Geospatial data-driven assessment of earthquake-induced liquefaction impact mapping using classifier and cluster ensembles

A 5.4 M_L earthquake occurred on November 15, 2017, in Pohang, South Korea. This earthquake was the second largest recorded earthquake in South Korea and had detrimental effects on the ground and infrastructure. Among all the ground deformations, hundreds of liquefaction-induced sand boils and ground failures observed near the epicenter were major issues. However, whether subsurface characteristics and liquefaction vulnerability indices trigger regional liquefaction manifestations and how local liquefaction occurs as a consequence remains elusive. In this study, we present a novel data-driven model for the analysis of site-specific liquefaction triggering that considers the spatial uncertainties of principal liquefaction vulnerability indices. This is achieved by establishing an advanced artificial intelligence technology that assembles optimization-oriented, supervised, and unsupervised machine-learning models. The phased decision-making process could develop unified liquefaction hazard zonation based on the clustering ensemble methodology and help identify feasible liquefaction impact mapping procedures via the optimized classification of their performance evaluation with liquefaction inventory. The alternative three-phase approach, depending on the feasibility of geo-data and geospatial modeling, consists of three zonation methods (macro-, micro-, and nano-zonation) based on a 3D grid network, which assigns the best-fitting machine-learning model. The resulting liquefaction impact map, which has a high resolution and is assigned nano-zonation-based clustered liquefaction indices, can assist in site-specific decision-making to zonate liquefaction-induced ground displacement.