Recent advances in foundation models (FMs) are transforming Earth observation (EO) by enabling scalable learning from multimodal geospatial data. FMs aim to unify representations across paltforms, sensors, spatial scales, and domains, supporting tasks such as land cover mapping, environmental monitoring, and predictive modeling. However, key challenges remain in generalizability, interpretability, benchmarking, and integration into scientific workflows. This session brings together researchers from various backgrounds, e.g., EO, AI, and Earth system science, to examine the role of FMs in advancing geoscience. We invite contributions on model architecture, benchamarking, and adaptation to geospatial data, including multimodal, multiresolution, and time series inputs. Both methodological advances and application-focused studies are welcome, as well as those linked to digital twins and decision-support systems. Submissions addressing data curation, evaluation, uncertainty quantification, and domain adaptation are also encouraged. The session aims to foster interdisciplinary collaboration and identify priorities for developing robust, scalable, and trustworthy geospatial AI.

• B - Biogeosciences
• GC - Global Environmental Change
• H - Hydrology
• OS - Ocean Sciences