← Back to Sessions

Improving Geospatial Liquefaction Prediction Models by Optimizing Non-Liquefaction Points Sampling: A Case Study of the 2023 Kahramanmaras, Turkey Earthquake Sequence

Description: 

To create geospatial liquefaction prediction models, we need to observe both liquefaction and non-liquefaction points to use this information to extract the value of explanatory geospatial proxies like VS30, distance to water, etc. Reconnaissance teams usually report only the location of liquefaction; therefore, to sample non-liquefaction points, we need to make different assumptions. In the literature, a donut-shaped area with an inner radius of 1 km and an outer radius of 15 km is used to create a fishnet with desired grid size and sample non-liquefaction points. This study looks at the critical aspect of non-liquefaction point sampling strategies and how the geometric properties of the donut-shaped non-liquefaction sampling region around the liquefaction observation affect the accuracy and generalizability of the resulting model. This study uses the extensive reconnaissance reports from the 2023 Kahramanmaras, Turkey Earthquake Sequence to compare different sampling strategies based on their impact on the accuracy of the logistic regression model. We use the inner and outer radii of the donut-shaped non-liquefaction sampling buffer as optimization variables to do a grid search over the different combinations of them. The resulting models are then compared in terms of balanced accuracy, area Under the ROC Curve (AUC), precision, and recall. The insights gained from this study will help refine the methodology for creating geospatial liquefaction prediction models and provide valuable guidance for optimizing earthquake risk assessment strategies.

Session: Regional-Scale Hazard, Risk and Loss Assessments [Poster Session]

Type: Poster

Date: 5/3/2024

Presentation Time: 08:00 AM (local time)

Presenting Author: Hooman

Student Presenter: Yes

Invited Presentation: 

Authors