A GLPI framework for gravelly soil liquefaction hazard assessment based on fuzzy mathematics

The Liquefaction Potential Index (LPI) is commonly used to evaluate liquefaction hazards in sandy soils, but is not applicable to gravelly soils. In addition, the choice of liquefaction prediction methods and the form of depth weighting functions may greatly affect the result of the LPI calculation. Meanwhile, the quartile thresholding method, for classifying the LPI values into hazard levels, suffers from boundary uncertainty, which can also affect the disaster assessment accuracy. Therefore, this study investigates the effects of different liquefaction prediction methods (simplified procedural method and its improvements, Bayesian network models) and different depth weighting functions (linear, logarithmic, exponential, hyperbolic, and power functions) on LPI results and disaster prediction accuracy. In addition, a new liquefaction potential index for gravelly soil (GLPI) evaluation is proposed, and the fuzzy mathematical method is used to handle the boundary uncertainty. The results show that the proposed GLPI hazard assessment method improves the prediction accuracy by 55 % over the LPI model. For the GLPI computational framework, the use of the Bayesian network method instead of the simplified procedure for liquefaction prediction was the most effective method for improving the accuracy of hazard assessment, with an improvement of 33 %; the depth weighting function has a relatively smaller effect, and the hyperbolic function has an improvement of 8 % compared with the linear function. In the hazard assessment based on GLPI values, fuzzy mathematics improved by 9 % over the quartile threshold method. Finally, the effectiveness of GLPI method was validated in new historical liquefied sites.