I have processed and interpreted side-scan sonar for survey projects and now build the processing chain in Dolphin Explorer. This is the sequence I use, why each step comes where it does, and what each one can get wrong. Where I describe specific behaviour, it is how Dolphin Explorer implements it.

What a ping actually records

A side-scan sonar sends a fan-shaped pulse to each side of the towfish or vessel, port and starboard, and records the strength of the returning echo against time. Early samples come from the water column directly below the sensor, where there is no seabed yet. The first strong return is the seabed under the sensor; after that, each sample comes from seabed further out to the side.

Two properties of that record shape everything that follows. Distance is measured along the slant from the sensor, not across the seabed, so near-range features are squashed. And the echo gets weaker with distance and changes with the angle at which the sound meets the seabed, so raw amplitude is not a direct measure of what the seabed is made of.

Bottom tracking comes first

Almost every correction needs the sensor’s height above the seabed, its altitude. Navigation files sometimes carry an altitude, but it can be badly wrong, so the dependable source is the record itself: the first seabed return in each ping.

A good bottom tracker is stricter than “the first loud sample”. Dolphin Explorer’s searches after a blanking zone for the first sample that crosses a threshold and is followed by a sustained return, so a lone spike, multipath or scatter in the water column is rejected. Each pick gets a confidence based on how far the return stands above the water just ahead of it, and low-confidence picks are flagged for review. Along the line, picks that jump away from their neighbours are rejected, and short gaps are bridged by interpolation.

Two rules matter in practice. One seabed per line: the waterfall view, the processing graph and the missing-altitude estimate all use the same detector and the same corrected seabed, so a ping gets the same answer wherever it is processed. And manual picks are never overwritten: once an interpreter has corrected the seabed, automatic detection works around those corrections.

Slant range to ground range

With altitude known, each sample can be placed on the seabed. If h is the altitude and r the slant range to a sample, the across-track ground range is √(r² − h²). Samples before the seabed (the water column) have no ground position and are masked rather than stretched into the image. This is what removes the dark band down the centre of a raw waterfall and makes near-range features their true shape.

If the bottom track is wrong, slant-range correction is wrong in a way that looks plausible: features near nadir are pushed or pulled and the centre of the mosaic tears. That is why the seabed is checked before anything else.

Radiometric corrections

Radiometric corrections make amplitude comparable across the swath and along the line, so that a brighter patch means a harder or rougher seabed rather than a closer one.

Time-varied gain (TVG)

Compensates for spherical spreading and absorption as sound travels out and back. Dolphin Explorer’s default spreading term is 20 dB per decade of range, with an optional absorption term in dB per metre.

Angle-range correction (ARC)

Seabed backscatter depends on grazing angle. ARC uses the altitude to work out that angle for each seabed sample and flattens the response across the swath. The water column has no grazing angle and is left alone.

Automatic gain control (AGC)

Evens out brightness along the line. In variable mode it uses a window of neighbouring pings, so it adapts as the seabed changes along the line.

Empirical gain normalisation (EGN)

Learns the average response of the line itself by altitude and range, and divides it out. Because it depends on geometry, it works on runs of pings with similar altitude.

In Dolphin Explorer, TVG and ARC are level-preserving: each correction curve is referenced to its own mean in decibels, so applying one does not brighten or darken the whole line, only reshape it. That makes it easier to switch corrections on and off and see what each one did.

Navigation and layback

A side-scan image is only as good as its positions. Raw GPS fixes are noisy, so the track is smoothed, with gaps respected rather than bridged blindly. For a towed sensor there is also layback: the towfish is some distance behind the antenna, so each ping’s position is moved back along the recorded heading (or the course over ground when there is no heading). Get layback wrong and the same boulder appears in two places on adjacent lines.

Enhancement is for the interpreter, not the record

After the corrections come steps that make the image easier to read: speckle filtering with a local median, percentile contrast stretching, histogram equalization, adaptive contrast (CLAHE) and destriping. These help the eye but change the amplitudes, so they belong at the end of the chain, and the corrected but unenhanced data should stay available. Dolphin Explorer applies them in its processing graph without modifying the original survey file.

Reading height from a shadow

Shadows are where side-scan earns its keep. An object standing on the seabed blocks the sound, leaving an acoustic shadow behind it, and the shadow’s length gives its height by similar triangles. If the sensor is at altitude h and the shadow runs from ground range r₁ to r₂, the object’s height is approximately h × (r₂ − r₁) / r₂.

For example, with the sensor 10 m above the seabed and a shadow from 40 m to 44 m out, the object is about 10 × 4 / 44 ≈ 0.9 m high. The same geometry is why Dolphin Explorer’s boulder detector weights the shadow most heavily: on many seabeds a boulder’s face barely shows, but its shadow does. I wrote more about detecting boulders automatically in my boulder detection research.

What to check before trusting the image

  • The seabed line: does it follow the first return on every line, especially over slopes and where the altitude changes quickly?
  • The centre of the mosaic: tearing or doubling along nadir usually means a bottom-tracking or altitude problem.
  • Adjacent lines: features that do not line up between overlapping lines point to navigation or layback errors.
  • Brightness across the swath: a bright stripe at a fixed range on every ping is a gain or angle problem, not geology.
  • What enhancement did: compare the enhanced image with the corrected one before calling a contact.

Further reading