Regression Models For Evaluating Liquefaction Probability

TLDR

The study develops statistical models to estimate liquefaction probability as a function of earthquake load and soil resistance parameters. Using binary regression on 278 case records, the authors derive four models—two based on cyclic stress ratio and two on magnitude‑distance functions—each employing corrected SPT N1(60) as the resistance metric. The models are recommended for practical use and are shown to compare favorably with existing liquefaction analysis methods.

Abstract

Statistical models are developed to express the probability of liquefaction as a function of earthquake load and soil resistance parameters. Results are obtained based on a catalog compiled for this study, which consists of 278 cases of liquefaction/nonliquefaction occurrences. A method of binary regression is used to derive four models recommended for use in assessment of liquefaction probability. Two of them use the cyclic stress ratio (Seed and Idriss 1971) as the earthquake load parameter; the other two use, as load parameters, explicit functions of earthquake magnitude and distance, similar to the function proposed by Davis and Berrill (1981). All four models measure liquefaction resistance through the corrected/normalized SPT (N1)60 value. Recommendations for practical implementation are provided, and comparisons are made with other methods of liquefaction analysis.