Landstreamer: Towable MASW Array for Rail Earthworks

• Landstreamer is a towable seismic array for MASW, integrating plate-mounted geophones and towing logistics for rapid, efficient data acquisition.
• The system reduces deployment time and crew requirements compared to conventional planted geophone spreads, while delivering comparable dispersion data.
• Coupling with Bayesian inversion quantifies uncertainty in shear-wave velocity profiles, supporting informed decision-making in railway infrastructure management.

Searching arXiv for the specified paper and closely related Landstreamer/MASW railway earthworks work. arxiv_search query="(Burzawa et al., 22 Jul 2025) Landstreamer railway earthworks MASW Bayesian inference" max_results=5 Landstreamer denotes a towed seismic array used for Multichannel Analysis of Surface Waves (MASW) in the mechanical evaluation of railway earthworks. In the reported railway application, the Landstreamer (LS) was designed to obtain shear-wave velocity ($V_s$) profiles for detecting Low Velocity Layers (LVLs) in disturbed railway-earthwork zones while reducing deployment time relative to a conventional planted-geophone spread. The same study couples LS-based acquisition with Bayesian inversion of dispersion data so that $V_s$ models are accompanied by quantified uncertainty, with the stated aim of supporting reliable decision-making in infrastructure management (Burzawa et al., 22 Jul 2025).

1. Concept and hardware configuration

In the reported configuration, the LS consists of 48 vertical-component geophones with low-cut frequency 4.5 Hz, each geophone mounted on a 0.25 kg steel plate. Receiver spacing is $\Delta x = 0.5$ m, and the total active cable length is $\simeq 23.5$ m. The geophones are wired via a rugged, inelastic strap carrying both signals and power along the array. The tow mechanism is fundamental to the system definition: the array is laid directly on the ballast and towed behind a small rail trolley, so no spiking or planting is required (Burzawa et al., 22 Jul 2025).

The same study frames the LS by direct comparison with a conventional spread composed of 96 geophones (14 Hz) planted by steel spikes, with $\Delta x = 0.25$ m and total length 23.75 m. That conventional arrangement requires manual insertion and removal at each roll position. In this comparison, the technical distinction is therefore not only receiver count or geophone frequency response, but also the replacement of repeated planting operations by a towable receiver line.

A plausible implication is that “Landstreamer” in this context names both a hardware architecture and a field logistics strategy: plate-mounted geophones, strap-based signal/power distribution, and towing-based repositioning are treated as a single integrated acquisition concept rather than as separable components.

2. Deployment on ballast and acquisition workflow

The LS was deployed both on the cess (shoulder) and on the track. On the cess, geophone coupling is reported as excellent because the steel plates sit directly on ballast stones. On the track, the array is placed directly on the ballast between sleepers, and small rubber pads can be added under each plate to prevent slipping. In roll-along mode, the LS is towed forward by 6 m along the track axis after each shot; this is explicitly contrasted with a stationary spread, where all receivers remain fixed while the source is moved (Burzawa et al., 22 Jul 2025).

Three acquisition modes were reported. First, the conventional cess survey was conducted at two positions, P0 (reference) and P1 (anomaly), using 96 geophones, $\Delta x = 0.25$ m, and shots at both ends (direct and reverse) with a sledgehammer (1.25 kg) on a metal plate. Second, the LS was tested on the cess at the same positions, P0 and P1, using $3\times$ hand-hammer/nylon-strike-plate impacts at each end. Third, the LS was used on the track for continuous profiling along 300 m, with $\Delta_{\mathrm{roll}} = 6$ m and a single direct shot per position produced by an automated 13.7 kg weight-drop system on a nylon plate. The total number of shot locations on track was 47, covering 300 m.

The acquisition parameters reported for these modes establish the LS as a higher-sampling, shorter-record system relative to the conventional cess deployment. The conventional setup used a sampling interval of 0.5 ms (2 000 Hz) and record length 2.0 s; LS cess and LS track used 0.125 ms (8 000 Hz) and 1.5 s records, with pre-trigger delay –0.01 s rather than –0.02 s. For LS track acquisition, the shot offset was 9 m, the stack was 2 impacts, and the roll shift was 6 m.

