Enhancing liquefaction analysis with data-driven techniques : a case study of the coastal area in Bengkulu city, Indonesia

Abstract

Bengkulu City of Indonesia located along the active Sumatra Subduction Zone, is highly vulnerable to liquefaction due to its extensive alluvial deposits. This study enhances liquefaction hazard assessment by integrating data-driven techniques with empirical models. A Deep Neural Network (DNN) was developed to predict shear wave velocity (Vs) profiles using spatial features, achieving high accuracy. The predicted Vs profiles supported the development of Vs30 and site classification maps, as well as a 3D subsurface model that was validated against borehole data. Seismic input scenarios included both deterministic events, such as the 2007 Mw 8.4 Bengkulu-Mentawai earthquake, and probabilistic seismic hazard analysis with a 10% probability of exceedance in 50 years. Peak ground acceleration (PGA) levels ranging from 0.1g to 0.6g were also considered. Liquefaction potential was evaluated using the Liquefaction Potential Index (LPI), which identified severe hazard zones in southern Ratu Agung and northern Gading Cempaka, with wider impacts observed under the probabilistic scenario. Ground settlement and lateral displacement were estimated based on volumetric and shear strain, respectively. The results showed that hazard severity and affected area increased with seismic loading up to a PGA of 0.4g, beyond which the response plateaued due to the inherent stiffness of the soil layers. This integrated framework offers a robust basis for hazard mapping and urban planning in data-scarce regions, supporting effective land-use regulation and disaster mitigation.