An enhanced hybrid approach for spatial distribution of seismic liquefaction characteristics by integrating physics-based simulation and machine learning

This study aims to propose an enhanced hybrid approach that combines physics-based simulation and machine learning to investigate the spatial distribution of seismic liquefaction characteristics. This innovative approach comprises two main components: Firstly, the physics-based frequency-wavenumber method is employed to construct the spatial-temporal field of ground motion in the study area, which provides ground motion quantities for assessing the liquefaction characteristic (e.g., liquefaction potential index) of the site. Subsequently, the seismic liquefaction parameters of the region are predicted using a machine learning (ML)-based SSA-XGBoost model. Due to the integration of physics-based simulation and machine learning techniques, which consider the effects of near-fault ground motion characteristics on seismic liquefaction, the proposed solution enables the evaluation of the spatial distribution of seismic liquefaction parameters under scenario earthquakes. In this study, the SSA-XGBoost model, constructed using the sparrow search algorithm (SSA) to automate and optimize the hyper-parameter tuning of the eXtreme gradient boosting (XGBoost), incorporates factors such as peak ground acceleration, magnitude scaling factor, ground water level, soil depth, vertical total overburden stress, vertical effective overburden stress, and fine content to evaluate their influence on liquefaction potential index. To demonstrate the effectiveness of the enhanced hybrid approach, the Jinnan district of Tianjin is taken as an example to evaluate liquefaction potential under various scenario earthquakes (M_w = 5.0, 5.5 and 6.0). The results show that the constructed SSA-XGBoost model has excellent predictive ability and is suitable for evaluating the liquefaction potential index of large-scale site soils. In the case of M_w 6.0 earthquake, most of the northern region of Jinnan district has the possibility of liquefaction, and some areas are seriously liquefied, and the liquefaction grade gradually decreases from the north to the south. These findings distinctly illustrate the spatial distribution of liquefaction characteristic parameters across the entire region, providing new insights and methods for similar studies and serving as a decision-making basis for the prevention and control of seismic liquefaction hazards.