Begin with the arrival, not the rock name
Did you know that an early seismic arrival does not announce the name or strength of the seabed rock? A marine seismic refraction survey records the time at which energy from a controlled source is first detected by a receiver. That earliest usable arrival is often called the first break or first arrival. Its time varies with source–receiver separation, and the resulting travel-time pattern helps constrain wave speed and subsurface structure. The scientific task is an inverse problem: infer plausible velocity distributions from measured travel times. The geological labels come later, after uncertainty and independent evidence have been considered.
Marine refraction can use seabed receivers, ocean-bottom seismometers, ocean-bottom nodes, or other acquisition layouts. Depending on source, offset and geology, the first arriving energy may travel principally through water, along or through sediment, or through a faster underlying layer. A first-break plot is not a photograph, a sediment core, or a direct measurement of a buried layer’s physical thickness. A report becomes trustworthy when it states what was observed, which phases were interpreted, what velocity model explains them, and what alternatives remain.
This guide translates time–distance plots, head-wave ideas and tomography into language readers can verify. Numerical exercises are hypothetical examples, not observations from a real site. It is educational rather than a geotechnical design approval. The method is valuable precisely because it exposes how a measured millisecond turns into a model-dependent depth, and how easily a false assumption can make a precise-looking contour misleading. Read the axis and geometry before reading the colour.
Identify the source and recording platform
A seismic source transmits energy that may propagate along several paths to the receiver. Marine acquisition differs from a land example because the water column is a propagation medium with its own sound-speed structure, tides, currents and source/receiver positioning challenges. An ocean-bottom receiver can record multiple components; a hydrophone primarily responds to pressure, while geophones or accelerometers may capture ground motion components. The instrument type, orientation and coupling influence which arrivals are visible and how they should be named.
Do not interpret the phrase ocean-bottom seismometer as a guarantee of perfect seabed coupling or identical response across all channels. A node resting on soft sediment can behave differently from one contacting rock. Clock drift, deployment position and instrument polarity require checks. A consistent time–distance trend can coexist with an unresolved deployment offset. A responsible translation states the acquisition layout, wave component and timing basis before attaching a physical explanation to the earliest visible line.
Source–receiver offset and navigation accuracy
Offset is the spatial separation used to explain the propagation path. In an ideal straight-line profile, it might be the horizontal distance between source and receiver projected onto the line. In real marine work, sources and seabed receivers can be separated crossline; the correct geometry may be fully three-dimensional. Vessel position alone is not enough, because the acoustic source has its own tow offset and the seabed instrument may not land directly under the deployment point. Velocity derived from an incorrect offset can look smooth while being systematically wrong.
For an illustrative source 800 m east and 600 m north of a receiver, horizontal separation is the square root of 800² + 600², exactly 1,000 m. Using 800 m as if the crossline displacement were absent creates a 20% distance shortfall. That error contaminates all speed estimates based on distance divided by time. Check the coordinate reference system, source navigation, instrument relocation estimates and units. Do not label a metre difference as milliseconds, or mix chart metres with geographical degrees.
Distinguish water arrivals from subsurface arrivals
A direct water-wave arrival travels predominantly through water and may be the earliest event at short offsets. A crustal or sub-bottom phase can become the first arrival at larger offsets if it samples a faster medium. The exact crossover depends on geometry, water depth and velocity structure. Refraction interpretation begins by identifying which observed timing branch corresponds to which physical path, not by fitting one straight line through every pick. For complex media, ray paths curve and require a numerical forward model.
At a simple water sound speed of 1,500 m/s, an entirely waterborne horizontal 1,500 m path would take about 1.00 s. A plotted event at 0.60 s is not automatically an error: the path may involve a faster medium, or the assumed travel path may be wrong. Likewise, an event arriving later than the direct wave may still contain useful information. The term first arrival is conditional on offset, receiver, noise and phase identification. Preserve these conditions when summarizing a survey.
