xml version 1.0 encoding UTF-8 standalone no
record xmlns http:www.loc.govMARC21slim xmlns:xsi http:www.w3.org2001XMLSchema-instance xsi:schemaLocation http:www.loc.govstandardsmarcxmlschemaMARC21slim.xsd
leader nam Ka
controlfield tag 001 001935419
007 cr mnu|||uuuuu
008 080423s2007 flua sbm 000 0 eng d
datafield ind1 8 ind2 024
subfield code a E14-SFE0002297
Dunn, Shane C.
Acoustic classification of benthic habitats in Tampa Bay
h [electronic resource] /
Shane C. Dunn.
[Tampa, Fla.] :
b University of South Florida,
ABSTRACT: The need for assessment of benthic habitat characteristics may arise for many reasons. Such reasons may include but are not limited to, habitat mapping, environmental concerns and identification of submerged aquatic vegetation. Oftentimes, such endeavors employ the use of aerial photography, satellite imagery, diving transects and extensive sampling. Aerial photography and remote sensing techniques can be severely limited by water clarity and depth, whereas diver transects and extensive sampling can be time consuming and limited in spatial extent. Acoustic methods of seabed mapping, such as the acoustic sediment classification system QTC are not hampered by water clarity issues. The acoustic sediment classification system QTC is capable of providing greater spatial coverage in fractions of the time required by divers or point sampling. The acoustic classification system QTC VIEW V(TM) was used to map benthic habitats within Tampa Bay.The QTC system connected in parallel to an echo-sounder is capable of digitally extracting and recording echoes returning from the seabed. Recorded echoes were processed using QTC IMPACT(TM) software. This software partitions echo waveforms into groups or classes based on their similarity to one another using multivariate statistics, namely Principal Component Analysis and K-Means clustering. Data was collected at two frequencies, 50 kHz and 200 kHz. Side-scan sonar data was collected coincident with the QTC data and used to produce mosaics of the various habitats in Tampa Bay. Side-scan sonar data was classified using QTC Sideview(TM) in an attempt to identify changes in benthic habitats. Sediment samples used for ground-truth were subjected to grain size analysis. Also, the percentage of organic matter and carbonate within samples was determined. Results of acoustic classification appear to accurately reflect changes in the sediment type and structure of the seabed.Grain size, particularly percent mud, appears to have a strong influence on classification. Carbonate hard bottom habitats were found to be acoustically complex, a characteristic useful for their identification. The QTC system was able to detect seagrass, although some misclassification occurred between vegetated and non-vegetated seabeds.
Thesis (M.S.)--University of South Florida, 2007.
Includes bibliographical references.
Text (Electronic thesis) in PDF format.
System requirements: World Wide Web browser and PDF reader.
Mode of access: World Wide Web.
Title from PDF of title page.
Document formatted into pages; contains 83 pages.
Advisor: Albert Hine, Ph.D.
x Marine Science
t USF Electronic Theses and Dissertations.
Acoustic Classification of Benthic Habitats in Tampa Bay Shane C. Dunn A thesis submitted in partial fulfillment Of the requirements for the degree of Masters of Science College of Marine Science University of South Florida Major Professor: Albert Hine, Ph.D. Stanley Locker, Ph.D. Lisa Robbins, Ph.D. Date of Approval: October 29, 2007 Keywords: QTC, echo classification, aqua tic vegetation, grain size, sediment, hard bottom Copyright 2007, Shane Dunn
Note to Reader The original of this document contains colo r that is necessary for understanding the data. The original dissertation is on file with the USF library in Tampa, Florida.
i Table of Contents List of Figures iii Abstract v Introduction 1 Background and Previous Work 3 Seafloor Acoustics 9 Methods 16 Acoustic Data Acquisition 16 Acoustic Sediment Classification (Data Processing) 16 QTC Sideview Acoustic Seabed Classification 20 Grab Samples 20 Grain Size Analysis 21 Pipette Analysis (Clay and Silt Size Fractions) 21 Carbonate Analysis 22 Organic Matter Analysis 22 Results 23 50 kHz QTC Data (Soft Sediments) 23 Class Complexity Analysis (50 kHz QTC Data) 31 200 kHz QTC Data (Soft Sediments) 33 Acoustic Classes 2, 3, and 5 33 Acoustic Classes 1, 6, and 7 35 Sediment Data and Acoustic Classification 35 QTC Sideview 40 Hard Bottom Surveys 41 50 kHz QTC Data Set (2006) 41 Acoustic Class 1 42 Acoustic Classes 2, 3, 4, and 5 43 Acoustic Class 6 44 Acoustic Class 7 44 50 kHz QTC Data Set (2007) 44 Acoustic Class 3 44 Acoustic Classes 1, 2, 4, 5, 6 and 8 45 Class Complexity Analysis (50 kHz 2007) 46 200 kHz QTC Data Set (2007) 47 Acoustic Class 5 48 Acoustic Class 2 49 Acoustic Classes 1, 3, 4, and 6 50 Sediment Data and Acoustic Classification (Hard Bottom Survey) 51
ii (Results continue) QTC Sideview 59 Submerged Aquatic Vegetation (SAV) Survey 60 Acoustic Class 1 60 Acoustic Class 2 60 Acoustic Class 3 61 Acoustic Class 4 62 QTC Sideview 63 Discussion 64 Echo Parameterization 64 Acoustic Classification of Soft Sediments (Safety Harbor) 66 Acoustic Classification of Carbonate Hard Bottoms 69 Acoustic Classification of Submerged A quatic Vegetation (SAV) 73 QTC Sideview 74 Assessment 75 Conclusions 77 List of References 78
iii List of Figures Figure 1. Locations of three survey areas 2 Figure 2. 50 kHz waveform and 200 kHz waveform 4 Figure 3. Graphs of porosity and density versus impedance and reflection coefficient 10 Figure 4. Graph of attenuation versus grain size 13 Figure 5. Graph of porosity versus grain size and attenuation 14 Figure 6. A cartoon showing a QTC ellipsoid 18 Figure 7. 3-dimentional Q-space 19 Figure 8. Map of all acoustic class in the Sa fety Harbor soft sediments 25 Figure 9. Grain size by percent weight for acoustic class 8 26 Figure 10. Grain size by per cent weight for acoustic class 4 27 Figure 11. Grain size by per cent weight for acoustic class 6 28 Figure 12. Grain size by percent weight for acoustic class 3 29 Figure 13. Side-scan sonar and 50 kH z data in the Safety Harbor soft sediments 30 Figure 14. Acoustic class complexity in the Safety Harbor soft sediments 32 Figure 15. Side-scan sonar and 200 kH z data in the Safety Harbor soft sediments 34 Figure 16. Distribution of gr ain size for acoustic class 8 36 Figure 17. Distribution of gr ain size for acoustic class 4 36 Figure 18. Distribution of gr ain size for acoustic class 6 37 Figure 19. Mean grain size 38 Figure 20. Percent sand, silt, clay and acoustic class 38 Figure 21. Percent mud (silt and Clay) 39 Figure 22. Carbonate, organics and mud by percent weight 39 Figure 23. Acoustic classification using QT C sideview in the Safety Harbor soft sediments 40 Figure 24. Acoustic class 1 42 Figure 25. Acoustic class 2, 3, 4, and 5 43 Figure 26. Acoustic class 3 45 Figure 27. Acoustic class 1, 2, 4, 5, 6, and 8 46 Figure 28. Acoustic class complexity in the Tampa Bay hard bottom 47 Figure 29. Acoustic class 5 200 kHz 48 Figure 30. Acoustic classes 2 and 7 ( 200 kHz) 49 Figure 31. Acoustic classes 1, 3, 4,and 6 50 Figure 32. Mean phi ( ) value for sediment samples collected in and around the Tampa Bay hard bottom 51 Figure 33. The relationship among mean phi ( ) and percent carbonate 52
iv Figure 34. Sediment grain size distribution within samples 52 Figure 35. The percentage of carbona te and remaindered material not dissolved by hydrochloric acid 54 Figure 36. The percentage of sand, mud, and gravel in sediments recovered from in and around the hard bottom 55 Figure 37. Percent calcium carbonate a nd gravel 56 Figure 38. Combined acoustic classes a nd grainsize 57 Figure 39. Combined acoustic classes and percent carbonate 58 Figure 40. Acoustic classification usi ng QTC sideview in the Tampa Bay hard bottom 59 Figure 41. All acoustic classes in the Northshore SAV 61 Figure 42. Side-scan sonar mosaic in th e Northshore SAV 62 Figure 43. Acoustic classification using QTC sideview in the Northshore SAV 63 Figure 44. 200 kHz acoustic class variability in the Safety Harbor soft sediments 69 Figure 45. Images depicting complexity in the Tampa Bay Hard Bottom 71
v Acoustic Classification of Benthic Habitats in Tampa Bay Shane C. Dunn ABSTRACT The need for assessment of benthic habitat characteristics may arise for many reasons. Such reasons may include but are not limited to, habitat mapping, environmental concerns and identification of submerged a quatic vegetation. Often times, such endeavors employ the use of aerial photography, satellite imagery, divi ng transects and extensive sampling. Aerial photography and remote sensin g techniques can be severely limited by water clarity and depth, whereas diver tran sects and extensive sampling can be time consuming and limited in spatial extent. Ac oustic methods of seabed mapping, such as the acoustic sediment classification system QTC are not hampered by water clarity issues. The acoustic sediment classification sy stem QTC is capable of providing greater spatial coverage in fractions of the ti me required by divers or point sampling. The acoustic classification system QTC VIEW V was used to map benthic habitats within Tampa Bay. The QTC system connected in parallel to an echo-sounder is capable of digitally extracting and r ecording echoes returning from the seabed. Recorded echoes were processed using QTC IMPACT soft ware. This software partitions echo waveforms into groups or classes based on their similarity to one another using multivariate statistics, namely Principal Co mponent Analysis and K-Means clustering. Data was collected at two frequencies, 50 kHz and 200 kHz. Side-scan sonar data was collected coincident with the QTC data a nd used to produce mosaics of the various habitats in Tampa Bay. Side-scan sonar data was classified using QTC Sideview in an attempt to identify changes in benthic habitats. Sediment samples used for ground-truth were subjected to grain size analysis. Also, the percentage of organic matter and carbonate within samples was determined. Results of acoustic classification appear to accu rately reflect changes in the sediment type and structure of the seabed. Grain size, partic ularly percent mud, app ears to have a strong influence on classification. Carbonate hard bot tom habitats were found to be acoustically complex, a characteristic useful for their identification. The QTC system was able to detect seagrass, although some misclassifi cation occurred between vegetated and nonvegetated seabeds.
1 Introduction There is a clear and esse ntial need for an increased ability to map benthic habitats within Floridas estuarine and coastal environments. Assessment of the seabed is vital to the successful management of fisheries, submer ged aquatic vegetation (SAV), sensitive hardbottoms, and a suite of environmental concer ns. Aerial photography and satellite imagery are commonly employed in habitat mapping. Wh ile both can be informative, neither is useful in deep or opaque waters. Extensive point sampling and divi ng transects are also common methods for benthic habitat mapping, although both are time consuming and typically limited in spatial coverage. Acoustical mapping methods are capable of providing detailed information about seabed characteristics in low visibility water, as well as in deep water. Acoustic methods such as side-scan sonar have proven effective in benthic habitat mapping. Side-scan sonar has been demonstrated to be particularly e ffective in mapping SAV, although information about sediment texture is of ten lacking. Acoustic sediment classification systems, on the other hand, have demonstrated an ability to di scriminate relatively subtle differences in sediment texture. At present, habitat mapping with acoustic classification systems in Floridas major estuaries is limited and absent within Tamp a Bay. Locker and Wright (2003), Locker and Jarrett (2006) and Locker (2006) demonstrated the effectiveness of acoustic classification systems in habitat mapping surveys conducted in southwest Florida. Although the survey conditions and benthic environment of the Ten Thousand Island region have commonalities with those of Tampa Bay, th e QTC (Quester Tangent Corporation) acoustic classification system remains unprove n within the state s largest estuary. Furthermore, acoustic classification of carbona te hard bottoms like those present within Tampa Bay is poorly understood. The extremely shallow depths of the surveys presented here represent another environmental conditi on not yet thoroughly explored in acoustic classification. Another significant component of this research is the comparison of singlebeam acoustic classification, side-scan sonar im agery, and swath acoustic classification. The three benthic habitats examined here are: (1.) soft sediments, mainly muds and fine sands; (2.) Carbonate hard bottom, consisti ng of exposed rock, sessile flora and fauna, and mixed siliclastic/carbonate sands and gravel and (3.) a shallow seagrass meadow. Hard-bottoms and seagrass meadows (submer ged aquatic vegetation, SAV) represent sensitive benthic habitat within Tampa Ba y. Locating and mapping the distribution of these regions within the bay is critical to their preservation and management.
