Seismic emissions from a passing train: turning ambient noise into a controlled source

Train traffic is a powerful source of seismic vibrations. Recent studies have shown that trains illuminate geological structures both at the crustal and the geotechnical scale. Existing works have been able to reconstruct approximately the spectral characteristics of the wavefield emitted by a passing train. In this work, we show that we can recover information on the train itself with high accuracy by looking only at the seismic recordings.

We record passing trains with seismic accelerometers less than 2 meters away from the track. We can isolate the signal emitted by each wheel, and thus reconstruct the trajectory of the train. This trajectory reconstruction is performed using a non-linear waveform inversion algorithm involving the varying train speed, the spacing between the wheels and an apparent wavelet emitted when the wheel hits close to the seismic sensor. After low-pass filtering the data below 15 Hz for passenger trains passing at around 100 km/h, we obtain harmonious waveforms suitable for our inversion technique. Especially, we are able to pick each wheel from the raw trace, which allows for a robust initial model avoiding local minima trapping during the non-linear inversion. The estimated parameters are minimally influenced by seismic wave propagation speeds, because the closest sleeper dominates the signal in this frequency band.

These results suggest that train traffic is a repeatable seismic source that can be can be characterized with good accuracy.  By having a better information about the source process, it might be possible to extract more information from the noise recordings, and thus gain in resolution in the imaging of the near surface. Especially, we expect enhanced repeatability of Rayleigh velocities measurements which is important for subsurface monitoring. Further, this also allows for railway traffic monitoring as trains can be identified and their speed measured as they cross seismic arrays.