Abstract

Knowledge of seabed properties away from investigated locations is often required, for example, when geotechnical surveys are sparse or when the field layout changes between investigation and construction phases. In such cases, design lines that appropriately incorporate the uncertainty of the seabed properties must be defined to ensure reliable (yet not overly costly) design. This paper explores how two different approaches, traditional engineering judgement and advanced statistical methods, fare at quantifying the uncertainty of a real offshore site. This is achieved by ‘hiding’ different data and ‘scoring’ the predictive performance of the methods against a range of criteria. The work reveals that complex geological sites are significantly more challenging to represent than the stationary random fields often examined in research and suggests that more advanced approaches incorporating broader data sets are required to reduce uncertainty.

Abstract

The further development of offshore windfarm areas in various countries plays a key role in the transition of energy production towards renewable sources. As offshore windfarm areas tend to expand and the amount of ground truth data is limited, the estimation of geotechnical parameters at unknown locations integrating other site investigation data becomes a necessary tool. This is especially relevant for cost efficient area wide site characterization. Here, the proper integration and correlation of geotechnical and geophysical data is a key factor for reliable ground model building. This study investigates different prediction methods, while presenting a modelling framework which incorporates geological, geotechnical, and geophysical information to derive synthetic Cone Penetration Testing (CPT) profiles using offshore windfarm site investigation data from the German North Sea. We combine geological interpretation, CPT data and 2D ultra high-resolution seismic reflection data. The geophysical and geological information are used to guide geotechnical parameter prediction. Additionally, seismic horizons constrain the prediction as structural information. For evaluation, we test and compare several prediction techniques, with different level of complexity, from geostatistical methods to machine learning. Seismic attributes are used as auxiliary information to improve CPT parameter prediction. To validate the results, CPT parameters are predicted onto a representative 2D seismic line and a leave-one-out cross-validation (blindtest) is performed. Though all methods struggle to replicate local extremes, results indicate a reduction of prediction uncertainty when implementing seismic attributes.