• Labels: two classifiers are trained on EEA PM_2.5 reference stations: station type (background/traffic; industrial stations excluded) and station area (rural/suburban/urban).
• Predictors: AlphaEarth 2025 satellite embeddings at the point and averaged within 25 m to 10 km; for type also GSWIFT modelled road traffic and distances to major roads.
• Models: XGBoost (type) and a neural network (area), evaluated with spatial cross-validation and on EEA stations not used in training. Probabilities are temperature-calibrated; confidence below 0.70 is flagged uncertain.
• Coordinates: precise coordinates are scored at the point. Rounded positions (the 0.002° grid of Sensor.Community, or 2–3 decimals) are averaged over random positions inside their cell; 0–1 decimals are not scored.
• Meaning: the class an EEA reference station at this coordinate would most likely have, not a verified description of the sensor's installation.

For more information, please see the accompanying paper (in preparation).