Operationalising EMS-98 Damage Classification: A UAV-to-GIS Pipeline for Macroseismic Survey Support

Giovanni Galli , Marco Dubbini, FIlippo Bernardini, Luca Arcoraci

Macroseismic surveys are central to post-earthquake damage assessment but remain labour-intensive, relying on in-person inspections that delay the production of structured, georeferenced damage information. While UAV imagery and deep learning have shown promise in accelerating façade-level damage detection, two limitations recur in the literature: most models are trained on damage taxonomies that depart from the European Macroseismic Scale (EMS-98), and outputs are typically delivered as standalone classifications rather than as informational products interoperable with the GIS environments used by emergency-management actors. This study presents an end-to-end pipeline that addresses both gaps. A YOLOv11x-seg instance-segmentation model is trained via staged transfer learning — fine-tuned on imagery from the INGV Database Fotografico Macrosismico (DFM) — to detect and classify façade damage according to the five EMS-98 grades. Per-image detections are then aggregated into a single building-level damage grade and georeferenced using UAV GNSS and camera-orientation parameters, producing a GeoPackage directly loadable in QGIS. The pipeline is operationalised through a UAV case study in Arquata del Tronto (Central Italy), an area severely affected by the 2016–2017 seismic sequence, using imagery acquired specifically for this work. Results characterise per-class detection performance, building-level grading accuracy against ground-truth survey records, and the practical viability of integrating the output into established Italian macroseismic-survey workflows