Comparison of Approaches to Data-Driven Site Characterization in an Offshore Carbonate Soil Environment in North Western Australia

This study compares the performance of three novel data-driven site characterization methods to current engineering practice for modeling geotechnical variability, in terms of net tip pressure from piezocone penetration tests at an offshore site. Comparison between the methods was conducted using cross-validation schemes and conventional predictive metrics, along with qualitative assessments of the practical utility of each method for engineering applications. Conventional metrics provide insights into key factors such as model practicality and uncertainty quantification. The results indicate that traditional engineering practice, which relies on integration of site investigation data and expert judgement, performs comparably well for the geologically complex site considered in this study. It is concluded that, although data-driven site characterization methods show promise for enhancing geotechnical site characterization, further research is needed to address gaps such as the integration of multisource data and improved uncertainty quantification.