Mini review of multimodal intelligence for air quality modeling

Single data sources can miss important information needed for air quality modeling. Ground stations provide accurate local measurements. Satellites, IoT sensors, traffic records, meteorology, and land-use data describe different parts of the same environmental process. These sources improve coverage and representation if they are combined. However, this combination does not guarantee better forecasting performance. Extra inputs may add noise, duplicate information, or become poorly aligned in time. This mini review examines 35 empirical studies published from 2024 to 2026. The review compares multimodal fusion, where information is combined through early, intermediate, and late fusion. It also compares how adaptive mechanisms change the contribution of individual modalities. The review then examines how reliable these models remain when sensors fail, satellite data is missing, or the model is used in a new place. Across the reviewed studies, the usefulness of a modality depends on the task, the quality of the input, and the conditions under which it is used. Future work should evaluate multimodal systems under missing inputs, changing environmental conditions, and physical constraints instead of relying only on complete benchmark datasets.