All acoustic classification was performed using QTC Impact processing software. In each case, side-scan sonar data was collected along with (QTC) single-beam acoustic classification data. Overlaying or superimposing acoustic classification data on side-scan sonar mosaics has proven very instructive in the interpretation of these data. Various types of ground-truthing were performed at the different survey locations, including sediment grabs, rock samples, and diver observations. These systems relatively rapid data acquisition rates allow for potentially drastic increases in mapping coverage at lower costs than diving operations or extensive point sampling. Demonstration of these systems (acoustic sediment classification) potential within Floridas estuaries and coastal waters is critical to the advancement of benthic habitat mapping in this region. Figure 1. Location of the three survey areas within Tampa Bay, Florida. 2
3 Background and Previous Work A brief overview of the theory behind acousti c sediment classification is presented here followed by a more explanatory section on the physics related to the classification of echoes. Acoustic sediment classification me thods operate under the premise that sound waves reflected from the seabed have imprinted on them a signature of the physical characteristics of the substrate (Von Szalay and McConnaughey, 2002). Physical characteristics of the seabed known to influence the shape of returning echoes include, grain size, porosity, sediment density, mi crotopography, and bent hic flora and fauna (Bornhold et al., 1999; Collins and Lacroix, 1997; Collins and McConnaughey, 1998; Quester Tangent, 2003). There are two dominant influences on the amplitude and shape of the returning sound wave; seabed roughness and contrast in acoustic impedance between the water column and the seafloor (Bornhold et al., 1999; Collins and Lacroix, 1997; Galloway and Collins, 1998; Preston et al., 2000). The general shape of the reco rded echo is comprised of an initial peak followed by a lower amplitude tail of variable duration. Th e peak of the echo is primarily related to initial specular reflection located at the center of the ensonified footprint. The amplitude and duration of the echos tail is governed in large part by sound scattered from the outer portions of the footprint (Hamilton, 2001; van Walree et al., 2005). Based on these principles some general trends can be expected in the nature of echoes returning from a particular type of seabed. An uncomplicated, smooth seabed would be expected to return a signal with an abrupt peak followed by brief tail. A complex seabed with pronounced texture would be expected to return a signal with a less abrupt peak and a tail of greater duration (Collins and Lacroix, 1999; Quester Tangent, 2003). Physical characteristics of the seabed are not the only factors cont rolling the shape and amplitude of the recorded echoes. The fre quency and beam-width of the outgoing pulse will also play a role in the acoustical char acterization of sediments (Collins and Rhynas, 1998). Low frequency transducers will generally possess larger beam-widths than higher frequency transducers (Galloway and Collins, 1998). Lower frequencies (less than 100 kHz) are capable of introducing greater am ounts of energy into the seabed than higher frequencies. This allows increased penetra tion on the part of lower frequencies. This coupled with the fact that most low frequency transducers have larger beam-widths results in lower frequency signals possessing th e ability to carry more information about the seabed (Collins and Rhynas, 1998; Prest on et al., 2000). Higher frequencies, although limited in their penetration, provide higher resolution and may detect more subtle changes in the seabed confined to the water-sediment interface (Galloway and Collins, 1998). In the end, the factors that contribute most to the re turning echo are the geometry and frequency of the outgoing pulse, texture or roughness of the seabed and the structure
of the sediments volume (van Walree et al., 2005). However, the frequency will dictate what characteristics of the seabed, e.g. grain size, microtopography, substrates volume, that carry more or less weight in the returning signal (Collins and McConnaughey, 1998). Figure 2. 50 kHz waveform (top), 200 kHz waveform (bottom). A marked difference can be seen in the characteristics of the two frequencies presented here. The lower frequency, wider beam-width 50 kHz signal possesses a wide peak followed by a pronounced tail. The smaller beam-width 200 kHz signal depicts an abrupt, sharp peak with a less pronounced tail. 4
5 The QTC (Quester Tangent Corporation) acousti c sediment classification system exploits the fact the seabeds characteristics beco me imprinted on the returning echo. The QTC system, while connected in para llel to the echo-sounder, extracts the analog signal of the returning echo, digitizes the waveform a nd records it for processing in QTC View software. The software partitions echoes into groups or classes based on their similarity to one another using multivariate statistics, namely principle component analysis (PCA) and cluster analysis. Echoes with similar characteristics are assumed to result from like sediments, thus the classes resulting from statistical analysis ar e thought to represent changes in sediment type and or struct ure (Collins and Lacroix, 1997; Collins and McConnaughey, 1998; Quester Tangent, 2003) Further explanation of the QTC processing and classification technique is contained in the methods section of this paper. QTC data acquisition and subsequent proces sing is focused exclusively on a sampling window containing only the first echo returning from the seafl oor (Preston et al., 2000). Alternative acoustic sediment classificati on technologies are avai lable from Marine Microsystems Ltd., marketed under th e name RoxAnn. The RoxAnn sediment classification method utilizes in formation from both the first and second echoes (Burns et al., 1989). The first echo is simply the primary reflection from the seafloor; the second echo is the first multiple of the primary return. The second echo (first multiple) upon being recorded has twice reflected at the seaf loor and once at the sea-surface (Hamilton et al., 1999; Wilding et al., 2003). Based on thes e two echoes the RoxAnn system derives two parameters used to classify the seabed, E 1 and E 2. E 1 values are calculated based on the tail of the first echo and are said to represent seabed roughness. E 2 values are calculated based on the entire second echo a nd are said to represent seabed hardness (Chivers et al., 1990). Once acquired, RoxAnn da tasets are typically viewed in an E 1 versus E 2 fashion. Statistical relationships, e.g. means, modes, medians, and standard deviation, among E values can be presented in scatter plots and used to classify differences in seabed characteristics (Gr eenstreet et al., 1997; Hamilton et al., 1999). Evidence in the litera ture suggests the single echo met hodology used in the QTC system may outperform the two echo approach of RoxAnn. Hamilton et al. (1999) conducted a comparison of the two systems around the Great Barrier Reef in Australia. Their research concluded that the QTC system gave the bette r classification of se diments even without post processing of the data. Furthermore, RoxAnns second echo appeared to display noise, variability, and was st rongly dependant on survey speed (Hamilton et al., 1999). Hamilton et al., 1999 concluded that QTC results appeared independent of survey speed, a conclusion verified in Von Szalay and McConnaughey (2002). Research involving acoustic clas sification of the seafloor ha s been conducted across the globe in a vast array of geol ogic settings. Survey objectives vary, ranging from relatively straight-forward sediment classification across simple seabeds to attempts at determining floral and faunal distributions on complex r eef terrains. Research reported on here is restricted to those surveys in which Ques ter Tangents proprietary acquisition system (QTC View ) and software processing pack age (QTC Impact ) were utilized. The Quester Tangent Corporation had upgrad ed and improved both the acquisition and
6 processing software related to echo classifica tion within the timeframe considered in the following review of previous work. Hamilton et al. (1999) provided a comp arative assessment of QTC and RoxAnn classification results. The survey was perfor med in the Cairns area of the Great Barrier Reef in Australia, across a diverse seabed comprised of reef, carbonate and terrigenous mud, carbonate sand with shells, and gravel The performance evaluation of the QTC system is of interest here. QTC data were ground-truthed via grab samples, box cores, diver observation, and underwater video. It wa s reported that QTC cl asses demonstrated an association with the known sediment grai n size distribution and porosity values. Both QTC and RoxAnn systems returned poor classi fication results over the rough portions of the survey area. Comparison of QTC classes with video observations was termed as excellent. There was an approximate corre spondence between QTC class boundaries and changes in side-scan sonar imagery. Bornhold et al. (1999) compared QTC classification results and side-scan sonar imagery in the Staits of Georgia, British Columbia, Canada. Water depths ranged from 5 to 45 meters and acoustic data were ground-truth ed with underwater video. Side-scan sonar imagery depicted seven classes, whereas th e QTC only classified five. The boundaries in the data sets were not always in agre ement. Classes included muddy sand, sand and gravel, rock and sediment cove red rock. Differences in thes e two types of acoustic data (side-scan sonar and QTC) were attributed in part to the angle at which they impart sound on the seabed, 90 in the case of the QTC ec ho-sounder and the variable but considerably lesser angle of incidence of side-scan s onar which accentuates rough seafloors. Morrison et al. (2001) reported on results of research aimed at the detection of acoustic class boundaries. The survey area was located in Kawau Ba y on the northeast coast of New Zealand, water depths ranged from 5 to 20 meters, and the seafloor was covered in soft sediments. QTC classification successfu lly identified differen ces in mud, sandy mud, and muddy sands with shell cover. Class boundaries were successfully detected where sediments moved from sandier to muddier. Transitional zones occu rred gradually and quickly along transects. Acoustic data were ground-truthed with video images of the seabed. More types of seabeds were identified with the video footage than with the QTC classification, also changes in acoustic classes (transitions) were not always in lockstep with what appeared on the video footage. Anderson et al. (2002) employed the QTC system in the coas tal waters off Newfoundland at depths ranging from 10 to 220 meters. They reported an ability to discriminate between seabeds consisting of mud, gravel, rock, cobble, algal cover, and w ood chips disposed of by industry. Ground-truthing of acoustic data was carried out with observation of the seabed made from a submersible. Submersible observations led to more bottom classifications than were discriminated w ith the QTC system and a high degree of variability in acoustic classes was observed along transect lines.
7 Ellingsen et al. (2002) detailed a QTC surv ey of a fjord in western Norway where acoustic data was ground-truthed with gravity box cores and grab samples. Water depths ranged from 5 to 72 meters. QTC acoustic classes were found to generally correspond to sediment grain size and sediment softness. Not all acoustic classes could be attributed to differences in grain size. Also, some sample locations were not easy to identify with a particular acoustic class, according to the auth ors. They attributed this to high seabed heterogeneity and or association with transi tion zones. The discussion of these data is keen to point out the relationship of acousti c classification and ground-truth. In essence, the acoustic classification data is only as us eful as the ground-truth is accurate and meaningful. Freitas et al. (2003) compared QTC acoustic classification re sults with a combination of sediment and benthic fauna data collected o ff the western coast of Portugal in 5 to 40 meters water depth. Seafloor conditions in the survey area were described as, a relatively monotonous sublittoral sandy plain. Sediment grain sizes were determined to range from very fine to coarse sands with a >25% gravel component. Acoustic classification was able to separate fine from very fine sands and sands from gravel. At coarser sand grain sizes the QTC classifi cation was unable to discriminate real differences determined by sampling. The author s speculated that this inability may have arisen from the similarity in compactness of the sediments. Meaningful comparison of biological data with acoustic classes was ach ieved after class part ition was reduced from the optimal level specified in the software to two classes. Foster-Smith et al. (2004) documented an at tempt at classifying biotopes in the English Channel based on data from acoustic classi fication systems (QTC and RoxAnn), sidescan sonar, grab samples, and video footag e. Following unsupervised classification (the type used in the classificati on of data contained herein) of QTC data, the authors reported a fair correspondence between results and actual seabed type. Coarse sand and gravel areas appeared clearly defined, whereas regions of the seafloor that were dredged did not associate well with any acoustic class. Multiple QTC classes were seen to occupy regions of the seafloor, where side-scan sonar sugge sted homogeneity. Aut hors hypothesized that QTC either incorrectly subdivi ded these seafloors or that side scan imagery did not display real changes in the acoustic nature of the seafloor. Overall assessment of the comparison between QTC classes and interpreted side-scan sonar imagery were that the two were similar, with QTC data presenting less definitive boundaries. Moyer et al. (2005) give an account of QTC classification combined with LADS (Laser Airborne Depth Sensor) and diver surveys across a relict reef tract located in 3-35 meters water depth off Broward Count y, Florida. Acoustic classification was able to identify changes in the sandy sediment areas. Shallow water sand with ripples and deeper water sands without ripples appeared separate as di d reef and rubble classes. Attempts to use classification results to discriminate between types of reefs were met with a marked decrease in accuracy. Greater accuracy was achieved at the le vel of classifying reefs, rubble, and sand.
8 Hutin et al. (2005) attempted a more specific application of QTC ac oustic classification, the detection of a scallop bed in 20 to 60 meters water depth in the Saint Lawrence estuary off Quebec, Canada. QTC acoustic classification data were compared to biological and sediment data, as well as phot ographs of the seafloor. Photographs reveal seafloor sediments to be predominantly gravel and coarse sands with little variability. Comparison of the biological data and acousti c classes did not agree and detection of the scallop bed was unsuccessful. It was reported that QTC classification did not represent sediment conditions either, and that this ma y be the result of the low variability of sediments across the survey area. Also, classifi cation results were reported to be strongly dependant on depth. Riegl and Purkis (2005) utilized multiple fr equencies (50 and 200 kHz) of QTC data and compared them to classes derived from analys is of satellite imagery. The survey site was located in the Arabian Gulf, offshore Dubai in the United Arab Emirates at depths around eight meters. Types of seafloors identified in satellite imagery were dense live coral, dense dead coral, sparse coral, seagrass, shallow algae, deep algae, hard grounds and sand. Both frequencies of QTC data produced on ly two meaningful classes. Classification of the 50 kHz data was capable of identifying hard and soft bottom types accurately. The 200 kHz data accurately separated regions of high and low rugosity. The 200 kHz classification proved useful fo r the identification of corals, whereas the 50 kHz data detected hard bottoms even when they were covered in a thin layer of sand, effectively masking their presence in the satellite imagery. Seagrass was present, although not densely concentrated in the survey area; ne ither the 50 kHz nor the 200 kHz resolved any seagrass signature. Weinberg and Bartholoma (2005) reported on the use of QTC sediment classification technology to monitor spoils re sulting from dredging off the German coast in the Weser estuary. Survey depths ranged from 6-20 mete rs and acoustic classification results are compared to sediment grab samples and si de-scan sonar. Most sediments in the area consisted of fine to medium sand with some gravel and <1% mud. The three QTC classes identified were found to correspond to fine to medium sand, medium sand, and medium to coarse sand containing low, moderate, and high shell content respectively. QTC classes were found to be in adequately explained with only sedimentological groundtruthing, causing speculation that roughness was a potential ke y player in classification. Interpretation of side-scan sonar data revealed three distinct classes of seafloor features (bedforms). QTC classes were strongly associ ated with the crests and troughs of the dunes revealed in the side-scan sonar data. Riegl et al. (2007) present data collected off the coast (10-40 meters water depth) of Cabo Pulmo, Mexico combining satellite imagery, grab samples, and underwater video with QTC classification to investigate coral co mmunities growing on intrusive dikes and the surrounding unconsolidated sediments. The QT C classes reflected well the difference between hard bottom and unconsolidated sediments. Differences in the unconsolidated sediments were not detected in the classifi cation results. The authors commented on the inability of the acoustic data to discern differences in the sandy sediments. Samples of the
9 sandy sediments were found to be variable in composition------although grain size, a characteristic of sediments known to influence acoustic classification, was similar. One advantage of using acoustic data in concert wi th satellite imagery was an ability of the QTC classes to identify hard bottoms covere d with a light dusting of sand; imagery suggested only a sandy bottom. QTC was also ab le to differentiate areas where corals had produced very coarse sediments. One of the least explored potential applica tions of QTC technology is the detection of submerged aquatic vegetation (SAV). Data acqui sition is carried out in much the same way as surveys aimed at the classificati on of sediments, although a minimum sounder frequency of 200 kHz is recommended (Prest on et al., 2005). Modi fications to data processing techniques and software supplemen ts are suggested by the Quester Tangent Corporation to optimize the effectivene ss of QTC Impact for SAV detection. The processing adaptations are mostly concerne d with picking echoes that represent the seafloor, not the top of the ve getation, and adjusting the window (in time) about the pick in which the waveform is analyzed (Quest er Tangent, 2005 and Preston et al., 2006). A detailed account of attempts at detecting SAV with the QTC system is given in Riegl et al. (2005). The survey was performed in th e Indian River Lagoon located on Floridas east coast in shallow water, less than tw o meters. There was some ambiguity between bare seabed and vegetation in the classes, but it is reported that the system identified seagrass, fairly accurately (Riegl et al., 2005). There is no indication of any modification to normal processing in QTC Impact, such as those mentioned above. The data were classified using calibration sites consisting of bare seabed, seagrass, and macroalgae. A calibrated classification scheme was also used in the survey presented in Preston et al., 2006, as were the modified QTC Impact techniques. Preston et al., 2006 reports not only an ability to discriminate vegetated from bare seabeds, but also differentiation between two species of seaweed. Seafloor Acoustics Some Background Information As was stated earlier in the introduction, there are two dominant char acteristics related to the seabed and overlying wate r column that influence the amplitude and shape of the returning sound wave: (1) seabed roughness and (2) contrast in acoustic impedance between the water column and the seafloor (Bornhold et al., 1999; Collins and Lacroix, 1997; Galloway and Collins, 1998; Preston et al., 2000). Data products made using acoustic methods, while in the end are often refe rred to as habitat maps or maps of the seafloor, are in point of fact only representa tions of changes in the physical properties of the seafloor which influence sound waves. Only after these (acoustic) data are interpreted, ground-truthed and are consider ed within the generally recognized framework of geology, oceanography, biology, ecology, etc., can they be represented otherwise. In an effort to better underst and what physical properties of the seafloor influence acoustic waves and how, some basi c underlying relations hips are addressed here.
