JIAN-YE CHINGIkumasa Yoshida2024-11-192024-11-192025-010266352Xhttps://www.scopus.com/record/display.uri?eid=2-s2.0-85207354139&origin=resultslisthttps://scholars.lib.ntu.edu.tw/handle/123456789/723156Previous investigations have shown that for the modeling the soil spatial variability, the Gaussian process regression (GPR) provides a more plausible trend model than the linear combination of basis functions. However, the effectiveness of the conditional random (CRF) simulation based on the GPR trend model (denoted by the t-GPR kriging) has not been investigated. This study first addresses the high computational cost issue of the t-GPR kriging for realisic 3D problems by deriving the Kronecker-product algorithms. Then, this study further investigates the effectiveness of the t-GPR kriging in CRF simulation using real case studies. It is shown that with the Kronecker-product derivations, the computational time can be dramatically reduced such that the t-GPR kriging can conduct CRF simulation for full-scale 3D problems.falseGaussian process regressionProbabilistic site characterizationRandom fieldSpatial variabilityjournal article10.1016/j.compgeo.2024.1068622-s2.0-85207354139