California Department of Fish and Game

Nearshore Ecosystem Database Project

Rikk Kvitek, Pat Iampietro, Eric Sandoval, Mike Castleton, Carrie Bretz,

Tally Manouki and Amanda Green

___________

SIVA Resource Center

Institute for Earth Systems Science and Policy

California State University, Monterey Bay100 Campus Center, Seaside, CA 93955

(831) 582-3529

• Assessment of habitat change due to natural or human impacts (e.g. climate change, oil spills, trawl disturbance)
• Monitoring and protecting important habitats (e.g. marine reserves, spawning areas, harvest closure areas)
• Design and location of marine reserves or aquaculture projects
• Species distributions and stock assessment

• Clear statement of purpose for the mapping project (e.g. well defined goals and objectives).
• Selections of scales for map extents and data resolution appropriate to the stated purpose.
• A universally accepted and broadly applicable hierarchical habitat classification system based on spatially nested physical and biophysical characteristics that control where species live.
• A means for acquiring data at appropriate resolutions and spatial scales for each of the relevant habitat characteristics.
• A means for combining, analyzing and displaying geospatial data sets collected in diverse formats, and at different scales and resolutions such that the habitat classification system can be applied.

1. Identify, collect, evaluate and convert all existing seafloor substrate and bathymetry data to digital GIS format for habitat classification. Special emphasis should be given to the 1986 Geology Maps of the California Continental Margin compiled by the USGS and California Department of Conservation Mines and Geology.
2. Convene a strategic planning workshop involving all parties having a vested interest in mapping California continental shelf habitats to:

• Identify opportunities for leveraging resources, combining missions and sharing data
• Define and adopt a universally applicable habitat classification scheme
• Develop criteria and standards for prioritizing sites to be mapped
• Develop criteria and standards for selecting mapping methods, scale and resolution
• Develop a prioritized list of sites to be mapped
• Draft a mission statement and strategic plan for funding

1. Create an initial set of "baseline" habitat maps for the continental shelf by applying the adopted classification scheme to existing seafloor habitat data in GIS format. The 1986 Geology Maps of the California Margin offer an ideal starting point.
2. Ground truth these baseline maps for accuracy and value.
3. Pursue in-house and multi-agency funding and support to carry out a strategic plan for mapping the habitats of the California continental shelf over the next decade
4. Develop partnerships with universities and other resource agencies as cost effective means for acquiring new data and developing new methods for data analysis and display.
5. Evaluate new technologies for more efficient and higher resolution habitat mapping in shallow nearshore environments. Testing these new techniques at sites where conventionally acquired data is already available or acquired simultaneously would be a logical first step in the assessment process.
6. Build up expertise and infrastructure for GIS analysis within the DFG marine group to make use of newly acquired and reprocessed geospatial habitat data.
7. Use GIS to combine geophysical habitat data (depth, slope, aspect & substrate) with new and existing species distribution and fishery data to test and refine the habitat classification scheme.
8. Explore links with NOAA and the military to reprocess existing data as well as collect new habitat data needed to complete the strategic plan.

A habitat is the place where a particular species lives or biotic community is normally found, and is often characterized by the dominant life form (e.g. kelp forest habitat) or physical characteristics (e.g. rocky subtidal habitat). Because habitats are repetitive physical or biophysical units found within ecosystems the same habitat may be found within different biogeographical provinces. Habitat mapping is typically undertaken by resource agencies to serve a variety of purposes including:

• Assessment of habitat change due to natural or human impacts (e.g. climate change, oil spills, trawl disturbance)
• Monitoring and protecting important habitats (e.g. marine reserves, spawning areas, harvest closure areas)
• Design and location of marine reserves or aquaculture projects
• Species distributions and stock assessment

While most subtidal species and resources can only be sampled directly using observational or other large scale (>1:10,000) survey techniques, it is often unreasonable to apply this level of effort to the entire coast of California. A major goal of habitat mapping, therefore, is to develop the ability to predict the distribution and abundance of species and resources from those physical and biotic parameters that define where species live and which can be remotely sampled.

The geographic limits to the distribution of many marine species result from barriers to migration, reproduction or survival. These biogeographic barriers result in ranges within which a species or community assemblage are likely to occur within the same habitat types. The habitat types can be defined in terms of those variables that control where a species lives within its range. Habitat parameters important to the distribution and abundance of benthic and nearshore species include:

• Water depth
• Substrate type
• Rugosity
• Slope/Aspect
• Void Abundance, Type & Size
• Sediment Type & Depth
• Exposure
• Vegetation
• Water Chemistry
• Water Temperature
• Biotic Interaction

Because the response of different species often varies with the spatial extent of these parameters, habitat scale is another factor important in defining where different species and biotic communities are likely to be found. For this reason, a benthic habitat classification system useful for defining species/habitat associations based on the parameters listed above, must also be hierarchically organized according to relevant spatial scales (see Habitat Classification Systems below).

