Air Quality Prediction

Air quality prediction estimates how pollutant concentrations such as PM2.5, PM10, ozone, and nitrogen dioxide vary across places and over time, helping inform public health alerts and exposure assessment. Methods range from statistical time-series models and recurrent networks to transformers, state-space models, and graph neural networks that combine station histories with weather, wind, terrain, satellite imagery, and land-use data. Physics-informed models represent processes such as pollutant transport and chemical change, while multimodal fusion, transfer learning, and lightweight edge models address sparse monitoring and deployment constraints. A central challenge is reliable performance beyond familiar regions and conditions: missing observations, geographic shifts, extreme events, and uncertain forecasts can undermine models that score well on conventional benchmarks, and predictive accuracy alone does not guarantee physically meaningful results.