What Controls Variability in DAS Earthquake Observations? Implications for Ground-Motion Models

Distributed Acoustic Sensing (DAS) provides dense, meter-scale ground-motion measurements along fiber-optic cables. However, developing ground-motion models (GMMs) from DAS data is challenging because observations are controlled by DAS-specific factors such as cable coupling, orientation, and channel correlation. In this study, we present the first regional, partially non-ergodic DAS-based GMM that explicitly identifies and quantifies cable-related contributions to ground-motion variability. We analyze strain-rate data from a 400-channel DAS array at the Milun campus in Hualien City, Taiwan, compiling peak strain rates and Fourier amplitudes (0.1–10 Hz) from 77 regional earthquakes (3<M<7, 45<R<170 km). Building on classical seismometer-based GMMs, we extend the variability framework to account for (1) cable coupling influenced by installation and environment types, (2) cable orientation, and (3) channel correlation inherent to DAS measurement principles and array geometry. Channel correlation is modeled using Matérn kernels parameterized by along-fiber and spatial proximity distances. The resulting DAS-based GMM shows magnitude-distance scaling comparable to classical models, while decomposing variability into physically interpretable components. Cable coupling emerges as a dominant broadband source of within-event variability, whereas orientation effects capture repeatable, frequency-dependent earthquake source radiation patterns. Modeling channel correlation significantly reduces channel-related standard deviations, demonstrating that treating DAS channels as independent observations biases uncertainty estimates. Overall, our results show that DAS-derived ground motions require a fundamentally different variability framework than that of classical GMMs, highlighting the importance of deployment metadata and correlation modeling. This approach provides a statistical and physical foundation for next-generation seismic hazard assessments using dense fiber-optic sensing.