Regional efficacy of a global geospatial liquefaction model

Abstract

Liquefaction hazard maps are important for both pre- and post-event planning and mitigation. The global geospatial liquefaction model (GGLM) proposed by Zhu et al. (2017) and recommended for global application results in a liquefaction probability that can be interpreted as liquefaction spatial extent (LSE). The GGLM uses ShakeMap's PGV, topography-based Vs30, distance to water body, water table depth, and annual precipitation as explanatory variables. The GGLM was originally developed and validated across 23 global earthquakes with most of the earthquakes in coastal settings. In this paper, LSE maps have been generated for 29 earthquakes around the world in a wide range of settings in addition to 23 of the original events to evaluate the generality and regional efficacy of the model. The GGLM was found to overpredict liquefaction spatial extent for earthquakes with large areas experiencing low PGA (below 0.1 g) and as a result, the GGLM has been modified to decrease over-prediction with the addition of a PGA threshold (no liquefaction when PGA < 0.1 g). The liquefaction intensity index (LII) is generated for each earthquake through the summation of LSE values across the event and compared with the liquefaction intensity inferred from the reconnaissance report. Using the LII as an indication of goodness of fit, the GGLM performs well across all regions with 5 earthquakes showing underprediction of LII and 4 earthquakes showing overprediction. When annual precipitation of a region exceeds 1700 mm, which is the upper quartile of annual precipitation in the development database, overprediction of liquefaction spatial extent is likely, and the use of threshold is recommended.