248-1 AI Foundation Models for Geosciences: Opportunities, Gaps, and Future Directions

Session: Expanding Geology’s Horizons: Geoinformatics, Open Science, and Open Data

Presenting Author:

Author:

Abstract:

Foundation models—such as large language models and vision foundation models—are powerful AI tools with the potential to transform big data geosciences. In this talk, I will explore the opportunities and challenges of applying vision-based geospatial foundation models (GFMs) to critical geoscience applications, including flood detection, natural feature recognition, landslide mapping, permafrost thaw monitoring, and crop classification. I will highlight the domain adaptability, generalizability, and few-shot learning capabilities of GFMs, and compare them with traditional task-specific models. While GFMs show promising results, I will also discuss key limitations and areas for future improvement. These include high computational and energy costs, the need for architectural enhancements to better capture semantic and spectral features, development of task-aware fine-tuning strategies, and the integration of additional modalities—such as text and other observational data—to enrich geospatial context and improve interpretability. Advancing these directions will help build a more flexible and domain-adapted GFM ecosystem, ultimately accelerating scientific discovery and enhancing decision-making in Earth science.

Geological Society of America Abstracts with Program. Vol. 57, No. 6, 2025

doi: 10.1130/abs/2025AM-6994

© Copyright 2025 The Geological Society of America (GSA), all rights reserved.