A novel machine learning approach for interpolating seismic velocity and electrical resistivity models for early-stage soil-rock assessment

Identifying near-surface lithological conditions is crucial for investigations such as building foundations, engineering projects, and groundwater resources, among others. Geotechnical drilling has limitations in collecting data from precise locations. Therefore, combining two geophysical techniques with machine learning (ML) algorithms for subsurface characterization yields better outcomes. Consequently, this novel approach was employed for the interpolation of SRT–ERT models and to develop the relationships between them for the geological terrain of the Kabota-Tawau area of Sabah, Malaysia. Two survey lines were established within a geologically favorable area of interest to evaluate and enhance the understanding of the study area’s near-surface lithologic units. The resistivity and seismic P-wave velocity (Vp) techniques were utilized to acquire the field data, after which the resulting models were interpolated. To improve subsurface lithological differentiation, the K-means clustering and simple linear regression algorithms were utilized to analyze the interpolated resistivity and Vp datasets. Via this approach, the area’s subsurface lithologies were identified as the clayey silt topsoil, along with weathered units characterized by stiff to very stiff clayey/silty material, very stiff to hard clayey/silty material, and hard to very hard clayey/silty unit. The developed velocity-resistivity empirical relation exhibits a practical prediction success rate exceeding 86% with high positive correlations, making it statistically significant and accurate in characterizing underlying geological variations. These findings underscore the efficacy of both ML approaches in accurately identifying distinct subsurface geological variations.