US/CHC97 Trials, Interpretation

Guide to the Creation and Interpretation of the
Comparative Data Products

A comparison of swath sonar systems,
demonstrated at the
1997 US/Canada Hydrographic Commission
Coastal Multibeam Surveying Course

Ocean Mapping Group, Dept. Geodesy and Geomatics Engineering
University of New Brunswick, CANADA
506-453-4568 (ph.), 506-453-4943 (fax.), jhc@omg.unb.ca

Choice of Benchmark Survey Area

All the data is acquired over the same terrains, which are two subtidal regions of the floor of the St. Croix River in water depths ranging from 0 to 40m. The sites were chosen because they have a variety of types of short wavelength topographic features developed in outcropping bedrock. Adjacent to the bedrock are featureless regions (believed to be muddy gravel pavements) in which no short wavelength relief can be resolved. These site thus allowed one to examine both:

- the limit of spatial resolution achievable by the sonars and

- the bottom detection noise level of the sonars in regions devoid of any true topography.

over the range of water depths examined.

The following figures show the regional bathymetry, bottom morphology and bottom backscatter strength within the two regions chosen:

Deeper subtidal region (20-40m)


Depth Encoded bathymetric map of the benchmark site and surrounds The colour scale runs from 34m (blue) to 3m (pink). All depths are tidally reduced (average tidal range is 7.5m).	Sun Illuminated Topographic map showing the contrast between the rough bedrock outcrop regions and the smooth sedimented areas.	EM1000 sidescan mosaic light tones are high backscatter (White = -15dB, Black = -35dB) This image illustrates the patchy nature of the true bedrock outcrop which is partly draped in places with unconsolidated sediments.

Shallower subtidal region (0-20)


Depth Encoded bathymetric map of the benchmark site and surrounds The colour scale runs from 23 (blue) to 3m (pink). All depths are tidally reduced (average tidal range is 7.5m).	Sun Illuminated Topographic map showing the contrast between the rough bedrock outcrop regions and the smooth sedimented areas.	EM1000 sidescan mosaic light tones are high backscatter (White = -15dB, Black = -35dB) This image illustrates the patchy nature of the true bedrock outcrop which is partly draped in places with unconsolidated sediments.

gridding methodology

Data is gridded using a weighted filter. For the deeper bedrock outcrop regions the grid nodes are spaced 1.0 metres apart (whereas a 0.5m node spacing was used for the inshore area). The data is interpolated onto the nodes using a 2nd order Butterworth filter. The weighted filter dimension, has a maximum interpolation radius of 2m (1m inshore) and a flat top with a radius of 0.5m (0.25m inshore). The method used is based on :

Slootweg, A.P., 1978, Computer contouring with a digital filter: Marine Geophysical Researches, v.3, p.401-405. 1978

The weighted gridding used is optimised for data that is spaced at, or closer than, the grid node spacing. This filter is designed to interpolate over no more than a single grid node. At the same time it acts as an anti-aliasing filter to prevent noise (or seabed information at wavelengths shorter than the grid spacing) biasing the gridded result. However for the case of data that has a variable density (as does the beam density across the swath of all the sonars considered) the filter has two detrimental side effects which need to be recognised as a limitation of the filter, rather than limitations in the sonars:

for the case of beams spaced wider than twice the grid nodes, incomplete surface interpolation takes place. This is especially noticeable for the 150 degree swath sonars such as the EM1000 in 1995 and the ELAC Mk11 this year.
If I open up the interpolation, I will smear out the fine detail which we are trying to resolve. For the purposes of target delineation there is no way the system can identify a shorter wavelength feature with a horizontal dimension of less than 2X the local beam spacing anyway.
as the beam spacing opens up, a terracing effect can be seen whenever steep slopes occur at locations corresponding to the point just before inter-beam interpolation is lost. This will occur whenever the beams are spaced more than a grid node but less than 2 grid nodes apart. One sees this in the outer edges of the EM3000, the ELAC and Seabat 8101 data as a series of ~swath parallel lineations.

Seabat 8101 ---- EM3000 ---- ELAC 150

the above three examples (from differing locations) show the limited interpolation and "terracing" that occurs as the beam spacing at the edges of the swaths goes beyond 1-2x the grid node spacing:

As stated above, in both cases, these artifacts are a limitation of the gridding algorithm used and should not be confused with limitations of the sonars themselves.

sun-illuminated images

In trying to separate the information content from the noise (artifacts due to various sources) there is rarely any point in looking at conventional depth encoded images (except for gross error sources).
Manufacturers commonly exhibit their data in spectacular multicoloured image formats which, while very pretty, tell the user little other than the fact that the sonar can indeed distinguish the regional depth distribution.

Figure : Colour coded image of a ~8km by 8km area with a depth range from 50 to 100m.

The limits of the sonar are most apparent at scales of about 2% of the local water depth or less. We thus require a method to highlight that scale of feature. Sun-illumination is an ideal method as it looks for the local directional depth derivative (i.e.: the slope rather than the depth). The characteristic short wavelength depth anomalies associated with imperfections in sonar bottom detection/motion compensation etc... are far more clearly revealed through the use of this method.

