A Novel Framework for Rapid Peak Ground Velocity Mapping Consistent with Physics-Based Simulations

We present a novel framework for rapidly generating peak ground velocity (PGV) maps that are consistent with physics-based simulations. The framework employs a two-step approach: first, using Green’s function reciprocity, it generates a low-spatial-density PGV map that captures the coarse spatial patterns of ground motion. In the second step, a machine learning model refines this preliminary output into a high-resolution, continuous map. The framework accounts for complex 3D geological and topographic effects and achieves accuracy up to the maximal frequency of physics-based simulations. Validated in the geologically complex San Francisco Bay Area, the framework is scalable, computationally efficient, and demonstrates strong potential for earthquake hazard assessment and near-real-time disaster response.