The amount or intensity of the energy reflected from the seabed is an important echo characteristic used in acoustic classification of sediments (Tegowski et al., 2003; Tegowski, J., 2005; van Walree et al., 2005). Energy levels associated with returning echoes are governed largely by the material property impedance. Impedance is determined by multiplying the density of a substance by its p wave velocity. Acoustic impedance is typically expressed as: I=V p. where is density and V is p wave velocity. The real importance of impedance with respect to seafloor echoes is not in its absolute value, but rather its contrast with the overlying water column. The amount of energy, in this case sound, reflected from the seafloor is determined by the magnitude of the difference in the two impedance values (Akal, 1972; Faas, 1969; Hamilton, 1970). The ratio of impedance contrast is quantified by the coefficient of reflection, first determined by Rayleigh, 1945 and at normal incidence is expressed: R= 2 V 2 1 V 1 / 2 V 2 + 1 V 1 where 1 and V 1 represent the water columns density and velocity respectively and 2 and V 2 represent the seabed materials density and velocity respectively. The relationship of impedance and coefficient of reflection (R) to physical properties of the seabed are investigated in Faas (1969), Hamilton (1970) and Akal (1972). The graphs depicted in Figure 3, adapted from Hamilton (1970), depict some relationships relevant to the classification of sediments. (a.) 10
(b.) Figure 3. The relationship of porosity and density to: acoustic impedance (a.), and reflection coefficient, R, (b.), adapted from Figures 2, 4, 5, and 6 in Hamilton, 1970. A clear relationship is depicted in Figure 3 where R increases with increased density and R decreases with increased porosity. Such a trend should be expected based in part on the relationship between porosity and density. As important sediment properties such as porosity and density change, these changes are recorded in the acoustic signal of seabed echoes. One way in which echoes record changes in the physical properties of sediments is through amplitude, measured in terms of dB and related to R through the following equation for bottom loss (BL) given in Hamilton (1970). BL= -20 log R Differences in the physical properties of sediments result in changes in impedance value. Impedance values determine the coefficient of reflection which controls the amplitude of the seabed echo. Changes in the amplitude, measured in dB, of echoes can be used as a method of classification (Tegowski et al., 2003; Tegowski, 2005; van Walree et al., 2005). 11
12 Along with acoustic impedance, roughness is an important characteri stic of the seabed controlling acoustic classificat ion. At the frequencies empl oyed throughout this research (50/200 kHz QTC data and 100/400 kHz side -scan sonar) seafloor roughness may generate a large percentage of the backscattered energy r ecorded in echoes. Seafloor roughness is a general term encompassing such things as sediment bedforms, e.g. sand ripples or waves, biologically generated features including borrows mounds, shells, and changes in sediments arising from locomoti on. Larger features such as corals, sponges and hard bottoms also contri bute to seafloor roughness. Controls on seabed roughness are complex, including weather, currents, geologi cal and biological factors (Jackson and Richardson, 2007). There is nearly always a positive correlation between backscatter and seafloor roughness, as well as backscatter and grain size (Jackson et al., 1986; Collier and Brown, 2005). While the aforementioned relationships are rather straight forward, it is important to mention the influence of frequency (w avelength) on roughness and by extension backscatter. Acoustic backscatter is the resu lt of seabed roughness on a scale similar to the wavelength of the acoustic signal (Jacks on et al., 1986). Based on a velocity estimate of 1500 m/s the wavelengths of the 50 kH z and 200 kHz signal should be 3.0 and 0.75 cm respectively. Acoustic backs catter resulting from sound waves having the previously mentioned wavelengths is not solely dete rmined by grain size and may be largely influenced by heterogeneity at the surface of the seafloor and within the sediments (Jackson et al., 1986 and Briggs et al., 2002). Heterogeneity at the surface of the seafloor and within sediments resulting in acoustic backscatter includes changes in the physical properties of th e sediments and irregularities buried within the sediments (Jackson et al., 1986; Jackson and Briggs 1992; Lyons et al., 1994; Jackson and Richardson, 2007). The cont ribution of the sediments volume to backscatter is a function of frequency, as it relates to attenuation, and the type of sediment. Jackson et al. (1986) and Jacks on and Briggs (1992) f ound that backscatter from sandy, coarse-grained seaf loors resulted mainly from roughness and heterogeneity at the sediment/water interface, whereas b ackscatter from heterogeneity within the sediments volume dominated in soft sedime nts. The impedance of fine-grained soft sediment is closer to that of the water column than is th e impedance value of sand, thus allowing greater penetration of the acoustic wave and more interaction with the sediments volume (Briggs et al., 2002). The shape and duration of echoes are dete rmined in part by the level of sound backscattered within seafloor, a fact that is exploited in the acoustic classification of sediments. It has been noted that increased signal penetration aff ects the duration or time spread of echoes, as well as backscatter or iginating from within the sediments volume (van Walree et al., 2005 and Briggs et al., 2002). Understanding the amount of penetration (or rate of attenuation) an acous tic signal is achieving provides insight into what sediment parameters may be co ntrolling acoustic classification.
Attenuation of acoustic signal within sediments is an important process to consider when classifying the seabed using echoes. The rate of attenuation experienced by a sound wave in sediments will have a large impact on the depth to which the acoustic signal penetrates. The thickness and or volume of sediment influencing the classification process is affected by the rate of attenuation. The attenuation of an acoustic pulse upon entering seafloor sediments is governed by the following equation: =kf n Where is attenuation given in dB/unit length, f is frequency, k is a constant and n is the exponent of frequency. For frequencies typically employed in marine geophysics n is generally close to one across a wide range of sediment types. In cases where grain size and acoustic wavelength are similar, attenuation may be based on n=4 (Hamilton, 1972). The constant k however, varies considerably with changes in the geotechnical properties of sediments. Some attributes of marine sediments that control the value of k and thus influence attenuation are: sediment structure, porosity, grain size, shape, contact among particles and physiochemical forces (McCann and McCann, 1969; Hamilton, 1972). The graph below, Figure 4 from Hamilton (1972) depicts the relationship between k from the above equation and mean grain size measured in phi units. Increasing phi () value denotes a decrease in grain size. Figure 4. The curve representing the relationship between attenuation and grain size is complex. Highest rates of attenuation occur in silty sands and sandy silts. Attenuation drops off rapidly following 4.5 phi and levels off around seven phi. Adapted from Figure 3 in Hamilton, 1972. 13
The relationship between k, which behaves like attenuation if frequency and the frequency exponent (n) remain constant, and grain size (, phi) is not straight forward. Values for k begin to increase rapidly around 2.5phi (fine sand) and continue until peak attenuation is reached near 4.5 phi (coarse silt) where k values begin to swiftly decline. According to Hamilton (1972) the highest k values, and thus greatest attenuation, occur in silty sands and sandy silts with grain size measurements between 3.5 phi and 4.5 phi. Notice the flattening out of the curve beginning around seven phi. It is in this size range that particles change from having non-active surfaces to active surfaces, this transition occurring with decreasing grain size. Thus, it is the physiochemical or cohesive properties of these sediments that result in decreased attenuation (McCann and McCann, 1969; Hamilton, 1972). The abrupt increase in the rate of attenuation around 2.5 phi and the abrupt decrease in attenuation rate at about 4.5 phi may be related to the way porosity changes with respect to grain size. The relationship of attenuation, porosity, and grain size are depicted in Figure 5 below, adapted from Hamilton (1972). Figure 5. The curve representing the relationship of attenuation and porosity (left) appears similar to the curve in Figure 4 suggesting a linkage. The two curves are related through the relationship of porosity and grain size (right). Adapted from Figures 5 and 8 in Hamilton, 1972. Hamilton (1972) reasons that if grain size decreases without a commensurate increase in porosity, such as in sands, then more sediment particles will be in contact with one another and greater attenuation will occur via intergrain friction. Conversely, if porosity increases substantially and grain size does not decrease much, as is the case in very fine sand and silts there will be less contact among particles, thus less intergrain friction and ultimately a decrease in attenuation. As an acoustic transducer moves over a varied seafloor, the physical changes in the properties of the seabed will affect the recorded signal. Changes in porosity, grain size, 14
15 density, roughness, and velocity will collectiv ely influence the interaction of acoustic energy with the sediments. The influence of these changing seafloor characteristics on echoes has a measurable affect, these affects can be used to separate echoes into classes representing simila r physical conditions. ~oOo~
16 Methods Acoustic Data Acquisition QTC and side-scan sonar data were acquired over the course of approximately one year. The soft sediment data sets were acquired over two consecutive days in October 2006 and consist of 50 and 200 kHz QTC data and a 100 kHz side-scan mosaic. Hard bottom data sets were collected during two separate periods, December 2006 and June 2007. The 2006 hard bottom data set consists of 50 kHz QTC data and a 100 and 400 kHz side-scan mosaic. The 2007 hard bottom data set consists of 50 and 200 kHz QTC data and a 400 kHz side-scan mosaic. The seagrass data set was acquired in July of 2007 and consists of 200 kHz QTC data and a 400 kHz side-scan mosa ic. All data was acquired from the R/V Price, a 25, outboard powere d, converted recreational fishin g vessel. Sea state was calm, three or less on the Beaufort scale. Acquisition operations were planned to coincide with monthly highs in the tidal cycle to ensure maximum accessibility to shallow areas during the soft sediment survey and the seagrass survey. Acoustic sediment classification (QTC) data was collected using a dual frequency (50 kHz and 200 kHz) Si-Tex echo-sounder. The transducer was pole mounted and located 25 cm below the waters surface. The beam -widths for the 50 and 200 kHz echoes were 18 and 7 respectively. Pulse length for the 50 kHz was 0.17 ms and 0.2 ms for the 200 kHz, the sounder was set to low power. Si de-scan sonar data was acquired using an Edgetech 272-TD towfish operated at 100 and 400 kHz. The towfish was suspended from the vessels bow to facilitate shallow water operations and positional accuracy. Digital acquisition, post processing, and mosaics were accomplished using Triton Elics ISIS Sonar and Delph Map GIS software and hardwa re. All acoustic data was located using a Trimble real-time differential GPS which records positions at sub-meter accuracy. Acoustic Sediment Classifi cation (Data Processing) Pre-processing quality control consisted of visually inspecting each echo trace for accurate bottom picks. Accurate location of the echo representing the seafloor is critical, as it dictates the portion of the recorded waveform that is subjected to analysis (Preston et al., 2000). All recorded echoes were stacked, a process used to increase the signal to noise ratio in the data (Quester Tangent, 2003) In this case every fi ve consecutive pings were averaged together to form a st ack. Stacking pings has consequences beyond improving data quality, most notab ly, altering the dimensions of the seafloor considered to represent a single data point. Von Szalay and McConnaughey (2002) have devised a formula to estimate what they term effective footprint length or EFL expressed as:
17 EFL = 2h tan ( ) + 4v/f where h is depth, is beam angle, v is vessel speed, and f is ping rate. The number of pings in the stack minus one is the constant associated with v thus the equation above represents a five stack ping. All QTC data was processed using QTC Impact (V) software. The processing technique begins by employing mu ltiple algorithms to determin e descriptive features of each waveform. The manufacturer (Quester Tang ent) reports that 166 echo shape features are determined in this process, including spectral and energy characteristics and information from both time and frequency domains (Collins and Lacroix, 1997). These 166 features are then reduced to the three mo st useful descriptors of the waveform via multivariate statistics, namely principle co mponent analysis. These three distinguishing features, referred to as Q-values, are then plotted in three dimensional mathematical space (Q-space). The result is that echoes with similar characteristics plot close to one another in Q-space, thus forming clusters. Cl asses of seabeds are determined based on the clustering of the data and assigned statistica l descriptors to indicate confidence in class assignment (Collins and Lacroix, 1997; Collins and McConnaughey, 1998; Collins and Ryhnas, 1998). Both data sets (50 kHz and 200 kHz) collect ed in the soft sediment survey (Safety Harbor) were processed using the automatic clustering engine (ACE) function contained within QTC Impact. According to the QTC Impact users manual, Auto Cluster is an automated clustering process using a Si mulated Annealing K-Means algorithm on an input QTC classification file in order to find an optimal numb er of classes. K-Means is a method of partitioning a multivariate data set into clusters without overlap in a way that minimizes the sum of the squared distance between data points and their closest centroid (Legendre et al., 2002). Initially, all da ta points are members of a single all encompassing cluster. Then at random, a pred etermined number of centers are identified within the original cluster. All data poin ts are then assigned to the randomly chosen center to which they are closest. Next, th e (randomly chosen) center points are redefined (moved) to the actual center of the data poi nts assigned to them. Following the move of the centers, again data points are assigned to the center point they ar e closest to and the process repeats until op timal clustering is achieved (A rthur and Vassilvitskii, 2007). The ACE function which produced meaningful results for the soft sediment survey did not perform well on any of the three hard bot tom data sets. In each case (three separate data sets) the ACE returned an optimal clusteri ng split at two or three classes, a result that seemed spurious for a seabed which had been confirmed by side-scan sonar, sediment/rock sampling, and diver observation to be quite complex and highly variable. As a result, an alternative manual clustering methodology was employed on each of the hard bottom data sets. The manual clustering method as prescr ibed in QTC Impact literature proceeds generally along the following logic. Initiall y, all data points are assigned to a single
cluster represented by an ellipsoid. An ellipsoid (cluster) can be manually split along any of its three axes, primary, secondary or tertiary. Figure 6. A cartoon representation of an ellipsoid taken from the QTC Impact manual. Clusters are split along one of the three axes shown. The initial all encompassing data cluster is split along its primary axis, resulting in two data clusters. A score is assigned to the split denoting its quality. This quality score is noted by the user and then the initial (primary axis) split is undone and the secondary axis is split. Again, a quality of split score is assigned to the split of the secondary axis, which is recorded and then the split is undone. Lastly, the tertiary axis is split, its quality score recorded and then undone. The split which receives the best quality score is then executed. This logic is then repeated on the two resulting clusters, continuing until the optimal split (number of classes) is achieved. This method produced meaningful and similar results across the three data sets acquired over the hard bottom. The seagrass survey was processed in the automatic cluster engine (ACE) manner described above. Initially the ACE was constrained to produce only two classes. Although it was known that more than two types of acoustically distinct seafloor were present in the survey area, the intent was to try to delineate only bare sediment from vegetation. The results bore no resemblance to the distribution of submerged aquatic vegetation (SAV) seen in side-scan sonar and aerial photography. Subsequently, the choice was made to cluster the data into four classes based on what was seen in the side-scan sonar imagery. Processing QTC data for the identification of SAV is the topic of ongoing research. Quester Tangent has issued supplemental literature containing suggested methodologies for optimizing success in locating seaweed. The term seaweed refers generally to tall forests of kelp in this context. Differences between standard echo classification and that of seaweed arise mainly in bottom picking and the positioning of the analysis window within the recorded waveform. Under normal operating conditions (classification of sediments) the bottom pick is determined by an amplitude threshold and is placed where 18
the likely seabed is. Only a very small amount of the echo prior to the large amplitude increase at the suspected seafloor is subject to analysis, five out of a total 256 samples. When vegetation is present, the bottom pick may occur at the top of the seaweed instead of the water-sediment interface. If waveform analysis proceeds in the standard way, both the signal from the vegetation and the sediments may become convolved and confuse the classification process. For the purposes of seaweed classification, the bottom pick is forced from the top of the vegetation to the likely position of the seabed through the use of record blanking and gates. Then the analysis window is moved earlier in time to include the influence of the vegetation and diminish that of the sediments (Preston et al., 2006; Quester Tangent, 2005). The recommended method of seaweed classification was not able to be employed in the seagrass data set reported here. Oftentimes it was impossible to determine where the seabed was in records where vegetation was suspected, thus the bottom pick could not be forced into position. Changes in bottom topography further complicated this endeavor. Interpreting where the sediment-water interface is located in a record where echoes from vegetation in the water column mask later returns would be greatly simplified on a nearly flat seafloor. In our survey area vegetation often occurred in concert with changes in bathymetry. A second complication involved the extreme shallowness of the seagrass meadows, often less than one meter deep. If the analysis window were pushed earlier in time, in an effort to capture the influence of the vegetation and diminish that of the sediments, it is likely that the ring down and or the out-going pulse may be captured. Figure 7. The image above is an example of data that has been subjected to PCA and cluster analysis. Three principal attributes of these data have been determined and used to discern differences among echoes. Values for the principle components are each plotted on their respective axis, X, Y and Z, resulting in the three dimensional Q-space depicted here. 19