Travel-time plots turn slope into apparent velocity
In the simplest homogeneous situation, travel time t increases as distance x divided by speed v, so t = x/v. On a graph with time on the vertical axis and distance on the horizontal axis, line slope dt/dx has units of seconds per metre and the inverse slope has units of metres per second. A flatter time–distance line corresponds to a higher apparent propagation speed. Apparent velocity is not always true material velocity; dipping interfaces, oblique paths and curved rays can affect the observed slope.
If travel time increases by 0.30 s when offset increases by 600 m on a well-defined straight branch, slope is 0.30/600 = 0.0005 s/m. Its inverse is 2,000 m/s. That mathematical result should be reported as a branch-derived apparent velocity until phase geometry is confirmed. If the time interval was mistakenly read as 300 s rather than 300 ms, the inferred speed would be out by a factor of one thousand. Units and axis labels protect the interpretation before any sophisticated inversion begins.
The near-offset branch is not always a unique surface layer
A near-offset first-arrival slope may be controlled by a shallow medium, but only if receiver geometry and propagation justify that assumption. In marine environments, the water wave, near-seabed sediments and receiver depth may all influence the early part of a plot. A short offset range can also make intercept and slope unstable. It is risky to convert the first observed segment into a definitive ‘layer 1 velocity’ without examining which path is actually first. A curved branch may indicate gradual velocity increase rather than a distinct horizontal interface.
Describe the near-offset branch in evidence language: travel times rise approximately linearly across a stated distance range, giving a certain apparent speed under a particular phase interpretation. Then identify what would distinguish alternative paths. Adding more near-offset coverage, examining components or comparing seabed geometry might help. The absence of a clear shallow segment does not prove the top sediment layer is absent; it may simply be unresolved or masked by a stronger arrival.
Head waves and the critical-angle idea
In a classical two-layer model where a faster lower layer underlies a slower upper layer, energy can travel down to the interface, propagate along it at an appropriate critical condition, and return to the surface as a head wave. At sufficient offset, that arrival can beat the direct wave. This is the familiar basis of intercept-time refraction analysis. It assumes approximately planar layering, identifiable phases and velocities increasing downward. The method does not work indiscriminately for every seabed sequence or strongly irregular interface.
The critical angle, defined under the usual ray-angle convention relative to the interface normal, satisfies sin(ic) = v1/v2 when v2 exceeds v1. If v1 = 1,500 m/s and v2 = 3,000 m/s, sin(ic) = 0.5 and ic = 30 degrees. That angle does not mean a visible reflector dips 30 degrees. It describes a ray condition in a simplified medium. Separating propagation angle from geological dip prevents a surprisingly common misreading of geometry illustrations.
Intercept time is not measured layer thickness
For a horizontal, two-layer idealization, a head-wave travel-time branch can be approximated by t = ti + x/v2. The intercept ti is where extrapolation of that branch meets the time axis. It reflects extra time spent travelling through the slower upper layer and depends on thickness and the two velocities. It is not a direct depth measurement and should not be read in metres. Recovering depth requires the correct model and angle relation, along with errors on the picked intercept and velocities.
With v1 = 1,500 m/s, v2 = 3,000 m/s and ideal intercept ti = 0.20 s, cos(ic) = sqrt(1 − 0.5²) = about 0.866. For a horizontal interface the inferred layer thickness h = ti × v1 /(2 cos(ic)), about 173 m. That value belongs to the ideal two-layer interpretation; a dipping or gradational boundary may not support it. A phrase like ‘173 m of sediment was measured’ overstates the evidence. Better: ‘the simplified intercept-time model estimates approximately 173 m under the specified velocities.’
Crossover distance has a useful physical meaning
The direct arrival and ideal head-wave arrival are equally fast at their crossover offset. With the direct branch t = x/v1 and refracted branch t = ti + x/v2, the crossover is x = ti /(1/v1 − 1/v2), provided v2 > v1. Before this distance, direct propagation may arrive first; beyond it the faster-layer path can dominate. The crossover is not automatically where the buried interface appears on the seafloor. It is a property of the time–distance relationship and the assumed velocity model.
