Improving operational use of post-disaster damage assessment for Urban Search and Rescue by integrated graph-based multimodal remote sensing data analysis

This work investigates the application of remote sensing technologies within the specific operational context of emergency urban search and rescue (USAR) efforts post-disaster. We thoroughly investigate two innovative methodologies, each tailored to meet distinct operational goals in a USAR setting. The first employs a belief propagation framework that is designed to provide prompt and robust initial damage assessments using sparse data, with the capability to incorporate additional on-site information as it becomes available. The second methodology introduces a modified graph convolutional network to quantify the uncertainty levels inherent in damage classification tasks. Three case studies were studied, using damage assessment data from the 2020 Beirut explosion, the 2021 Haiti earthquake and the 2023 Türkiye-Syria earthquake. Our experimental results demonstrate the potential of these approaches to achieve operational objectives, particularly in terms of robustness and scalability in disaster scenarios. The versatility offered by graph-based methodologies establishes a solid foundation for addressing these dynamic challenges, suggesting a promising direction for continued research in this field. •Satellite analysis inform better post-disaster Urban Search and Rescue decisions.•Multimodal analysis leverages more information than the sum of single sources.•Analysis of satellite data post-disaster must be robust, scalable and fast.•Dynamic and updatable methods include more information as the response progresses.