A condition-based hybrid model for estimating liquefaction-induced lateral spreading

In the event of soil liquefaction, the subjection of the ground to topographic elevation differences, such as in free field or gently sloping ground conditions, leads to lateral spreading, which is one of the most severe forms of liquefaction-induced ground deformation. This study investigates the predictive capabilities of the two widely adopted empirical models by Zhang et al. (2004) and Youd et al. (2002) for predicting liquefaction-induced lateral spreading. Utilizing high-resolution aerial LiDAR measurements and 83 Cone Penetration Test (CPT) data from the 2011 Christchurch earthquake, the study identifies significant predictive limitations in both models when applied outside their optimal conditions. The Zhang model often overpredicts steep slopes, while the Youd model tends to under-predict massive displacements in deep, clean sand deposits. To minimize these mispredictions, tiered hybrid model-selection scheme was developed using a Decision Tree machine learning algorithm. This condition-based selection approach uses free face ratio (W), liquefiable layer thickness (T15), and CPT tip resistance (qc) to alternate between the two methodologies. The findings show that combining the two methods yields better results than using only one method within study limitations.