• Variational Causal Bayesian Inference for Seismic Multi-hazard and Impact Estimation from Satellite Images: [Xu, S., Dimasaka, J., Wald, D., Noh, H.] Inputs are rasterized building footprints from OpenStreetMap, USGS ground shaking maps, prior models of landslide and liquefaction, and NASA Damage Proxy Maps derived from Synthetic Aperture Radar.; Outputs are posterior models of building damage, landslide, and liquefaction.; Groundtruth are publicly available datasets of geotagged observations.; Case studies are 2016 Central Italy Mw-6.2 earthquake, 2018 Hokkaido Mw-6.6 earthquake, 2019 Ridgecrest Mw-7.1 earthquake, and 2020 Puerto Rico Mw-6.4 earthquake.
    pdf
    Zenodo
    GitHub

  • Probabilistic Earthquake Risk and Loss Estimation of the Greater Metro Manila Area, Philippines: [Dimasaka, J., Baker, J.] Exposure Inputs are 2014 GMMA-RAP exposure dataset of land use, building inventory, structural typology, and estimated building valuation.; Hazard Inputs are OpenQuake probabilistic seismic hazard analysis for varying return periods (100, 250, 500, 750, 1000, 1500, 2000, 2500 years) and VS30 soil map.; Fragility and Vulnerability Inputs are from HAZUS and UPD-ICE methodology.; Risk Outputs are regional damage states of over 1,750 local administrative zones; Annualized economic loss, casualty estimates (slight, severe, life-threatening, and fatalities), and damaged floor area for varying levels (none, slight, moderate, extensive, complete without collapse, and complete with collapse).
    pdf
    GitHub