Four teams from different countries have been recognized for their breakthrough work in using artificial intelligence to detect earthquake damage from space, which marked the end of the global competition, organized by the European Space Agency in cooperation with the International Charter "Space and Major Disasters".
The winning teams are TelePIX from the Republic of Korea, Datalayer from Belgium, DisasterM3 from Japan and Thales Services Numériques from France were recently honored at the ceremony, which took place at the 54th meeting of the Board of Directors of the Charter in Strasbourg, and the French Space Agency (CNES) took over the leadership of the Charter for the next six months.
ESA Φ-lab Challenges, collected 143 participants from 40 countries, to investigate, how far artificial intelligence can go in automating disaster detection from space.
Contestants trained artificial intelligence models, able to distinguish between damaged and undamaged buildings, using one of the largest Earth observation datasets, ever collected for this purpose - over 200 high-resolution images of five earthquakes.
This image above shows Team TelePIX's winning model forecast for Mandalay, Myanmar, after the earthquake in March 2025 year. Mandalay was chosen as one of the final test sites in the competition. Red shapes represent predicted damage. The blue dot indicates where the photo was taken, which is also shown below.
knowing, that a single operator or satellite cannot meet the needs of disaster relief, ESA and CNES in 1999 launched the International Charter on Space Issues and Elimination of the Consequences of Major Disasters. IN 2000 In 2008, the Canadian Space Agency joined them. Now it's a collaboration 17 space agencies, which provides free satellite imagery to support disaster response around the world.
The "Artificial Intelligence for Earthquake Response" competition was developed and implemented by the Φ-Lab of the European Space Agency (THIS) together with the industrial team, which created the environment, tools and evaluation system for participants, so they can develop and test their models.
Data set, used in the contest, included over 200 high-resolution images of five major earthquakes and 13 points - in general 475 GB of data - obtained from the operational archives of the Charter, global cloud platform, implemented by ESA and operated by an industrial consortium from Italy and Poland 2018 year.
This data came from a global virtual constellation of satellites, including the Pleiades (CNES/Airbus), WorldView and GeoEye (USGS/Maxar), COMPSAT-3 (KARI), Global (BlackSky) таGaofen-2 (CNSA), making it one of the most diverse datasets, ever created for artificial intelligence-based damage mapping.
Behind the scenes, these efforts reflected the spirit of international cooperation, provided for by the Charter. Luxembourg Institute of Science and Technology and ACRI-ST (France) coordinated the competition, providing scientific oversight and guaranteeing the quality and relevance of the dataset. Terradue (Italy), developer of the ESA Charter Mapper, provided global data access through the ESA Φ-lab Earth Observation Training Data Laboratory, giving all teams equal starting opportunities.
Participants faced challenges, similar to those, as in real-world emergency operations: multisensory images, variable resolution, complex co-registration and extreme class imbalance—how, example, in Mandalay, Myanmar, where only damage was done 0,2% of almost half a million buildings.
Among the best participants, the European finalists stood out with their advanced approaches. Datalayer used scalable cloud-based machine learning pipelines to efficiently process massive data sets, while Thales Services Numériques applied deep learning and robust AI techniques from the aerospace industry to accurately identify structural damage.
Source: https://www.esa.int/Applications