Foundation Models (FMs), the latest breakthrough in Artificial Intelligence (AI) following the advent of Deep Learning, promise to revolutionize various fields, including Earth Observation (EO) and Earth Sciences. These models are trained on vast amounts of unlabeled data using self-supervised learning (SSL), allowing them to capture complex patterns and inherent information within the data. Once pre-trained, FMs can be adapted to numerous downstream tasks through fine-tuning or learning with little to no additional training samples, unlocking AI's full potential across a wide range of applications. This paradigm shift will transform the information value chain, significantly impacting the industry, research and development, and the broader scientific community. 

Incorporating Foundation Models into the Earth Observation and Earth Sciences ecosystem holds great promises for enhancing the analytical and predictive capabilities of geospatial applications including potentials for prescriptive AI. Recent advances, such as the use of Foundation Models in weather prediction, have demonstrated their ability to significantly improve the accuracy of forecasting by analyzing vast amounts of climate and atmospheric data. These models can uncover intricate patterns in weather systems and offer more reliable predictions across different time scales. Additionally, Foundation Models introduce new capabilities for geospatial semantic data mining by leveraging latent space representations and embeddings, which allow for the extraction of meaningful patterns and relationships from complex datasets. This approach not only enhances the interpretation of geospatial information but also significantly reduces the volume of raw data required for analysis. In addition, FMs reduce the need for large amounts of task-specific training data, which is particularly advantageous in EO and remote sensing where acquiring large amounts of labeled data can be challenging (e.g., laborious, expensive, and time-consuming).   

However, the integration of FMs into the EO ecosystem is not without challenges. One of the primary issues is adapting these models to the unique characteristics of EO data, which often involves multi-modal, multi-resolution, all-spectral-bands, and heterogeneous datasets. The continuous synchronization of AI models with physical entities in digital twins requires sophisticated updating mechanisms. Additionally, the vast volumes of data generated by EO systems necessitate the development of AI architectures capable of processing and analyzing this data efficiently, which remains a significant technical challenge. A crucial aspect of deploying FMs in Earth Observation and Earth Sciences is the evaluation of their performance. Existing benchmarks provide valuable initial assessments but are limited in scope and complexity. Another major challenge is the computational intensity required for training large-scale FMs on extensive datasets. The experience of using supercomputers to train these models highlights the need for optimizing training strategies to suit different hardware configurations, ensuring that the benefits of FMs can be fully realized across various platforms and institutions. 

This workshop on Foundation Model for Earth Observation and Earth Sciences aims to address these challenges by bringing together experts from AI, geospatial science, climate, weather and Earth observation communities. The goal is to foster interdisciplinary collaboration, share the latest research and technological advancements, and discuss practical solutions for effectively integrating FMs into the Earth Observation and Earth Sciences ecosystems. Topics to be covered include data curation, advancements in the AI architecture, fine-tuning, model evaluation, and challenges of integrating AI with geospatial digital twins, upstream and downstream applications. Additionally, the workshop will explore how Foundation Models can be leveraged in the context of commercial Earth Observation products and services, driving innovation in the private sector. Through this workshop, we aim to pave the way for the responsible and impactful use of AI in advancing our understanding of Earth's systems and enhancing global environmental monitoring and response efforts.