Modeling of urban air quality dynamics using hyperparameter-tuned boosted regression

Rapid urbanization in the Gulf Cooperation Council (GCC) region has intensified air quality challenges, particularly elevated concentrations of fine particulate matter (PM₂.₅). While meteorological drivers have been studied, the combined influence of land cover and socioeconomic factors remain underexplored in arid environments. This study employs a Bayesian-optimized XGBoost model to predict PM₂.₅ levels across ten major GCC cities (2012–2024). The framework integrates satellite-derived meteorological variables, land surface temperature and vegetation indices, and nighttime light radiance as a proxy for anthropogenic activity. Results show high PM2.5 concentrations (120–140 µg·m−3​​​​​) along the Arabian Gulf coast, with Kuwait City most affected. SHAP (SHapley Additive exPlanations) analysis identifies surface pressure and wind speed as key predictors. With strong performance (Coefficient of variation, R² > 0.75, and Root Mean Square Error, RMSE < 15 µg·m−3), the study substantiates the value of interpretable machine learning for evidence-based air quality management in desert urban regions.