Figure : Sun-Illuminated image of the same ~8km by 8km area with a depth range from 50 to 100m. In this case a whole variety of instrument artifacts are clearly visible (bottom detection, systematic beam pointing errors, roll time delays, roll biases, navigation timing errors, tide merging errors etc...).

For all the images presented as part of this comparative study, the following parameters are used:

Data is illuminated assuming 1 X vertical exaggeration, a sun from the top right with an elevation of 45 degrees. The surface properties are modelled in a lambertian-like manner. The intensity is proportional to the cosine of the angle between the sun-vector and the normal to the local surface.

Interpretation of sun-illuminated images

Things to look for here include:

- evidence of swath parallel lineation indicating a systematic bias in the bottom detection
- zones within the swath where similar bottom detect noise characteristics exist. This might indicate the location of differing bottom detect methods across the swath.
- swath orthogonal lineations indicating that vessel motion compensation is imperfect.
(Remember this could be a failing of either the sonar implementation of the attitude, or conversely the data supplied (or even both)).
- isolated point like targets which do not show up in any of the other systems. These would indicate "outliers" which are isolated false bottom detects.
(one has to be careful here because different systems have different levels of spike filtering enabled so that lesser or more isolated spikes are likely to get through).
- how does the system handle steep slopes:
- how sharp is the resolution of a particular seafloor feature (edge of bedrock scarp or fault lineation) as imaged by each of the sonars in turn. This will reflect the combination of three main attributes of the sonar system:

difference from mean surface

Ideally one would like to come up with a measure of the uncertainty associated with the sounding solutions derived by a particular system. The problem is that this uncertainly often contains a large systematic component due to the imperfect integration of ancillary sensors e.g.:

biases in roll, pitch and heading
biases in positioning
imperfect refraction solution

This is very likely the case for the systems compared herein. I would like to isolate those errors due to :

- imperfect short period motion compensation.
- bottom detect noise

One can do this reasonably by selecting a location on the seafloor which you believe to be devoid of any short wavelength seabed roughness. We have a great advantage here in that we view most of the benchmark seabed, more that 10 times with 5 different sonars and can thus confidently select regions that meet the above stated criteria. Additionally some of these sonars provide sidescan-like imagery, confirming the lack of short wavelength targets.

For regions that exhibit these smooth seabed characteristics, one can take a subset of one sonars estimates of the bathymetry (for example just the near nadir data), and smooth the data to remove any apparent roughness with wavelengths shorter that the noise. Because the seabed is really smooth (we assume) then the smoothed version of the bathymetry will contain just the real topography.

	showing swath of data used to calculate statistics crossing narrow nadir-only corridor or orthogonal survey line (N-S aligned rectangle represents area from from data in figure below is derived)

showing cross section through instantaneous soundings and DTM derived from nadir-only data. It is the bias and standard deviation of the instantaneous soundings w.r.t. the nadir-only gridded surface which is calculated (NOTE the mean DC bias for the whole swath is just the tide and is removed)

We can now look at the deviation of a set of intersecting, instantaneous sounding solutions of another sonar from the smoothed estimate and plot these as a function of beam number/grazing angle. The dominant gross difference will be due to systematic errors sources. For the short duration of the intersection of the two sonar solutions, the systematic error sources will result in fixed shift of bias for each of the beams. If this bias is removed, what we have left is an estimate of the short wavelength sounding errors from the system . These are dominated by bottom detection noise and imperfect short-period motion compensation (and of course any true seabed roughness at wavelengths shorter than the beam footprint).

The errors are separated into a mean bias and a standard deviation for each one degree bin :

the mean depth difference is calculated to estimate the systematic bias that can be a result of a number of factors:

In addition, because we have chosen a region in which there is minimal change in regional slope, any gyro bias will show up as an apparent roll bias. Also if there is a small systematic position offset error this will show up as a constant bias for all beams (swath displaced laterally on a regionally sloping surface).

Thus most sources of systematic error can be separated out into the mean bias. What is left is a mixture of the sounding detection noise and the short wavelength seabed roughness. Because, as stated above, we choose an area which we believe to be devoid of short wavelength roughness, the standard deviation, should predominantly reflect the noise (uncertainty) in the bottom detection process (either estimation of slant range for a given angle or estimation of angle for a given slant range).

While the bias estimates are relatively easy to understand and interpret in terms of sources of systematic error, care needs to be taken in the interpretation of the standard deviations. If the sonar system were perfectly calibrated (and positioning (3D) errors were minimised) then these standard deviation estimates might be used as a benchmark to see if the sonar is performing within the (rather ambiguously specified) IHO standards.

The standard deviation numbers, however, can be dramatically affected by the way in which the bottom detection algorithm is implemented. For the systems that use amplitude detection methods for discrete beams for each bottom detection, the statistics are generally valid. In contrast however, bottom detect solutions that are derived from phase measurements (either inter-row or within beam) can be very misleading.

