• The estimated ground structures are validated using the available borelog data. • BNN + single station microtremor enables fast, uncertainty-aware Vs30 estimation. • Reduces site characterization time from days to seconds; scalable for microzonation. • Provide cost-effective seismic site response assessment using microtremors. Accurate estimation of the time-averaged shear wave velocity in the upper 30 m ( Vs 30 ) is essential for seismic microzonation, ground motion simulation, and earthquake risk assessment. Traditional microzonation methods often rely on dense seismic arrays and extensive geotechnical investigations, which are costly and time-consuming. This study presents a practical Bayesian Neural Network (BNN) framework that integrates horizontal-to-vertical spectral ratio (HVSR) curves from single-station microtremor measurements with site-specific physical constraints to achieve high-accuracy Vs 30 predictions. The framework further incorporates Monte Carlo (MC) Dropout to provide statistically calibrated uncertainty intervals alongside the predictions. A case study on Penang Island, Malaysia, demonstrates the applicability of the proposed approach. Although Penang has is traditionally considered aseismic, it is affected by far-field earthquake waves from neighboring regions. The model achieves an R 2 of 0.941, MAE of 7.394 m/s, and MAPE of 2.87% on independent tests. At the 95% confidence level, the prediction interval coverage probability (PICP) reaches 95.12% with a mean prediction interval width (MPIW) of 52.326 m/s, closely matching the nominal confidence level without excessive conservatism. These results indicate that the framework can capture the spatial distribution of Vs 30 with a relative error below 5% while providing reliable uncertainty quantification. Owing to its non-invasive nature, computational efficiency, and transferable architecture, this method offers a powerful and scalable tool for seismic microzonation, earthquake scenario modeling, and risk-based decision-making in regions worldwide.
Liu et al. (Sun,) studied this question.