Barcelona Supercomputing Center

Foundation models are reshaping not only business processes, but increasingly also the scientific and engineering landscapes, including atmospheric prediction and Earth observation. Through self-supervised learning, these models generate rich semantic representations of raw data in the form of tokens and embeddings. These latent representations are emerging as the new lingua franca in an era of AI-driven knowledge discovery and value creation.

In this keynote, I will share insights from our latest research into multi-modal Earth Observation Foundation Models (i.e. TerraMind+), leveraging token-level correlation learning to enhance climate and transition risk assessments. By fusing satellite imagery with atmospheric variables, we aim to unlock new levels of precision and insight in environmental applications. In addition, I will explore how embeddings can fundamentally reshape data workflows, through compression, reducing storage and transfer costs, enabling data federation, and accelerating data exploration, inference, and model adaptation to domain-specific tasks.

These innovations hold transformative potential not just for environmental science, but for any domain navigating vast and complex data landscapes.

In an increasingly complex security landscape, the fusion of diverse data and intelligence methods such as OSINT and Earth Observation (EO) data is becoming critical for effective threat assessment and response. As the nature of global threats evolves, so too must our methodologies for gathering, analysing, and utilising intelligence. This keynote will dive into the practical application of combining traditional and more innovative intelligence techniques, with real-world use cases from the security and intelligence sectors. 

I will discuss the integration of OSINT and EO data in intelligence operations, particularly for addressing transnational organised crime, cyber-enabled threats, and geopolitical instability. Criminal actors are increasingly incorporating GenAI into their workflows, enhancing the scale and sophistication of transnational organised crime. 

This evolution requires organisation to adapt their strategies and reinforces the need for enhanced inter-agency cooperation and information sharing to improve threat detection and cross-border intelligence coordination. This session will provide an overview of the intelligence and security domain, along with insights into the operationalisation of advanced methodologies, offering strategies to enhance the efficiency, accuracy, and overall effectiveness of security measures.