For v1 = 1,500 m/s, v2 = 3,000 m/s and ti = 0.20 s, the crossover is 600 m. At 900 m, the direct branch predicts 0.60 s while the head-wave branch predicts 0.20 + 900/3,000 = 0.50 s. The head wave arrives 0.10 s earlier under this model. If acquisition only covers offsets shorter than 500 m, the faster layer might remain hidden from first-arrival data. A short survey lacking a refracted branch cannot establish that no faster layer exists.
A gradient can create curvature without a sharp boundary
Real marine sediments may have compressional-wave speeds that increase gradually with compaction. Instead of two clean straight lines on a travel-time plot, one may observe curvature associated with continuously bending rays. Forcing a layered straight-line fit onto that pattern invents a sharp interface the measurements do not demand. Gradient and discrete-layer models can both explain limited noisy data; their relative plausibility should be tested by forward modeling and geological context. Use the term velocity gradient when the evidence supports progressive change rather than assuming every curvature break is a formation boundary.
A useful exercise is to compare predicted arrival times for a two-layer model and a smooth-gradient model at actual receiver offsets. If their differences are smaller than picking uncertainty, the survey cannot resolve the choice. In that case publish an interval of compatible structures or a regularized velocity profile, not one exact contact line. A smooth inversion image may still contain model-imposed smoothness. The colour gradient can be algorithmic even when the data constrain only average travel times.
First-break picking is a measurement with uncertainty
Picking first arrivals means marking the onset of the earliest defensible wave energy in a recorded trace. Noise, low-frequency emergence, filtering, receiver coupling and source signature can obscure the beginning. Different analysts or automated detectors can choose slightly different instants. The pick’s numerical resolution may be limited by sampling interval, but sample spacing alone does not describe onset uncertainty. A 1 ms sample interval does not guarantee a 1 ms timing accuracy in a noisy marine trace. Preserve distinction between digital precision and physical confidence.
If a first break is visibly ambiguous over a 15 ms transition, recording a pick to six decimal places will not resolve it. Mark a quality flag or uncertainty interval and test how alternative choices affect inversion. A bad pick can sometimes produce a locally attractive model change. Comparing picks against raw, filtered and adjacent-channel waveforms is more defensible than minimizing mathematical residuals alone. Translation should distinguish a human or machine selection from an independently known propagation instant.
Preprocessing can help timing while creating bias
Band-pass filtering, deconvolution, gain and noise suppression can make first breaks easier to see. However, a zero-phase filter may introduce pre-event energy, and certain operations may shift or distort onset appearance. Phase treatment matters if a picker defines the first break by crossing a threshold or correlation peak. A processing script can reproduce a result perfectly while reproducing a systematic time bias. Keep unprocessed or minimally processed traces accessible for timing verification, and document any timing corrections.
Suppose the same trace’s picked onset moves from 1.234 s to 1.220 s after filtering. That 14 ms difference could matter for shallow high-resolution tomography. It should trigger examination of the filter’s phase response and detection threshold, not an automatic statement that the faster pick is more accurate. An accountable workflow stores original traces, selected processing, pick version and rationale. Readers need to know whether plotted times represent untouched observations or carefully corrected measurements.
Clock drift and timing synchronization
Ocean-bottom instruments may operate autonomously and depend on clocks that drift over the deployment period. Source timestamps and receiver timestamps must be reconciled consistently. A constant shift across all records can affect intercept times; a varying drift can distort arrival slopes across time if sources are shot sequentially. GPS synchronization at deployment and recovery, instrument-specific timing correction and uncertainty documentation are essential where applicable. A beautiful inversion cannot infer an absolute clock offset uniquely from geology in every configuration.
