2025/10/29
1649 view
32 like
Groundbreaking research using artificial intelligence to detect seismic damage from space has been recognized by four teams from different countries, marking the conclusion of a global competition organized by the European Space Agency in collaboration with the International Charter on Space and Catastrophes (commonly referred to as the Charter).
54th Council Meeting of the International Charter in Strasbourg
The winning teams – South Korea's TelePIX, Belgium's Datalayer, Japan's DisasterM3 and France's Thales Services Numériques – were recently recognized at a ceremony held at the Charter's 54th Executive Board meeting in Strasbourg, with French space agency CNES taking the lead on the Charter for the next six months.
The Earthquake Response AI Challenge, part of the ESA Φ-lab Challenges initiative, combined Charter's operational experience with ESA Φ-lab's commitment to innovation, bringing together 143 participants from 40 countries to explore how far artificial intelligence can go in automating post-disaster damage detection from space.
AI predicts earthquake damage in Mandalay
The competitors trained an AI model that can distinguish between damaged and undamaged buildings using more than 200 high-resolution images of five seismic events, one of the largest Earth observation datasets ever collected for this purpose.
This image above shows the TelePIX team's winning model prediction for Mandalay, Myanmar after the March 2025 earthquake. Mandalay was selected as one of the final test sites for this challenge. Red shapes represent predicted damage. The blue dots indicate the location of the photos, which are also featured below.
ESA Charter Head Philip Barry said: “When an earthquake occurs, every minute counts. By accelerating the production of reliable building damage maps from satellite data, these models will one day help rescue teams reach affected areas faster.”
Earthquake damage Mandalay
Global cooperation for faster disaster response
Recognizing that no single operator or satellite alone can meet the demands of disaster management, ESA and CNES launched the International Charter on Space and Catastrophes in 1999. In 2000, the Canadian Space Agency joined. The agreement is now a collaboration between 17 space agencies to provide free satellite imagery to support disaster response around the world.
Under a six-monthly rotation, CNES has taken over as the lead agency, co-hosting the latest Charter Board meeting and AI Challenge awards ceremony with ESA.
The AI for Earthquake Response Challenge was designed and implemented by ESA's Φ-lab and an industry team that created the environment, tools, and evaluation framework for participants to develop and test their models.
Earthquake response challenge using AI
The dataset used in the competition includes over 200 high-resolution images from five major earthquakes and 13 locations, totaling 475 GB of data. These data were taken from the operational archive of Charter, a global cloud-based platform implemented by ESA and operated by an Italian and Polish industrial consortium since 2018.
These are obtained from global virtual satellite constellations such as Pleiades (CNES/Airbus), WorldView and GeoEye (USGS/Maxar), KOMPSAT-3 (KARI), Global (BlackSky), and Gaofen-2 (CNSA), making it one of the most diverse datasets ever built for AI-driven damage mapping.
Behind the scenes, this effort reflected the Charter's spirit of international cooperation. The Luxembourg Institute of Science and Technology and ACRI-ST (France) coordinated the competition, provided scientific oversight, and ensured the quality and relevance of the dataset. Terradue (Italy), developer of ESA Charter Mapper, enabled global access to data through the Earth Observation Training Data Lab at ESA Φ-lab, giving all teams an equal starting point.
Participants faced challenges similar to real-world emergency operations, including multi-sensor imagery, variable resolution, complex cross-registration, and extreme class imbalances. For example, Mandalay, Myanmar, where only 0.2% of approximately 500,000 buildings were damaged.
Among the top performers, the European finalists stood out for their cutting-edge approach. Datalayer leveraged a scalable cloud-based machine learning pipeline to efficiently process large datasets, while Thales Services Numériques applied deep learning and trusted AI techniques from aerospace to accurately identify structural damage.
next step
CNES, operator of the Pleiades cluster and current lead agency for the Charter, is currently spearheading a post-challenge evaluation to assess how best-performing AI models can be integrated into operational damage mapping workflows.
Combining the experimental spirit of ESA Φ-lab and the humanitarian mission of the Charter, this effort demonstrated how space data and AI can work together to improve rapid disaster response. This is a clear example of innovation and international cooperation.