Given these considerations, a successful, regional habitat mapping program needs to include the following elements:

• Clear statement of purpose for the mapping project (e.g. well defined goals and objectives).
• Selections of scales for map extents and data resolution appropriate to the stated purpose.
• A universally accepted and broadly applicable hierarchical habitat classification system based on spatially nested physical and biophysical characteristics that control where species live.
• A means for acquiring data at appropriate resolutions and spatial scales for each of the relevant habitat characteristics.
• A means for combining, analyzing and displaying these various geospatial data sets collected in diverse formats, and at different scales and resolutions such that the habitat classification system may be applied.

Each of these elements is discussed in the following sections. In Section 2.2 we give a brief overview of the purposes for and general approach to benthic habitat mapping. We then cover some of the issues pertaining to scale and georeferencing habitat data in Section 2.3. Requirements and recommendations for a suitable benthic habitat classification system are discussed in Section 3. We then review and provide examples from a wide range of habitat data acquisition methods in Section 4, covering the advantages and limitations of standard methods as well as those of emerging new technologies. Information on specifications, manufacturers, and service providers using these data acquisition tools have been compiled into an extensive database, and summarized in tables presented in Section 5.

In our discussion of the types of final product options available for habitat mapping projects in Section 6, we give only a brief overview of the various approaches available for data fusion, analysis and display of habitat data. Recent advances in Geographic Information Systems (GIS) have now brought spatial data analysis and display capabilities to virtually every desk top computer. While we use GIS extensively in our own habitat mapping work, and will make use of several of our GIS products as examples in this report, we will leave the review and assessment of GIS systems and applications to other authors. This decision is consistent with DFG's request that we focus our efforts on reviewing the specific technologies for the acquisition and classification of seafloor substrate and depth data.

There are four key considerations related to the display and georeferencing of habitat data:

• The scales at which the data are to be displayed and applied
• The selection of base maps to which the data are to be georeferenced
• The methods and objects used to depict the data (raster imagery, points, lines and areas).
• The coordinate system, datum and projection the data are to be used or displayed in.

With the advent of geographic information systems (GIS) it is now possible to merge, layer and display virtually all geocoded habitat data at any desired scale. Unfortunately, data collected at one scale may lose its meaning when displayed at a scale that is inappropriate for either the resolution (spatial density) or extent of the data set. Thus, while data collected at a particular resolution within a given area may be adequate for one purpose, it may not be suitable for other habitat mapping needs. For example, polygon features representing habitat classes measuring < 100 m² within a small coastal marine reserve can be accurately displayed at large map scales (>1:10,000). These same features will shrink to lines, points or disappear entirely on smaller scale maps (< 1:50:000) such as those used for displaying the regional distribution of fisheries or habitats (Table 2.1). Although GIS can circumvent this issue of display scale to some extent by providing the user with the ability to zoom in and out, the utility of hardcopy products are severely effected by the scale of display.

Table 2.1 Standard mapping scales and resulting display resolutions (adapted from Booth et al. 1996, and Greene et al., in press).

There is also the relationship between map scale and data resolution. While it is possible to collect high-resolution data over vast areas, the cost of doing so, and the size of the resulting data sets may be impractical if the primary purpose is to provide a regional overview of gross habitat types. Consequently, the selection of map scale depends on two factors: 1) the scale of the base map to be used (see below) and 2) the purpose of the study.

Table 2.2 General categories of methods for sampling coastal subtidal habitats and the scales at which they can be used (after Robinson et al. 1996).

Because the way in which a variable is sampled will affect the scale at which it can be meaningfully displayed or classified, it is important to match how habitats are sampled with the overall scale of the project. Robinson et al. (1996) reviewed the sampling methodology presently available for sampling subtidal environments (Table 2.2).

California coastal habitats within the 0 - 30 m depth range exist within a narrow zone often extending no more than a kilometer from shore. As a result, many of the coastal features such as reefs and islands are lost at smaller mapping scales (<1:100,000) and must be mapped and displayed at larger scale.

The published California Continental Margin maps (Greene and Kennedy 1986) drawn at the 1:250,000 scale, show the major geophysical seafloor features for the California continental shelf. While the sediments and substrate types depicted on these maps are relevant to the classification of marine habitats, the scale at which they are depicted limits their utility within the shallow subtidal. At this scale, habitat elements within the 0-30 m depth range are reduced to line features at best. These maps are nevertheless an excellent reference data set for megahabitat or regional scale habitat mapping, and correspond to the 1:250,000 mapping scale recommended as a standard for mapping coastal resources at the "Provincial" (regional) scale in Booth et al.'s 1996 technical report to Fisheries and Oceans Canada. Larger map scales (>1:50,000), however, are required for mapping and displaying most of the habitat features within the 0-30m depth zone.