The environmental note attached to the track deployment is central to interpreting LS behavior on railway ballast. Reflecting sleeper resonances and ballast attenuation introduce higher noise, and the 8 kHz sampling rate is justified as a way to capture both low- and high-frequency content under these conditions.

3. Dispersion analysis and probabilistic inversion

The MASW workflow is based on transformation of seismograms $u(x,t)$ into the frequency–phase-velocity domain by slant-stack, with dispersion energy defined as

$$D(f,c)\;=\;\bigl|\!\!\sum_{x}u(x,t)\,\exp\!\bigl[-\,\mathrm{i}\,2\pi f\,(t - x/c)\bigr]\bigr|\;.$$

From $V_s$0, dispersion images are picked for the fundamental mode $V_s$1 and first higher mode $V_s$2 within the wavelength bounds

$V_s$3

The measured dispersion curve is denoted $V_s$4, and the pick uncertainties $V_s$5 are stated to follow a Lorentzian distribution at low frequencies (Burzawa et al., 22 Jul 2025).

Forward modelling uses a layered parameterization $V_s$6, with phase velocities $V_s$7 computed via the Thomson–Haskell reflectivity matrix. The posterior probability density of model parameters given the dispersion data is written

$V_s$8

with

$V_s$9

The prior $\Delta x = 0.5$0 is uniform over parameter bounds for 4 layers + half-space. The reported bounds are: $\Delta x = 0.5$1 m, $\Delta x = 0.5$2 m, and an infinite half-space; $\Delta x = 0.5$3–$\Delta x = 0.5$4 m/s and $\Delta x = 0.5$5 m/s; with $\Delta x = 0.5$6 and $\Delta x = 0.5$7 kg/m$\Delta x = 0.5$8. Inference is performed by reversible-jump Markov-chain Monte Carlo (RJ-McMC) using 5 independent chains, 150 000 iterations each, and burn-in 30 000. Proposal standard deviations are 5 m/s for $\Delta x = 0.5$9 and 0.05 m for $\simeq 23.5$0, chosen to target $\simeq 23.5$1 acceptance. From the accepted ensemble, the study extracts a median model $\simeq 23.5$2, a best-fit layered model $\simeq 23.5$3 with minimum $\simeq 23.5$4, and 10–90 % credible intervals on $\simeq 23.5$5.

This inversion framework is not merely an adjunct to LS acquisition. In the reported use, the LS provides rapid dispersion data, while Bayesian inference supplies posterior structure, credible intervals, and marginal PDFs that constrain how anomaly signatures are interpreted.

4. Comparison with planted geophone spreads

A direct comparison on the cess constitutes the main qualification step for the LS. At positions P0 and P1, dispersion curves from the LS and the conventional spread, after the conventional data were degraded to the same $\simeq 23.5$6 and $\simeq 23.5$7, showed $\simeq 23.5$8 picks within mutual error bars. The associated spectrograms and seismograms exhibited similar surface-wave energy (20–50 Hz). The conclusion stated in the study is explicit: LS on cess yields equivalent dispersion data to planted geophones (Burzawa et al., 22 Jul 2025).

This comparison addresses a common technical concern associated with towable arrays on granular support: whether omitting steel spikes compromises coupling to the point of degrading surface-wave analysis. In the reported cess tests, that concern is not borne out by the extracted dispersion curves. The steel-plate mounting on ballast stones is presented as sufficient to obtain comparable signal-to-noise ratio and comparable dispersion resolution.

The reported efficiency difference is operationally significant. Conventional profiling on the cess, with no roll-along, achieved $\simeq 23.5$9 m/h, whereas the LS in roll-along mode on ballast achieved $\Delta x = 0.25$0 m/h. For 300 m, the stated acquisition times are 10 h vs. 6 h, corresponding to –40 % acquisition time. Crew size also differs: the conventional setup required 4 persons for planting geophones, whereas the LS required 2–3 persons for towing, shooting, and processing. Data quality is summarized as comparable S/N and dispersion resolution, with repeatable picks within $\Delta x = 0.25$1.

