Accurate 24-hour forecasting of ground-level ozone is crucial for public health, as exposure to this harmful air pollutant links to respiratory, cardiovascular diseases, and premature death. However, predicting ozone across the entire United States is difficult due to the complex, constantly changing nature of weather and air pollution. In this study, we used a new type of artificial intelligence called a Temporal Fusion Transformer to make ozone forecasts more accurate. This model was trained using data from around 8,000 monitoring stations across the U.S., along with weather and pollution data from federal sources.

Compared to the standard government forecasting system, our model proved substantially more accurate, reducing errors by over 24%. It also helped pinpoint which weather conditions and pollutants most influence ozone levels. This new approach, leveraging advanced AI like the Temporal Fusion Transformer, promises more precise, timely air quality warnings. Improved forecasts empower communities to respond faster to dangerous pollution, enhancing public health protection.