Evaluation of a cone penetration test thin-layer correction procedure in the context of global liquefaction model performance

Abstract

Engineering geologists routinely perform liquefaction hazard assessments using data from the cone penetration test (CPT). However, the volume of soil mobilized by the CPT acts as a low-pass filter on the true stratigraphy, potentially removing information such as the data defining a thin layer of soil or the interface between two dissimilar soils. The Boulanger and DeJong (2018) CPT inversion procedure, which aims to correct these effects, is herein evaluated in the context of CPT-based liquefaction model performance. Using over 15,000 case-histories from 24 earthquakes parsed into 2 datasets, 18 different liquefaction models are studied, resulting in 36 performance trials. In 1 of these trials, the CPT inversion procedure increases model efficiency to a statistically significant degree, but in 23 others it significantly decreases efficiency. This decline in performance tends to grow as profiles become more stratified. To explore remedies, a liquefaction triggering curve is rederived from inverted CPT data, such that its training and forward implementation are made consistent. Nonetheless, this exacerbates the decline in prediction efficiency. Ultimately, the results of this study are not a direct assessment of the pioneering Boulanger and DeJong (2018) procedure. However, the results do provide evidence that this procedure – when applied to existing CPT-based liquefaction models – may provide no demonstrable performance benefit.