Session OverviewGeospatial data sits at the heart of some of the most pressing decisions of our time — from disaster response and climate monitoring to urban planning. Artificial intelligence is reshaping how that data is collected, interpreted, and acted upon. Yet the field is at an inflection point: transformative capabilities are emerging alongside equally serious questions about hype, environmental cost, accessibility, and the human pathways into the discipline. This session brings together four perspectives that together span the full arc of GeoAI today. Prof. Ali Mansurian examines how Geospatial Large Language Models are democratising access to spatial intelligence for geographic analysis. Dr. Andrea Manara presents the outcomes of a multi-institutional challenge on Geospatial Foundation Models. Prof. Gregory Giuliani offers a critical audit of the field. And PhD researcher Julia Leonardi brings the early-career perspective. Together, they address a central question: is the field living up to its promise — and if not, what must change?
Moderator: Prof. Ludovico Biagi, Politecnico di Milano
The Panel
- Prof. Ali Mansurian, University of Lund, Sweden
Talk: GeoLLMs and the future of spatial intelligence
Abstract
A new generation of Geospatial Large Language Models (GeoLLMs) is beginning to reshape how society interacts with geographic information. For decades, the power of geographic information systems (GIS) has remained largely in the hands of specialists capable of navigating complex tools, data structures, and analytical workflows. GeoLLMs introduce a transformative paradigm in which natural language becomes the interface to spatial intelligence, allowing users to ask questions and receive maps, analyses, and insights in return. This shift has the potential to democratize access to geospatial knowledge and significantly expand its role in policy, planning, science, and public decision-making. At the same time, it invites a broader reflection on how geospatial data infrastructures, governance models, and digital ecosystems must evolve to support reliable, transparent, and trustworthy AI-mediated spatial analysis. This talk explores the emerging vision of AI-enabled geospatial ecosystems and discusses what this transformation may mean for the future of spatial knowledge and evidence-based policy.
- Dr Andrea Manara, Senior System Analyst, International Telecommunication Union
Talk: Geospatial foundation models in practice
Abstract
Manara opens with a presentation of several webinars on Geospatial Foundation Models (GFMs) he co-organised in the GeoAI Discovery Series, tracing how a collaboration involving ITU, ESA Φ-lab, KTH, and Politecnico di Milano took shape and led to the public launch of a GeoAI challenge in March 2025. The challenge — titled “Reaching New Heights with GeoFM Embeddings” — asks participants to leverage embeddings from state-of-the-art Geospatial Foundation Models such as Alpha Earth, TerraMind, Tessera, and Thor to perform Digital Surface Models (DSM) and Digital Terrain Models (DTM) estimations and cover-type segmentation over unseen geographic regions- the embeddings are produced with open source satellite imagery only. The output has a wide range of real-world applications relevant for meeting the SDGs; for ITU, in particular, this dataset is crucial to enhance the accuracy of some ITU-R radio-wave propagation prediction methods. A core scientific question is whether GFMs trained on a limited set of areas can generalise to entirely different regions. Manara then presents the results submitted by participants, discussing what the outcomes reveal about the current capabilities and limitations of GFMs for multi-task learning at global scale. - Prof. Gregory Giuliani, University of Geneva & UNEO/GRID-Geneva
Talk: Are we really delivering the promises of GeoAI or is it just a Hype?
Abstract
A critical assessment of whether AI tools are truly delivering scientific advances in geospatial analysis — or whether the field is caught in a paradox of overpromising while overlooking its own environmental costs. Giuliani interrogates the assumption that ever-more-complex models represent meaningful scientific progress, asking whether marginal gains in classification accuracy justify the computational and carbon costs required to achieve them. He challenges the research community to be honest about the gap between benchmark performance and real-world utility, and to confront an uncomfortable irony: that AI systems built to monitor and protect the environment are themselves accelerating the environmental degradation they are meant to address. His intervention is a call not for pessimism, but for a more rigorous, self-aware, and ecologically responsible GeoAI research agenda.
- Ing. Julia Leonardi, PhD Researcher, Politecnico di Milano
Talk: The young researcher's perspective on GeoAI
Abstract
How the next generation enters GeoAI — through environmental science or computer science — and why the field urgently needs more integrated, accessible, and internationally inclusive academic pathways? Leonardi maps at least two distinct entry routes into the discipline: students arriving from Geography and Environmental studies who know the “why” of spatial problems but may lack methodological depth, and those from Computer Science who know the tools but often struggle to contextualise them within real-world Earth applications. She argues that this dual-track structure is not merely a pedagogical inconvenience but a structural barrier that limits both the quality of research and the diversity of who participates in it. Drawing on her own experience and that of peers across European doctoral programmes, she calls for genuinely co-designed Geoinformatics curricula, greater visibility for summer schools and early-career initiatives, and a more deliberate effort to lower barriers for students from under-resourced or underrepresented regions.