Physics-Anchored Short-Term Earthquake Hazard Characterization Using Large-Scale Seismic Data

Overview

California will inevitably experience future damaging earthquakes, but the exact timing and location of large events remain unpredictable. This study asked whether near-term hazard can be estimated probabilistically from evolving seismicity and continuous waveforms. I developed a physics-anchored, interaction-aware continuous space-time model that combines an ETAS baseline, neural sequence learning, and distilled waveform-derived inputs. The learned interaction term adapts hidden-influence ideas from coordinated-anomaly modeling to clustered seismicity. Applied to ~340 TB of SCEDC and NCEDC continuous waveform data, the model improved short-term hazard characterization beyond catalog-only baselines, achieved AUC-ROC = 0.79 on a held-out observed-catalog test, and enabled 75× faster inference after waveform distillation. These results suggest that continuous waveforms preserve short-term hazard information not fully retained in standard catalogs, showing that earthquake hazard is shaped not only by long-term structural conditions but also by evolving short-term sequence dynamics reflected in continuous seismic waveforms.

Competition history

• CSEF 2026 Earth & Environmental Sciences(Senior Division) · Entry S-08-18

Related projects

Closest projects by meaning, across every fair and year in the corpus.

Browse more like this

• All Earth and Environmental Sciences projects
• CSEF Earth and Environmental Sciences projects
• Every CSEF 2026 project