For those methods relying on phase, there need not be any correlation between the number of phase estimates possible (controlled by digitising rate) and the number of independent bottom detection solutions that can be derived (controlled by the quantitisation of the phase, the noise levels, the pulse length etc..). In general, for these systems many more phase angle estimates can be provided than independent solutions. In this case, the final product may (or may not) be a running average or median filter of the solutions. Depending on the amount of filtering applied to the phase solutions, the standard deviation of the estimates will vary widely. Thus the standard deviation can be artificially high (reflecting oversampling in a noisy environment) or low (reflecting strong filtering).

In this case, a far more reliable estimate of the inherent bottom detection noise can be obtained by qualitatively viewing the finely gridded products (e.g.: SE rock outcrop or NW rock outcrop or Shallower rock outcrop ) . The sharpness with which edges and small targets are resolved is a better indicator of the limit of resolution of the sonar, than statistics of the large volume of (perhaps not independent) phase angle estimations.

Plots provided show the standard deviation and bias of the soundings as a function of grazing angle. All plots are provided with the same scale.

Sounding Density/ Coverage Patterns

Swath bathymetric sonars are commonly presented as the hydrographic tool of the future in part because they have the potential to provide: "Full Bottom Coverage". Conventional single beam sounding patterns are generally inadequate to provide this coverage. This term has now been proposed as a qualifier for the latest draft of the IHO S-44 document. With the more rigorous definitions for this term, however, most of the swath bathymetric sonars operational today would not strictly provide 100% coverage at the 3dB level at survey speeds above 10 knots. This looks at the performance of the sonar demonstrated under these rigorous conditions.

This section is an extension of "Are you really getting full bottom coverage?" (a technical discussion to accompany the recent paper in the Hydrographic Journal entitled "How effectively have you covered your bottom?").

To graphically demonstrate the effect of motion during forward propagation on swath coverage patterns, a series of coverage plots have been generated using a common motion time history. For all the models a constant velocity of 12 knots is used. The vessel actually propagates in a straight line (i.e.: no surge or sway), but is free to rotate about any of the three axes and to heave up and down. The water depth selected is 20m and the sonar model uses the angular sector, beam widths and ping repetition rates observed from the swath sonars demonstrated at the USCHC course.

The image of the seafloor exhibited is a patch of dimension 200 x 200m. The image presented is a 1200 by 1200 pixel representation of the seafloor with a pixel dimension of 17 by 17cm.

Areas coloured blue were not ensonified within the 3dB limits. Areas in greyscale were ensonified, the greyscale representing the range -3dB (black) to 0dB (white).

MOST IMPORTANTLY, the ping rates used for each of the presented models is that observed to be used by the sonar in 20m of water at the USCHC 1997 Multibeam Course in June 1997.

Figure: 32 second Motion Time Series

Figure showing the 32 second time series of roll, pitch yaw and heave used while the model sonar propagates over a 200 by 200m square of the seafloor at 20m depth. (The time series is in fact a real sample generated from a 36m vessel in sea state 4).

multiple overlapping views

You will note that, for many of the sonars, the ability to resolve short wavelength features (such as bedrock ridges and boulders) varies as a function of grazing angle. For a single swath, one cannot directly compare this as the two regions (high and low grazing angle or near and far range) are mutually exclusive.

For most of the data provided, there is sufficient overlap to view the same target from different parts of the swath. For some of the sonars, views at different azimuths are also available.

I find that such comparative plots enable the observer to assess graphically the variation in target resolution that the sonar is realistically capable of. In addition, because the same bedrock outcrop pattern is viewed by all the sonars, the relative spatial resolution capability of each of the sonars can be directly compared and contrasted.

It is important to bear in mind the angular sector covered by each sonar. When contrasting two data sets, one should compare the sonar performance at the same grazing angle. i.e.: don't compare the far range data of a wide swath system directly with the near nadir data of a narrower swath system.

The sonars use a wide variety of differing bottom detection types that are implemented over different angular sectors. By looking at the closeup sun-illuminated images it is often possible to see the transition zone between different bottom detection methods. Similarily it is easy to view the noise level in the sounding data as one varies slant range and grazing angle.

back to the USCHC97 index.

Last modified December 28th, 1997, by John E. Hughes Clarke (jhc@omg.unb.ca)

*Figure: 32 second Motion Time Series*
	Figure showing the 32 second time series of roll, pitch yaw and heave used while the model sonar propagates over a 200 by 200m square of the seafloor at 20m depth. (The time series is in fact a real sample generated from a 36m vessel in sea state 4).

Guide to the Creation and Interpretation of the Comparative Data Products

A comparison of swath sonar systems, demonstrated at the 1997 US/Canada Hydrographic Commission Coastal Multibeam Surveying Course

Ocean Mapping Group, Dept. Geodesy and Geomatics Engineering University of New Brunswick, CANADA 506-453-4568 (ph.), 506-453-4943 (fax.), jhc@omg.unb.ca