Even at the larger mapping scale of 1:50,000, important coastal habitat features such as kelp forests, offshore rocks and reefs become reduced to one dimensional line features rather than polygons. More appropriate for nearshore habitat mapping of coastal features is the 1:24,000 scale common to the USGS topographic 7.5 minute quadrangle maps. This scale and set of map boundaries have already been used to provide the base maps for:

• U.S. Fish and Wildlife National Wetlands Inventory
• USGS digital ortho quads (DOQ)
• California coastline maps used by DFG, the California State Lands Commission, the TEALE data center, and the California Coastal Commission

At this scale, features down to 24 m in linear dimension can be easily depicted. Given the wide application of the 7.5 minute quad scale and footprint, we recommend its extension to nearshore coastal habitat mapping at the local scale.

Much of the physical detail important to many species occurs at the meter and sub-meter scale (e.g. substrate texture, grain size, void spacing and size). As a result, data collection and mapping capable of depicting this detail is critical to habitat classification at the Macro- and Micro-habitat scales (Figs. 2.1 and 2.2).

Figure 2.1. Biological microhabitats of hydrocorals and sea anemones with lingcod (Ophiodon elongatus) and young of the year rockfish (Sebastes spp.) on top of rock pinnacle mesohabitat (photo courtesy of Greene et al. in press).

Figure 2.2. Examples of Micro- and Macro-habitats. (Left) Pebble microhabitat in offshore Edgecumbe lava field, southeast Alaska (Greene et al. in press). (Right) Crevice in the Pliocene Purisima Formation that has been differentially eroded along the walls of Soquel Canyon, Monterey Bay, California (photos courtesy of Greene et al. in press).

Secondly, not all "GIS" type programs are capable of accurately merging data having different geodetic parameters. ArcView, the most popular GIS viewer program, cannot be used to reproject geospatial data. Once an ArcView project file has been created for a specific set of geodetic parameters, only those data sets stored in the same coordinate system, datum and projection as the project file can be accurately added as a theme. Here again, while it may be possible to import data sets having different geodetic parameters into ArcView as themes, they will not be correctly georegistered. ArcView, however, is a rapidly evolving program, and may eventually have the ability to reproject and co-register data from different projections, datums and coordinate systems. Until this capability is added, data will have to be initially collected or reprocessed using a true GIS program to be compatible with existing ArcView data sets. This consideration is especially important when sharing data between organizations using different geodetic parameters for their geospatial products and data.

Habitat mapping is being increasingly relied upon by resource management agencies as a tool for predicting the real or potential distribution of species or communities that are difficult to survey directly. To facilitate effective data sharing between organizations seeking to leverage their resources, a single, universal benthic habitat classification system is needed to insure that results from different studies can be efficiently and effectively combined.

Booth et al. (1996) have identified the following principles that should be included in a subtidal habitat classification system:

• Subtidal habitats must be identifiable, repeatable environmental units, divided into types or classes.
• Classes must represent the full range of subtidal habitats located within the region to be mapped.
• The classification system must be of use to resource managers. Classes must have biological meaning so factors that determine the biotic community structure (or those that control suitability of the habitat for a particular biotic resource) should be incorporated into the classification scheme, preferably at as high a level as possible.
• The classification system must be hierarchical with application at various scales depending on the intended use and data sources. The top levels must be based on characteristics that can be mapped at a small scale using remote sensing methods and will define the boundaries within which other levels are subdivisions.
• All types of sampling techniques should result in the same habitat classes or community definitions. The level to which a habitat can be classified will, however, be determined by the resolution of the sampling technique.
• The classification system should recognize time scales over which variables change. Habitat variables that change over shorter time scales should be incorporated at a lower level than variables that vary over longer time scales. For example, rock substrate changes over a longer time frame than sediment type, which changes less rapidly than kelp canopies or eel grass beds.
• The system must attempt to incorporate established classifications wherever possible to aid in the incorporation of existing data sets and compatibility with other studies.
• The system must be able to respond to foreseeable changes in information requirements and advances in processing and presentation technology.
• The system must be sensitive to existing sampling programs and be able to respond to foreseeable advances in data collection methods.

