Register here
The International Telecommunication Union (ITU) is launching the fourth edition of the GeoAI Challenge, inviting students and professionals worldwide to tackle real‑world geospatial problems using artificial intelligence (AI) and machine learning (ML) to support advance solutions to solve global challenges.
This session introduces the challenge problem statement “Reaching new heights with GeoFM embeddings”, developed in cooperation with the European Space Agency (ESA) Φ‑lab Challenges, KTH Royal Institute of Technology, Politecnico di Milano, and the Group on Earth Observation (GEO).
The project represents an important first step toward building a global open dataset of Digital Surface Models (DSM), Digital Terrain Models (DTM), and ground‑cover categories derived solely from open satellite imagery. Participants will train models on selected regions and assess their ability to generalize across diverse geographical areas. Such datasets enable a wide range of SDG‑relevant applications.
For ITU, global geospatial datasets are essential to improve the accuracy of ITU‑R radio‑wave propagation prediction methods, which support spectrum sharing studies, radio‑interference assessment, and coverage estimation for point‑to‑area services such as mobile and broadcasting systems.
Participants will explore how embeddings from state‑of‑the‑art Geospatial Foundation Models (e.g. AlphaEarth, TerraMind, Tessera, Thor) can be leveraged for multi‑task learning, including land‑cover segmentation and height estimation.
The webinar will present the challenge objectives, dataset and platform, best practices, and conclude with a live Q&A session.
Session Objectives:
By the end of this workshop, participants will be able to:
- Explain the objectives and structure of the ITU–ESA GeoAI Challenge and its relevance to solving global challenges.
- Describe the challenge problem “Reaching new heights with GeoFM embeddings”, including the dataset, tasks, and evaluation approach.
- Compare the role of Geospatial Foundation Model (GFM) embeddings in supporting multi‑task learning for land‑cover segmentation and height estimation.
- Identify best practices for training and generalizing AI/ML models across multiple geographic regions using open satellite imagery.
- Apply knowledge of the challenge platform and workflow to prepare for participation in the GeoAI Challenge.
Recommended Mastery Level / Prerequisites:
This session is intended for participants with an intermediate level of experience in Python and applied data science. Participants are expected to:
- Use Python for data analysis and machine‑learning workflows (e.g. NumPy, pandas, PyTorch or TensorFlow).
- Apply basic machine‑learning concepts, including supervised learning, regression, and classification.
- Work with geospatial or raster data, or demonstrate familiarity with multidimensional arrays and image‑like data structures.
- Understand fundamental concepts of model training, validation, and generalization across datasets.
