Backscatter you already own: turning MBES returns into seabed maps

The dataset already on your drives

When a multibeam echosounder measures depth, it records the strength of the returned acoustic signal in the same instant. That amplitude – backscatter – is written to the same file, on the same survey line, at no additional acquisition time. The recent Hydro International article on hydrographic backscatter puts the commercial case plainly: the global blue economy sits at roughly US$2.5 trillion a year, offshore wind is heading toward hundreds of gigawatts within the decade, and subsea cables, power interconnectors and carbon capture systems are being deployed at unprecedented rates. Each of those developments needs to know what the seabed is made of, not just how deep it is. Grab sampling, coring and video inspection give accurate point data and nothing between the points. Backscatter gives continuous coverage across the entire survey box.

The article’s core technical point is one every processor recognises: two areas at the same depth can return very different acoustic energy. Bathymetry describes geometry; backscatter responds to composition (mud, sand, gravel, rock), surface roughness, heterogeneity, acoustic frequency, incidence angle and environmental conditions. NOAA’s operational guidance and the wider hydrographic community have documented this repeatedly. The data holds the answer to the composition question. The difficulty, as the source correctly frames it, is not acquisition but processing, interpretation and integration into decisions.

One point the article does not make, and that we want on the table from the outset: IHO S-44, in its current 6th edition, governs bathymetric uncertainty – total horizontal and vertical uncertainty, feature detection, coverage – and sets no acceptance criteria for backscatter at all. A survey can be fully S-44 compliant and still deliver backscatter that is worthless for classification. That gap is the root of most disappointment on this subject.

Why the by-product outvalues the depth on some jobs

For a nautical charting survey, depth is the deliverable and backscatter is genuinely secondary. For a large share of blue-economy work, the priority inverts. Seabed type is what drives the engineering.

On a cable or interconnector route, the burial assessment depends on sediment character: thermal resistivity governs the cable’s current rating, and sediment mobility governs whether the trench stays buried. Backscatter maps show where you cross from soft mud into sand waves or exposed rock long before the plough finds out the hard way. On a wind farm, foundation design rests on geotechnics, but a calibrated backscatter mosaic is the most efficient tool for siting the boreholes and cone penetration tests – you place expensive intrusive sampling where the acoustic facies change, not on a blind grid. For environmental consent, habitat mapping under regimes such as the EU Habitats Directive requires the spatial extent of features like Annex I reef, and that extent has to come from full-coverage acoustics tied to ground truth. Point samples alone cannot delineate a boundary.

The commercial exposure is asymmetric. The marginal cost of recording usable backscatter during a survey you are running anyway is small – some calibration discipline and disciplined logging. The cost of going back for it is a full remobilisation: vessel, crew, weather standby, mobilisation of the same sensor to the same site. If the backscatter you already paid to acquire cannot be classified because nobody specified how it was to be collected and stored, you pay twice for information you had in the first pass.

What actually determines whether the data can be classified

The decision of whether to invest in seabed classification from existing backscatter turns on a handful of engineering realities. Understand these and the go/no-go becomes straightforward.

Calibration state – relative versus absolute. Raw backscatter from a survey sonar is a relative measurement. To turn it into a physically meaningful quantity you correct for source level, time-varying gain, transmit and receive beam patterns, and acoustic absorption along the two-way path. Absolute (radiometrically calibrated) backscatter, expressed in decibels referenced to a known standard, can be compared across surveys, systems and years. Uncalibrated backscatter can only be classified relatively, within a single mosaic. Most survey sonars leave the factory without a rigorous absolute calibration, which is why the GeoHab Backscatter Working Group guidelines (Lurton and Lamarche, 2015) remain the reference document for anyone serious about this. If you need to compare a baseline survey against a monitoring survey three years later, relative data will not do it.

Angular dependence. Backscatter strength varies strongly with grazing angle – high near nadir where reflection is specular, falling and flattening across the outer swath. If you mosaic without correcting this, you get the classic banding: bright nadir stripes and dark outer beams that have nothing to do with the seabed. Angular Range Analysis exploits the same physics deliberately, using the shape of the angular response curve over a patch as a discriminator between sediment types. Either you correct the angular effect to make a clean mosaic, or you model it to classify – but you cannot ignore it.