20 The soft sediment (Safety Harbor) and the Tampa Bay hard bottom (2007, 50 kHz) data sets were also subjected to class complexity analysis. The class complexity analysis is available in Quester Tangents visual ization and mapping software called QTC CLAMS. The results of the complexity analysis reflect the level of heterogeneity of the seafloor. The QTC CLAMS manual defines cl ass complexity analysis as, a measure of the classes represented in the search radi us as a percentage of the total number of classes. In the case of both the soft sedi ment and hard bottom data sets, grid node spacing, search radius, and search size were ten meters, 25 meters, and ten members respectively. QTC Sideview Acoustic Seabed Classifi cation forSside-Scan SonarIimagery Side-scan sonar data from each of the su rvey areas were classified using QTC Sideview to determine changes in the characteristics of the seaf loor. The 100 kHz sidescan data from the soft sediment survey in Safety Harbor, the 400 kHz data collected over the hard bottom in 2007 and the 400 kHz seagrass data were individually classified in the Sideview software. QTC Sideview classifies side -scan sonar data based on the statistical properties of backscatter imagery (Quester Tangent, 2004). The amplitude and texture information associated with side-scan imagery data can be used effectively to discriminate differences in seabed characteristics. One obstacle to the classification of side-scan data is the fact that backscatter properties ar e not controlled solely by ge ology. The design and user controlled parameters of the s onar systems themselves also in fluence backscatter levels in side-scan imagery (Preston et al., 2004). Anot her major control on backscatter levels is the grazing angle (Collier and Brown, 2005; Preston et al., 2004). In order for classification results to only re flect changes in the physical pr operties of the seafloor, the influence of the previously mentioned factors must be removed or mitigated. QTC Sideview contains a method of removing or reducing such artif acts referred to as image compensation, the details of which ar e contained in Preston et al., 2004. Following compensation, the imagery data is divided into rectangular sections of user controlled dimension for classifi cation. Each rectangle will re present a single point in Qspace following principle component analysis (PCA). A series of algorithms, including statistical moments, power spectral ratio s, grey-level co-occurrences and fractal dimension values, are used to generate features which are then subjected to PCA and automatic clustering-----similar to that descri bed earlier, automatic cluster engine (ACE) (Quester Tangent, 2004; Preston et al., 2004). Grab Samples Sediment samples were acquired within the soft sediment survey area on October 6, 2006 using a Ponar grab sampler. Sample locations were chosen based on the QTC (50 kHz) sediment classification data which was acqui red and processed the previous day. Sidescan sonar imagery (100 kHz) was also c onsidered when choosing the locations of
21 sample sites. Twelve samples were taken in all and positions were recorded with a differential global positioning system. Sediment samples were placed into plastic bags, sealed, and transported back to the lab where they were refrigerated. Sediment samples were also obtained in a nd around the hard bottom survey site. After processing the 2006 acoustic classification data and side-scan sonar imagery it become apparent that sediments were abundant within the hard bottom area. It appeared that sediments, their presence or absence and possibly their thickness and physical properties might be influencing acoustic cl assification. Fourteen sediment samples were taken in the hard bottom area targetin g the different acoustic classes, no t all attempts (via Ponar grab sampler) were successful in recovering enough sediment for laboratory analysis. Sample locations were recorded to sub-meter accura cy using a Trimble differential GPS. Both sets of sediment samples (soft and hard bottom) were subjected to the following laboratory analysis. Grain Size Analysis Initial preparation of the samples entaile d a three-way split, split #1 for grain size analysis, split #2 for carbonate and organic matter analysis and split #3 for archival. The grain size analysis began with samples being soaked in Clorox overnight to remove organic material, followed by rinsing with RO (reverse-osmosis) water three times. Samples were then wet sieved to remove all material 63 microns and finer, thus separating the sand from mud. The mud frac tion was washed through the 63-micron sieve with dispersant (NaPOx 180mg/l) and co llected in a 1000 ml graduated cylinder. The sand fraction was then drie d overnight in an oven at 50 C. The dry samples were weighed and sieved through a sieve stack from -t wo phi to four phi, at half phi intervals. The contents of each sieve in the stac k were then weighed to 1/10 000 accuracy. Pipette Analysis (Clay and Silt Size Fractions) The fraction of the sample 63 microns a nd less or mud was collected in 1000 ml graduated cylinders and diluted with dispersa nt so that levels in the cylinders were precisely 1000 ml. One by one in a rigorously timed precession each cylinder was thoroughly mixed (one min), and allowed to settle for 20 seconds. Immediately following the 20 second interval, 20 ml of the mixture was extracted via pipette at a depth of 20 cm. The 20 ml mixture was then placed in pre-weighed beakers and dried. Following this first extraction at 20 cm, a temperature dependant time interval was allowed to elapse, approximately one hr 50 min for normal room temperatures. A second extraction was then made via pipette of 20 ml at ten cm depth in the cylinder. This 20 ml sample was also placed in a pre-weighed beaker and dried. After drying, both sets of beakers were a llowed to acclimate (four hours) or accumulate moisture from the air and then weighed. The first extraction of 20 ml at 20 seconds after
22 mixing represents the total mud content. The second extraction of 20 ml at ten cm depth approximately one hr 50 min following mixi ng represents the clay component. The difference between the two gives the silt fraction. Carbonate Analysis The carbonate analysis is straightforward and simply entails the dissolution of the carbonate material within the sample. The samp les were first dried in an oven at 50 C overnight. These samples are from the second sp lit so they are not treated with Clorox. Once dry the samples were placed in pre-weighed beakers and weighed. Ten percent HCl was then added to the samples, samples were mixed, and the samples were allowed to soak overnight. The HCl was then decanted from the sample and the sample washed three times with RO water. The samples were pl aced back into the oven (50 C) and dried overnight. Once dry, the samples were re-weighe d and the difference in weight calculated to represent the carbonate lost. Organic Matter Analysis Using the same material as that used in th e carbonate analysis, the samples were placed into pre-weighed crucibles and baked in mu ffle furnace for two and a half hours at 550 C. Following baking, the sample were allowed to cool and reweighed. The difference was recorded as organic matter or Loss on Ignition. ~oOo~
23 Results Soft Sediment Survey (Safety Harbor) 50 kHz QTC Data (Soft Sediments) Processing of the 50 kHz acoustic sediment cl assification data in QTC Impact revealed the presence of eight significant classes (F igure 8). The number associated with each class (1 through 8) has no bearing on the relationship among cl asses. That is to say that Class 1 does not necessarily occur next to Class 2 geographically, nor do the two numerically consecutive classe s necessarily share similar acoustic characteristics. It appears that three acoustic classes emerge as dominant within these data, Class 8, Class 6 and Class 4, in order from fine to coarse grained. The remaini ng classes, with the possible exception of Class 3, appear to be subsets of the dominant classes or representative of tran sitional zones, with respect to grain size. Class 8, depicted in yellow in Figure 9 is the dominant class representing the finer grained sediments within the survey area. Thr ee sediment samples were taken within this classs spatial extent, average grain si ze ranged from 7.15 to 7.63 phi, identifying these sediments as mud according to the Wentwort h (1922) scale. The range of grain sizes sampled within Class 8 corresponds to fine and very fine silts. Class 8 is dominant within the central region of Safety Harbor and is ab sent from the shallower shoreline areas and is not present within the la rge sandy shoal on the eastern side of the harbor. When overlain onto side-scan sonar data, Class 8 a ligns well with regions of low backscatter (Figure 13). Class 4, depicted in blue in Figure 10, is the dominant class representing the larger, coarser grained sediments in the survey area. Five sediment samples were taken within this classs spatial extent, average grain size ranged from 2.61 to 3.21 phi, identifying these sediments as sand according to the We ntworth (1922) scale. The range of grain sizes sampled within Class 4 corresponds to fi ne and very fine sands. Class 4 is dominant within the region directly adj acent to the shoreline as well as the entirety of the sandy shoal located on the eastern shore of the surv ey area. Class 4 is also prevalent in the southern extent of the survey area, where th ere is an apparent coarsening of sediments. When overlain onto side-scan sonar data, Class 4 aligns well w ith regions of high backscatter (Figure 13). Class 6, depicted in light green in Figure 11, is the dominant cl ass representing the intermediate grain sized sediments within th e survey area. Four sediment samples were
24 taken within this classs sp atial extent, average grain si ze ranged from 3.74 to 5.95 phi, identifying these sediments as transiti onal between sand and mud according to the Wentworth (1922) scale. The range of grain sizes sampled within Class 6 corresponds to very fine sand, coarse silt, and medium silt. Class 6 is dominant in south-western region of the survey area and in areas located be tween the shallow shoreline and the deeper, central portions of Safety Harbor. When overlain onto side-scan sonar data, Class 6 aligns well with areas of mode rate backscatter (Figure 13). Class 3, depicted as light blue in Figure 12, is somewhat prevalent within the survey area although not to the extent of the dominant three aforementioned classes (8, 4 and 6). Based on its association with Class 4, which re presents the coarsest sediment samples in the survey area and location with in areas of higher backscatter, Class 3 likely represents fine to very fine sands. Class 3 is mostly f ound in the southern region of the survey area amidst higher levels of backscatter and also in an isolated pocket of high backscatter located in the northern portion of Safety Harbor (Figure 13). ~oOo~
Figure 8. Aerial photograph of Safety Harbor, Tampa Bay overlain with 50 kHz acoustic classes. 25
Figure 9. Aerial photograph of Safety Harbor, Tampa Bay overlain with 50 kHz Class 8. 26
Figure 10. Aerial photograph of Safety Harbor, Tampa Bay overlain with 50 kHz Class 4. 27
Figure 11. Aerial photograph of Safety Harbor, Tampa Bay overlain with 50 kHz Class 6. 28
Figure 12. Aerial photograph of Safety Harbor, Tampa Bay overlain with 50 kHz Class 3. 29
Figure 13. 100 kHz side-scan sonar and the dominant acoustic classes. 30
31 Class 1 and Class 7 appear to be associated with regions of larger grained, coarser sediments. Both classes occur in proximity to the shoreline, and are associated with the sandy shoal on the eastern side of the survey area. Both classes are found in areas of moderate to high backscatter and although no samples were taken directly within these classes, sediments properties are probably similar to those in Class 4 or sand. Class 2 and Class 5 appear to be associated with regions of finer grained sediments, predominantly in the central portion of Safety Harbor. Both classes are absent from the shoreline area and the sandy shoal on the easte rn side of the survey area. Generally speaking, both classes are found in regions of low backscatte r, although in some isolated instances----particularly in the southern por tion of the survey area--there is some association with moderate backscatter. Both classes are found in close proximity to sampling station SH-11, which had a mean gr ain size of 7.63 phi (the largest phi or smallest grain size measured among the twelve samples). It is likely that these two classes, strongly aligned with Class 8, represent the smalle r end of the range of grain sizes found within the survey area. Their association with low backscatter in what appears to be a relatively deeper channel in the side-scan data lends further support to this idea (Figure 13). Class Complexity Analysis (50 kHz QTC Data) Complexity analysis sheds light on the level of heterogeneity of the seafloor. The lighter regions (Figure 14, white) correspond to area s of the seabed containing more acoustic diversity. The darker areas repr esent less diverse seabeds. The complexity is a function of the number of different acoustic classes found w ithin a user specified search radius about a grid node, relative to th e total number of classes in the survey area. ~oOo~
Figure 14. Map of class complexity Safety Harbor, Tampa Bay, grid node spacing = 10m, search radius = 25 m, search size = 10 m. 32
33 200 kHz QTC Data (Soft Sediments) Processing of the 200 kHz acoustic sediment classification data in QTC Impact revealed the presence of seven statistically significant classes. One class, Acoustic Class 4, was very minor (less than 30 members) and is not considered here. Acoustic Class 6 (Figure 15, orange) clearly dominates these da ta with 4542 members, nearly 56% of the total. The next most populous class is a dist ant second with 990 memb ers. The reason for this trend comes from a bias in track line lo cation. The majority of the 200 kHz data was collected near the shoreline in water that was depth prohibitive to 50 kHz acquisition. The shallow shoreline region of Safety Harbor is dominated by coarser sediments (very fine to fine sand) which ar e associated with Class 6. There appears to be general agreement betw een the distribution of the 200 kHz acoustic classes and backscatter levels in side-scan sonar imager y. Areas of high backscatter within Safety Harbor are concentrated along the shoreline and in the south where a sandy shoal extends westward into th e harbor. Within these regions (high backscatter) Acoustic Classes 1, 6, and 7 clearly dominate and sedi ment sampling indicates very fine to fine sandy conditions. Very low backscatter levels are generally confined to the deeper, central region of Safety Harbor where muds dominate the seabed. Here, Acoustic Classes 2, 3, and 5 are clearly dominant. The distribution of the 200 kHz acoustic classes depict a greate r level of complexity than what might be inferred from sediment samples and side-scan sonar imagery. Areas of the seafloor which appear somewhat homogeneous in side-scan imagery, i.e. the region of low backscatter in the center of Safety Harbor, contain several acoustic classes. Also, sediment samples with very similar physical characteristic are associat ed with more than one acoustic class. Acoustic Classes 2, 3, and 5 Acoustic classes 2, 3, and 5 have a clear asso ciation with the finer grained sediments found in Safety Harbor. Average grain size for sediments occurring within these classes is > seven phi which corresponds to mud. Acous tic Class 3 (Figure 15, green) is likely to represent the coarsest sediments of the three classes. Class 3 appears to be a transitional class between silts and fine sands; it is frequently associ ated with Acoustic Class 6 (Figure 15, orange) which is generally repres entative of sands. These three classes are most prevalent in the deeper, central region of Safety Harbor where sediment samples indicate a pronounced increase in mud content. Also, these classes show a clear association with regions of very low backscatter in side-scan sonar imagery.
Figure 15. 200 kHz acoustic classification data overlain on 100 kHz side-scan sonar imagery. 34
35 Acoustic Classes 1, 6, and 7 These three acoustic classes have a clear asso ciation with the coarser grained sediments found in Safety Harbor. Average grain size for sediments occurring within these classes is generally < 4.0 phi which corresponds to very fine and fine sands. Classes 1 and 7 represent the coarsest sediments identified w ithin the Safety Harbor survey area, these classes nearly always occur near the shorel ine and within the sandy shoal located on the southeast side of the embayment. Acoustic Cla ss 6 is by far the most prevalent class. This is partially due to a shallow track line bias, but it is also interesting that Class 6 is associated with sediment samples ranging in grain size from medium silt to fine sand. Sediment Data and Acoustic Classifica tion (Soft Sediment s, Safety Harbor) Analysis of the sediment samples revealed that when viewed in terms of mean grain size, the samples were relatively similar. That is to say that all samples were found to be fine grained and nearly free of gravel. According to the Wentworth Size Scale the samples ranged from fine sand to fine silt. Comparis ons of sediment properties and acoustic class in this section refer only to the 50 kHz QTC data set acquired in Safety Harbor, Tampa Bay. Figures 16, 17, and 18 demonstrate the similari ty of sediment grain size distribution within a particular acoustic class. The grain size distribution of the sediments in Acoustic Class 8, (Figures 9 and 16) is unimodal. Thes e sediments appear to lack any significant amount of material less t6an 2.5 phi, this corresponds to a pronounced lack of sand. The grain size distribution of these samples wei ghs heavily toward finer grained sediments classifying them as mud. The grain size distribution of the sediments in Acoustic Class 4, (Figures 10 and 17) is also unimodal. The sediments contained in th e five samples taken within this acoustic class appear very similar to one another with a dominant grain size of about three phi or fine sand. The mud content in this acoustic cl ass is limited to a sma ll percentage (< 10%) in each of the samples. The grain size distribution of the sediment s in Acoustic Class 6, (Figures 11 and 18) differs from Classes 8 and 4 in that it is bimodal. One mode represents fine sand ( three phi) and a second mode represents mud, with a strong clay component. More variability is seen within this acoustic class, with re spect to grain size, th an in Class 8 or 4.
