FuXi-Air: air quality forecasting based on emission-meteorology-pollutant multimodal machine learning

Abstract

Abstract
Air pollution has emerged as a major public health challenge worldwide. Numerical simulations and single-site machine-learning approaches in air quality forecasting faced multiple limitations. It is urgent to develop a low-cost, efficient air quality forecasting model. FuXi-Air has been constructed based on multi-modal data fusion to support high precision and air quality forecasting. The model successfully completes 72-h forecasts for six major air pollutants at an hourly resolution across multiple monitoring sites within 25–30 s, outperforming the numerical air quality models applied in operational forecasting. Key influencing factors analysis shows the integration of meteorological, emission and observational data significantly improves the precision and ensures the reliability of forecasting under differing pollution mechanisms, varied with megacities. This study provides both a scientific reference and a practical example for applying deep machine learning to support rapid air pollution risk warning.