Cross‐Period Ground Motion Spatial Correlation With Path and Site Effects: Bayesian Inference and Risk Implications

ABSTRACT Accurately characterizing cross‐period spatial correlation of ground motion intensity measures is essential for regional earthquake risk assessment. We propose a model that represents multi‐period within‐event residuals via a principal‐component decomposition and, for each latent component, specifies a spatial kernel that separates path and site influences through epicenter‐to‐station azimuth differences and contrasts. Parameters are inferred in a Bayesian framework, yielding full posterior distributions, and thus uncertainty quantification. Posterior predictive checks and log pointwise predictive density (LPPD) demonstrate consistent improvements over isotropic baselines. Two synthetic regional case studies illustrate risk implications: in a near‐epicenter setting with spatially heterogeneous , traditional isotropic models over‐aggregate local extremes and produce heavier‐tailed scenario loss‐exceedance curves, whereas the proposed model reduces tail inflation; in a far‐epicenter setting with relatively uniform , the exceedance curves are nearly indistinguishable, with minor corrections to the baseline's slight underestimation. By jointly capturing cross‐period structure and distinct path/site effects, the formulation mitigates these biases and delivers improved portfolio‐loss estimates while remaining compatible with standard scenario‐based and probabilistic regional portfolio‐loss workflows.