Search papers, labs, and topics across Lattice.
This work evaluates how representations from Earth observation foundation models鈥攕pecifically AlphaEarth, AnySat, and TerraMind鈥攃an be repurposed to predict socioeconomic indicators and urban livability. Addressing the challenge of severe ground-truth label scarcity in developing regions, the authors benchmark embedding utility while explicitly accounting for complex spatial dependencies. Their framework isolates the most predictive physical features latent in these embeddings, substantially improving the accuracy of downstream socioeconomic outcome prediction from satellite imagery.
Satellite foundation models already encode the latent physical signals needed to map ground-level urban livability, bypassing the need for costly ground surveys in data-scarce regions.
While accurate measurement of socioeconomic indicators remains challenging in data-scarce regions, which limits policy interventions and resource allocation, high-resolution geospatial data is widely available and can contain information on various livability statistics. We investigate which physical features are encoded within foundation model embeddings, such as AlphaEarth, AnySat, and TerraMind, and provide a systematic framework for identifying the most predictive geospatial indicators. By analyzing how different types of geospatial data influence urban livability predictions, our approach enables researchers to prioritize the most informative features for their specific applications. Additionally, we demonstrate how to leverage foundation model embeddings to enhance prediction performance for these outcomes. This work contributes a principled methodology for extracting actionable information from satellite imagery while accounting for complex spatial dependencies, with applications in predicting urban livability in regions with limited observation data.