If an instrument clock gains 40 ms over its operation, uncorrected picks may gradually appear early relative to the source record. The resulting systematic pattern can be mistaken for a lateral velocity change when the survey progresses in one direction. Confirm synchronization and drift-correction signs with instrument logs. The phrase time-corrected should identify correction method and reference clock. Do not silently hide a timing adjustment in a spreadsheet and then present the velocity image as if every pick were an unmodified raw observation.
Positioning seafloor receivers
A seafloor node may land away from its nominal deployment position because of currents, water depth and descent dynamics. Position can be estimated using acoustic ranging, inversion of arrival times, shipboard observations or other methods. The uncertainty may be anisotropic and can be correlated with the velocity model. Treating the drop point as the receiver coordinate may bias geometry, especially at short offsets where a small distance error is proportionally large. Position uncertainty belongs beside timing uncertainty in the inversion narrative.
Imagine a receiver nominally 100 m east of its actual resting place. For a 1,500 m/s path, the corresponding travel-time scale for 100 m is approximately 0.067 s, although the actual change depends on path direction and geometry. That is much larger than millisecond-level pick precision. A location estimate derived from the same arrivals used in the velocity inversion must not be counted as an entirely independent confirmation. Clearly distinguish measured navigation, estimated placement and model-adjusted position.
Wave type: P waves versus S waves and converted energy
Marine refraction commonly emphasizes compressional or P-wave first arrivals, but seabed instruments can also record shear and converted phases under suitable conditions. Wave speed depends on elastic properties, density and wave type. A P-wave velocity is not a shear-wave velocity, and neither directly measures undrained shear strength or friction angle without additional assumptions and tests. Fluid-rich sediments can transmit compressional pressure waves while lacking an equivalent propagating shear wave in the water column. Always state which phase is being modeled.
A data table labeled 2.2 km/s should not be translated as an elastic stiffness without specifying the wave type, density and model. If converted phases are present, confusing them with first P arrivals can create wrong apparent velocities. Multi-component records and polarization analysis may help phase identification. Geological interpretation requires a chain connecting the seismic parameter to the material quantity of interest. A velocity anomaly is evidence about wave propagation; it is not an unrestricted label for strength, permeability or fracture density.
Reversed shots and reciprocal reasoning
Recording energy along a line from more than one direction can reveal asymmetry caused by dipping layers, lateral velocity change and acquisition effects. In an ideal reciprocal medium, travel time between the same source and receiver locations should be reciprocal. The geometric condition matters: ‘reversed survey’ is not merely a different vessel heading with unmatched positions. Comparing reciprocal paths provides a useful QC check on timing, phase identification and heterogeneity. Apparent velocities measured in opposite directions can differ across a dipping refractor without violating basic wave physics.
Before averaging two slopes, confirm whether they refer to comparable rays and offsets. If one direction has poor coupling or a different first-arrival phase, a mean number may be physically meaningless. Document the two observed branches and identify why they disagree. In a simple dipping-layer interpretation, forward and reverse apparent velocities can help estimate dip, but that inference needs explicit equations and assumptions. For complicated geology, use appropriate ray tracing rather than forcing a two-line textbook solution.
Refraction can miss a hidden low-velocity zone
The classical head-wave method is best suited to increasing velocity with depth. A lower-velocity layer beneath a faster layer may not generate a first arrival that dominates the travel-time envelope. Such a hidden layer can be invisible to first-arrival refraction even though it exists. Therefore an interpretation that shows no low-speed layer is not evidence of its absence unless other measurements constrain it. This limitation is essential for geotechnical, groundwater and structural questions in which a weak or highly weathered layer could matter.
A careful summary might say: ‘The first-arrival model provides no resolved indication of a low-velocity layer over this depth range; the method is insensitive to some hidden low-velocity configurations.’ That differs from claiming homogeneous strong rock. Reflection data, boreholes or other complementary observations may test the gap. The general lesson is that inversion can make only distinctions that the measured wave paths are sensitive to. A model cell without a contrasting colour does not automatically mean a uniform earth.
