The “Site Recognition Challenge” in Data-Driven Site Characterization

One distinctive feature of geotechnical engineering is site uniqueness or site-specificity. However, there is no data-driven method to quantify site uniqueness. The corollary is that it is not possible to identify “similar” sites from big indirect data (BID) automatically and no method to combine sparse site-specific data with big indirect data to produce a quasi-site-specific model that is less biased compared to a generic model and less imprecise compared to a site-specific model. This “site recognition” challenge is difficult because site-specific data is MUSIC-X (Multivariate, Uncertain and Unique, Sparse, Incomplete, and potentially Corrupted with “X” denoting the spatial/temporal variability). This paper presents the application of 4 methods (hybridization, hierarchical Bayesian model, record similarity method, site similarity method) to construct a quasi-site-specific transformation model between the undrained shear strength and normalized cone tip resistance. The similarity methods are “explainable”, because a list of “similar” sites can be generated explicitly for inspection.