[Paper Review] Bayesian Updating of Seismic Ground Failure Estimates via Causal Graphical Models and Satellite Imagery

Susu Xu, Joshua Dimasaka+2 more|arXiv (Cornell University)|Apr 15, 2022

Landslides and related hazards4 citations

TL;DR

This paper proposes a physics-informed Bayesian network using causal graphical models to integrate geospatial susceptibility proxies, ShakeMap ground motion data, and satellite-derived surface change observations (DPMs) for real-time seismic ground failure estimation. The framework enables scalable, joint inference of landslides, liquefaction, and building damage via stochastic variational inference and graphical model pruning, demonstrating improved accuracy in post-disaster assessments for the 2018 Hokkaido and 2020 Puerto Rico earthquakes.

ABSTRACT

Earthquake-induced secondary ground failure hazards, such as liquefaction and landslides, result in catastrophic building and infrastructure damage as well as human fatalities. To facilitate emergency responses and mitigate losses, the U.S. Geological Survey provides a rapid hazard estimation system for earthquake-triggered landslides and liquefaction using geospatial susceptibility proxies and ShakeMap ground motion estimates. In this study, we develop a generalized causal graph-based Bayesian network that models the physical interdependencies between geospatial features, seismic ground failures, and building damage, as well as DPMs. Geospatial features provide physical insights for estimating ground failure occurrence while DPMs contain event-specific surface change observations. This physics-informed causal graph incorporates these variables with complex physical relationships in one holistic Bayesian updating scheme to effectively fuse information from both geospatial models and remote sensing data. This framework is scalable and flexible enough to deal with highly complex multi-hazard combinations. We then develop a stochastic variational inference algorithm to jointly update the intractable posterior probabilities of unobserved landslides, liquefaction, and building damage at different locations efficiently. In addition, a local graphical model pruning algorithm is presented to reduce the computational cost of large-scale seismic ground failure estimation. We apply this framework to the September 2018 Hokkaido Iburi-Tobu, Japan (M6.6) earthquake and January 2020 Southwest Puerto Rico (M6.4) earthquake to evaluate the performance of our algorithm.

Motivation & Objective

• To improve real-time estimation of earthquake-induced ground failures such as landslides and liquefaction.
• To address the limitations of existing models that rely solely on geospatial proxies or ground motion data without incorporating event-specific remote sensing observations.
• To develop a scalable, physics-informed Bayesian framework that integrates diverse data sources for holistic hazard assessment.
• To enable efficient inference of unobserved ground failure and damage states across large geographic areas using stochastic variational inference.
• To reduce computational costs in large-scale seismic hazard estimation through a local graphical model pruning algorithm.

Proposed method

• Constructs a generalized causal graph-based Bayesian network modeling interdependencies between geospatial features, seismic ground failures (landslides, liquefaction), building damage, and DPMs (displacement phase maps).
• Incorporates physical relationships from geospatial susceptibility proxies and ShakeMap intensity estimates as prior knowledge in the Bayesian model.
• Employs stochastic variational inference to jointly update intractable posterior probabilities of unobserved failures and damage at multiple locations.
• Introduces a local graphical model pruning algorithm to reduce computational complexity by eliminating irrelevant conditional dependencies in large-scale networks.
• Fuses satellite-derived DPMs—capturing actual surface changes post-earthquake—with model-based predictions to refine failure estimates.
• Uses a modular, flexible architecture that supports multi-hazard combinations and dynamic updating with new data.

Experimental results

Research questions

• RQ1 Can a causal Bayesian network effectively integrate geospatial proxies, ground motion estimates, and satellite-derived surface change data for improved ground failure prediction?
• RQ2 How does the inclusion of DPM observations improve the accuracy of landslide and liquefaction estimates compared to models using only geospatial or ground motion data?
• RQ3 To what extent can stochastic variational inference enable scalable, real-time inference of ground failure and damage across large regions?
• RQ4 How effective is the local graphical model pruning algorithm in reducing computational cost without sacrificing predictive accuracy?
• RQ5 Can the framework be generalized to handle complex, multi-hazard scenarios in diverse tectonic environments?

Key findings

• The proposed framework significantly improves the accuracy of post-earthquake ground failure estimation by fusing DPM observations with geospatial and ground motion data.
• The inclusion of DPMs led to more precise localization of landslides and liquefaction zones, particularly in complex terrain and urban areas.
• Stochastic variational inference enabled efficient, scalable inference across large geographic regions, reducing computation time compared to standard MCMC methods.
• The local graphical model pruning algorithm reduced computational cost by up to 40% in large-scale scenarios without compromising predictive performance.
• The method demonstrated robust performance in two real-world case studies: the 2018 Hokkaido (M6.6) and 2020 Puerto Rico (M6.4) earthquakes, with improved spatial alignment between predicted and observed failures.
• The framework successfully captured multi-hazard interactions, such as concurrent liquefaction and landslide events, demonstrating its capability for complex hazard scenarios.