GeoAI forEmergency Management: a multi-modal suite of tools for rapid rubble detection and structural damage assessment

After a major disaster, rapid mapping and assessment of structural damage are essential for situational awareness and for supporting emergency response and recovery. Remote sensing, increasingly integrated with deep learning, can enhance the speed and accuracy of post-event damage identification. Yet much of the literature still treats damage detection as stand-alone models, rather than interoperable tools designed around emergency stakeholders’ workflows. This research develops, conceptually and practically, a programmatic suite of Geospatial AI tools deployable across the emergency management cycle in Italy, a seismically active setting that remains under-served by locally tailored model development. The thesis proposes and evaluates the Earthquake Emergency Suite, a set of GeoAI tools supporting actors at different phases of the cycle. During response, the Rapid Collapse Mapper (RCM) identifies rubble and building collapse from post-event optical satellite imagery and outputs GIS-interoperable vector layers to prioritize search-and-rescue. During recovery, the Macroseismic Survey Mapper (MSM) supports macroseismic survey workflows by detecting façade damage in UAV-derived imagery and classifying it according to the European Macroseismic Scale (EMS-98). Both tools align with the informational workflows and institutional structures of the Italian civil protection system. Across experiments, performance was moderate (F1=0.35 for RCM; F1=0.28 for MSM), consistent with limited and heterogeneous training data. While results do not yet support fully automated deployment, the research demonstrates the feasibility of a national-context GeoAI suite delivered through a reproducible, GIS-compatible pipeline that can serve as a blueprint for future work. Priorities include expanding and diversifying datasets, improving annotation consistency and cross-event representativeness, and further validating MSM on UAV imagery to reduce viewpoint variability and improve robustness.