Advances in operational air quality forecasting for Mexico City: Integration of updated emissions, temporal profiles, and initial conditions

Abstract

This study presents an updated operational ozone forecasting system for central Mexico, incorporating improvements such as a revised VOC model-ready emissions, cycling of previous-day forecasts as initial conditions, and changes to temporal profiles for Sunday emissions. A refined evaluation methodology was also introduced, in which the maximum hourly ozone concentration predicted over a selected urban area is compared against the maximum observed value reported at any RAMA monitoring station, excluding background sites. This approach is proposed to better reflect population exposure to peak ozone levels and improve the relevance of forecast verification for air quality alerts. Model of over the period April 14-July 2, 2025, shows that the system captures high-ozone events (≥ 135 ppb) with moderate skill. Categorical metrics indicate a gradual decline in forecast performance from day 1 to day 3, primarily due to an increase in false alarms. The Probability of Detection (POD) remains relatively high (0.76 on day 1 to 0.67 on day 3), but the False Alarm Ratio (FAR) and Probability of False Detection (POFD) increase with lead time. The Gilbert Skill Score (GSS) decreases from 0.31 to 0.16, reflecting reduced forecast skill with lead time and highlighting a tendency to overpredict exceedances. These results underscore the need for further improvements, including the use of urbanized models, improvements in emissions inventories, ensemble approaches, and real-time data assimilation.