Geospatial Foundation Models: Workshop Takeaways

Geospatial foundation models, under the broader umbrella of GeoAI, are gaining traction in geospatial analytics, driven by advances in large-scale representation learning and models like Google's Population Dynamics Foundation Model. Despite promising developments, challenges remain in scaling data access, designing robust architectures, and integrating models into analytical workflows. The recent Geospatial Foundation Models Workshop in Barcelona, organized by CARTO and the Barcelona Supercomputing Center, brought together industry and academic experts to discuss these issues, highlighting the diversity of approaches and the importance of bridging the gap between experimental results and practical applications. Participants emphasized the need for scalable retrieval strategies, interpretability, and the development of benchmarks to assess models' readiness for deployment. CARTO's efforts focus on integrating third-party embeddings, developing analytical tools, and collaborating with the Barcelona Supercomputing Center to advance population dynamics models. The workshop underscored the necessity of sustained collaboration to transition from exploratory phases to mature applications in geospatial foundation models.