A spatial building damage inventory of the 2022 Luding Earthquake and its preliminary vulnerability analysis

Spatial seismic vulnerability assessments are primally conducted at the community and grid level, using heuristic and empirical approaches. Building-based spatial statistical vulnerability models are rare because of data limitations. Generating open-access spatial inventories that document seismic damage and building attributes and test their effectiveness in assessing damage would promote the advancement of spatial vulnerability assessment. The 2022 Mw 6.7 Luding earthquake in the western Sichuan Province of China provides an opportunity to validate this approach. The local government urgently dispatched experts to survey building damage, marking all buildings with damage class stickers. In this work, we sampled 2889 buildings as GPS points and documented the damage classes and building attributes, including structure type, number of floors, and age. A polygon-based digital inventory was generated by digitizing the rooftops of the sampled buildings and importing the attributes. Statistical regressions were created by plotting damage against shaking intensity and PGA, and Random Forest modeling was carried out considering not only buildings and seismic parameters but also environmental factors. The result indicates that statistical regressions have notable uncertainties, and the Random Forest model shows a ≥79% accuracy. Topographical factors showed notable importance in the Random Forest modeling. This work provides an open-access seismic building damage inventory and demonstrates its potential for damage prediction and vulnerability assessment.