A combined Bayesian with ensemble learning method for the determination of geological cross-sections using multi-source geophysical and borehole data

An accurate determination of subsurface stratigraphy is crucial for the design and construction of underground infrastructure. However, the uncertainties involved in geological cross-sections delineated using sparse and limited borehole data can be significant. Seismic wave-based non-invasive geophysical methods excel in identifying stratigraphic patterns and offer valuable insights for geological profiling. Thus, this study proposes a combined Bayesian with the ensemble learning method to determine geological cross-sections from multi-source geophysical and borehole data. Calibration of geophysical thresholds for distinct geological interfaces using Markov Chain Monte Carlo (MCMC) simulation based on borehole data enables the automated transformation of multiple geophysical profiles into geological cross-sections. Subsequently, a robust geological cross-section is developed using an ensemble learning method. The effectiveness of this method was validated by the data from a site in Singapore, where three types of seismic geophysical surveys and five boreholes were utilized. The results show that the MCMC-based Bayesian calibration can reduce significantly the uncertainty in geophysical threshold estimation, achieving about 80 % to 90 % reductions in most posterior standard deviations. The ensembled geological cross-section presents fused stratigraphic features with more reasonable variations in geological interfaces and formation thicknesses. Compared to training and validation boreholes, it achieves accuracies of 92.40 % and 91.23 %, respectively, with a low average information entropy of 0.04, indicating effective uncertainty reduction. These results have demonstrated the capability of the proposed Bayesian method to determine geological cross-sections from multi-source geodata with enhanced reliability and reduced uncertainty.