University of Houston AI predicts ozone levels up to 14 days ahead

University of Houston's Air Quality Forecasting and Modeling Lab has created an AI system that predicts tropospheric ozone levels up to two weeks in advance, a massive improvement over the current state-of-the-art of just three days. Led by atmospheric chemistry professor Yunsoo Choi and doctoral student Alqamah Sayeed, the system tackles a long-standing challenge: ozone is a secondary pollutant that forms near the surface, causing lung and heart damage. Traditional numerical models rely on physics equations but lose accuracy rapidly after day three. The team instead built a machine learning algorithm using a unique loss function called the index of agreement (IOA), which compares forecasted versus actual outcomes. By feeding it four to five years of historical ozone data, the AI learned to recognize patterns the way a child learns not to touch a hot cup—through repeated input and refinement.

The system's practical impact is significant. Ozone alerts currently offer only two to three days of warning, limiting the ability of vulnerable populations (children, the elderly, asthmatics) to prepare or avoid exposure. With two-week forecasts, public health agencies can issue longer-term advisories, reduce panic responses, and even plan interventions to curb ozone-forming emissions. The researchers published their findings in Scientific Reports, emphasizing that this is the first successful attempt at 14-day surface ozone forecasting. Beyond health, the AI could also inform climate models, as ozone plays a complex role in atmospheric chemistry and warming. The team plans to further refine the algorithm and test it in different regions and conditions to validate its generalizability.