Simulation-Based Ground Motions for Earthquake-Triggered Landslide Hazard: Validation Against Historical Events and Application to Scenario Earthquakes

Fengjiang Ju, Te-Yang Yeh, Josh West, & Yehuda Ben-Zion

Submitted August 30, 2026, SCEC Contribution #15259, 2026 SCEC Annual Meeting Poster #TBD

Co-seismic landslide probability models rely on ground-motion parameters such as peak ground velocity (PGV). In regions with abundant historical earthquake recordings, empirical Ground Motion Models (GMMs) are well constrained and products such as USGS ShakeMap provide reliable PGV. Scenario-earthquake simulations provide an alternative approach that can augment these methods.

To evaluate this approach, we use the 1989 Loma Prieta Mw 6.9 and 1994 Northridge Mw 6.7 earthquakes, which have well-constrained GMMs and mapped landslide inventories. For each event we simulate seismic wavefields from several finite-fault source models, compute PGV, and predict landslide probability on a grid using the Nowicki Jessee et al. (2018) model. Classifying cells with probability above a threshold as predicted landslide cells, we define precision as the fraction of predicted cells that coincide with mapped landslides, and recall as the fraction of mapped landslide cells that are correctly predicted. Grid resolution and threshold level dominate the performance: refining the grid from coarse (~0.925 km) to fine (~0.111 km) raises recall but lowers precision by about an order of magnitude (~15-30% to ~1-5%), driven by more non-landslide cells rather than weaker discrimination. Across both events a consistent trade-off emerges: simulated PGV yields higher recall, whereas ShakeMap gives comparable or more stable precision, with the gap widening at high thresholds; the best-performing source model depends on the event.

We further apply the approach to two scenario earthquakes: a M9.0 Cascadia megathrust and a M7.8 southern San Andreas Fault event. Preliminary simulations suggest contrasting patterns: Cascadia produces broadly distributed, lower-probability hazard, whereas southern San Andreas produces more localized, higher-probability hazard near urban corridors, with hazard more sensitive to ground motion for Cascadia and to terrain slope for San Andreas. These results show the potential of simulations where empirical ground-motion constraints are unavailable. Updated results will be presented in the meeting.

Citation
Ju, F., Yeh, T., West, J., & Ben-Zion, Y. (2026, 08). Simulation-Based Ground Motions for Earthquake-Triggered Landslide Hazard: Validation Against Historical Events and Application to Scenario Earthquakes. Poster Presentation at 2026 SCEC Annual Meeting.

Related Projects & Working Groups
Ground Motions (GM)