Deep learning-based mapping of conflict damage

March 2025 to February 2028

Rapid and effective humanitarian response depends on maintaining a clear,
up-to-date understanding of how armed conflicts impact civilian populations, including the location and extent of infrastructure damage. Satellite imagery is vital for such assessments, yet manual mapping remains time-consuming and resource-intensive. This project aims to create a tool that automates key parts of the damage-mapping process by applying deep learning techniques to open-source satellite data. The project brings together an interdisciplinary team of experts and researchers in conflict studies, machine learning, and remote sensing from CSS, University of Zurich and ICRC to develop technology-driven humanitarian solutions that better utilise satellite data. The project is funded by Innosuisse.

The tool developed during this project automatically detects and visualizes conflict-related building damage and guides users to areas requiring manual assessment. Its main technical innovation lies in integrating an active learning approach, ensuring that the model improves over time from user feedback. Validation through quantitative testing, qualitative assessment, and case studies will demonstrate the tool’s accuracy and operational value

ETH PI: Dr. Valerie Sticher

ETH project team: Dr. Jenniina Kotajoki

Partners: Prof. Dr. Jan Dirk Wegner (PI, University of Zurich), Thomas Gazel-Anthoine (University of Zurich), Dr. Lukas Drees (University of Zurich), Dr. Thao Ton-That Whelan (ICRC) and Mitchell Paquette (ICRC)