Grain size by % Weight for Acoustic Class 80.0010.0020.0030.0040.0050.0060.00-2-1.5-1-0.500.511.522.533.544 to 88 to12Phi intervals% Weigh t Sample SH-5 Sample SH-4 Sample SH-11 Figure 16. Distribution of grain size within sediments associated with Acoustic Class 8 in Safety Harbor. Grain Size by % Weight for Acoustic Class 40102030405060-2-1.5-1-0.500.511.522.533.544 to88 to12Phi interval% Weigh t Sample SH-9 Sample SH-8 Sample SH-12 Sample SH-7 Sample SH-1 Figure 17. Distribution of grain size within sediments associated with Acoustic Class 4 in Safety Harbor. 36
Grain Size by % Weight for Acoustic Class 60102030405060-2-1.5-1-0.500.511.522.533.544 to 88 to12Phi intervals% Weigh t Sample SH-10 Sample SH-3 Sample SH-6 Sample SH-2 Figure 18. Distribution of grain size within sediments associated with Acoustic Class 6 in Safety Harbor. Figure 19 depicts the relationship between mean grain size (measured in phi) and acoustic class. These data appear to show a continuum in the distribution of grain size relative to acoustic classification. This plot suggests that mean grain size may not be the dominant control on acoustic classification. Figure 20 shows percent sand and mud plotted for each sample within acoustic classes. When sand is removed from the plot, (Figure 21), leaving only silt and clay (mud) the visible trend in the data suggests three modes corresponding to acoustic classification. Figure 22 depicts the relationship of carbonate and organic content to acoustic class. Again these data suggest three facies, which in turn correspond to three different acoustic classes. These data (percent weight of carbonate and organics) plotted with percent weight of mud (yellow line) suggest both carbonate and organic material are fine grained. ~oOo~ 37
Figure 19. Mean grain size of sediments in the three dominant acoustic classes in Safety Harbor. Figure 20. Percentage of sand, silt and clay found in sediments associated with the three dominant acoustic classes in Safety Harbor. 38
Figure 21. Percent mud contained within sediment samples associated with the three dominant acoustic classes in Safety Harbor. Figure 22. The relationship of carbonate material and organic matter to mud. 39
QTC Sideview The results of the side-scan sonar imagery classification appear to correlate well with changes in backscatter levels in the Safety Harbor mosaic. Figure 23. Acoustic classification of Safety Harbor, Tampa Bay using QTC Sideview. 40
41 The region of increased mud c oncentration, centrally located in Safety Harbor (Figure 23, green), is clearly depicted in the Sideview data. The coarser-grained sandy shoal (light blue) on the eastern shore of Safety Harbor is another obvious featur e identified in these data. Intermediate grain size sediments, thos e occurring between the fine sands and fine silts, are also captured in the Sideview clas sification. The olive-drab colored region in the south of the harbor and the isolated pocket in the north ar e likely composed of intermediate grain sized sediments, mainly ve ry fine sands. Some rather subtle features interpreted in the side-scan sonar mosaic, su ch as northwest-southeas t trending channel in the southern stretch of the ha rbor, are quite evident in th e Sideview classification. The acoustic classification of the northeast region of Safety Harbor appears scattered and suggests acoustic diversity. This trend that is in line with the results of the complexity analysis performed on the 50 kHz single-beam data. Hard Bottom Surveys As was stated earlier in the Methods section, the hard botto m survey incorporates three acoustic classification data sets two at 50 kHz and one at 200 kHz. Each of the surveys were conducted over the same general area identified as hard bo ttom habitat, although their size, location, and track line position ar e not identical. To facilitate analysis, interpretation, and most im portantly comparison among these data, an area of best overlap in coverage among the th ree surveys (two frequencies) was chosen. This area of investigation is identic al between surveys, contains a similar amount of data points from each survey, and represents all acoustic classes. The hard bottom acoustic classification data described here are presented in a GIS environment, each data point is geo-referenced and each acoustic class is represented by a different color. In most cases these data ar e superimposed on side-scan sonar imagery to help illustrate their relationship to the inte rpreted hard bottom. For all side-scan imagery presented here, dark areas represent high backscatter----generally indicating coarser sediments and or exposed rock. Lighter regions in the imagery represent low backscatter and typically indicate fine-grained sediments. 50 kHz QTC Data Set (2006) Processing of the 2006 50 kHz hard bottom data set in QTC Impact resulted in the identification of seven acoustic classes. Little acoustic diversity is seen outside the area identified as hard bottom habitat. Here, outside the hard bottom, a single acoustic class (Class 1) is clearly dominant. Some acous tic diversity is identifiable outside the prominent hard bottom area, where patchiness oc curs in the data. Although, even in areas (outside the hard bottom) where this acoustic diversity or patchiness occurs members of acoustic Class 1 are typically present and in some cases dominant. Four of the seven acoustic classes identified (2, 3, 4 and 5) appear to comprise much of the diversity in the hard bottom habitat. Visual interpretation of the data w ithin the hard bottom does not reveal any striking relationship among the four dominant acoustic classes. These classes
are concentrated within the interpreted hard bottom, but their distribution relative to one another appears scattered. Acoustic Class 1 Acoustic Class 1, (Figure 24) clearly dominates the vast majority of the survey area, comprising 61.1% of the data set. Only within parts of the interpreted hard bottom (dark regions in the imagery) does this class become less prevalent and in some cases absent. Acoustic Class 1 does not appear to carry a discriminatory signal for the identification of hard bottom. Acoustic Class 1 is interpreted to represent unconsolidated sediments (sand). Figure 24. Acoustic Class 1 (pink) (50 kHz) superimposed on a 400 kHz side-scan sonar mosaic. The dark regions (high backscatter) indicate hard bottom or coarse sediments. Lighter regions represent finer-grained sediments (sand). 42
Acoustic Classes 2, 3, 4, and 5 These four acoustic classes, (Figure 25) collectively representing 33.3% of the data set exhibit a clear association with the area interpreted to be hard bottom habitat. Figure 25. Acoustic classes 2, 3, 4, and 5 (50 kHz) interpreted to represent the hard bottom habitat. A clear association is seen between the presence and concentration of these classes and higher levels of backscatter in the 400 kHz side-scan sonar mosaic. 43
44 The presence of these classes is, with a few exceptions, confined to the hard bottom region. Their abundance and concentration incr eases with backscatter level, further strengthening the interpretation that these ac oustic classes carry information useful for the identification of hard bottom habitat. Acoustic Class 6 Acoustic Class 6 is widely distributed thr oughout the survey area, representing 4.8% of the data set. This class does not appear to be associated with the hard bottom area. There is some evidence (grab samples) that suggests this class may repr esent unconsolidated sediments of smaller grain size than Class 1. Acoustic Class 7 Acoustic Class 7 is the least significant class identified and represen ts only 0.8% of the data set. Class 7 is widely distributed th roughout portions of the survey area interpreted as soft sediments. Class 7 is nearly absent inside the interpreted hard bottom. It seems likely that this class may be representative of soft sediment. Its us efulness in identifying location or interpretation of the nature of se diments is limited due to its widely dispersed distribution and scarcity. 50 kHz QTC Data Set (2007) Processing of the 2007 50 kHz data set in QTC Impact resulted in the identification of eight acoustic classes. As was the case with the data set acquired in 2006, one acoustic class (Class 3) emerged as dominant throughout the survey area. Again, nearly all of the acoustic diversity is seen in the area interpre ted to represent the hard bottom habitat. The dominant class appears clearly aligned with areas of sediment cover. The remainder of the classes show an association with what ha s been identified as hard bottom habitat. There is a clear distinction in these data be tween hard bottom habitat and areas dominated by unconsolidated sediments. It is less clear what distin ction or relationship can be discerned among classes interpreted to repr esent the hard bottom. Acoustic classes located within the hard bottom s how a great deal of scatter. Acoustic Class 3 Acoustic Class 3 (Figure 26), comprising 58.3% of the data set clearly dominates the survey area. Its presence is ubiquitous with the exception of some of the most robust regions of the hard bottom hab itat. This class somewhat diminished presence within the hard bottom, correlation with low backscatter, and overall pervasiveness in the survey area would indicate its ali gnment with unconsolidated sediment cover.
Figure 26. Acoustic Class 3 (50 kHz) clearly dominates the survey area; Class 3 is representative of unconsolidated sediment cover. Acoustic Classes 1, 2, 4, 5, 6 and 8 These six acoustic classes, (Figure 27) collectively representing 41% of the data set appear to carry a hard bottom signal. The frequency and concentration of each of these acoustic classes increases within the hard bottom habitat. These classes are nearly absent outside the hard bottom except in what appear to be small patches of high backscatter in the side-scan sonar mosaics. Similarity is seen between these data and the 2006 data set with respect to scatter among the classes representing the hard bottom. A clear hard bottom signal is present within these data distinguishing it from the surrounding unconsolidated sediments; determining differences among the classes representing the hard bottom is a more opaque matter. 45
Figure 27. Acoustic Classes 1, 2, 4, 5, 6, and 8 (50 kHz) interpreted to represent the hard bottom habitat. Much scatter is apparent in these data. Taken collectively these classes align well with the high backscatter of the 400 kHz side-scan sonar mosaic. Class Complexity Analysis (50 kHz 2007) Complexity analysis is an indicator of the level of heterogeneity of the seafloor. The lighter regions (Figure 28, white) correspond to areas of the seabed containing more acoustic diversity. The darker areas represent less diverse seabeds. The complexity is a function of the number of different acoustic classes found within a user specified search radius about a grid node, relative to the total number of classes in the survey area. This type of analysis is particularly instructive in a carbonate hard bottom environment where the seafloors acoustic properties change rapidly over small spatial scales. This analysis clearly defines the region of hard bottom habitat (light color) amidst the surrounding unconsolidated sediments (Figure 28, dark areas). 46
Figure 28. Map of class complexity hard bottom habitat, Tampa Bay, grid node spacing = 10m, search radius = 25m, search size = 10. 200 kHz QTC Data Set (2007) Processing of the 200 kHz hard bottom data set in QTC Impact resulted in the identification of seven acoustic classes. Among the classes Class 5 is dominant, although not to the extent of which a single class dominated in the two earlier discussed 50 kHz 47
surveys. It seems likely that this dominant class (5) is again associated with unconsolidated sediments; however its presence extends into some of the higher backscatter (interpreted as possibly exposed hard bottom) regions. The data appears less scattered than the previous two data sets. This owing to the presence of two moderately dominant classes, which combined with the clearly dominant Class 5 represent 90% of the data. The remaining four classes are relatively small and sparsely distributed making their interpretation difficult. Acoustic Class 5 Acoustic Class 5 (Figure 29), comprising 50.2% of the data set appears throughout the survey area. Class 5 is likely associated with unconsolidated sediments (low backscatter). The concentration of Class 5 members increases in areas interpreted to have sediment cover, although this class is clearly evident within the hard bottom habitat (high backscatter). Figure 29. Acoustic Class 5 (200 kHz) interpreted to represent unconsolidated sediments. Class 5 is present even in regions of increased backscatter (interpreted to be possible hard bottom) suggesting the presence of sediments within the hard bottom habitat. 48
Despite the fact that Class 5 is present throughout the survey area, including regions of higher backscatter, it does not appear to carry any signal useful for the identification of hard bottom. Instead, its appearance within the hard bottom habitat is likely due to the fact that there are unconsolidated sediments of various thicknesses present there. Acoustic Class 2 Acoustic Class 2 (Figure 30), comprising 26.3% of the data set appears to carry a signal useful for the identification of hard bottom habitat. Although possessing a clear affinity for regions of higher backscatter, Class 2 is not absent from areas of moderate to low backscatter known to contain relatively fine grained sediments. The spatial distribution of this class is such that it does not conform neatly to patterns in backscatter. However, in certain instances its concentration increases with backscatter levels. Figure 30. Acoustic Classes 2 and 7 (200 kHz) appear to show some indication of hard bottom habitat and or coarse grained sediment. This is based on the fact that these two classes are clearly absent to the west of the hard bottom where soft sediments are present. 49
Acoustic Classes 1, 3, 4, and 6 The remaining acoustic classes in the 200 kHz survey represent a collective 9.3% of the data set. These classes (1, 3, 4, and 6) shown below in Figure 31, do appear to be more prevalent in areas of higher backscatter. It is possible that each of these classes may carry a signal useful for the identification of hard bottom habitat. This possibility rests on the fact that each of the classes distribution has a diminished presence in low backscatter regions dominated by finer-grained sediments located on the west side of the survey area. Figure 31. Acoustic Classes 1, 3, 4, and 6 associated with elevated backscatter, containing a potential hard bottom signal. 50
Sediment Data and Acoustic Classification (Hard Bottom Survey) Despite having been identified as a hard bottom habitat comprised of an exposed hard substrate (limestone) capable of supporting sponge and soft coral colonies, the survey area is not without sediment cover. The presence and behavior of sediments within hard bottom environments is thoroughly examined in Riggs et al., 1996 and Riggs et al., 1998. With respect to the current survey area, the presence of sediments within the hard bottom habitat was first suggested in side-scan sonar and QTC data and then confirmed with diver observation and Ponar grab samples. Generally speaking, sediments recovered on the periphery as well as within the hard bottom were fine to medium mixed quartz/carbonate sands with a variable gravel component and very little mud (mostly < 2%). Average grain size among samples fell within a narrow range, 1.47-2.18 phi (Figure 32). Sediments with greater average grain size tended to contain a larger percentage of carbonate material, (Figure 33). Grain size distribution within sediments, pictured below in Figure 34, are similar across the survey area and appear to be unimodal at about three phi (). Most of the variability in the sediments with respect to grain size appears to be located toward the small end of the phi value range (larger grain size). Little variability is seen in the large end of the phi value range, where nearly all samples contain only trace amounts of silt and clay (mud). Mean phi value (Hardbottom survey)0123HB-13HB-2HB-1HB-11HB-5HB-7HB-9HB-10HB-3HB-12HB-14Sample IDphi valu e mean phi Figure 32. Mean phi () value for sediment samples collected in and around the Tampa Bay hard bottom. 51
Percent carbonate and mean phi05101520253035404550HB-13HB-2HB-1HB-11HB-5HB-7HB-9HB-10HB-3HB-12HB-14Sample ID% weigh t 00.511.522.5phi % CO3 mean phi Linear (mean phi) Figure 33. The relationship among mean phi () and percent carbonate in sediment samples, the trend indicates increases in carbonate material generally coincide with larger mean grain size. Grainsize Distribution0.005.0010.0015.0020.0025.0030.0035.0040.0045.00-2-1.5-1-0.500.511.522.533.544 to88 to12phi value % HB_1 HB_2 HB_3 HB_5 HB_7 HB_9 HB_10 HB_11 HB_12 HB_13 HB_14 Figure 34. Sediment grain size distribution within samples taken in and around the hard bottom survey area. Grain size distribution appears unimodal at about 3 phi, corresponding to the boundary between fine and very fine sand on the Wentworth scale. More variability is seen in the low range of phi, larger grain size. 52
53 Percent gravel and percent carbonate material ap pear to increase in regions suspected to be hard bottom. A clear relati onship can be seen in side-scan sonar data with respect to backscatter levels and percentage of car bonate material (Figure 35).Similarly, the percentage of gravel in a sample appears to increase in regions of elevated backscatter within side-scan sonar data (Figure 36). The two previously mentioned relationships, increasing carbonate material in high backscatter regi ons and increasing gravel component in high backscatter regions, s uggests a possible linka ge between percent carbonate and gravel component The relationship between these two variables (Figure 37) is that sediments containing a relatively large gravel component will also have an increased percentage of carbonate material, although the opposite is not always true. A relatively high concentration of carbonate material does not necessarily indicate an increase in gravel component. The sediment data and analysis presented he re were determined from samples collected following the initial 50 kHz QTC (2006) surve y, thus the location of sample sites was determined based upon the results of said survey. A comparison of acoustic data and sediment properties is therefore most m eaningful on the 50 kHz 2006 survey, as the position of tracklines in subsequent surveys doe s not coincide with sample locations. The effective footprint length (E FL) or dimensions over whic h the stacked echo signal is integrated for classification is no greater than about 1.3 meters across track by 3.3 meters along track in the hard bottom survey area. He terogeneity of the seafloor in the hard bottom habitat is known to cha nge considerably on scales much smaller than the EFL. This fact, coupled with the dimension of th e seafloor represented by the sediment grab samples (0.0225m 2 ), requires precision in ground -truthing and does not allow for assumptions to be made about relationships between sediment samples and nearby acoustic classes from other surveys. As was previously reported, processing of the 50 kHz (2006) data set resulted in the identification of seven acoustic classes. No meaningful relationship was able to be determined among individual acoustic classe s and sediments samples. However, if acoustic classes are separated into two categor ies, those representing hard bottom (2, 3, 4, 5) and those representing finer-grained uncons olidated sediments (1 and 6) and compared to grain size characteristics and percent car bonate there does appear to be a linkage. The data show an increase in both gravel c ontent and percent carb onate within acoustic classes interpreted to represent hard bottom. Samples associated with acoustic classes interpreted to be fine-grained sediments show a marked decrease in percent carbonate and gravel component. Figures 38 and 38 dem onstrate the aforementioned relationship between acoustic classifi cation, gravel component and percent carbonate. ~oOo~
Figure 35. The percentage of carbonate and remaindered material not dissolved by hydrochloric acid in sediments recovered in and around the hard bottom. The white number at the top of each pie is the sample ID and the precise location of collection is indicated by the red arrow. These data are projected upon a 100 kHz side-scan sonar mosaic in which dark areas represent regions of high backscatter. These darker regions have been interpreted to represent hard bottom habitat with various thicknesses of sediment cover. The three sample locations where no sediments were recovered are thought to represent exposed rock. The percentage of carbonate material increases in regions interpreted to be hard bottom based on elevated backscatter levels. 54
Figure 36. Trends in percent sand, mud, and gravel with respect to backscatter in side-scan sonar imagery. Areas of suspected hard bottom appear darker (high backscatter) in the 100 kHz side-scan mosaic. Exact position of each sample is referenced with the black arrows. Sample sites where no sediments were recovered are interpreted to be areas of exposed rock. 55
Percent CaCO3 and gravel05101520253035404550HB-13HB-2HB-1HB-11HB-5HB-9HB-7HB-10HB-3HB-12HB-14Sample ID% CaCO 3 0123456789% grave % CO3 % gravel Figure 37. The relationship of % carbonate and gravel, where large gravel components require an increase in the percentage of carbonate material, but elevated levels of carbonate are not necessarily reflected in the gravel component. ~oOo~ 56
Figure 38. Combining acoustic classes to represent only hard bottom and soft sediment removes much of the scatter in the data, revealing a possible linkage between gravel component and acoustic classification. Samples 1, 2, 3 and 13 each posses a very small percentage of gravel and are located in regions of the survey area interpreted as soft sediments. 57
Figure 39. High percentages of carbonate indicate the presence of the hard bottom habitat. Areas with diminished levels of carbonate correspond well with acoustic classes interpreted to be soft sediments. 58
QTC Sideview Sideview acoustic classification was performed on the 400 kHz side-scan sonar data collected over the hard bottom in 2007. Figure 40. QTC Sideview classification of 400 kHz side-scan sonar data. The hard bottom habitat is clearly depicted, as are the unconsolidated sediments found to the north and west. 59
60 The results of the classification show a clear delineation of the hard bottom habitat. There is some evidence of acoustic noise and artifact classes within these data. Despite this fact, the acoustic signal of the hard bottom is obviously separated from the surrounding unconsolidated sediments. The class shown in red in Figure 40 (above) ap pears to be representative of the hard bottom, although this may not always be the ca se. In some instances, particularly in the southeast region of the hard bot tom, the red class appears to be an artifact. One can see the linear trend of the red class in this area of the data. In other ar eas of the hard bottom the red class shows no linear trend and is clear ly related to changes in backscatter levels associated with the presence of sediments. Submerged Aquatic Vegetation (SAV) Survey The 200 kHz QTC seagrass survey was processed so as to result in four separate classes of seafloor. An initial attempt at clustering th e data into only two classes with the intent of separating bare sediments from vegetated seafloor failed to achieve meaningful results. The choice of four clusters was based main ly on side-scan sonar imagery where four main types of seafloor were apparent, namely SAV, sandy sediments associated with a shoal, deeper-water sediments to the east of the shoal in Tampa Bay, and amorphous patches of higher backscatter located in a dredged area near the beach. Side-scan sonar (400 kHz) imagery was very successful at detecting vegetated and non-vegetated seabeds (Figure 42). Acoustic Class 1 Class 1 appears to be the st rongest candidate for possessing an acoustic signal indicating SAV. Class 1 aligns fairly well with seagrass meadows interpreted from aerial photography, (Figure 41, green). Class 1 is al so present in regions of side-scan sonar imagery where SAV is located. Class 1 is clearly absent in the bare sandy patches within the shallow shoal area, as well as in the deeper -water bare sediment region to the east of the shoal in Tampa Bay. Class 1 is also absent within a channel dredged through the sandy shoal. One possible inconsistency w ithin an interpretation of Class 1 as representing SAV is in the deeper dredged area located just off the shoreline. Here in the aforementioned area, Class 1 is present, yet nothing in the side-scan sonar imagery would suggest the presence of SAV. Acoustic Class 2 Class 2 does not have any association with SAV; rather it appears to carry a strong sediment signal. Class 2 is clearly aligned with the bare sediment seabed located in the deeper regions of the survey area, (Figure 41, blue). Class 2 is nearly the sole class occupying the deeper waters to the east of th e shallow sandy shoal. Class 2 is also present
in the channel running through the shoal, as well as in the dredged area off the sandy beach. Figure 41. The distribution of the 4 acoustic classes resulting from the SAV survey, Northshore, Tampa Bay. Acoustic Class 3 61Class 3 shows some alignment with regions of seagrass in the aerial photography and Class 3 (Figure 41, red) may possess some signal related to the presence of SAV. It appears that Class 3 exhibits what can best be described as a mixed signal. It may be the case that both SAV and sediments are influencing the echoes contained in this class.
