Regional and site-specific liquefaction evaluations of the 2017 M5.5 Pohang earthquake in Pohang-si

Abstract

The 2017 Pohang earthquake with a moment magnitude (M) of 5.5 caused liquefaction in Pohang-si, South Korea, with approximately 600 sand boils. This contrasts with most historic cases of liquefaction, typically triggered by earthquakes of M ≥ 6.0. This study conducts regional and site-specific liquefaction evaluations for the M5.5 Pohang earthquake in Pohang-si. For the regional assessment, 13 geospatial liquefaction probability models are developed using logistic regression (LR) and four machine learning algorithms: random forest (RF), gradient boosting, extreme gradient boosting, and artificial neural network. These models consider nine variables, including two seismic loading variables (peak ground acceleration (PGA) and peak ground velocity), five geospatial variables (slope-derived average shear wave velocity (VS) of the upper 30 m, compound topographic index, distance to the coast, distance to the nearest water body, roughness), and two geotechnical variables (averaged standard penetration test N values up to 20 m depth and depth to rock). The model performances of this study and previous studies are evaluated using four indicators: accuracy, balanced accuracy, F1 score, and area under the receiver operating characteristic curve. The machine learning-based models outperform the LR-based models, with the RF-based model showing the best performance. Finally, the best model is applied to the entire Pohang-si. The estimated liquefaction probabilities agree well with the observed distributions of the sand boils and settlements. These models provide a valuable tool for continuous liquefaction hazard mapping in regions with similar geological and geotechnical conditions to Pohang-si. Large sand boils were reported at the Songdo Pine Forest, although it is far from the epicenter of the Pohang earthquake. For site-specific liquefaction evaluations at this site, deep VS profiles are estimated using microtremor array measurements and multi-channel analysis of surface waves. Based on the best VS profile, one-dimensional nonlinear ground response analyses are performed in conjunction with pore water pressure models (i.e., effective stress analyses). To account for uncertainties in nonlinear soil properties and ground motions, eight cases for nonlinear soil curves are included and 48 rock outcrop motions are used. At this site with soft and deep soil layers, short-period components are de-amplified, while intermediate- and long-period components are amplified. In addition, several cases result in the maximum pore water pressure ratio of 1 (i.e., liquefaction occurrence). Based on the results of ground response analyses, liquefaction potential index (LPI) values are also calculated using a VS-based simplified method. To compute the cyclic stress ratio, the maximum shear stress ratio ( 𝜏𝑚𝑎𝑥 ′ ), PGA of surface motion multiplied by stress reduction factor (rd), and PGA estimated by ground response analyses (PGAGRA) are utilized. The mean LPI values derived from 𝜏𝑚𝑎𝑥 ′ and PGA × rd are zero, while those derived from PGAGRA range from 0.013 to 3.202, indicating minor liquefaction-induced damage. Although the mean LPI values underestimate the observed maximum sand boil area of 84.7 m2, this study demonstrates that reliable liquefaction assessments in deep soft soil sites such as the Songdo Pine Forest require comprehensive analyses incorporating multiple evaluations rather than relying on a single approach.