Hierarchical Bayesian modelling of quasi-region-specific soil porosity

Abstract

Determination of soil parameters is considered crucial to the safety of marine geotechnical engineering. Due to the limited number of borehole samples that are usually available, the process of parameter estimation involves great uncertainty. A conventional practical approach is to conduct regression modelling of the region-specific data to develop an empirical equation, such as one that relates acoustic impedance (I) to porosity (n). However, most of the regions have insufficient measured data for such modelling. This paper proposed a porosity prediction method based on hierarchical Bayesian modelling (HBM), which aims to predict n in data-limited regions. A quasi-region-specific I–n relationship is established by simultaneously considering the measured data from other regions and the target region. Moreover, eight published I–n datasets from different regions are compiled to verify the proposed method. The comparative advantages and improvements in porosity estimates are discussed by comparison with two alternative models. From the extracted results, it is confirmed that the proposed method can predict n for specific regions with small datasets and effectively reduce the predictive uncertainty. Ten region-specific datapoints are sufficient to achieve a balance between the measurement cost and the predictive reliability.