Incorporating uncertainty in susceptibility criteria into probabilistic liquefaction hazard analysis

Most conventional approaches for assessing liquefaction triggering hazards generally rely on simplified procedures that involve identifying liquefaction susceptible layers and calculating a factor of safety against liquefaction (FSL) in each layer. Such procedures utilize deterministic semi-empirical models for standard penetration test (SPT), cone penetrometer test (CPT), or shear wave velocity (Vs)-based subsurface data. This general approach largely neglects considerable uncertainties in ground shaking, as well as aleatory variabilities and epistemic uncertainties inherent to liquefaction susceptibility and triggering prediction. A more robust methodology introduced by Kramer and Mayfield (2007) is known as probabilistic liquefaction hazard analysis (PLHA), which integrates the full ground motion hazard space with probabilistic forms of liquefaction triggering models (e.g., Boulanger and Idriss 2014), resulting in the computation of FSL profiles with consistent return periods. Multiple PLHA computational platforms have been developed over the years, with the computational framework from Makdisi (2021) serving as the basis for a new Liquefaction Hazard Tool under development at the U.S. Geologic Survey (USGS).
Despite significant improvements in recent years to the availability of seismic hazard data and probabilistic triggering and effects models, the issue of incorporating uncertainty in characterizing liquefaction susceptibility remains a challenge. Most compositional susceptibility criteria (i.e., whether or not the soil exhibits sand-like behavior) currently in use are presented as deterministic bounds based on in-situ or laboratory test data; similarly, determination of soil saturation is often based on a single groundwater level from in-situ testing. As a result, the same types of binary decisions must be made in PLHA as in more conventional methods. With the expansion and availability of field and laboratory data pertaining to liquefaction through resources such as the Next Generation Liquefaction (NGL) project, there exists the potential for an improved set of susceptibility models for CPT, SPT, and Vs-based applications. Presented here is a brief discussion on how probabilistic susceptibility modeling can be accommodated in PLHA calculations, as well as how the use of multiple models can be leveraged within a logic tree to improve the representation of epistemic uncertainty in liquefaction hazard analysis.