Next‐Generation Probabilistic Liquefaction Model Building at the Regional Scale

Free‐field liquefaction manifestations (i.e., ejecta, ground cracking, or settlement occurring away from buildings or slopes) can have a significant impact on infrastructure systems and lifelines on a regional scale. Prior cone penetration test (CPT)‐based regional hazard analysis procedures generally rely on applying existing manifestation models, which are not directly developed with regional data. In this paper, we present a methodology for building probabilistic liquefaction free‐field manifestation models that: i) uses modern CPT database(s) with thousands of case histories; ii) predicts the probability of surface manifestation of a given severity (instead of just “yes” or “no” manifestation); and iii) incorporates a spatial correlation component to account for densely spaced case histories. We also demonstrate how cross validation and proper scoring rules can be used to evaluate the probabilistic predictions of competing models (e.g., comparing different ground motion intensity measures). We show that a model with the spatial correlation component performs significantly better than an otherwise identical model without the spatial component () and thus should be considered when developing regional models. We also highlight opportunities for future research directions based on some surprising trends observed in this work.