Generalization of Satellite-Based Building Damage Mapping Across Earthquake Events

Yifan Sun, & Lingsen Meng

Submitted August 30, 2026, SCEC Contribution #15152, 2026 SCEC Annual Meeting Poster #TBD

Rapid and accurate post-earthquake damage assessment is critical for emergency response. This study evaluates the spatial and cross-event generalizability of two complementary satellite-based paradigms: a physics-based Synthetic Aperture Radar (SAR) framework and a data-driven deep learning approach. We apply both methods to recent earthquake events, including the 2025 Myanmar and 2026 Venezuela earthquakes, under a unified spatially disjoint and cross-event evaluation framework. In the physics-based track, Damage Proxy Maps (DPMs) are derived from Sentinel-1 InSAR coherence loss and processed with Variational Bayesian Causal Inference (VBCI) to separate total surface change into landslides, liquefaction, and building damage. In the data-driven track, we employ a multimodal MetaFormer-based network that classifies per-building damage by fusing post-event optical imagery with seismic and SAR-derived information.

Through cross-event and spatially disjoint validation, we quantify the generalization limits of both approaches. The physics-based VBCI pipeline effectively isolates building damage in dense urban environments, distinguishing it from co-located hazards, while comparisons with rapid imagery assessments identify environmental controls on detection performance across diverse landscapes. The data-driven MetaFormer framework demonstrates strong building-damage classification capability, and its performance under increasingly stringent validation protocols provides insight into how data-driven models transfer across regions and event types. Together, these results highlight the importance of rigorous out-of-distribution evaluation for reliable operational earthquake damage mapping.

Key Words
Venezuela Earthquake, Myanmar Earthquake, Damage Proxy Map, liquefaction, building damage, Remote Sensing, seismic multi-hazard modeling, coseismic deformation

Citation
Sun, Y., & Meng, L. (2026, 08). Generalization of Satellite-Based Building Damage Mapping Across Earthquake Events. Poster Presentation at 2026 SCEC Annual Meeting.

Related Projects & Working Groups
Fault and Rupture Mechanics (FARM)