Geospatial Site Terms for Central and Eastern U.S. Ground-Motion Models Using Physiographic Province and Sediment Thickness

Abstract

Most recent Central and Eastern United States (CEUS) ground-motion models (GMMs) rely on the time-averaged shear-wave velocity in the upper 30 m of the subsurface (VS30) as the primary site parameter. Site terms in GMMs are known to reduce site-to-site residuals (δS2Sj), which quantify the repeatable, systematic deviations between observations and predictions at a station. However, the inputs for site terms are often unavailable both at ground-motion recording stations and at locations where GMMs are used to predict shaking levels. This highlights the need for spatially continuous site terms. We use mixed-effects regression to obtain δS2Sj from ground-motion predictions using the Next Generation Attenuation-East (NGA-East) GMMs for the CEUS and two different ground-motion datasets. These δS2Sj values show strong regional trends consistent with large geologic structures, such as the Mississippi Embayment, the Superior Uplands, and sections of the Coastal Plain. In this study, we evaluate a range of continuous and categorical site variables for use in GMMs to capture both local and regional site effects. These variables serve as proxies for sediment stiffness, shallow velocity structure, sediment/bedrock impedance contrasts, basin effects, and topographic effects. From these variables, we develop a new two-parameter linear amplification model using (1) a categorical variable based on physiographic provinces (Fenneman and Johnson, 1946) and (2) a continuous sediment thickness variable from a high-resolution regional dataset (Boyd et al., 2024) and a global dataset (Pelletier et al., 2016). The physiographic province parameter captures large-scale trends in amplification that effectively regionalize the site term, whereas sediment thickness accounts for local variation. When tested on ground motions from M>4 earthquakes, this model reduced the standard deviation of δS2Sj by an average of 23.4% across all periods and significantly decreased spatial trends in residuals relative to the current NGA-East amplification model (Stewart et al., 2020).