Comparative Assessment of RF and XGBoost Machine Learning Models for Urban-Scale PM2.5 Estimation in Quito, Ecuador Using Satellite and Ground-Based Observations

Estimating PM₂.₅ exposure in high-altitude Andean cities is challenging due to limited monitoring coverage and the complex interactions among topography, meteorology, and atmospheric chemistry. This study presents a comparative assessment of Random Forest and Extreme Gradient Boosting (XGBoost) for urban-scale PM₂.₅ estimation in Quito, Ecuador. Ground-based observations from the Metropolitan Atmospheric Mon-itoring Network of Quito (REMMAQ) were integrated with Sentinel-5P TROPOMI products, meteorological variables, and topographic predictors processed in Google Earth Engine. Models were developed separately for wet and dry seasons to account for seasonal variability. XGBoost achieved the highest predictive accuracy during the wet season (R² = 0.73), when topographic controls dominated PM₂.₅ variability and the DEM emerged as the most influential predictor. In contrast, RF demonstrated greater robustness during the dry season (R² = 0.63), when photochemical interactions became increasingly important and the CO–SO₂ combustion index was the dominant predictor. Spatial predictions identified a persistent north–south pollution corridor within the urban core of Quito. These findings indicate that PM₂.₅ dynamics in inter-Andean val-leys are governed by seasonally shifting physical and chemical controls. The proposed framework provides a scalable approach for generating spatially continuous air-quality estimates in mountainous urban environments with limited monitoring in-frastructure.