How many CPTs does it take to make a synthetic CPT?

The rapid growth in demand for offshore renewables has driven significant interest in linking near-surface geophysical and geotechnical data to develop integrated ground models. Using a machine learning-based workflow to link parameters we have predicted synthetic cone penetrometer (CPT) responses from geophysical data. We have sought to determine how many physical CPTs are required to train the machine learning algorithm to predict the magnitude and uncertainty of a synthetic CPT appropriately. This is a complex, multivariate analysis influenced by variable ground conditions and the types of infrastructure being installed. We provide insight into this project-critical issue, illustrating the influence of the depositional environment, loading history for a particular soil unit, and the foundation concept design on the quantity of CPT information required to obtain an adequate prediction of the ground conditions from geophysical inputs. While the results are not definitive, they provide valuable indicators for optimization of survey plans.