Upscaling 1 500 000 synthetic CPTs to voxel CPT models of offshore sites

In 2020, the Dutch government published 1 500 000 synthetic CPT profiles for use in development of the Hollandse Kust (west) Wind Farm Zone, offshore Netherlands. The scale of this approach was novel at that time and possibly first-ever. The synthetic CPT profiles were derived from a training data set of 122 actual CPTs and ultra-high-resolution (UHR) seismic reflection traces, using machine learning by a convolutional neural network. The synthetic CPT profiles were limited to positions along the 162 UHR survey track lines (2D) and were limited to cone resistance to a depth of 50 m below seafloor. The UHR track lines were spaced at about 400 m. This paper explores upscaling the synthetic CPT approach to voxel (3D) models and adding (S)CPT-based parameters such as shear modulus at small strain G_max . Future added-value is expected from post-2020 improvements seen in seismic reflection data resolution, attribute extraction and neural network architecture.