FOUNDATION MODELS FOR CLIMATE AND SOCIETY (FM4CS)

Summary

The project “Foundation Models for Climate and Society” (FM4CS) seeks to leverage the breakthrough of Foundation Models (FMs)  in shaping the future of Earth observation. During the last year, FMs have revolutionized the understanding and processing of data, in particular language data, paving the way for significant advancements in various fields.  FM4CS aims to establish a multimodal  FMs for Earth observation (EO), with a focus on data from the Sentinel 1-3 satellites, using the latest advancements in self-supervised learning.

The FM4CS project aims to demonstrate the capabilities of FM in a range of climate and society downstream tasks: snow monitoring, flood zone mapping, sea ice mapping, iceberg detection and monitoring, and draught mapping and monitoring.

Project Objectives

The main objective of the FM4CS project is to unleash the potential of self-supervised learning and large-scale multimodal Foundation Models for EO use cases related to climate science and society applications and services. In particular, we aim to:

1. Define and develop an EO-specific FM using large-scale EO archives.
2. Develop AI4EO downstream tasks using the developed FM for the EO use-cases: snow mapping, flood zone mapping, sea ice mapping, iceberg detection and draught monitoring.
3. Develop a roadmap for the use of FM in EO, with emphasis on scaling up, preparation for operationalization, and adoption of this technology in EO.
4. Engage and disseminate project results with the players and potential users of AI FMs within the EO community. Downstream climate and society applications

Downstream climate and society applications