Reading a crossline check as a map, not a pass rate
A crossline analysis is one of the few QC products on a bathymetric survey that compares the dataset against itself. Two survey lines cross the same patch of seabed, acquired at different times, on different headings, under different tide and sound-velocity conditions. The depth they report at the crossing point should agree. How closely they agree is a direct, measurable statement about how much the dataset can be trusted.
In practice, this check is often reduced to a single number in the final report – a percentage of crossings within tolerance – and read the way one reads a passing grade. That reduction is where the useful information gets lost.
The tolerance envelope
IHO S-44 does not prescribe a crossline procedure. What it defines is the Total Vertical Uncertainty (TVU) permitted at a given depth, at the 95% confidence level:
TVU = √( a² + (b × d)² )
where d is the depth and a and b are the constants set by the survey order. For Order 1a, the order most commonly specified for offshore wind site investigation:
- a = 0.50 m (depth-independent component)
- b = 0.013 (depth-dependent component)
Which gives, across a typical Southern North Sea site:
| Depth | Permitted TVU (Order 1a) | Crossing threshold (√2 × TVU) |
|---|---|---|
| 20 m | 0.56 m | 0.80 m |
| 30 m | 0.63 m | 0.90 m |
| 40 m | 0.72 m | 1.02 m |
| 50 m | 0.82 m | 1.16 m |
| 60 m | 0.93 m | 1.31 m |
Tap the chart to enlarge it
The second column is the point that gets skipped. A crossing difference is the disagreement between two independent measurements, each carrying its own uncertainty. Comparing that difference against a single TVU value is the wrong test – the two uncertainties combine in quadrature, so the envelope for the difference is √2 × TVU. Specifying which of these two thresholds applies is a contractual decision, not a technical detail, and it should be written into the specification before acquisition rather than argued about after delivery.
A worked example
The figures below are synthetic – built to illustrate the method, not taken from client work – but the shape is one we see repeatedly.
A site investigation survey over a wind farm development area, water depths from 22 m to 58 m, acquired to Order 1a. The crossline analysis produces 1,240 crossing points. The summary line in the report reads:
By almost any project's acceptance criterion, that passes. It is a good number. The dataset gets signed off.
Now the same result, resolved spatially. The 45 failing crossings are not distributed randomly across the block. They fall into three groups:
| Group | Crossings failing | Depth range | Seabed character |
|---|---|---|---|
| A | 6 | 24–31 m | flat, sandy – isolated, no pattern |
| B | 8 | 44–58 m | flat – all from one line, one day |
| C | 31 | 26–34 m | mobile sand wave field |
Tap the chart to enlarge it
Three different findings, three different consequences.
Group A is noise. Six isolated crossings out of 1,240, no spatial or temporal pattern. This is what a healthy dataset looks like at the tails.
Group B is a systematic error. All eight failures come from a single line acquired on a single day. That signature – clustered in time, not in space – points at the vertical reference: a tide correction applied from the wrong station, a draft change not logged after bunkering, a sound velocity profile that had aged past its useful window. It is a real defect, and it is also the easiest of the three to fix, because a systematic offset on a known line can be quantified and corrected rather than reacquired.
Group C is the one that matters. Thirty-one failures concentrated in a mobile sand wave field, over a depth range of 26–34 m. Here the two survey lines genuinely disagree, and both may be correct: the seabed moved between the two passes. No reprocessing will reconcile them, because there is nothing to reconcile – the measurements describe two different states of a surface that changes.
Before attributing all of it to mobility, though, one component has to be subtracted, and it appears in exactly the same places. On a sloping surface, a horizontal offset between the two lines turns into a vertical difference even if the seabed has not moved at all: on a 15° slope, one metre of horizontal offset produces 27 cm of height difference on its own. On the flanks of a sand wave, the steepest gradients are precisely where the crossings fall. This is why several specifications exclude high-gradient areas from crossline statistics, or require the difference to be normalised against the local slope before it is compared with the threshold.
Separating the two components is what makes the attribution defensible. If the disagreement survives once the slope contribution has been removed, then the seabed really did move – and at that point the statement holds up even in front of a contractor with an interest in dismantling it.
Why the pass rate hides the finding
The 96.4% headline treats all 45 failures as equivalent. Resolved spatially, they are three separate findings requiring three separate responses: accept, correct, and – for Group C – escalate.
Group C matters because of where it sits, not how large it is. Thirty-one crossings is 2.5% of the dataset. But if a proposed foundation location or a cable route crosses that sand wave field, the survey has just produced quantitative evidence of seabed mobility in the exact area where burial depth and scour protection will be designed. That is not a QC failure to be dispositioned and closed. It is a geohazard finding, and it belongs in the engineering discussion, not in an appendix.
The pass rate cannot tell you this. It is a scalar summary of a spatial phenomenon, and the spatial structure is the entire finding.
What to ask for in the specification
Most of what makes a crossline check useful is decided before the vessel sails:
- 01State which threshold applies – TVU or √2 × TVU – and at what confidence level. Do not leave it to be inferred from the standard.
- 02Require the crossline result as a plotted surface, not only as a summary statistic. A difference map with the tolerance envelope applied shows structure that a percentage cannot.
- 03Require failures to be grouped and attributed – noise, systematic, or real change – rather than listed. The attribution is the analysis; the list is only the input to it.
- 04Define what happens next for each attribution. A systematic offset gets corrected. Genuine seabed change gets escalated to the engineering team. Without this defined in advance, both outcomes tend to receive the same treatment: a note in the report.
- 05Set the crossline density – a common baseline is crosslines totalling around 5% of mainline length, but the right figure depends on the site and the decisions the data will support.
The underlying point
A crossline check answers a narrower question than it appears to. It does not tell you whether the data is good. It tells you where two independent measurements of the same seabed disagree, and by how much – and the value is in reading that as a map of where confidence is lowest, not as a grade.
A dataset that passes at 96.4% is not uniformly 96.4% reliable. It is highly reliable across most of the block and least reliable in one specific area – and in this example, that area is where the engineering happens.
CLEGAR provides independent QC and technical assurance on geophysical datasets for offshore developers, contractors and asset owners. If you are specifying a survey, or reviewing one you have received, we are glad to talk it through.