Characterizing the spatial variability of marine soil properties with site-specific sparse data using a Bayesian data fusion approach

Sparse site-specific test data complicates soil spatial variability characterization, resulting in substantial statistical uncertainty in model parameters. Rare studies explicitly address this uncertainty, a more pronounced issue in offshore wind engineering due to large and multi-source yet sparse and non-co-located data. This study proposes a Bayesian conditional co-simulation (BCCS) method for spatial variability characterization of marine soils in offshore wind farms. Utilizing primary (e.g., internal friction angle, ϕ) and secondary (e.g., standard penetration test, SPT N values) variable measurements, the BCCS method employs a Bayesian framework to infer variogram model parameters and to quantify statistical uncertainty. Notably, the statistical uncertainty is considered in subsequent conditional co-simulation of the primary variable. The proposed approach is applied to characterizing the spatial variability of ϕ based on measurements of ϕ and SPT N in a sand layer in an offshore wind farm. The proposed methodology effectively characterizes marine soil spatial variability using sparse non-co-located primary and secondary datasets. Neglecting statistical uncertainty in variogram model parameters underestimates the prediction uncertainty for the primary variable. This can be mitigated by incorporating an informative prior distribution, assimilating secondary data, and increasing primary data volume. Efficacy depends on existing knowledge and data quality.