62re coustic Class 4 lass 4 (Figure 41, pink) appears to be associated with the sandy sediments of the the the raph side-scan sonar. Class 3 is absent in sand patches on the shoal and in the deeper offshoarea and the channel. Where Class 3 becomes inconsistent with seagrass location is mainly in the deeper dredged area along the shore. Here, Class 3 is present in high numbers where nothing suggests the likelihood of seagrass. A C shallow shoal and the beach. Class 4 clearly dominates the bare sediment patches tosouth of the channel within the shallow shoal. Class 4 also has an increased presence within the dredged area along the sandy beach shoreline. Class 4 is nearly absent fromdeeper region to the east of the shoal into Tampa Bay. Class 4 does appear scattered throughout the shallow shoal in regions interpreted to be seagrass in the aerial photogand in the deeper dredged area off the beach. Figure 42. Side-scan sonar mosaic (400 kHz, 15cm resolution) of the North Shore seagrass meadow.
63TC Sideview he Sideview acoustic classification of the Northshore side-scan sonar data (400 kHz) e ent SAV Q T was partially successful in identifying areas of vegetated seabed. The linear band of SAV trending southwest-northeast is clearly represented in the Sideview data. This belt of SAV marks the eastern edge of the sandy shoal. Either side of this SAV belt appears to bbare sediments, a trend depicted in the Sideview data. The SAV located atop the shallow shoal is also depicted, although not in the manner predicted from aerial photography and the side-scan mosaic. The acoustic classes interpreted to represdisplay a higher level of discontinuity or patchiness than was evident in the mosaic imagery. Figure 43. QTC Sideview classification of 400 kHz side-scan sonar data of the Northshore survey area.
64 Discussion Echo Parameterization know what physical characteristics of th e seafloor influence choes, but also what the measurable effect s are. The results of acoustic classification ere processed using QTC pact. QTC Impact is commercially ava ilable processing software developed and uester d g ata is s ubjected to principle omponent analysis (PCA) to id entify the three most useful descriptors. At no point is the ation as s is given in, Tegowsk i et al. (2003), van alree et al. (2005) and Tegowsk i, J. (2005). The basis for echo shape parameterization The It is not only important to e surveys are not always strai ghtforward and adequate ground -truthing data may not be available to explain trends. Acoustic classification data does not have the strong visual component inherent in seismic sections or si de-scan sonar mosaics that allows a certain level of intuitive inte rpretation. Understanding the classi fication process at each of its steps is vital to accurate inte rpretation of these results. All of the acoustic classification data re ported on here w Im distributed by Quester Tangent Corporation located in British Columbia, Canada. Q Tangent reveals little detail about the exact parameters used in echo classification. QTC literature states, generally, that a series of algorithms is used to identify 166 descriptive features of each stack, in this case, 5 su ccessive pings. These 166 features are derived from spectral and energy characteristics in both time and frequency domains (Collins an Lacroix, 1997). The QTC Impact training cour se manual goes into more detail, listin some statistics involved in echo classifica tion. According to the manual, cumulative amplitude, amplitude quantiles and histograms and power spectrum information goes into the derivation of the 166 descriptors. A more detailed description of QTC Impact processing is given in the Met hods section of this text. From the point of identifying these 166 descri ptors, the d c user aware of what the 166 descriptors represent and which of them were identified as explaining most of the variance in the data set determined by PCA. This is quite understandable with respect to protecting proprietary methods from competition in the marketplace. However, from a research standpoin t it leaves some room for specul to how echoes are classified a nd one must look elsewhere to ga in insight into the variou parameters other researchers have employed. An excellent account of echo parameterizati on W involves the use of statistical and spectral moments as we ll as fractal dimensions. Statistical moments describe the clustering te ndencies and overall shape of a distribution of data points. The first moment of a statistical distribution is the mean or average. second moment describes how wide the distribu tion is or its variance. The third moment
65 ts to the echo envelope of coustic data. van Walree et al. (2005) used envelopes in the time domain to calculate s a alree l. (2003), van Wa lree et al. (2005) and Tegowsk i, J. (2005) used spectral oments calculated in the frequency domain to discriminate among sediment types in,. ge volve m easurements of the energy and or intensity f returning echoes. One such measurement, described in van Wa lree et al. (2005) as n el, he as is a measure of skewness or how asymmetri cal the distribution is Positive skewness denotes a distribution leaning hea vy in the positive direction of x with negative skewness being the opposite. The fourth moment of a distribution is called kur tosis and represen its tendency toward peaking or fl attening out (Press et al., 1996). Shape parameters related to statistical distribu tions can be applied a time spread (second moment), defined as th e temporal duration of the echo and echo envelope skewness (third moment), a measure of asymmetry, to classify echoes. Time spread increases with incr easing backscatter (roughness) and or penetration. The skewness of the echo envelope in the time doma in is another shape parameter useful for discerning sediment types. van Walree et al. (2 005) reports that in general, there i positive skewness associated with echoes from the seabed. This tendency arises from the fact that initial specular reflection generate s an abrupt, high amplitude peak which is followed by sound returning from backscatte r and reverberation in the sediments volume. Variation in backscatter associat ed with the sediments surface and the contribution from the sediments volume can influence an echos skewness (van W et al., 2005). Tegowski et a m These parameters include but are not limited to spectral width and spectral skewness. The following description of spectral width is given in Tegowski et al. (2003), The spectral width parameter is defined by the mean frequency and the concentration of spectral power density around it. If spectral energy is widely shared amongst the ran of frequencies then its spectral width is grea ter. Sands tend to have a narrow spectral width, whereas silts and clays tend to distribute spectral energy more widely (Tegowski 2005). Spectral skewness refers to how evenly the power spectral dens ity is distributed about the mean frequency, the larger the skewness the greater the asymmetry. Soft, muddy seafloors generate echoes with greater spectral skewness than do sandy or gravel seafloors (van Walree et al., 2005). Other methods of parameterization in o echo energy, is derived from the intensity of the sound reflected from the seafloor. The echo energy parameter is a function of ac oustic impedance and backscatter strength, which encompasses seafloor hardness as we ll as roughness (van Walr ee et al., 2005). Va Walree et al. (2005) reported that changes in th e seafloor corresponding to sand, grav and mud were discernable usi ng the echo energy parameter and that greater differences in dB levels were observed at higher frequencie s. Tegowski (2005) employed the parameter of integral backscattering strength, described as, the logarithmic measure of the energy value integrated for th e total echo signal duration. Th is parameter is related to the hardness of the seafloor and include s volume backscattering (Tegowski, 2005). T normalized moment of inertia of the echo inte nsity was used by Tegowski et al. (2003) a means of echo classification. This para meter describes how an echos energy is
66 ., ublications is fractal dime nsion (van Walree et al. (2005), Tegowski, J. (2005), and s in s nd to choes af ety Harbor) data, it appears at the acoustic response of sediments in nor thern Tampa Bay is governed in large part d fety obably does not veal the true scope of the variables c ontrolling acoustic response. Sediment sample ity, hness distributed about its center of gravity. Small values of this parameter indicate short echo durations, where as larger values indicate greater echo pulse time s (Tegowski et al 2005). This relationship was used to determ ine the presence or ab sence of vegetation. One method of echo parameterization co mmon throughout the three aforementioned p Tegowski et al. (2003)). Fractals are shapes or forms that are invariant across change scale (self-similar); fractal geometry is the mathematics used to describe such forms a they are much too complex for Euclidean geometry (Mandelbrot, 1982). The fractal dimension is used as means of determining waveform complexity. Complexity, in this case, is determined by repeatedly measuri ng a shape (echo envelope) at smaller and smaller scales and calculating the speed at which length, surface or volume increases (Peitgen, 2004). Tegowski, (2005) reported that, The sound backscattering by the oceanic bottom obeys fractal laws. This statement is based in part on Yamamoto, T., 1996, where small scale vertical fluctuations in density and sound velocity were fou have fractal geometry. Tegowski reasons that this fractal nature is imparted on the returning echo. In practice, th e fractal dimension has proven valuable in discerning soft sediments from hard sediments, such as m ud and gravel (van Walr ee et al., 2005). E penetrate further in softer sediments and are thereby more affected by the small vertical changes in density and sound velocity giving them greater complex ity (greater fractal dimension value) than harder substrates, (T egowski, 2005 and van Walree et al., 2005). Acoustic Classification of So ft Sediments (Safety Harbor) Based on analysis and interpreta tion of the soft-sediment (S th by sediment grain size, particularly pe rcent mud. Visual comparison among acoustic classification data and side-scan sonar data suggest a strong corre lation between the two. Changes in sediment classification (acousti cally) appear to coin cide nicely with contrasting levels of backscat ter in side-scan sonar mosaics. There also appears to be a relationship between acoustic sediment cl asses and water depth. These data were subjected to depth compensation processing, so this relationship (dep th and class) shoul be related to sediment properties and not physical changes in the echos waveform experienced at variable depth, e.g. spreadi ng or attenuation. Sediment sampling confirms that in general, the finer gr ained sediments (mud) are found in the deeper parts of Sa Harbor. The shallower shoreline region and the shoal in the southeas t of Safety Harbor are primarily comprised of the coar ser grained material (fine sand). The assessment that grain size is the main control on classifica tion pr re analysis was limited primarily to grain si ze, percent carbonate, and percent organic matter; thus leaving many variables known to control acoustic response, such as poros density, and roughness unmeasure d. Precise quantific ation of porosity, density, roug and velocity structure requires measurement in-situ or lab analysis of undisturbed samples, grain size analysis does no t (Jackson and Richardson, 2007).
67 in sediments ssigned to a particular acoustic class. Grain size distribution within Acoustic Class 4 and ss ts a mong nt bility tic cl assification, although erhaps not uniformly across phi values. When sediment mean grain size data is plotted stic ic matter a nd acoustic classification is bserved. Changes in the per centage of carbonate material and organic matter in the t is positive tic response, also may have played a part the classification of these data. This inference has been drawn from the obvious a are There appears to be marked similarity among grain size distributions with a 8, representing fine sand and mud respectively, are unimodal. In the case of Class 4 this mode is centered at approximately three phi or fine to very fine sand. Each of the five samples contained in this class share this trend. The single mode present in Class 8 is located about the 8-12 phi range, or clay. Agai n, the three samples contained in this cla share this trend. The grain size distribution of sediments contai ned in Class 6 represen departure from the unimodal tr ends of Class 4 and 8. The grain size distribution of sediments in Class 6 is bimodal. The two m odes present are at about three phi and 8-12 phi, which corresponds to a mode in both sand and clay. There is greater diversity a grain size distributions within sediments contained in this cl ass. That being said, it is important to note that all samples in Class 6 share this bi-modality trend. Also, in each of the four samples these modes occur at the sa me phi value. The sand and clay compone of the samples within this class occur at similar percentages, with the exception of Sample 10. Generally speaking, Class 6 sediments have comparable percentages of sand and mud within individual samples. Class 6 is not as restricted with respect to varia in sediment properties as are classes 4 and 8. While there may be greater diversity in this class, the sediments that compri se it are in fact similar. Sediment grain size appears to exhibit co ntrol over acous p against acoustic class, there appears to be a continuum in grain size upon which acou classifications do not occur at pronounced changes. When these same data are plotted as percent silt and clay (mud), three somewhat distinct m odes occur which correspond to acoustic classification. This suggests that pe rcent mud may have a stronger influence over acoustic classification than mean grain size. Also association between carbona te content, organ o sediments does correspond to changes in acoustic classification. However, this association appears to be related more closely to grain size rather than composition. I clear in these data that levels of carbonate material and organic matter display a correlation with mud content. Such a relatio nship suggests that carbonate material and organic matter represent constituents of silts and clays. Mud in turn, appears to be a dominant influence on acoustic classification. Roughness, another dominant control on acous in agreement between the geographic distribution of acoustic classes and side-scan mosaics depicting contrasting levels of backscatter. Backscatter levels in side-scan sonar dat known to be controlled at least in part by the seabeds texture or roughness. If changes in classes of acoustic data are in agreement with contrasting backscatte r regions in side-scan mosaics, perhaps these linkages are related to roughness of the seabed. Since grain size plays a part in seabed roughness, this may be what the side-scan sonar data reflects.
