Dynamic post-earthquake updating of regional damage estimates using Gaussian Processes

• Published in: Reliability engineering & system safety Vol. 234; p. 109201
• Main Authors: Bodenmann, Lukas, Reuland, Yves, Stojadinović, Božidar
• Format: Journal Article
• Language: English
• Published: Elsevier Ltd 01.06.2023
• Subjects: Gaussian Process models; Post-earthquake damage assessment; Regional earthquake risk models; Uncertainty reduction; Gaussian Process models; Uncertainty reduction; Regional earthquake risk models; Post-earthquake damage assessment
• ISSN: 0951-8320; 1879-0836
• DOI: 10.1016/j.ress.2023.109201

The widespread earthquake damage to the built environment induces severe short- and long-term societal consequences. Better community resilience may be achieved through well-organized recovery. Decisions to organize the recovery process are taken under intense time pressure using limited, and potentially inaccurate, data on the severity and the spatial distribution of building damage. We propose to use Gaussian Process inference models to fuse the available inspection data with a pre-existing earthquake risk model to dynamically update regional post-earthquake damage estimates and thereby support a well-organized recovery. The proposed method consistently aggregates the gradually incoming building damage inspection data to reduce the uncertainty in ground shaking intensity geographic distribution and to update regional building damage estimates. The performance of the proposed Gaussian Process methodology is demonstrated on one fictitious earthquake scenario and two real earthquake damage datasets. A comparison with purely data-driven methods shows that the proposed method reduces the number of building inspections required to provide reliable and precise damage predictions. •Encode domain knowledge in Gaussian Process models to estimate post-earthquake damage.•Constrain estimates of the earthquake-induced ground shaking with inspection data.•Simultaneously update building vulnerability functions and missing inventory data.•Demonstrated on one simulated and two real earthquake damage datasets.•Reliable and precise predictions with small amounts of post-earthquake data.