Here we present two example classification schemes developed for the subtidal environment. The system proposed by Booth et al. (1996) for the shallow subtidal habitats of British Columbia, Canada incorporates those classes found to be in current usage (Table 3.1). The more broadly applicable and detailed subtidal habitat classification system being developed and applied by Greene et al. (in press) also satisfies virtually all of principles listed by Booth et al. (1996). We present this latter scheme here as an example and possible starting point for the development of a universal benthic habitat classification protocol, and one ideally suited for nearshore marine habitat classification in California.

Table 3.1. Proposed physical habitat variables with examples of habitat classes for creating a coastal subtidal benthic habitat classification system (Booth et al. 1996).

The horizontal resolution, or posting, of singlebeam acoustic data is defined by the sampling interval along the track lines and the spacing between track lines. Because it is generally impossible or too costly to space survey lines as close together as the interval between soundings along the track lines, most older bathymetry data sets tends to have much higher resolution along track than across track. This situation necessarily leads to considerable interpolation between track lines when constructing contours or gridded DEMs. As a result, the DEMs are generally either too course (postings at > 50m) or inaccurate for fine grain mapping at macro- or micro-habitat scales.

One advantage of single beam depth sounders however, is the ability to interface them with acoustic substrate classifiers. These co-processors correlate the intensity values from the single beam echo returns with seafloor substrate hardness and roughness.

The most accurate method of bottom classification is that of in situ testing. Direct observations by SCUBA divers, drop or ROV video, or submersible provide substrate classifications with very high confidence levels, as do grab samples or cores; the latter two methods are especially useful for classifying sediments. However, application of these high-resolution, high-confidence methods of substrate classification in large area mapping projects can be quite costly in terms of money and effort. While class resolution of core and grab samples can be extremely high, the samples must be very closely spaced in order to give appreciable spatial (x,y) resolution. Similar obstacles exist for application of direct visual observation or video imagery to large areas; because of the limitations imposed by visibility underwater, cameras and/or observers must be placed in close proximity to the seabed that is to be classified, and achieving good bottom coverage becomes logistically difficult. In essence, drop camera samples are analogous to cores and grabs in that they are point samples, while ROV and submersible observations and video surveys may provide swath or area information within the visibility and physical range limits of their traveled course. Logistical constraints (in terms of cost, equipment required, support, etc.) can be quite high for ROV and especially submersible work. Towed camera systems may offer a considerably lower cost alternative to ROV or submersible observations while giving greater aerial coverage than drop cameras, but are also difficult to deploy in complex bathymetric settings, owing to the fact that they must be "flown" quite near the bottom due to visibility limitations. Over relatively flat bottom, or with very good visibility, however, these systems may be quite useful. All of these factors make direct observation of bottom type a much more appropriate tool for groundtruthing classifications derived from a remote sensing method with higher efficiency in covering large areas and lower cost per unit effort. Indeed, groundtruthing using the above methods is crucial when employing remote sensing techniques. In addition to providing greater coverage efficiency, bottom classifiers can help automate the classification process to some degree, especially relative to the human interpretation that must be applied to sidescan sonar or video imagery in order to map large areas. The primary means of remotely sensing and classifying substrate in the marine environment are acoustic methods.

The following text discussing acoustic substrate classifiers is drawn primarily from "Bottom Sediment Classification In Route Survey" (Mike Brissette, Ocean Mapping Group, Department of Geodesy and Geomatics Engineering, University of New Brunswick, http://www.omg.unb.ca/~mbriss/BSC_paper/BSC_paper.html#Bottom Sediment Classification). Additional text has been added, but the bulk of this section is quoted directly from that report.

This section will discuss two such sonars, namely Marine Micro System's 'RoxAnn', and Quester Tangent's 'QTC View'. Each discussion will look at the theory of operation behind each sonar as well as performance size requirements and costs.

RoxAnn is manufactured by Marine Micro Systems of Aberdeen Scotland. RoxAnn uses the first and second echo returns in order to perform bottom sediment classification. The first echo is reflected directly from the sea bed and the second is reflected twice off of the seabed and once off of the sea surface (Fig. 4.4). This method was first used by experienced fishers using regular echo sounders [Chivers et al, 1990]. The fishers observed that the length of the first echo was a good measure of hardness in calm weather.

Figure 4.4. Diagrammatic representation of first and second returns (from Chivers et al, 1990).