68 ent classification system is apable of discerning rather subtle difference in sediments with respect to grain size. All e 200 kHzs enhanced ability to work in allow water. When operating the transducer at 200 kHz, we were able to record usable t ion appears to be cont rolled in part by sediment grain size. hanges in the grain size distribution of sediments from sample to sample are e bed here Hz sediment classification data and e side-scan mosaic it overlies (Figure 44). Lit tle contrast is evident within the overall f Oo~ There are other contributors to roughness such as mi crotopography, small sediment ripples, and biological material which may be important. It appears, based on these data, that the QT C acoustic sedim c of the sediment samples mean grain si ze fell within 2.61 and 7.63 phi, a relatively narrow range across the Wentworth scale. These mean grain sizes span the size classes of fine sand, very fine sand, coar se silt and medium to fine silt. It appears that this technology possesses a discriminatory power far greater than an abil ity to discern gross differences in sediment grain size composition. One lesson clearly illuminated in this survey is th sh echoes in water less than one meter deep. Interference between the outgoing pulses, transmit ring down, and the returning echo limited our ability to operate at 50 kHz in very shallow water. Dissimilarity among 50 kHz and 200 kHz track lines has limited direc comparison between frequencies. However, some conclusions may be drawn from the limited 200 kHz data. The 200 kHz classificat C accompanied by changes in acoustic class. Acoustically speaking, there appears to b more diversity in the seabed than sedi ment data would suggest. The grain size distribution as well as the pe rcent carbonate and organic ma tter are very similar among samples SH-4 and SH-5 (Figure 44), yet the acoustical representation of the sea is somewhat different. This suggests echo cl assification is being influenced by factors other than grain size, perh aps small-scale roughness. Also noteworthy is the comparison between the 200 k th low backscatter levels presen t in both images, yet acoustic classification data suggests diversity. It is possible that the two methods are responding to different characteristics o the seabed. The difference in frequency and angle of incidence between the two instruments (100 kHz Side-Scan Sonar and 200 kHz vertical incidence transducer) may explain some of the discrepancy seen here. ~o
Figure 44. 200 kHz acoustic class variability conflicts with 100 kHz side-scan sonar imagery that suggests homogeneity. Also, sediment samples SH-6 and SH-5 are very similar. The along-track variability or patchiness of the 200 kHz acoustic classification data is unable to be directly explained by the analysis contained here. A more intensive ground-truthing campaign comprised of multiple techniques, i.e. porosity, velocity, density, and roughness measurements may be instructive in determining the cause of the acoustic complexity seen in sediments of like grain size composition. Possible seabed characteristics controlling classification include the influence of organisms, e.g. bioturbation and algal cover. Changes in microtopography or roughness, such as small sediment ripples, may also contribute to this patchiness in the data. Despite the difference in track locations between the 50 kHz and 200 kHz datasets, comparison between the two is possible. Comparison between the two frequencies (50 and 200 kHz) shows some marked differences. One prominent distinction between the two data sets is that the 200 kHz data discriminated more diversity in sediments where the 50 kHz data set suggested homogeneity. This is especially evident in the central portion of Safety Harbor where very fine grained sediments are found. Class 8 in the 50 kHz data clearly dominated this region, whereas several acoustic classes are evident in the 200 kHz data. Several factors may have contributed to this result, the wider beam-width of the 50 kHz signal and thus larger footprint may have had a greater averaging effect on sediment classification. Furthermore, the higher frequency data (200 kHz) possesses a much smaller wavelength and may be more discriminatory of subtle changes in seabed characteristics. Acoustic Classification of Carbonate Hard Bottoms The most prominent trend in the three hard bottom surveys is the presence of a single, dominant acoustic class pervasive throughout the survey area. This trend is somewhat 69
70 diminished in the 200 kHz survey yet sill appa rent with respect to Class 5, representing slightly more than 50% of the total data. In itially, the fact that a single acoustic class dominated much of a seabed thought to cont ain a high degree of diversity appeared problematic. Broadening the investigation to include side-scan sonar, sediment sampling, diver observations and rock collection led to a reasonable explanation for this trend in the data. It appears that in each survey the dom inant acoustic class repr esents unconsolidated sediments, more specifically medium to fine grained mixed quartz/carbonate sand. Importantly, the single dominant acoustic class from each survey over-simplifies classification of the sediments. The sediments associated with the hard bottom are rather diverse, a characteristic not captured in the acoustic classes presented here. It is possible that if further splits were conducted, beyond wh at was indicated as statistically warranted, that these dominant classes would partition in a manner that would depict differences in the sediments connected with the hard bottom. Nothing appears in the literature specific to the hard bottoms of Tampa Bay, nor their interaction with sediments or role as sediment producers. Obrochta et al. (2003), describes hard bottoms ranging in age from Miocene to Quaternary located offshore of Tampa Bay and proposes them to be a sediment source. Based on the data presented here it would appear that th e hard bottom surveyed in Tampa Bay is also a sediment source. The presence of lithoclasts w ithin sediments in and around th e hard bottom confirms this idea. Research on similar carbonate hard botto ms, analogs to those of Tampa Bay is not lacking. Extensive work has been publis hed on the relationship between sediment distribution, hard bottom charac teristics and associated flora and fauna off the North Carolina coast. Hard bottom habitats on the Carolina shelf are frequently buried or have their exposed surface area modified by mobile sheets of Holocene sediments (Riggs et al., 1996). The time scales on which these hard bottoms are buried and exposed may be on the order of days or substantially longer periods (R iggs et al., 1998). The thickness of mobile sediments and the extent to which they redu ce exposure of the hard bottoms determines in part, the type of benthic community that exists in these habita ts (Renaud et al., 1996; Riggs et al., 1996; Ri ggs et al., 1998). There are definite differences between the hard bottom environments of the Atlantic shelf and Tampa Bay, e.g. geologic, biologic, a nd hydrodynamic factors. Despite this, the results of the research presented here suggest the possibility of similar conditions with respect to the general interactions between sediments and hard bottom habitats. Strong storm events have demonstrated the capability to drastically alter sediment distributions over the North Carolina hard bottoms (Rena ud et al., 1997; Riggs et al., 1996; Riggs et al., 1998). Tampa Bay experiences summertime convective and tropical storms, as well as winter frontal systems which are capable of transporting benthic sediments leading to changes within the hard bottom habitat. Interpretation of acoustic cl assification data and side-scan sonar mosaics suggests the presence and movement of sediments on the Tampa Bay hard bottom. Small dunes appear in both the 100 and 400 kHz side-scan data, their long axis orientated generally
north-south. These features may represent flow traverse bedforms in which case they could be classified as small two-dimensional dunes in accordance with Ashley (1990). If the prevailing current flow was north-south during these features formation they might be classified as sand ribbons. Regardless of the flow regime under which these features formed, based on these data, active sediment transport may be occurring across this particular hard bottom. The strongest evidence demonstrating the presence of mobile sand sheets within the hard bottom comes from diver observations. Divers not only confirmed the presence of sediments atop the hard bottom, but also witnessed sediments covering the stalks of gorgonian colonies. The presence of sediment, especially at centimeter thicknesses, is known to limit the recruitment of juvenile gorgonians (Gotelli, 1988). It seems reasonable that large gorgonians having their bases covered in several centimeters of sediment would indicate ongoing sediment transport. Similar conclusions were based on this same logic in Gotelli (1988). Figure 45. Images depicting the complex nature of the hard bottom. A. Very coarse gravel and shell material. B. Cross section of the hard bottom rock depicting a very heterogeneous structure. C. Example of the hard bottoms rugosity. D. Gorgonian soft coral attached to the hard substrate. E. Examples of fine and coarse grained sediments recovered around the hard bottom. F. Hard bottom rock encrusted with mollusks with brown algae. 71
72 The trend in these data of a single dominant class appears to be explained by the presence of sediments within the hard bottom habitat. Simply, these classes are pervasive in the survey area because sediments are pervasive in the survey area. What is less conspicuous is an explanation for the remaining classes which appear highly scattered. The small scattered classes interpreted as hard bottom, if left classified in the manner achieved from the process described in the methods secti on (manual clustering), do not appear to be clearly associated with readily identifiable changes in the seabed. Because hard bottom habitats are by definition, a hard substrate at the seafloor, they are capable of supporting epifauna and flora. This fact is in part what makes them so important and why they (hard bottoms) share a large part of the focus of this research. The presence of various organisms on the hard bottom and the results of their activities serve to complicate the surface of the substrat e on which they reside. This addition to seabed heterogeneity serves to complicate matters pertaining to acoustic classification and frustrates efforts to di scern exactly what conditions on the seafloor are controlling classification. The ability to accurately correlate an acoustic class with a particular type or characteristic of the seafloor is directly related to ground-truthing (Collins and McConnaughey, 1998). Many physical characteristics of the seabed have been demonstrated to influence echo classification, i.e. grain size, porosity, se diment density, microtopography, and benthic flora and fauna (Bornhold et al., 1999; Collins and Lacroix, 1997; Collins and McConnaughey, 1998; Quester Tangent, 2003). The more heterogeneous a seabed is with respect to influencing factors, it stands to reasons, the mo re involved is the nature of acoustic response and thus classification. Di scovery and explanation of the seabed characteristics governing classification in a diverse setting may require multiple sophisticated ground-truthing sc hemes not conducted here. Reports on the performance of acoustic classification technologies demonstrate the shortcomings of the QTC system in accurately classifying reef covered seafloors. The idea is put forward that it is not the reefs composition but rather the extreme rugosity of these features that cause instab ility in the classification of echoes. It is suggested that such instability and the resultant chaotic natu re of the acoustic classes be used as an indicator of such areas (Hamilton et al., 1999) It appears at least possible and perhaps likely that a similar phenomenon may be occurr ing in the data acquired over the hard bottom habitat. Much scatter is seen in these data, particularly in re gions with the highest backscatter levels in side-scan sonar data. It may be the ca se that high variability in acoustic classes may serve as a proxy for identifying carbonate hard bottoms in and around Tampa Bay. The class complexity analysis was successful in identifying the extents of the hard bottom habitat, a result that lends credence to the ideas put forward here.
73 Acoustic Classification of Submerged Aquatic Vegetation (SAV) Detection of submerged aquatic vegetation (S AV) or seagrass with the QTC system was not straight-forward. Acoustic classes de termined from QTC Impact processing showed alignment with known areas of SAV. Class 1 produc ed in the seagrass survey agreed well with SAV distributions seen in aerial photography. Class 1 (potential seagrass) was absent from the deeper areas offshore and within a ch annel, furthering the case for its association with SAV. Its pres ence was abruptly terminated within sand patches located throughout the seagrass meadows. However, Class 1 appears in the deep dredged area just off the beach where there is little evidence for vegetation. Both sediments and vegetation are likely in fluencing the classification of echoes. Ambiguity in the classes potentially repr esenting SAV appears to be linked to the influence of the sediments. Vegetation can influence the durati on of echoes, as can increased penetration into sediments. It is possible that confusion in classification may result between echoes of greater duration resulting from decreased attenuation within sediments and those from vegetation. Quester Tangent has addressed some of the p itfalls associated with classifying SAV in Preston et al. (2006) and Quester Tangent ( 2005). One principally im portant issue is the placement of the bottom pick at the echo which actually represents the seabed, more precisely, the sediment water in terface. This becomes an issu e in regions with SAV, as echoes resulting from vegetation may have greater amplitude than the underlying sediments. Because the portion of the echo subj ected to analysis typically begins just above the bottom pick and extends much further in time afterward, if the pick is left at the top of the SAV, the waveform may include both vegetation and sediment signal. This convolved signal, when classified, may lead to degradation in the accuracy of the resulting classes. It may be the case that this precise error is occurring in the data presented here. Unfortunately, it was not possible to implement all of the suggestions put forward in Quester Tangents literature. Had the su rvey conditions been favorable for the implementation of the recommendations, perhap s the results of the SAV survey would have dramatically improved. An inability to determine the echo representing the seabed plagued attempts to modify the normal pr ocessing routine. Oftentimes, the echo representing the sediments was masked by the overlying SAV. Using nearby echoes from un-vegetated seafloors to estimate where the bottom should be was unsuccessful; owing to the fact that frequently the presen ce of SAV coincided with abrupt changes in bathymetry. If one were able to make the assumption that the sediment water interface would be at the same depth on either side of a vegetated region of the seafloor, as well as
74 across the extent of the vegetation, then gue ssing at where to put the pick would be straight-forward. This assumption would have been false in the data presented here. Another problem resulted from the extreme shal lowness of the survey area, particularly where SAV occurred. Typically, the bulk of th e waveform subjected to analysis comes after the bottom pick. In the case of SAV, the pick is placed at the sediment water interface and then the time window for analysis is shifted above the pick. This allows for incorporation of the vegetation signal in the overlying water column and diminishes that of the sediments. In the very short acoustic r ecords, resulting from very shallow water (< 1m), moving the analysis record earlier in time may have captured the signal of the outgoing pulse and or the transmit ring-down. Lessening the extent to which the window is moved earlier in time may not have removed the influence of the sediments. QTC Sideview The results of the QTC Sideview side-s can imagery classification were quite promising. This technique was particularly effective in identifying the carbonate hard bottom habitat. When compared with the sing le-beam classification of the hard bottom, the swath systems classification appears more obvious----lacking th e scatter associated with the single-beam transducer data. The Sideview classification was also successf ul in the soft sediment environment of Safety Harbor. Here, in the soft sediments, the swath systems classification was more similar to that of the single-beam than was the case in the hard bottom. Similar trends were present in the single-beam and side-scan classification, with gr ain size appearing to play a dominant role in both. The Sideview classification clearl y shows the area of increased mud in the central portion of Safe ty Harbor, the same region present in the single-beam data, and confirmed with sedi ment sampling. Coarser-grained sediments associated with the shoreline and shoal area are also clearly iden tified by the Sideview acoustic classes. Changes in sediment grai n size have a clear association with the Sideview acoustic classes. However, smallscale roughness is almost certainly another influencing factor in the cl assification of these data. The least instructive of the Sideview clas sification maps was that of the submerged aquatic vegetation (SAV). The classification process was successful in identifying some areas of SAV, although not to the degree that the acoustic cl asses represented the true coverage of the vegetation. A linear belt of SAV bounding the easte rn edge of the shallow shoal was clearly delineated in th e Sideview data. The SAV occurring atop the shallow, sandy shoal is represented as patchy and discontinuous in the Sideview classification results. The SAV coverage in this region appears more continuous in sidescan sonar backscatter imagery and aerial photography. This result may be explained by the complex backscatter respons e of SAV in the side-scan im agery. Backscatter response, a primary determinant of acoustic classes, does not appear to be uniform for SAV.