Ray coverage determines where the model is informative
A travel-time tomography image is assembled by fitting many source–receiver paths through a computational model. Some cells are sampled by numerous independent paths at varied angles; others may be barely reached. The displayed colours can cover the entire rectangular model even where rays provide little constraint. Edge regions and deeper parts often deserve special caution. A high-resolution grid does not equal high-resolution data, and a smoothly interpolated velocity contour does not reveal how much of it depends on regularization.
Ask for ray-density, path-coverage and sensitivity displays alongside the final velocity image. A claimed deep anomaly outside reliable ray coverage should be marked weakly constrained. Compare recovery tests, such as inserting synthetic features into the model and seeing whether the survey geometry would resolve them. A checkerboard recovery illustration is helpful but does not automatically validate every shape, amplitude or true geological interface. The exercise must use realistic noise and model assumptions and be reported without exaggeration.
Tomography is an inverse problem with multiple solutions
Travel-time inversion adjusts a model so predicted arrivals approach observed arrivals. Different velocity structures can produce very similar times, especially with limited offset, sparse receivers or restricted ray directions. Regularization selects among those alternatives by imposing smoothness, damping or another preference. The objective function therefore often balances data misfit and model complexity. A smooth tomographic section is one defensible model, not a uniquely photographed interior. When translating inversion terms, make explicit which features the data require and which are primarily stabilized by assumptions.
A shallow high-velocity body and a deeper broad speed increase might both reduce some travel times. An inversion can trade depth against velocity if ray geometry is limited. Present alternative starting models or regularization strengths where that tradeoff affects the conclusion. Calling an algorithm ‘advanced’ does not remove nonuniqueness. The useful question is whether the reported property remains robust when analyst choices change within reasonable physical limits. The data should govern the conclusion, not the model palette.
Travel-time residuals need a numerical interpretation
A residual is the difference between measured and predicted arrival time under a declared sign convention. It can be expressed as observed minus calculated or vice versa; never assume the sign without checking the legend. Residual maps help diagnose systematic travel-time errors, but a small average residual can conceal positive and negative misfits that cancel. Root-mean-square residual is often more informative about typical magnitude, although it still does not prove a unique or geologically correct model.
For four synthetic residuals of +20, −10, +30 and −20 ms, the mean is +5 ms. The RMS is square root of (400 + 100 + 900 + 400)/4, approximately 21.2 ms. Reporting only the mean would understate the variation. Compare the RMS with justified pick uncertainties and inspect residuals against offset, source time and position. If residuals exhibit a coherent slope, possible geometry or velocity bias should be investigated rather than declaring success based on one aggregate statistic.
Overfitting gives a lower error without more geology
An inversion with many unconstrained cells can fit timing noise, receiver-location mistakes or systematic biases. Increasing model complexity may reduce residual RMS while making the geology less believable. Conversely, overly strong smoothing can erase a real fault or steep velocity gradient. A credible process considers the expected noise level, model resolution, independent constraints and stability across a range of regularization strengths. Good model selection does not simply maximize the number of visible colour boundaries.
When comparing two runs, record the same data selection, pick weights, starting model and geometric reference where possible. If one run introduces far more parameters, a small misfit improvement is not automatically worth the additional structure. Independent validation, cross-validation and forward prediction can reveal whether the apparent detail is stable. Translate results as ‘a model consistent with the available arrivals’ until independent evidence narrows the alternatives. The word fit is not synonymous with proof.
Seawater speed and bathymetry are part of the model
Water-column sound speed depends on temperature, salinity and pressure and can vary vertically and temporally. Bathymetry affects travel paths and where receivers sit relative to the seabed. An inaccurate water layer may be compensated by a wrong shallow sediment speed in inversion. Similarly, a mistaken seabed elevation can shift the inferred depth of a faster refractor. Refraction and multibeam bathymetry are complementary because one constrains arrival paths and the other helps define geometry, but they are not substitutes for each other.
