The proposed UAV-integrated machine learning framework provides a novel approach for high-resolution, three-dimensional estimation of roadside air quality.