Utilizing train-induced prelude and coda wavefields from DAS for phase velocity dispersion imaging

The deployment of telecommunication fiber-optic cables along railway tracks facilitates the recording of strain-rate wavefields for subsurface characterization. However, the broadband seismic noise generated by the train itself often masks site-specific signals, complicating the extraction of local geological characteristics. To address this, recent research has utilized the ambient wavefield—recorded both immediately before (prelude) and after (coda) train passage—to compute cross-correlograms and derive phase velocity dispersion curves along the fiber.

In this study, we conduct a systematic comparative analysis of train-induced prelude and coda wavefields to assess the consistency of synthesized dispersion curves. The dataset consists of 24 h of DAS recordings acquired over a 1,250 m fiber, with a gauge length of 20 m, channel spacing of 5 m, and a sampling rate of 500 Hz. Homogeneous segments along the fiber are first identified by examining resonant site frequencies in the 1–20 Hz range using autocorrelation functions. Cross-correlograms are subsequently computed between a selected virtual source and neighboring channels within these homogeneous sections to reconstruct phase velocity dispersion curves.

Preliminary results indicate that comparing resonance frequency maps for the prelude and coda of the same train reveals channel-specific discrepancies of 2–3 Hz. Furthermore, the lateral stability of these resonance frequencies fluctuates depending on the wavefield phase analyzed. Even in segments identified as laterally homogeneous, the resulting correlograms vary significantly; wave propagation is generally clearer in the prelude, whereas the coda more frequently excites higher modes. Interestingly, the fundamental modes on the dispersion curves remain consistent across both phases, even when the underlying correlograms differ substantially. These variations are highly dependent on the train type. For instance, trains with higher RMS amplitude values produce more laterally stable resonance maps—particularly within the coda—though a discrepancy in absolute values between the prelude and coda persists. These trains also yield more consistent correlograms and dispersion curves along the fiber. These findings highlight the sensitivity of the results to the specific source signal and suggest that combining the prelude and coda data from multiple trains could significantly improve the stability and reliability of the seismic imaging of the sites along the fiber.