Key points are not available for this paper at this time.
Due to the expanding scale of offshore wind farms, deriving the representative/characteristic profiles of soil properties required for turbine foundation design often relies on cone penetration tests (CPTs) and the associated CPT-based empirical statistical models. The small strain shear modulus G max of soils, a crucial design parameter, has seen a notable enhancement in both the quality and quantity of available data measured in situ thanks to the extensive use of seismic CPTs in offshore site investigation. However, the conventional statistical modelling approach (i.e., complete pooling) has not adequately exploited the increased quantity of G max data and may result in misleading probabilistic predictions and representative profiles for G max , as it ignores potential variations between both CPT locations and soil units in the data when calibrating a CPT-based empirical model. This paper introduces a Bayesian cross-classified hierarchical model that accommodates both sources of variation above in the CPT-based empirical relationship, allowing for a “high-resolution” prediction of representative G max profiles for each soil unit at each CPT location across a wind farm. Through a real-world example application, we demonstrate that the proposed hierarchical model offers significantly improved probabilistic predictions of G max across the entire site and hence more reasonable representative profiles compared to the complete pooling model. A leave-one-group-out analysis further indicates that the hierarchical model provides notably better prior predictions for the left-out G max data, with meaningful posterior updating and uncertainty reduction after observing the new data.
Feng et al. (Wed,) studied this question.