Fostering Advancements in Foundation Models via Unsupervised and Self-Supervised Learning for Downstream Tasks in Earth Observation

The FAST-EO project seeks to revolutionize Earth Observation (EO) by adapting Multimodal Foundation Models (FMs) for specific EO data types like synthetic aperture radar, multispectral, and hyperspectral sensors. This adaptation aims to address the complexity of data-driven problems in science and engineering, enhanced by the data proliferation in the information era. The project focuses on enhancing multimodality, incorporating text-based and semantic prompts, and overcoming computational barriers to make these models more accessible and operational. It also emphasizes affordable fine-tuning for global-scale deployment. The effectiveness of these tailored FMs will be validated through various practical applications, including weather and climate disaster analysis, methane leak detection, changes in forest biomass, soil property estimation, land cover change detection, and monitoring mining field expansion. This comprehensive approach aims to democratize and optimize the use of FMs in EO, paving the way for more robust insights and decision-making capabilities across diverse applications.