Integrated Ground Modelling from a Geotechnical Perspective-Paper

Paper

Pdf Summary

The paper presents a workflow for creating quantitative integrated ground models that combine geotechnical, geological and geophysical data to estimate soil properties and their uncertainty across offshore development sites. Using public data from the Holland Kust Zuid wind farm in the North Sea, the authors predict cone penetration test (CPT) tip resistance, qc, beyond borehole locations.<br /><br />Two geostatistical approaches use seismic interpretations to define soil-unit boundaries and guide interpolation. The first assigns a constant normalized tip resistance to each layer; the second interpolates down-sampled qc profiles. The models are assessed through leave-one-CPT-out cross-validation, which tests predictions at locations excluded from calibration and helps reveal overfitting. Although the second approach fits the training data more closely, its accuracy does not carry over to unseen data. Both approaches show substantial uncertainty and a tendency to overpredict, indicating that model simplification—not just spatial variability—is an important source of error.<br /><br />The study then uses predicted CPT profiles and their uncertainty to estimate jack-up spudcan penetration across the site. The results illustrate how an integrated model can support preliminary foundation design and help identify areas where further investigation may be valuable.<br /><br />The authors emphasize that the simple models are demonstrations of an evaluation method, not definitive prediction tools. More advanced interpretation and seismic inversion could improve estimates, while independent validation is essential to avoid overfitting. Overall, quantitative ground models offer a structured way to integrate diverse site data, map soil properties and uncertainties, and inform early engineering decisions.

Keywords

quantitative integrated ground models

geotechnical data

geological data

geophysical data

cone penetration test

CPT tip resistance

seismic interpretation

leave-one-out cross-validation

uncertainty estimation

jack-up spudcan penetration