A survey should state whether the sound-speed profile was measured, modeled or adopted from a representative reference. If water velocity is fixed at 1,500 m/s for convenience, quantify whether reasonable variation could influence the claimed sediment anomaly. Compare ray tracing with alternative water models and check source depths. A map of derived sub-bottom velocity has a dependency on the water model that should be acknowledged in the methods and any high-stakes summary.
Velocity-to-depth conversion is model dependent
A travel-time inversion may present velocity values in a geographic depth grid, but the depth coordinate still depends on assumed ray paths, seabed topography and model parameterization. In some refraction workflows, interface depth is estimated by intercept-time methods; in others, velocity is distributed continuously in cells. These products answer different questions. A sharp interface line drawn onto a smooth tomography image can be a later interpreter’s choice rather than a discontinuity required by data.
Suppose an ideal top layer is inferred as 173 m thick under the two-layer example. A competing gradient model may explain the same picks without a single boundary at precisely 173 m. The responsible translation offers the estimated value within the stated model and explains why competing parameterizations are plausible. If a construction decision requires actual interface depth, additional investigation may be necessary. Mathematical inversion cannot create physical sampling where none occurred.
Apparent velocity does not directly identify lithology
The speed of compressional waves depends on elastic moduli, density, stress, saturation and many details of rock and sediment fabric. Different lithologies can share similar P-wave speeds, while one lithology can exhibit a wide range under different conditions. Therefore ‘3.5 km/s means sandstone’ or ‘2.0 km/s means clay’ is not a valid universal key. Velocity ranges are contextual clues, not dictionary translations. A strong geological report combines seismic velocity with cores, logs, depositional history and other constraints.
If a survey reports a faster zone beneath slower sediments, describe a relative speed increase first. It may be consistent with greater compaction, cementation, lithological change or a different physical state. Decide among explanations using independent evidence. For families and students, a useful analogy is that knowing a car’s speed does not reveal its make; likewise measuring wave speed does not uniquely reveal material identity. The scientific conclusion should not exceed the parameter actually measured.
Resolution versus numerical cell size
Tomography software can output a grid with cells only a few metres wide even when acquisition paths cannot resolve differences at that scale. Model cell size controls how an answer is represented and parameterized, not how much information the field data contain. Resolution depends on frequency, ray coverage, path geometry, noise, regularization and sensitivity. Converting pixels into metre-level geological boundary accuracy is one of the most misleading uses of a polished subsurface image.
Imagine a report that uses 5 m model cells but has receiver spacing of 250 m and sparse deep ray crossings. The output may be displayed at 5 m increments, yet many neighboring values are strongly correlated and model-determined. Do not announce a 5 m lateral resolution solely because that is the grid dimension. Use synthetic recovery and independent constraints to provide a more honest spatial resolving scale. State what the acquisition can distinguish, not merely what the file can store.
Uncertainty should separate several error sources
Timing picks, source navigation, node position, water speed, phase interpretation, velocity parameterization and inversion regularization contribute different uncertainties. Some are random; others are systematic and correlated. A confidence interval from repeated inversions under one unchanged wrong assumption can omit the dominant bias. Report uncertainty in a form appropriate to the user decision: for example, an estimated speed range over a well-sampled region or alternative possible depths for a refractor. Avoid false precision in both text and figures.
If a model speed is quoted as 2,350 m/s with uncertainty of ±100 m/s, indicate the meaning of the interval and which errors were included. It is not necessarily a 95% statistical range. If an interface depth is assigned to the nearest metre but uncertain by tens of metres, use appropriate rounding. Reproducible uncertainty language makes the interpretation useful to engineers, researchers and non-specialists without hiding technical complexity.
QC before publishing the velocity image
A practical review should read raw first-arrival gathers, pick quality flags, timing and navigation corrections, travel-time plots, residual maps, ray coverage, starting models, alternative inversions and crossline consistency where available. Ask whether the result is stable under reasonable parameter choices. A contour extending beyond acquisition coverage is a cartographic estimate, not additional field evidence. Each figure should identify coordinate system, depth datum, colour scale, units and processing version.