A plausible implication is that the LS does not simply trade data quality for speed in the tested setting; rather, the study treats the speed gain as compatible with the core MASW observables used for inversion.

5. Detection of low-velocity layers in railway earthworks

The principal diagnostic target is the Low Velocity Layer (LVL). In the track survey, 47 dispersion images were acquired along 300 m with $\Delta x = 0.25$2 m. The $\Delta x = 0.25$3 mode was reliably picked from 10–40 Hz at all positions, whereas $\Delta x = 0.25$4 was identifiable between 25–210 m. Despite higher noise, dispersion remained traceable up to 50 Hz. The resulting pseudo-2D $\Delta x = 0.25$5 sections revealed an inverse-dispersion signature—described as $\Delta x = 0.25$6 increasing with $\Delta x = 0.25$7 then decreasing—within the anomaly zone (Burzawa et al., 22 Jul 2025).

Bayesian inversion examples at P0 and P1 illustrate how the LS-derived dispersion data were translated into subsurface velocity structure. At P0 (reference zone), the posterior envelope on $\Delta x = 0.25$8 is described as narrow, and both median and best-fit theoretical dispersion curves match $\Delta x = 0.25$9. The corresponding 1D profile has shallow $\Delta x = 0.25$0 m/s, $\Delta x = 0.25$1 m/s, $\Delta x = 0.25$2 m/s, and half-space $\Delta x = 0.25$3 m/s; thicknesses are $\Delta x = 0.25$4 m and $\Delta x = 0.25$5 m, while $\Delta x = 0.25$6 is unconstrained. Posterior PDFs are reported as well peaked. At P1 (anomaly zone), the inversion shows a velocity inversion at $\Delta x = 0.25$7 m with a $\Delta x = 0.25$8 drop to $\Delta x = 0.25$9 m/s and thickness $3\times$0 m; the posteriors are broader, indicating higher uncertainty.

When the median $3\times$1 profiles are juxtaposed, they produce a pseudo-2D $3\times$2 section with reliable depth stated as $3\times$3 m in the sense of $3\times$4. In that section, a Low-Velocity Layer ($3\times$5 m/s) thickens to $3\times$6 m from 180–220 m along the line, and this thickening is reported to correlate with track geometry degradation. The marginal PDFs of $3\times$7 are used to highlight spatial variations and uncertainty.

Within the scope of the reported survey, Landstreamer is therefore not merely a transportable receiver array. Its significance lies in enabling pseudo-2D LVL delineation along railway track while preserving an inversion framework capable of distinguishing relatively well constrained zones from those with broader posterior uncertainty.

6. Operational recommendations, limitations, and prospective extensions

Several practical recommendations are reported for LS-based railway MASW. The steel plates should be cleaned of ballast fines to ensure consistent coupling. On ballast track, the study recommends high sampling (8 kHz) and a weight-drop source to mitigate ballast attenuation. It further recommends a roll-along step $3\times$8 array length to maintain $3\times$9 coverage; in the reported field case, $\Delta_{\mathrm{roll}} = 6$0 m for a 23.5 m array. It also recommends including both fundamental and higher modes in dispersion picking to improve shallow resolution. At sections with concrete slabs, the LS plates should be temporarily removed or spacers should be used to avoid noise from rigid layers (Burzawa et al., 22 Jul 2025).

The limitations are equally explicit. On-track LS deployment exhibits elevated noise from sleepers. The study therefore notes that multisource (direct+reverse) stacking or ambient-noise MASW could improve signal-to-noise ratio. Non-uniqueness in layer thickness remains a limitation of the inversion, and joint inversion with GPR or ERT (cross-gradient constraint) is suggested as a way to reduce it. For broader monitoring, integration of distributed acoustic sensing (DAS) along fiber is proposed as a route toward continuous monitoring without roll-along. Further acceleration is associated with automated dispersion picking (e.g. machine-learning) and real-time inversion pipelines.

These limitations qualify the role of the Landstreamer. The reported results do not imply that towing eliminates coupling, noise, or inversion ambiguity issues; rather, they indicate that within the tested railway setting, those issues remain manageable enough for LS-based MASW to support efficient earthwork characterization and LVL detection with quantified uncertainty.