The second echo, which mimicked the first echo, was much less affected by rough weather. RoxAnn uses two values, E1 and E2, in order to estimate two key parameters of the sea floor, namely roughness and hardness. The first echo contains contributions from both sub-bottom reverberation and oblique surface backscatter from the seabed. It has been shown that oblique backscattering strength is dependent on the angle of incidence for different seabed materials. At 30 degrees there is almost a 10 dB difference in scattering level between mud, sand, gravel and rock [Chivers et al, 1990]. The first part of the first echo contains ambiguous sub-bottom reverberations and is therefore removed (Fig. 4.5). Most or all of the remaining portion of the first echo is then integrated to provide E1, the measure of roughness. The exact parameters within which E1 is integrated are difficult to estimate and is therefore based on empirical observations in a number of different oceans [Chivers et al,1990]. The entire second echo is integrated, which is the relative measure of hardness and is designated E2 [Schlagintweit, 1993]. A processor is used to interpret E1 and E2 such that bottom characteristics may be determined [Rougeau, 1989]. Looking at E1, on a perfectly flat sea floor, non incident rays would be expected to reflect away from the transducer. As the sea floor is not perfectly flat, the returning energy from non incident rays coincides and interferes with the incident rays and indicates the roughness of the sea floor [Chivers et al, 1993]. The specular reflection of the sea floor is a direct measurement of acoustic impedance relative to the sea water above it. Hardness can be estimated using E2 because the acoustic impedance is a product of the density and speed of longitudinal sound in the sea bed [Chivers et al, 1990].

Figure 4.5. First and Second Return Waveforms (from Schlagintweit, 1993)

Schlagintweit [1993] conducted a field evaluation of RoxAnn in Saanich Inlet off of Vancouver Island using two frequencies, 40 kHz and 208 kHz. RoxAnn was deployed over a ground-truthed area that had been previously inspected by divers. A supervised classification method was used and a "modest" correlation was found at both frequencies. Classification differences between the two frequencies were due to the different sea bed penetration depths of these frequencies on various sea floor types. That is, the frequency dependent penetration factor into the sea floor depended on the local sea floor itself. Schlagintweit felt that the frequency should be chosen according to the application. Schlagintweit believed that an unsupervised classification method would be the best alternative, i.e., let the system select the natural groupings and then look at ground truthing. Both the Chivers et al [1990] and Rougeau [1989] articles support this method of an initial calibration. In separate tests, Kvitek et al [in press] found quite good agreement between classes created from sidescan sonar interpretation and those created using unsupervised classification of RoxAnn E1 & E2 values at the Big Creek Ecological Reserve in Big Sur, CA (Fig. 4.3). Using sidescan imagery and video groundtruthing, Kvitek et al found that RoxAnn successfully classified sand, rock, and coarse sand/gravel between 6-30m depth in a 2-3 sq. km area in this study.

The RoxAnn system is very compact. The entire unit consists of a head amplifier (not shown) which is connected across an existing echosounder transducer in parallel with the existing echo sounder transmitter, and tuned to the transmitter frequency. The parallel receiver accepts the echo train from the head amplifier [Schlagintweit, 1993]. The installation requires no extra hull fittings, simply room for the processing equipment. The required processing equipment includes an IBM compatible computer and an EGA monitor [Rougeau, 1989]. Software which is specifically written to handle RoxAnn data must then be installed on the computer for processing analysis. The RoxAnn Seabed Classification System retails for about $15,000 US and the additional RoxAnn software costs about $10,000 US. Other programs such as Hypack, which retails for US$ 11,000, are also compatible with the RoxAnn hardware [Clarke, 1997]. These prices do not include taxes, installation expenses or services of a technician for calibration and sea trials.

QTC View is manufactured and distributed by Quester Tangent Corporation of Sidney, BC [Quester Tangent Corporation, 1997]. Like RoxAnn, Quester Tangent's QTC View uses the existing echo sounder transducer; however, QTC View does not examine two different waveforms. Instead, analysis is performed on the first return only. Quester Tangent's other classification system ISAH-S (Integrated System for Automated Hydrography) is also available, and uses the same approach as QTC View in wave form analysis. However, ISAH-S offers multiple channels for multi-transducer platforms, integration with positioning and motion sensors, and helmsman displays. QTC View is more of a standalone system accepting GPS input for georeferencing of echo sounder data. QTC View operates in the following manner. First, both the transmitted echo sounder signal and return signals are captured and digitized by QTC View. Second, the sea bed echo is located (bottom pick), and an averaged echo from several consecutive returns is computed [Prager 1995]. Next, the effects of the water column and beam spreading are removed such that the remaining wave form represents the seabed and the immediate subsurface [Collins et al, 1996]. Quester Tangent's echo shape analysis works on the principle that different sea beds result in unique wave forms. Through principal component analysis, complex echo shapes are reduced into common characteristics. Each wave form is processed by a series of algorithms which subdivides it into166 shape parameters [Collins et al, 1996]. A covariance matrix of dimension 166 x 166 is produced and the eigen vectors and eigen values are calculated. In general, three of the 166eigenvectors account for more than 95 per cent of the covariance found in all the wave forms. The 166 (full-feature) elements of the original eigen vector are reduced to three elements ("Q values"). These reduced feature elements will cluster around locations in reduced feature space corresponding to a sea bed type [Prager, 1995]. Test Results QTC View was designed to operate in both the supervised and unsupervised classification modes. If no ground-truthing has taken place in an area of interest, QTC View will still cluster-like areas such that some type of calibration or ground truthing may be performed after the survey. In a test conducted by the Esquimalt Defense Research Detachment, QTC View was found to have produced very good results. QTC View was used over the same area where the RoxAnn tests were conducted off of Vancouver Island in the unsupervised classification mode. QTC View was able to discriminate between eight different seabed types. After a calibration, QTC view was found to agree with each ground truthed area and showed good transition from seabed type to seabed type [Prager, 1995].

