Regionalized Bayesian site amplification models for Türkiye

Accurate representation of site amplification is essential for seismic hazard analysis, particularly in regions with complex geological structure. This study develops a Bayesian hierarchical site amplification model that captures nonlinear, period-dependent, and regionally variable behavior using an extensive strong-motion dataset from Türkiye. The model employs a piecewise VS30 functional form derived from nonparametric scaling and clustering, and incorporates nonlinear scaling with reference rock motion intensity (PSArock). Regional variability is introduced through random-slope adjustments, enabling spatial differences in amplification to be represented while maintaining statistical stability through partial pooling. Results indicate pronounced nonlinear and regional variability for soft soils and reduced regional differences for stiff sites (VS30 > 600 m/s). A key observation is a systematic amplification peak within the 400–550 m/s velocity range, where the model exhibits consistent underprediction across periods. Because this feature persisted even for stations located on flat basin interiors, a detailed investigation of stations in this VS30 interval was conducted and is presented in the study. Mapping and site-specific examination reveal that many of these stations are located near slope breaks, basin edges, or transitional geological settings, suggesting that 2D/3D wave-propagation or impedance-transition effects—rather than VS30 alone—may drive the increase in the amplification. Additionally, residual underprediction persists at high-VS30 rock sites, highlighting the need for improved empirical characterization of stiff-soil and rock conditions. Overall, the proposed Bayesian framework provides a flexible, data-driven, and uncertainty-aware approach for modeling site amplification, with direct applications in probabilistic seismic hazard analysis and performance-based earthquake engineering.