Investigating the origin of azimuthal site-response variability using KiK-net data: a twofold spectral ratio approach

Abstract In this study, we show that the widely observed azimuthal dependence of the surface-to-borehole spectral ratios (SBSRs) is governed primarily by local, site-specific complexities rather than regional path-specific effects, resolving a fundamental ambiguity in site response analysis. This ambiguity arises from the significant event-to-event variability seen in empirical data, which contradicts the stable, time-invariant theoretical transfer function predicted by one-dimensional (1D) site response analysis and introduces considerable epistemic uncertainty into ground motion models (GMMs). A key challenge has been the indeterminate origin of this variance whether attributable to the propagation path or the local site. To systematically decompose these contributing effects, we developed an analytical framework based on the twofold spectral ratio (TSR) method. Applying this framework to a large dataset from 293 nearby vertical array station pairs from Japan's Kiban–Kyoshin network (KiK-net), our analysis was designed to isolate and nullify the common path-effect term, enabling a direct, quantitative evaluation of site-specific variability. The analysis yielded two empirical results. First, the path-effect term isolated by the TSR showed high correlation between surface and borehole records, with the median correlation coefficient remaining above 0.5 across the entire 1–10 Hz frequency band, validating the successful separation of this component. Second, the residual event-to-event SBSR fluctuations between adjacent stations exhibited a near-zero spatial correlation across the entire frequency range, which is inconsistent with the hypothesis of a shared, regional cause. This reclassification of variability from a random uncertainty to a deterministic, site-specific attribute has direct implications for advancing seismic hazard assessment. Our dual-dataset approach provides a framework to validate the decomposition of site-specific variability into its aleatory and epistemic components. This provides the physical basis for developing non-ergodic GMMs, which can reduce overall uncertainty by explicitly modeling the predictable epistemic component, facilitating more sophisticated, physics-based site characterization.