Please indicate if you will participate onsite or on line at the event 2nd ESA-NASA Workshop on AI Foundation Model for Earth Observation (EO).

Please note that all speakers presenting at the event must be present in person.

To ensure you have the best experience at the workshop, we’d love to know which sessions interest you the most. Please let us know your preference.

Building Agentic Earth Intelligence: A Hands-On Tour of the EVE Platform and Tool Ecosystem (2 Hours by  À. R. Atrio,  A. Lopez, J. Rohit – Pi School):  This hands-on session provides a practical tour of Earth Virtual Expert (EVE), an open, agentic platform  for Earth Observation (EO) and Earth Sciences funded and supported by ESA Φ-lab. In this session, we will create an end-to-end agent workflow from scratch, starting only from a real geospatial tool’s API and an LLM API. Participants will see how EVE combines a domain‑focused LLM (EVE‑Instruct) with structured tool use, connects to external geospatial services, runs tool calls with transparent traces, and supports community‑added tools via a standard MCP interface.

From Operational Needs to FM Design: A Co-Design Lab for Disaster-Ready EO FMs (2 Hours by J. Van Den Hoek – OSU,  J. Jakubik – IBM, H. G. Pankratz - UAH, Myscon Truong - Spatial Informatics Group, P. Parthelme and C. Scher – OSU, J. Maharjan and H. Yin – KSU, F.  Meyer and W. Horn – ASF): This hands-on session is meant to bridge interests of natural hazard and disaster domain experts and deep learning and FM experts. The session is designed as a co-design lab where participants will translate disaster impact assessment needs into explicit architectural constraints and benchmarking conditions suitable for guiding future EO FM development.

Earth Embeddings for EO: Retrieval, Discovery, and Change-Oriented Search (2 Hours by F. Yu, R. Madhok – TerraByte AI): This hands-on session introduces Earth embeddings – vector representations of EO products learned by foundation models and demonstrates how they enable scalable search, discovery, and comparison across space, time, and sensors. We connect common training paradigms (masked/self-supervised learning, contrastive learning, and multimodal alignment) to the retrieval behaviors practitioners observe in real EO workloads.

Operational Geospatial AI: Fine‑Tuning, Inference, and Scalable EO Model Serving (4 Hours by F. Otieno and M. Gazzetti – IBM): This session will focus on fine‑tuning pretrained GeoFMs, curating datasets, and running inference workflows. This will be done with GEOStudio, a platform for orchestrating geospatial machine‑learning workflows that supports dataset onboarding, model training, and deployment for inference. Attendees will learn how to serve GeoFMs using vLLM, the de facto standard for serving open‑source AI models in production environments. Finally, we will introduce the cloud‑native technologies and practical skills needed to support reliable and scalable deployment of GeoFMs in real‑world environments.

Building a GeoAI Agent: A Hands-On Tutorial on Agentic Foundation Models for Earth Observation (4 Hours by L. Thomas, S. Ranjan, W. Ji – Development Seed): This hands-on tutorial introduces participants to the design and implementation of an agentic Earth Observation (EO) foundation model workflow, or “GeoAI Agent,” capable of translating conversational scientific intent into spatially explicit, reproducible geospatial analyses.  The tutorial explicitly addresses the unique challenge of combining map-based geospatial interaction paradigms with conversational reasoning systems, demonstrating patterns for spatial grounding, tool orchestration, and scientific guardrailing.

Building Scalable AI EO Workflows: TerraTorch Embedding Workflows & TerraTorch Iterate for Zero-Invasive HPO and NAS (4 Hours by R. Kienzler and I. Wittman - IBM): This hands-on session will cover a full scalable AI EO workflow: from generating EO embeddings to building and training downstream models and automating large-scale experimentation for any EO AI pipeline.  It will provide an end‑to‑end introduction to building such pipelines using TerraTorch and TerraTorch Iterate, introducing geospatial embedding workflows and general-purpose AI EO automation for HPO and NAS.

Quantitative Evaluation and Science-Driven Use of Weather Foundation Models (4 Hours by L. Ott, K.H. Breen and M.L. Carroll - NASA):  This hands-on session introduces a science-driven benchmarking framework for evaluating foundation models for mid-range weather prediction. Participants will generate multi-day forecasts and quantitatively assess performance against reanalyses and observations across atmospheric levels, latitudinal bands, geophysical variables, and forecast lead times.

Please indicate below if you will attend the Social Dinner.