Ground truth. Acoustic classification is calibration against physical samples. Without grabs, cores or seabed video tied to the mosaic, you have segmentation, not classification – clusters of similar backscatter with no name attached. The samples anchor the acoustic facies to a recognised scheme such as Folk or Wentworth, and let you assign a confidence to each class boundary.

System and frequency consistency. Backscatter is frequency dependent; a 200 kHz response and a 400 kHz response over the same mud are not interchangeable, because higher frequencies interact more with fine-scale roughness and penetrate less. Change the sonar, the frequency or even the pulse length mid-project and you have introduced a discontinuity that will read as a false facies boundary.

Where teams get it wrong

1. Treating backscatter as spoil to be discarded

The most common and most expensive error is upstream of any processing. Survey specifications written around S-44 bathymetry acceptance say nothing about backscatter, so the data is logged with whatever settings the online team chose, gains are changed line to line for a clean depth solution, and the raw amplitude is never archived in a recoverable form. By the time a client asks for seabed type, the information needed to calibrate is gone. The fix costs nothing at acquisition – it is a specification and a logging discipline, not a hardware purchase.

2. Promising absolute classification off uncalibrated data

The opposite failure is over-promising. A processor delivers a colourful sediment map from a single uncalibrated survey and labels the classes in absolute geotechnical terms. When a second survey from a different vessel or sonar produces different numbers over the same seabed, the client concludes the method is unreliable. It is not – the deliverable was mis-scoped. Relative classification within one internally consistent mosaic is defensible. Absolute, cross-survey classification demands radiometric calibration that was never done.

3. Under-resourcing ground truth

Acoustic facies without physical samples are unnamed and unvalidated. Teams routinely allocate a token number of grabs, place them without reference to the acoustic pattern, and then cannot separate two classes that look distinct in the mosaic. Ground truth should be planned after a first-pass segmentation of the backscatter, with samples targeted at each distinct facies and at the boundaries between them, not scattered on a regular grid that may miss a class entirely.

4. Mosaicking away the very signal that carries the information

Aggressive line-to-line normalisation and empirical seam matching can produce a visually seamless mosaic that has had the real seabed contrast flattened out of it. The result looks professional and classifies badly. The seams you are fighting are usually the angular and gain effects described above; correct those physically rather than smoothing them cosmetically, or the classifier trains on artefacts.

How to decide

The decision is job-specific. Work through it in this order.

Establish the deliverable first. If the client needs relative habitat zones within one survey, uncalibrated backscatter with proper angular correction and ground truth is enough. If they need cross-survey comparison, change detection or absolute sediment properties, you must specify radiometric calibration up front – it cannot be retrofitted.
Write backscatter into the survey specification, not just the bathymetry. State the sonar, frequency, pulse length and gain regime to be held constant, require that raw backscatter (snippet or time series) be logged and archived, and prohibit undocumented online gain changes. Reference the Backscatter Working Group guidelines (Lurton and Lamarche, 2015) as the acceptance basis, since S-44 gives you nothing here.
Run a system calibration where absolute values are required. Acquire over a stable, homogeneous reference area at the start and end of the campaign to characterise the angular response and check for drift. This is the seabed equivalent of a bar check, and it is what lets you defend the numbers later.
Design ground truth around the acoustics, not a grid. Produce a preliminary unsupervised segmentation, then place grabs, cores or drop-camera stations to sample every distinct facies with a minimum of several validated points per class, and put stations on the boundaries. Report classes against Folk or Wentworth so the output speaks the language of the geotechnical and consent teams.
Keep the processing chain auditable. Log every radiometric correction applied – source level, TVG, beam pattern, absorption – and the angular treatment, so a third party can trace a class boundary back to the physics. For consent and design work this audit trail is what makes the classification stand up to scrutiny.
Cost it as a small acquisition add-on, not a separate campaign. Because the data is collected regardless, the incremental cost is calibration time and processing hours. Weigh that against the price of a remobilisation to recover seabed information you already sailed over. On most modern blue-economy projects that comparison decides itself.

The strategic point in the source article is correct: the backscatter is not the challenge, the discipline around it is. Treat it as a core hydrographic dataset from the specification stage, resource the calibration and ground truth honestly, and you convert a discarded by-product into a seabed map you already paid to collect.

Based on: Utilizing Hydrographic Backscatter Data