Found 750 publications. Showing page 1 of 75:
The AlphaEarth Foundations model, recently released in Google Earth Engine as annual satellite embeddings, provides a new way to work with multi-sensor Earth observation data. Each 10-m pixel is summarized as a 64-dimensional vector that captures the yearly trajectory of surface conditions using information learned from optical, radar, LiDAR, and other datasets, including climatic model outputs and digital terrain data. Rather than representing physical measurements directly, these embeddings condense complex spatial and temporal patterns into compact descriptors that can be used as inputs for machine-learning regression models. This allows researchers to explore environmental patterns—such as air quality—that are influenced by geographical, environmental, and meteorological conditions in cities.In this study, we evaluate whether these annual embeddings, represented as 64 bands (A00–A63), can describe spatial patterns of urban NO₂ without explicitly supplying additional land-use, meteorological, or emission datasets. We present first results from two contrasting environments: Quito, a high-altitude Andean basin in Ecuador, and Essen, a dense urban–industrial region in western Germany. Models trained only with the embedding bands and ground-based NO₂ observations reproduce meaningful spatial gradients in both cities, suggesting that the embeddings encode attributes relevant to emission intensity, urban structure, and pollutant dispersion.These early results highlight the potential of foundation-model satellite embeddings as lightweight, scalable predictors for urban air-quality analyses. They also show how these embeddings can be combined with advanced AI-based regression models, offering a new option for studying air pollution patterns in cities where data availability is often limited by the small number of air-quality monitoring stations.
2026
2026
2026
2026
2026
2026