Assessing prediction uncertainties in quantitative ground models for offshore wind farm areas

In the German North Sea, where sedimentary successions were shaped by repeated glacial–interglacial cycles, transgressive flooding, and fluvial reworking, the subsurface consists of very heterogenous deposits. This includes e.g. boulders and paleo-channels, which pose geotechnical risks and therefore challenges for the offshore infrastructure installation and planning. Therefore, offshore wind farms demand reliable information on the shallow subsurface to plan and design support structures safely and economically in those geologically complex, highly variable near‑surface deposits. To investigate the shallow subsurface, geotechnical and geophysical measurements are performed and later integrated in order to create ground models which serve as planning basis for wind farm layouts. Conventional ground models for this purpose remain largely qualitative, focusing on the interpretation of soil units as well as the identification of geohazards and lack information at unexplored locations. For more informed ground models, there is an emerging interest in uncertainty aware quantitative ground models which provide estimates of continuous or categorical information apart from directly explored locations. However, studies’ recommendations on different methodologies as well as their uncertainties and their further use in ground modeling and wind farm planning are missing, leaving the question unanswered how reliable the predicted data is and if it can be used as replacement for real measurements. This cumulative doctoral thesis aims to assess uncertainties in predictive models and how they support quantitative ground modeling as well as their impact on the following design and decision process. In this interdisciplinary work, the utilized datasets combine ultra-high resolution two‑dimensional multichannel seismic data, stratigraphic interpretations, acoustic impedances derived from post‑stack inversion, and cone penetration tests from geotechnical campaigns. In this setting, seismic data delivers continuous information on subsurface layering, while in-situ tests provide sparse pointwise depth profiles of soil response. Bringing these sources of information together is the key to models that are both geologically meaningful and quantitatively usable for engineering. On this basis several predictive methods are tested to estimate data at unexplored positions, while implementing and improving data integration and prediction workflows. First, cone penetration testing profiles are predicted and compared to measured profiles using various methods to assess their performance and related uncertainty. Given the data density and attribute quality available in the case studies, different methods perform comparable to each other at larger scales, and no single algorithm dominates across all conditions. Incorporating acoustic impedance and related attributes generally reduces uncertainty in predicted geotechnical parameters and helps reproduce lateral variations within soil units. Probabilistic predictions are carried further into engineering design to demonstrate how predicted geotechnical profiles can be applied in the design of offshore foundations and to assess if they can be used equivalently to measured data when uncertainty is handled explicitly. This gives insights into future applicability of predicted data as potential replacement for measured data. Beyond continuous parameters, this thesis addresses the lateral uncertainty of discrete subsurface heterogeneities by quantifying probabilities of buried channels present between 2D seismic lines and their uncertainty in terms of potential location. Channels are common in glaciogenic settings and can complicate pile installation as well as stability due to property contrasts, affecting wind farm layout planning. The predictions indicate that fidelity declines as line spacing increases, highlighting the strong control of survey design while giving insights into future survey planning. In summary, this thesis contributes to a better understanding of predictive models and their use in quantitative ground models for offshore wind farms. It clarifies their implications for development, especially engineering design, fills key knowledge gaps, and offers insights into model uncertainty. This work builds a foundation for future research in a young field while complementing standard qualitative interpretation. Together, these contributions show that quantitative ground models add practical value to offshore wind development, by delivering continuous and categorical estimates of engineering-relevant information at locations without direct measurements and by providing location- and depth-specific uncertainty that can be carried into deterministic or reliability-based design. These capabilities can be used to optimize site survey planning, wind farm layout, and foundation design. By moving ground models from qualitative to quantitative ones, this work provides a foundation for safer, more efficient offshore wind development in the framework of geologically complex areas.