Extracting dispersion characteristics of the subsurface under a railway line from passively recorded DAS data

Two developments of this century have allowed for a greatly increased potential for monitoring of the near surface in geotechnical applications. First, Distributed Acoustic Sensing (DAS) allows glass fibre cables attached to an interrogator to be utilised for sensing seismic vibrations with dense spatial sampling. This allows for robust, permanent recording installations that cover large spreads. 

Second, the theory of seismic interferometry shows that with the use of ambient noise, under certain conditions, any recorded trace can be turned into a virtual-source position. In the most common application, the same event, recorded at multiple positions that share a travelpath, must be crosscorrelated to obtain a virtual shot. For the full response, this must be repeated for all sources on a surface effectively surrounding the medium of interest and the results stacked.

The combination of DAS and seismic interferometry for real-time monitoring require large amounts of passive data to be collected. This does mean that subsequent processing workflows have to be adapted according to computational capabilities. Even for relatively simple workflows, high-performance computing concepts must be applied to keep processing speed aligned with data collection.

Given the large amounts of recorded data, it becomes tempting to adopt the mindset that better results are obtained by simply stacking more data. However, for seismic interferometry, a proper selection of useful noise is essential in retrieving good results.

One of the proposed monitoring applications of seismic interferometry on DAS data is for monitoring the subsurface under railway lines. The shear modulus is used to monitor the strength of the soil. As such, surface-wave analysis methods are the seismic investigation method of choice. The advantage of monitoring close to active railway lines is that passing trains provide strong noise sources. When a train passes directly past the sensors, the wavefield is very complex, but waves generated by the train propagate both backwards and ahead. These waves can be used for seismic interferometry. As different trains generate different source spectra for the wavefield, multiple different trains must be included in the data and stacked after seismic interferometry to obtain a broader frequency band.

The dataset that we use is from an 8-km-long straight section of DAS cable along a rail line between Rotterdam and Delft in the Netherlands. We estimate the passage of a train along the DAS line with the envelope of the energy of the data. Then, we select windows ahead and behind the train that capture the generated waves. As the location of the train is known, we can use the trace closest to the train as a master trace and only the causal parts of the result are summed with the total stack. Finally, the dispersion spectrum is computed from the virtual shots to extract dispersion information along the line.

Together with intermediate results, we show that consideration of the noise sources that are present and how to utilise these leads to improved results. This requires more preprocessing but also finally decreases the amount of data that must be crosscorrelated.