QTC View is comprised of a head amplifier and PC with a DX2/66 processor. The head amplifier is connected in parallel across the existing transducer and to the PC via a RS232 cable. The PC also accepts the GPS data in NMEA-0183 standard GGA or GGL format for georeferencing of data [Collins et al, 1996]. The PC displays three windows: one for the reduced vector space, one for the track plot and classification and the third for seabed profile and classification. Figure 4.6 illustrates the QTC View screen output.

Figure 4.6. QTC View Screen Display (from Quester Tangent, 1997)

QTC is presently working with Reson, Inc. on adaptation of QTC View for use with multibeam depth sounders. This development will greatly increase survey efficiency by supplying substrate class data over most or all of the multibeam swath, but it is unknown when this product will be available. At present, however, QTC View will work with the Reson 8101 multibeam head, although it uses only the nadir beam data. QTC View retails for approximately US $15,000 [pers com J. Tamplin] [Lacroix, 1997] whereas ISAH-S retails for approximately $35,000 [Collins, 1997]. Unlike RoxAnn, the QTC View purchase price includes the software, and like RoxAnn the user must supply the computer. Hypack is not yet capable of acquiring raw QTC View data, but Coastal Oceanographics has provided support for recording the reduced dataset (3 "Q" values) processed in realtime by QTC view. The above prices do not include taxes or installation.

Both products discussed above have been shown to be useful tools for acoustic bottom substrate classification. The levels of success achieved in past studies using these tools is a function of the inherent qualities of the tools themselves, the operator and processor/analyzer expertise of those involved, the methods used, and the specific conditions of the areas studied. For this reason, true between-product comparisons are difficult. By far the most important fact to remember when using either of these tools (or any remote sensing method, for that matter) is that classifications created using these methods must be groundtruthed using one of the direct observation methods discussed above. Only with independent verification can confidence be placed in remotely sensed data.

During the last 10-15 years, use of multibeam bathymetry in hydrographic mapping has become increasingly common and accepted. Initially fraught with considerable accuracy and precision issues, multibeam sonar technology has improved vastly and rigorous testing has established its reliability. The ability to acquire denser sounding data while surveying fewer tracklines (with greater spacing between lines), and simultaneously acquiring backscatter imagery using the same sensor, has made multibeam a popular tool. Using this technology, however, requires attention to a number of considerations that are less crucial when using single-beam technology.

Multibeam depth sounders, as their name implies, acquire bathymetric soundings across a swath of seabed using a collection of acoustic beams (Fig. 4.7 left), as opposed to a single beam, which ensonifies only the area directly below the transducer. The number of beams and arc coverage of the transducer varies among makes and models, and determines the swath width across which a multibeam sounder acquires depth measurements in a given depth of water (Fig. 4.7 and 4.8). It is important to note that effective swath width is often somewhat less than potential swath width, as data from the outer most beams is often unusable due to large deviations induced by ship roll and interference from bottom features such as pinnacles. The potential swath width shown in Figure 4.8 may only be realized under calm conditions over a relatively flat bottom. Swath width is depth dependent, requiring closer line spacing in shallower water if full coverage is to be maintained. The mechanics and physics of how the beams are formed varies as well among makes and models, and may be a consideration of importance if extremely high resolution, precision, and accuracy are required.

In order for the multibeam system to calculate accurate x, y, and z positions for soundings from all off-nadir (non-vertical) beams (every beam other than the center beam), precise measurement of ship and transducer attitude is required. This includes measurement of pitch, roll, heading, and (preferably) vertical heave. Thus, a motion sensor must be interfaced to the unit, so that its output may be used to adjust and correct the multibeam data in either real time or post-processing.

Figure 4.8. Relationship between multibeam bathymetry transducer beam angle and swath coverage. For example: with a 90 degree beam angle swath width will be twice the water depth.

