Regionalized Geospatial Liquefaction Model for California Using Bayesian Logistic Regression
Description:
Liquefaction poses a significant geotechnical risk during earthquakes, threatening infrastructure and communities. Global geospatial liquefaction models have been used effectively in rapid response and risk studies to quantify regional liquefaction extent but have been found to have regional bias as a result of variation on the quality of global parameters. Developing accurate, regionalized models for predicting liquefaction is crucial for effective risk mitigation. This study introduces a Bayesian logistic regression approach to create a regional geospatial liquefaction model for California. By starting with a global liquefaction dataset as a prior, we integrate detailed observations from California-specific seismic events, including the M6.9 1989 Loma Prieta, 1994 M6.6 Northridge, 2003 M6.6 San Simeon, 2014 M6.0 Napa, 2019 M6.4 and M7.1 Ridgecrest, and the recent 2024 M7 Offshore Cape Mendocino earthquakes where liquefaction was observed and documented. Earthquakes such as 2000 M5.0 Yountville, 2008 M5.4, Chino Hills, 2015 M4 Piedmont, and M7.1 Hector Mine earthquakes where no liquefaction was observed are also included to constrain the model. Our methodology uses a logistic regression model trained on the global dataset as a baseline. We then apply the Laplace approximation to derive the covariance matrix, which serves as the prior for our Bayesian model. The model is subsequently refined using California's regional data, examining the impact of various weighting strategies on the balance between global insights and local specifics to avoid overfitting. This approach offers a framework for creating more accurate regional geospatial liquefaction models. This research underscores the critical balance between leveraging prior knowledge and incorporating new data to produce reliable, context-specific hazard assessments.
Session: Earthquake-triggered Ground Failure: Data, Hazards, Impacts and Models [Poster]
Type: Poster
Date: 4/17/2025
Presentation Time: 08:00 AM (local time)
Presenting Author: Laurie
Student Presenter: No
Invited Presentation:
Poster Number: 1
Authors
Hooman Shirzadi hooman.shirzadi@tufts.edu Tufts University |
Laurie Baise Presenting Author Corresponding Author laurie.baise@tufts.edu Tufts University |
Babak Moaveni babak.moaveni@tufts.edu Tufts University |
|
|
|
|
|
|
Regionalized Geospatial Liquefaction Model for California Using Bayesian Logistic Regression
Session
Earthquake-triggered Ground Failure: Data, Hazards, Impacts and Models