Kriging-Based Conditional Random Field for Regional Liquefaction Potential Mapping Considering Statistical Uncertainty

Regional liquefaction potential assessment usually requires spatial interpolation based on probabilistic models (e.g., conditional random field, CRF). Accuracy of spatial interpolation relies highly on the number of testing data and stochastic model parameters. Since testing data is often insufficient, statistical uncertainty on model parameters is inevitable. Moreover, efficient CRF simulation across a large region is also of practical importance in engineering applications. In this paper, regional probabilistic characterization of the liquefaction severity index (LSI) calculated from cone penetration test (CPT)-based simplified procedure (SP) is presented based on Kriging-based CRF. With the proposed approach, the spatial variability and statistical uncertainty are, explicitly and simultaneously, considered through the ancestor sampling method (ASM) under a Bayesian framework. The proposed method is illustrated and validated using real CPT data. Results show that the proposed method provides reasonable spatial interpolation results of LSI values based on a limited number of CPT data, and the spatial variability and statistical uncertainty are taken into account in a quantifiable and rational way without compromising the computational efficiency of CRF simulation. Ignoring the statistical uncertainty might lead to underestimation of the prediction uncertainty.

Keywords: Bayesian, conditional random field, cone penetration test, liquefaction potential