Evaluating and Optimizing Air Quality Forecasting for Critical Particulate Matter Episodes in the Santiago Metropolitan Region, Chile

Severe wintertime particulate pollution (PM10 and PM2.5) affects the Santiago Metropolitan Region in Chile and is intensified by basin topography and frequent thermal inversions. Local authorities rely on the Critical Episodes Management (CEM) forecasting system, yet its predictive performance is variable. This study assesses CEM to identify operational vulnerabilities and propose data-driven improvements for urban air-quality governance. About ~1.2 million hourly meteorological and air-quality records (2017–2022) were analyzed using Generalized Additive Models (GAMs) to characterize key nonlinear relationships, and we evaluated the operational skill of the Cassmassi-1 PM10 model and the WRF-Chem-based PM2.5 forecasting component used by the system. Cassmassi-1 missed more than 50% of critical episodes and showed a false-alarm rate above 60%, consistent with limitations associated with static or incomplete emission representations. By contrast, the WRF-Chem-based component achieved episode prediction accuracy above 70%. GAM results indicate that wind speeds below 2 m s−1, high diurnal temperature range, and relative humidity below 65% are strongly associated with extreme events. Considering the results, we recommend transitioning to nonlinear forecasting approaches that explicitly incorporate these meteorological thresholds and vertical stability indicators to improve alert reliability, strengthen urban resilience, and reduce population exposure.