Climate Change AI side event at ESA Φnnovation Summit: AI for Earth Observation & Climate – What Actually Matters?

Climate Change AI is organizing a side event at the European Space Agency Φnnovation Summit. The side event “AI for Earth Observation & Climate – What Actually Matters?” is taking place on June 23, 2026 in Frascati, Italy.

Overview

AI-powered Earth Observation (EO) is advancing rapidly, and space agencies are at the forefront of that momentum. Yet as capabilities multiply – foundation models, AI-generated products, planetary-scale monitoring pipelines – it is not always clear what moves the needle for climate. This workshop takes a critical look at current technical development to ask which advances translate into meaningful climate impact, which trade-offs are underexamined, and how EO agencies and the broader community can better position their work to support effective climate action. The session is structured around key choices – model size, data distribution, product integration, and communication – where clearer thinking leads to better outcomes.

Objectives

Participants will leave with a sharper framework for evaluating AI and EO from a climate impact perspective; a clearer picture of where geo-foundation models genuinely fit – and where their limitations matter most; and practical guidance on how to best develop AI-generated geospatial products and integrate them with existing data ecosystems to effectively support (AI-for-)climate applications.

Speakers

• Clinton Oduor
• Evan Shelhamer
• Ana Martinez

Moderators

• Nikola Milojevic-Dupont
• David Rolnick

Schedule

14:37	Start of event
14:37–14:45	Welcome participants and short introduction
14:45–15:05	Block 1: Fireside chat with Clinton Oduor
15:05–15:15	Block 1: Q&As with Clinton Oduor
15:12–15:35	Block 2: Fireside chat with Evan Shelhamer
15:35–15:45	Block 2: Q&As with Evan Shelhamer
15:45–16:05	Block 3: Fireside chat with Ana Martinez
16:05–16:15	Block 3: Q&As with Ana Martinez
16:15–16:30	General discussion
16:30	End of event

Structure

Block 1 — Model Size and Climate Cost: Getting the Trade-offs Right. The EO community has enthusiastically embraced large models, sometimes leading to sizable improvements, but a model’s climate relevance is not always a function of its size. This block examines where large models incur overhead for climate-relevant EO tasks, where lighter approaches are underrated, and how to think about key trade-offs. The goal is not skepticism about scale, but precision about when it pays off.

Block 2 — Geo-Foundation Models & Earth Embeddings for Climate. Geo-foundation models mark a real shift in EO capabilities, but their role in climate tasks is still emerging. This block assesses them realistically—what they do well, where they fall short, and how the EO community can engage in ways that keep climate applications central. At the core of this discussion is the role of Earth embeddings as a key innovation for geospatial climate analytics. We will explore current calls to action: how to address challenges and opportunities, and the importance of building an accessible, interdisciplinary, and impactful Earth-embedding community.

Block 3 — AI-Generated Climate Products: Integration for Real-World Impact. This block addresses the “last-mile” gap between EO outputs and actionable climate decisions. As EO increasingly generates derived datasets, such as building footprints and land-use maps, a key challenge is integrating them with existing sources, such as cadastral records and citizen science data. The block focuses on the resulting challenges, including uncertainty quantification and communication, principled data integration, and usability for downstream users. The aim is to show how adherence to good practices can enhance the usability and real-world impact of EO products for climate applications.