Keep negative findings precise. Saying ‘no high-velocity body detected’ is a statement about sensitivity under this survey, not proof that no such body exists. If the available offset range cannot sample a target depth, say that clearly. If a slower layer would be hidden by first-arrival physics, acknowledge it. These boundaries are part of accurate scientific translation, not footnotes to be removed when writing a shorter public summary.
Checked worked case: two-layer head-wave interpretation
Take a synthetic horizontal two-layer setting with upper speed 1,500 m/s, lower speed 3,000 m/s and picked intercept 0.20 s. The critical angle is 30 degrees from the vertical normal. Thickness is 0.20 × 1,500 /(2 × cos30°), or approximately 173.2 m. The direct and head-wave curves meet at x = 600 m. At x = 900 m the two predicted times are 0.600 s and 0.500 s respectively. The numbers are mutually consistent with the ideal equations.
Now change the intercept to 0.22 s without changing speeds. The inferred thickness becomes about 190.5 m, a difference of roughly 17.3 m. This sensitivity explains why a seemingly minor onset or phase-picking difference can matter. No core was collected in this exercise; all values are assumed. The conclusion is a model-dependent calculation, not a factual statement about a real ocean-bottom interface.
Checked worked case: residuals and competing models
Imagine five observed first-break times compared with predictions from two velocity models. Model A residuals in milliseconds are +10, −20, +15, −5 and 0. Their squared values sum to 100 + 400 + 225 + 25 + 0 = 750, giving RMS sqrt(750/5), about 12.2 ms. Model B residuals are all +13 ms, giving RMS 13 ms. Model A has slightly lower RMS, but Model B’s uniform bias suggests a correction such as a timing offset might be relevant. Selection requires more than choosing the smallest raw RMS.
The uniform bias in Model B is also information about the process. A simple clock or source-time correction could improve it if independently justified, while Model A’s mixed residuals might reflect unmodeled structure or random noise. Do not ‘fix’ the data by adjusting timing without records. Explain what diagnostic additional evidence would decide. The scientifically sound language is about current fit, possible causes and tests, rather than declaring a winner based on decimal places.
Checked worked case: when a fast layer is not observed
Suppose receivers cover only x = 0 to 400 m while a plausible two-layer model predicts a crossover at 600 m. Across the observed offsets, the direct branch is faster than the head wave. The absence of a clear refracted first-arrival branch does not refute the faster lower layer; the acquisition does not reach the expected discriminating offsets. This is a fundamental observational limitation, not a processing failure that can be cured by colourful interpolation.
A reader preparing a summary should say the recorded first arrivals predominantly constrain shallower propagation under the model and do not uniquely test the proposed deeper layer. A wider offset range or independent geophysical information could help. Equally, do not assume the faster layer exists; the exercise demonstrates why neither assertion is justified by that short survey alone. Good translation communicates what the experiment can decide.
Transfer problem 1: slope
A first-arrival branch increases by 0.24 s over 480 m. Calculate apparent velocity. Answer: slope is 0.24/480 = 0.0005 s/m, so apparent velocity is 2,000 m/s. The number does not by itself identify one formation.
Transfer problem 2: milliseconds
A pick is 850 ms. Express it in seconds. Answer: 0.850 s. Using 850 seconds in a travel-time equation would create a thousandfold timing error.
Transfer problem 3: critical angle
Let upper speed be 1,800 m/s and lower speed 3,600 m/s. The ideal critical angle relative to the normal has sin(ic)=0.5, so ic=30°. That is not the interface’s geological dip.
Transfer problem 4: direct arrival
Under ideal constant propagation at 1,500 m/s, what is the time across 750 m? Answer: 750/1,500 = 0.50 s. Real bent or mixed paths need an appropriate propagation model.
Transfer problem 5: phase identification
Two coherent straight segments cross in a first-arrival plot. Is their intersection a buried rock boundary position? Answer: no. It is a crossover of predicted arrival times, not automatically a mapped geological contact.