In addition, because of longer travel times for off-nadir beams, variations in the speed of sound in water (SOS) can induce relatively large errors in these beams; especially if temperature stratification exists in the water column. For this reason, sound velocity profiling should be conducted on site during a survey, and the SOS data used to adjust depth soundings. Controlling for variations in SOS is of increasing importance as depth increases. Multibeam surveying also requires more rigorous system calibration to account for systemic variations in, and improve the accuracy of, heading, roll, and pitch sensor values, as well as any adjustment to navigation time tags that will reduce timing errors between navigation and sonar data. This calibration, known as a "Patch Test", is typically conducted by running a series of survey lines over the same area with relative orientations that allow assessment of the variables listed above.

Multibeam bathymetric surveying generates orders of magnitude more data than single-beam surveying, resulting in greater storage requirements, longer processing times, and the need in some cases for greater processing power. Gigabytes of data may be generated daily, (as opposed to megabytes in single-beam surveys), especially if backscatter imagery is being recorded as well. The removal of bad sounding data during the editing process is, accordingly, a much larger task in multibeam than in single beam surveys, although some processing packages allow some degree of automation of this process.

The considerations and requirements listed above make multibeam surveying a much more complex and expensive undertaking relative to single beam, but the benefits in cost per unit effort and resolution can well outweigh the hardships, especially if extensive surveying is planned. Survey speeds of up to 30 knots are now possible with some systems. Minimal costs for setting up a multibeam system range from $75,000-$150,000 US for equipment alone, not including vessel, installation, and maintenance costs. Higher precision equipment with greater capabilities and more features can cost substantially more.

Sidescan sonar is the only technology capable of producing continuous coverage imagery of the seafloor surface at all depths. (Blondel and Murton [1997] give an excellent and comprehensive review of sidescan sonar theory, technology, imagery and application in their recent book, Handbook of Seafloor Sonar Imagery.) These systems transmit two acoustic beams, one to each side of the survey track line. Most sidescan systems use transducers mounted on a towfish pulled behind the survey boat (Fig. 4.2 & 4.9), but some are hull mounted. Because towfish can be deployed well below the water’s surface, they can be used in deeper habitats than hull mounted systems.

Sidescan sonar beams interact with the seafloor and most of their energy is reflected away from the transducer, but a small portion is scattered back to the sonar where it is amplified and recorded. The intensity of the backscatter signal is affected by the following factors in decreasing order of importance:

• Sonar frequency (higher frequencies give higher resolution but attenuate more quickly with range than lower frequencies)
• The geometric relationship between the transducer and the target object (substrate slope)
• Physical characteristics of the surface (micro-scale roughness)
• Nature of the surface (composition, density)

Figure 4.9. Klein sidescan sonar towfish about to be deployed from stern of survey vessel, and Klein 595 recorder printing hardcopy image (sonograph) of seafloor. Note black, port transducer running down the left side of the towfish (Klein Associates).

For each sonar pulse or ping, the received signal is recorded over a relatively long-time window, such that the backscatter returned from a broad swath of seafloor is stored sequentially. This cross-track scanning is used to create individual profiles of backscatter intensity that can be plotted along track to create a continuous image of the seafloor along the swath (Fig. 4.9).

Swath width is selectable but maximum usable range varies with frequency. High frequencies such as 500kHz to 1MHz give excellent resolutions but the acoustic energy only travels a short distance (< 100 m). Lower frequencies such as 50kHz or 100kHz give lower resolution but the distance that the energy travels is greatly improved (>300 m). Typical systems used for nearshore mapping have frequency ranges from 100 to 500 kHz with resolution as fine as 20 cm. Resolution also varies with swath width. Thus, while a 500 kHz system set at range of 75m will cover a 150m swath at 20 cm resolution, a 100 kHz system set at a range of 250m will cover a 500m swath but at a resolution closer to 1m. There is also a direct relationship between maximum allowable survey vessel speed and range. The shorter the range, the slower the speed and the more survey lines required to cover a given area. (Typical sidescan sonar survey speeds are around 4-5 knots, but with newer systems have been increase to 10 knots.) Thus, the trade-off between swath width, resolution, survey speed, and financial resources must be considered when planning a survey. The choices will depend on: 1) the size of the area to be surveyed, 2) what resolution of substrate definition is required, and 3) how much time and money is available for the survey. Interactive survey time estimate calculation tables such as the Hydrographic Survey Time Estimate Worksheet shown below can be easily constructed in a spreadsheet program such as Microsoft Excel. These tables can be used to construct what-if scenarios to explore the relative time requirements and costs for different survey parameters.

