A 1.What is the difference between supervised and unsupervised classification?
Unsupervised - Data are processed from a number of raw formats and
logged as a single class, and then clustered into different acoustic classes.
This provides unique flexibility in defining a classification scheme.
Supervised - If the bottom type is known before classification, data from
the areas of known sediment type can be used to build a catalogue, which would
then be used to classify subsequent or archived data.
A 2.What is Quester Tangent's approach to seabed classification?
Seabed classification can be done visually, mechanically, and acoustically. All visual methods (divers, video, photography) and mechanical methods (divers, grab samples, cores, probes) are slow and manually intensive, thus expensive and not suited to extensive survey work. Also, the visual methods are subjective. Acoustic methods, however, can cover large areas quickly as there is no need to stop the survey vessel. The power of acoustic seabed classification is the ability to apply visual or mechanical classifications over much larger areas than point data alone would allow; that is, the sediment properties obtained from the point samples can be applied with confidence over entire regions that have been mapped acoustically. In QTC software products, the data are processed using statistics and produce results that represent acoustic diversity.
A 3.Why do I have to ground truth the results?
Sediment classification using acoustics alone is possible only in specific and unusual situations. The information in seabed echoes and backscatter is not specific enough to allow one to state with confidence that the sediment is sand, mud, gravel, or whatever classes are of interest. The classification results represent acoustic diversity.
Ground-truthing is key in determining seabed type. Data collection provides information on characteristics and physical properties. There are numerous methods to collect information, the most common of which are presented below.
- Grab sampling, the simplest method, will likely be the primary source of ground-truth data. However, the limitations of grab sample data should be recognized and additional techniques employed. For example, grab samplers are not very useful in areas where there is a gravel armour - samples are often lost during recovery because the clasts catch in the jaws - and they are practically useless on bedrock. There are numerous types of samplers including Van Veen, Shepec and Eckman, all which penetrate less than 15 or 20 centimetres to collect loose sedimentary material from the uppermost portion of the seabed. At least two samples should be taken to ensure good representation.
- Coring works well with low frequency sounders (24-33 kHz) as the acoustic response can represent 60 centimetres or more subsurface penetration. Note that core data from previous investigations may be available to supplement the data set. There are numerous types of corers such as gravity corers and piston corers. A core barrel is dropped through the water column and into the seabed. Loose sedimentary material is forced into the tube and retained by a special valve or core catcher. Penetration, usually less than two metres, is highly dependent on substrate and becomes limited as grain size increases, making corers ineffective in gravel or cobbles. A vibrocorer can sample very coarse sand and small gravel, but this special corer requires extensive logistics.
- Video of the calibration area in conjunction with a series of grab samples is the most effective method. An underwater video camera mounted on a frame with a light source films and records the seabed. The camera must be stationary or towed at a very low speed for the process to work.
- Still photography is not practical during real-time calibration due to the delay in photographic processing. However, photographs can be valuable in a post-processing mode, particularly on surfaces such as bedrock outcrops where grab samples are ineffective. One camera, or two for stereo photography, is lowered to the seabed. The camera usually has a bar scale suspended below it acting as a trigger and providing a reference to estimate surficial sediments grain size.
A 5a).If groundtruthing has been done in a surveyed region, can the same information be used for further surveys of the same section over the next few years?
Yes, providing the same echo sounder is operating in the same way. It is particularly important that the pulse length be the same.
A 5b).And if sand (or other sediments) has the same backscatter if surveyed at the same depth and frequency but in different waterways?
No. It is important to realize that acoustic classification involves more than just grain size. A sediment is characterized by over 8 variables: mean grain size and its distribution (at least two variables), density ratio, sound speed ratio, surface roughness (two variables), attenuation coefficient, and volume scattering coefficient. At high frequencies surface scattering dominates, so surface roughness can be very important. Sand with a very smooth surface will backscatter quite differently from sand with ripples or waves. At low frequencies volume scattering can be important, so homogeneity of the sediment is emphasized. Pure sand will backscatter differently than sand with shell fragments and densely packed sand will backscatter differently that bioturbated sand. There is no frequency for which the grain size is the only important variable. For this reason, our classifications are based on 'acoustic diversity'.