Transfer problem 6: residual RMS
Residuals are +10 ms and −10 ms. Their mean is zero but their RMS is 10 ms. A zero mean must not be reported as a perfect fit.
Transfer problem 7: hidden low velocity
A first-arrival tomogram has no slow band under a fast layer. Can you say none exists? Answer: no. Low-velocity layers may be poorly detectable in conventional first-arrival refraction.
Transfer problem 8: ray coverage
A model cell has no traversing rays but a definite colour. Is its value directly constrained? Answer: no. It may be an interpolation or regularization-driven estimate; show coverage and sensitivity.
Transfer problem 9: receiver position
A node’s drop location and estimated landing location differ by 80 m. Which belongs in travel-time geometry? Answer: the best supported actual receiver position, with positioning uncertainty recorded; the drop point is not automatically the sensor location.
Transfer problem 10: interval versus apparent speed
A travel-time line gives 2,700 m/s apparent speed. Should you label all material traversed 2,700 m/s? Answer: no. State the phase, geometry and simplifying assumptions before interpreting local velocity.
Transfer problem 11: model comparison
Two inversion models fit within pick uncertainty but have different interface depths. Which exact depth is observed? Answer: neither depth is directly observed. Report the unresolved model alternatives and identify what further measurements could separate them.
Transfer problem 12: report sentence
Write a public interpretation of a shallow high-speed zone with moderate ray coverage. Model answer: ‘First-arrival travel times are consistent with a relatively faster shallow zone under the stated inversion settings; its thickness and material identity remain model dependent and require independent constraints.’
FAQ: Does seismic refraction measure soil strength?
No. It primarily constrains wave propagation and, through modelling, velocities. Engineering strength is a different property and needs appropriate tests or calibrated local relationships. Avoid universal velocity-to-strength conversions.
FAQ: Why use ocean-bottom nodes instead of a single streamer?
Seabed recording can provide geometry and components useful for refraction and wide-offset paths. The actual benefit depends on the scientific objective, deployment layout, coupling, instrument quality and processing.
FAQ: What is a first break?
It is an interpreted onset of the earliest detectable arrival in a trace under a defined picking method. It is not necessarily the strongest peak, and it carries timing and phase-identification uncertainty.
FAQ: Is low RMS enough to prove an inversion?
No. Many velocity models may fit the times. Model complexity, assumptions, ray coverage, independent information and sensitivity to alternative starting conditions are also essential.
FAQ: Why do the edges of a tomogram look smooth?
Regularization and boundary constraints can fill poorly sampled cells with smooth values. A smooth edge is not necessarily well measured. Inspect ray coverage and resolution estimates.
FAQ: Can a refraction line alone establish a fault?
A velocity discontinuity may be compatible with faulting, lithological contrasts or other structures. Independent mapping, reflection data or borehole evidence may be needed to support a fault label.
FAQ: Is a velocity model a geological map?
No. A velocity model estimates a physical wave-propagation parameter under assumptions. Geological interpretation requires an additional evidence-based reasoning step. Keep the two products distinct.
FAQ: What belongs in the caption?
Include acquisition geometry, first-arrival phase, velocity units, inversion method, vertical datum, model resolution limitations and any region lacking adequate ray coverage. The reader should know what the colours can support.
Final interpretation checklist
Confirm actual source and receiver positions, wave types, first-break definitions, timing and drift corrections, water-column assumptions, direct versus refracted phase picks, travel-time slopes, head-wave applicability, residual statistics, ray coverage and regularization choices. State which properties are measured, inferred or merely hypothesized. Verify every unit and conversion; do not let a high-resolution image make claims beyond the recorded paths. The strongest interpretation is the one whose limits remain legible.
Primary sources and complementary verified lessons
USGS: Nonlinear refraction and reflection travel-time tomography
USGS: Two-dimensional inversion of seismic refraction first arrivals
USGS: Marine multichannel seismic reflection processing
Google Search Central: People-first information
eduKateSG: Translation series index