Another variable important to survey time is the amount of overlap desired between adjacent track lines. Most sidescan sonar systems cannot "see" the seafloor directly beneath the towfish. (Klein’s new multibeam sidescan system is an exception.) As a result, if complete coverage of the seafloor is required, it will be necessary to have up to 100% overlap of the sidescan swaths, such that the port side of swath along one track line is completely covered by the starboard side of the swath from the adjacent track line. In this manner, the outer range of one swath can be used to "fill-in" the missing inner-range of the adjacent swath during post-processing.

An additional advantage of designing overlap into the survey is to provide different views of the seafloor. This approach is especially important in areas of high relief, where features such as rock pinnacles may block the acoustic beam from striking and reflecting off that part of the seafloor hidden from towfish view. This interruption of the acoustic beam will create shadows or blind spots in the record, which can be filled with information from adjacent tracklines if there is sufficient overlap. Running track lines at different angles over the survey area can also be used to give a more complete picture of what the habitat looks like. For example, the acoustic appearance of canyons, pinnacles and exposed rock strata can vary greatly with approach angle.

Once the survey is completed, the swath images or sonographs can then be combined into a composite image or mosaic of the entire area surveyed (Fig. 4.10). Traditionally, these sonographs were created as hardcopy originals by the sidescan recorder, but are now more often recorded in digital form. As a result, all post-processing, including image enhancement, mosaicking and GIS product creation can be done electronically. Interfacing the sidescan with a differential GPS navigation system can produce georeferencing and imaging accuracy at submeter resolutions. To obtain this accuracy, however, requires that the off-set or "layback" between the sidescan sonar transducer and the GPS antenna is accurately determined and recorded throughout the survey.

Figure 4.10. Sidescan sonar mosaic of Big Creek Ecological Reserve, Big Sur, California produced with an EG&G 260 100 kHz towfish sidescan sonar system (authors’ unpublished data).

The sonographs and mosaics are used to create what is know as a sidescan interpretation. This process involves tracing polygons around regions of similar substrate as identified on the sonograph (Fig. 4.11). While it is relatively easy to differentiate between rock and sediment on the sonograph, caution must be exercised in the interpretation of the substrate based solely on the sidescan imagery if finer division of the substrate type is required (e.g. cobble, gravel, coarse sand, fine sand, silt, clay, etc.). As a result, it is often necessary to augment the sidescan data with some form of direct sampling (scuba, video, ROV, bottom grabs, etc.) in order to groundtruth the interpretation.

Groundtruthing is especially critical when image analysis software first developed and refined for use with satellite imagery is used to automate the classification and interpretation of the sidescan imagery. Classification involves identifying different features or classes in an image based on their reflectance characteristics. There are two principal methods for performing a classification of an image. "Unsupervised classification" is a method for grouping pixels in an image into classes or "clusters", based on their statistical properties, without the user supplying any prior information on the classes. Once the unsupervised classification has been performed, the clusters that the classifier has identified can be examined and labeled according to what class they represent in the real-world as determined via groundtruthing.

"Supervised classification" involves the user first "training" the system in recognizing different classes by selecting representative samples of each class or habitat type from the image: these samples are known as training sets and should be groundtruthed prior to performing the supervised classification. The system then assigns each pixel in the image to one of these pre-determined classes. Some groundtruthing is essential for accurate classification results regardless of the method used. While highly effective in processing aerial imagery of terrestrial habitats, development of classification techniques is still in its infancy for application to acoustically derived images of marine habitats. These classification routines are available in stand-alone image processing software packages such as ERDAS and DIMPLE, as well as accessories or modules for some GIS software packages including those offered by ESRI and MicroImages.

Once processed and correctly georeferenced, the sidescan imagery and interpretations can also be draped over DEM’s to give a 3D representation of the seafloor (Fig. 4.12).

Figure 4.12. Sidescan sonar mosaic draped over DEM of Big Creek Ecological Reserve, (authors’ unpublished data).

Geohazards are also more of a consideration in shallow waters because towfish altitude above the seafloor is often limited by water depth. Towfish altitude should be kept between 10% and 40% of the range if full coverage of the selected swath width is desired. Less than 10% will result in loss of signal from the outside part of the range, and greater than 40% will produce a large gap in coverage directly below the fish. In water depths of > 40m a towfish could be kept up to 40m off the bottom while still maintaining a range of 100m on a side. This margin of safety is not available, however, in water depths of 10 to 30 m, where the towfish must be kept at least 10m off the bottom but cannot be raised more that the water depth. Thus, a 20m pinnacle in 30m of water presents a very serious hazard to sidescan operations. For this reason, it is always advisable to conduct a bathymetric survey prior to the sidescan work in areas of uncertain seafloor morphology.