75 In general, the Sideview classification tec hnique performed well across the board. This technique appears to provide as much, and in some cases more, information about the seabed as the single-beam data. This is partially true due to the vast increase in coverage provided by the swath of the side-scan sonar. The near 100% coverage achieved with the side-scan acoustic classificati on method greatly reduces the need for data interpolation between track lines. In each of the surveys, the Sideview classification did result in some artifact classes. In most cases these classes (a rtifact) were clearly identif iable and able to be removed with editing methods. In some cases, i.e. the red class in the hard bottom data, an acoustic class appeared to have both artif act properties as well as true correlation with seabed type. This is probably the result of the fact that backsc atter levels ar e a function of sonar specifications and grazi ng angle, as well as changes in seafloor characteristics. Assessment Based on the research presented here, the QT C system operated most effectively in the soft sediment environment (Safety Harbor) in Tampa Bay. Acoustic classes appear to correspond well with grain size distributions particularly mud concentration. Soft unconsolidated sediments appear to represen t non-complex acoustic conditions. The lack of complexity in the soft sediment envir onment appears to facilitate straightforward classification of the seabed. Considering th at QTC Impact processing software reduces variability in waveforms to three principle components, the fewer variables contributing to differences in echo characteristics, the mo re meaningful the classification results. The hard bottom survey stands in contrast to the soft sediment survey with respect to its complex acoustic conditions. It would appear th at the natural characte ristics of the Tampa Bay hard bottom do not lend themselves to straightforward acoustic classification. It seems likely that the hard bottom is an ephemeral feature changing on unknown timescales. Sediments are imported into th e hard bottom from surrounding areas adding to those produced from the hard bottom its elf. These sediments vary in grain size, chemical composition and perhaps most impor tantly in gravel content and thickness. Variations in sediment thickness may determin e the extent to which the acoustic signal interacts with the rocks surface. This could lead to fluctuations in echo amplitude and influence classification. Also, sediments have the potential to change the seabeds roughness by masking the rocks surface, agai n possibly affecting classification results. The presence of vegetation, sponges, and soft corals attached to the substrate only serve to further complicate the acoustic set ting of the Tampa Bay hard bottom. Detection of SAV with the QTC system wa s not without ambiguity. The influence of vegetation is seen in the classification re sults. Acoustic classe s align themselves reasonable well with seagrass interpreted from aerial photog raphy. However, these same classes that appear to trac k along with vegetation also occupy non-vegetated seabeds, although to a much lesser degree. It seems th at in some cases the vegetation signal may become convolved with that of the sediment s and lead to misclassification. If aerial
76 photography or side-scan sonar imagery is considered when interpreting acoustic classification data, the results may be useful for detecting SAV out of the range of optical detection. To definitively say that the QTC system performed better or worse across various seabeds may be misleading. The limiting factor in the performance of the QTC system is as much a function of ground-trut hing as anything internally related to the technology. It is important to remember that regardless of how well acoustic classes correlate to features on the seabed that researchers are interest ed in, i.e. sediment type, vegetation, and substrate structure, these data represent only changes in the physical/acoustical properties of the seafloor. The physical changes in the s eabed that are identifie d in the classification may or may not be readily discernable with practical ground-truthing. Acoustic classification data does not stand alone and a comprehensiv e understanding of the method is required for accurate interpretation, part icularly when ground-truth or acoustic imagery is lacking. ~oOo~
77 Conclusions Acoustic sediment classification is a useful tool for benthic habitat mapping in shallow estuarine environments Acoustic classification data and side-scan sonar imagery compliment one another well, each delivering information where the other is lacking. Acoustic classification can be very diagnostic in soft sediments Grain size, particularly pe rcent mud, appears to be a strong influence on acoustic classification Complexity/heterogeneity of the hard bottom environment makes diagnostic predictions based on classes impractical. Carbonate hard bottoms represent comp lex acoustic conditio ns; use of this condition appears to be effective for their location Sediment grain size, particularly mud a nd gravel, appears to strongly influence acoustic classes. Acoustic classification is able to identify carbonate hard bottoms; acoustic complexity appears to be a good indicator of hard bottom habitats. The Tampa Bay hard bottom is a sediment source and is subject to reduced exposure from sediment transport. Detection of SAV is possible with unm odified processing, although there may be some misclassification of bare sediments as being vegetated. 200 kHz data displayed greater acoustic he terogeneity than was depicted in 50 kHz acoustic data, side-scan sonar im agery and sediment sampling. Also, 200 kHz data was not able to identify the hard bottom as well as the 50 kHz data. The importance of meaningful ground-tr uth cannot be oversta ted. Interpretation and usefulness of acoustic cl assification is a direct function of the ability to ground-truth the survey area. A relatively minor presence of gravel a ppears to influence acoustic classification
78 List of References Akal, T., 1972. The relationship between the ph ysical properties of underwater sediments that affect bottom reflection. Marine Geology, vol. 13, p. 251-266. Anderson, J.T., Gregory, R.S., and Collins, W. T., 2002. Acoustic classification of marine habitats in coastal Newfoundland. ICES Jour nal of Marine Science, vol. 59, p. 156-167. Ashley, G.M., 1990. Classification of Larg e-Scale Subaqueous Bedforms: A New Look at an Old Problem. Journal of Se dimentary Petrology, vol.60, no. 1, pp. 160-172. Bornhold, B.D., Collins, B. and Yamanaka, L., 1999. Comparison of seabed characterization using sidescan sonar a nd acoustic classification techniques. Proceedings Canadian Coastal Conference 1999, 15 pp. Arthur, D. and Vassilvitskii, S., 2007. k-m eans++ The Advantages of Careful Seeding. 2007 Symposium on Discrete Algorithms (SODA). New Orleans, Louisiana. Briggs, K.B., Williams, K.L., Jackson, D.R., Jones, C.D., Ivakin, A.N., Orsi, T.H., 2002. Fine-scale sedimentary structure: implica tions for acoustic remote sensing. Marine Geology, vol. 182, p. 141-159. Bornhold, B.D., Collins, B., Yamanaka, L., 1999. Comparison of seabed characterization using sidescan sonar and acoustic classification techniques. Proceedings of the Canadian Coastal Conference, 1999. Briggs, K.B., Williams, K.L., Jackson, D.R., Jones, C.D., Ivakin, A.N., Orsi, T.H., 2002. Fine-scale sedimentary structure: implica tions for acoustic remote sensing. Marine Geology, vol. 182, p. 141-159. Burns D.R., Queen C.B., Sisk H., Mullarkey, W., Chivers, R.C., 1989. Rapid and convenient acoustic sea-bed discrimination for fisheries applications. Proceedings of the Institute of Acoustics 11, Part 3, p.169-178. Chivers, R. C., Emerson, N., and Burns, D. R., 1990. New acoustic processing for underway surveying. The Hydrographic Journal, vol. 56, p. 9-17. Collier, J.S. and Brown, C.J., 2005. Correlation of sidescan backscatte r with grain size distribution of surficial seabed sedi ments. Marine Geology, vol. 214, p. 431-449.
79 Collins, W.T. and Lacroix, P., 1997. Operational philosophy of acoustic waveform data processing for seabed classification. In: Proc eedings of Coastal Open Space Utilization (COSU 97) International Conf erence, Singapore, 1, p. 225-234 Collins,W.T. and McConnaughey, R.A., 1998. Acousti c classification of the sea floor to address essential fish habitat and marine protected area requirements. Proceedings of the 1998 Canadian Hydrographic Conferen ce, Victoria, Canada, p. 361-368 Collins, W.T., and Rhynas, K., 1998. Acoustic seabed classification using echo sounders: operational considerations and strategies Proceedings of Canadian Hydrographic Conference Victoria, B.C., 7 pp. Ellingsen, K.E., Gray, J.S., and Bj rnbom, E., 2002. Acoustic classification of seabed habitats using the QTC VIEW system. ICES Journal of Marine Science, vol. 59, p. 825835. Faas, R. W., 1969. Analysis of the relati onship between acoustic reflectivity and sediment porosity. Geophysics, vol. 34, no. 4, p. 546-553. Folk, R.L., 1980. Petrology of Sedimentary Rocks. Hemphills, Austin, TX, 159 pp. Foster-Smith, R.L., Brown, C.J., Meadows, W.J., White, W.H., and Limpenny, D.S., 2004. Mapping seabed biotopes at two spatial scal es in the eastern En glish Channel. Part 2. Comparison of two acoustic ground discrimination systems. Journal of the Marine Biological Association of th e United Kingdom, vol. 84, p. 489-500. Freitas, R., Rodrigues, A., and Quintino, V ., 2003. Benthic biotopes re mote sensing using acoustics. Journal of Experimental Ma rine Biology and Ecology, vol. 285-286, p. 339353. Galloway, J.L., and Collins, W.T., 1998. Dual frequency acoustic classification of seafloor habitat using the QTC View. Proceedings of IEEE Oceans, Nice, France. Gotelli, N.J., 1988. Determinants of recruitmen t, juvenile growth, and spatial distribution of a shallow-water gorgonian. Ecology, vol. 69, p. 157-166. Greenstreet, S.P.R., Tuck, I.D., Grewar, G. N., Armstrong, E., Reid, D.G., Wright, P.J., 1997. An assessment of the acoustic survey technique, RoxAnn, as a means of mapping seabed habitat. ICES Journal of Marine Science, vol. 54, p. 939-959. Hamilton, E. L., 1970. Reflection coefficients and bottom losses at normal incidence computed from Pacific sediment prope rties. Geophysics, vol. 35, no. 6, p. 995-1004. Hamilton, E. L., 1972. Compressional-wave attenuation in marine sediments. Geophysics, vol. 37, no. 4, p. 620-646
80 Hamilton, L. J., Mulhearn, P. J., and Poeckert, R., 1999. Comparison of RoxAnn and QTC-View acoustic bottom classi fication system performance for the Cairns area, Great Barrier Reef, Australia. Continental Shelf Research, vol.19, p.1577-1597. Hamilton, L. J., 2001. Acoustic seabed cl assification systems. Department of Defense, Defense Science and Technology Organization. 150 pp. Hutin, E., Simard, Y., and Archambault, P ., 2005. Acoustic detection of a scallop bed from a single-beam echosounder in the St. Lawr ence. ICES Journal of Marine Science, vol. 62, p. 966-983. Jackson, D.R., Winebrenner, D.P., Ishimaru, A., 1986. Application of the composite roughness model to high frequency bottom backs catter. Journal of the Acoustical Society of America, vol. 79, no. 5, p.1410-1422. Jackson, D.R. and Briggs, K.B., 1992. High frequency bottom backscattering: roughness versus sediment volume scattering. Journal of the Acoustical Society of America, vol. 92, no. 2, p. 962-977. Jackson, D.R., and Richardson, M.D., 2007. Monogr aph Series in Underwater Acoustics: High-Frequency Seafloor Acous tics. Springer, New York. Legendre, P., Ellingsen, K.E., Bjrnbom E., Casigrain, P., 2002. Acoustic seabed classification: improved statistical method. Ca nadian Journal of Fisheries and Aquatic Sciences, vol. 59, p. 1085-1089. Locker, S.D. and Wright, A. K., 2003. Benthic habitat mapping for habitat suitability modeling in Rookery Bay National Estuarine Re search Reserve. Fina l report, College of Marine Science, University of South Florida, St. Petersburg, Fl., 84 pp. Locker, S.D., 2006. Mapping submerged aquatic vegetation using side-scan sonar, Cape Romano Shoals, Florida, Final report, College of Marine Science, University of South Florida, St. Petersburg, Fl., 18 pp. Locker S.D. and Jarrett, B.D., 2006. Benthi c habitat mapping in Pumpkin Bay, Florida: establishing baseline conditions prior to Picayune Strand Restoration. Final report, College of Marine Science, University of South Florida, St. Petersburg, Fl., 60 pp. Lyons, A.P., Anderson, A.L., Dwan, F.S., 1994. Acoustic scattering from the seafloor: modeling and data comparison. Journal of the Acoustical Society of America, vol. 95, no. 5, p. 2441-2451. McCann, C. and McCann, D.M., 1969. The atte nuation of compressional waves in marine sediments. Geophysics, vol. 34, no. 6, p. 882-892.
81 Mandelbrot, B., 1982. The Fractal Geometry of Nature. Freeman, San Francisco. Morrison, M.A., Thrush, S.F., and Budd, R., 2001. Detection of acoustic class boundaries in soft sediment systems using the seafloor discrimination system QTC VIEW. Journal of Sea Research, vol. 46, p. 233-243. Moyer, R.P., 7, B., Banks, K., and Dodge, R.E., 2005. Assessing the accuracy of acoustic seabed classification for mapping coral reef environments in South Florida (Broward County, USA). International Journal of Tropical Biology and Cons ervation, vol. 53, p. 175-184. Obrochta, S.P., Duncan, D.S., Brooks, G.R., 2003. Hardbottom development and signifigance to the sediment-starved west-centr al Florida inner contin ental shelf. Marine Geology, vol. 200, p. 291-306. Peitgen, H., Jurgens, H., and Saupe, D., 2004. Chaos and Fractals: New Frontiers of Science. Springer, New York. Preston, J.M., Rosenberger, A., Collins, W. T., 2000. Bottom classification in very shallow water. Proc Oceans 00, Newport. Press, W. H., Flannery, B.P., Teukols ky, S.A., Vetterling, W.T., 1989. Numerical Recipes. Cambridge University Press, Cambridge. Preston, J.M., Christney, A.C., Collins, W.T ., and Bloomer, S, 2004. Automated acoustic classification of side-scan images, Oceans 2004, Kobe, Japan. Preston, J.M., Inouchi, Y., Shioya, F., 2006. Ac oustic classification of submerged aquatic vegetation. Proceedings of the Eighth Europ ean Conference on Underwater Acoustics, Carvoeiro, Portugal, 12-15 June. Quester Tangent Corporation, 2003. Mapping intertidal oysters with QTC View. Technical Report, SR41-03. Quester Tange nt Corporation, Sidney, B.C. Canada. Quester Tangent Corporation, 2005. Classifyi ng seaweed with QTC Impact: Report on research on seaweed discrimination and advice to users of QTC Impact modified for seaweed mapping. Document number: DTN-SNGL-0112, Quester Tangent Corporation, Sidney, B.C. Canada. Rayleigh, J.W.S., 1945. The Theory of Sound. Dover, New York, N.Y., Vol. 2, p. 504. Renaud, P.E., Riggs, S.R., Ambrose, W.G., Schmid, K., and Snyder, S.W., 1997. Biological-Geological Interac tions: Storm Effects on Macroalgal Communities Mediated
82 by Sediment Characteristics and Distributi on. Continental Shelf Research, vol. 17, no. 1, p. 37-56. Riegl, B.M., Halfar, J., Purkis, S.J., and Godinez-Orta, L., 2007. Sedimentary facies of the eastern Pacifics northernmost reef-lik e setting (Cabo Pulmo, Mexico). Marine Geology, vol 236, p. 61 and 77. Riegl, B., Moyer, R.P., Morris, L., Virnstein, R., and Dodge, R.E., 2005. Determination of the distribution of shallow-water seagra ss and drift algae communities with acoustic seafloor discrimination. International Journa l of Tropical Biology and Conservation, vol. 53, p. 165-174. Riegl, B.M., and Purkis, S.J., 2004. Detection of shallow subtidal corals from IKONOS satellite and QTC View (50, 200 kHz) singlebeam sonar data (Arabian Gulf; Dubai, UAE). Remote Sensing of Environment, vol. 95, p. 96-114. Riggs, S.R., Snyder, S.W., Hine, A.C. and Mearns, D. L., 1996. Hardbottom Morphology and Relationship to the Geologi c Framework: Mid-Atlantic Continental Shelf. Journal of Sedimentar y Research, vol. 66, no. 4, p. 830-846. Riggs, S.R., Ambrose, W.G., Cook, J.W ., Snyder, S.W., and Snyder, S.W., 1998. Sediment Production on Sediment-Starved C ontinental Margins: The Interrelationship Between Hardbottoms, Sedimentological and Benthic Community Processes, and Storm Dynamics. Journal of Sedimentary Research, vol. 68, no. 1, p. 155-168. Tegowski, J., 2005. Acoustical classification of the bottom sediments in the southern Baltic Sea. Quaternary In ternational, vol. 130, p. 153-161. Tegowski, J., Gorska, N., and Klusek, Z., 2003. Statistical analysis of echoes from underwater meadows in the eutrophic Puck Ba y (southern Baltic Sea). Aquatic Living Resources, vol. 16, p. 215-221. van Walree, P.A., Tegowski, J., Laban, C ., and Simons, D., 2005. Acoustic seafloor discrimination with echo shape parameters: A comparison with ground-truth. Continental Shelf Research, vol. 25, p. 2273-2293. von Szalay, P.G., and McConnaughey, R.A., 2002. The effect of slope and vessel speed on the performance of a single beam acoustic seabed classification system. Fisheries Research, vol. 54, p. 181. Wentworth, C.K., 1922. A scale of grade and cl ass for clastic sediments. Journal of Geology, vol. 30, p. 377-392.
83 Wienberg, C., and Bartholoma, A., 2005. Acoustic seabed classification in a coastal environment (outer Weser Estuary, German Bight): a new approach to monitor dredging and dredge spoil disposal. Continen tal Shelf Research, vol. 25, p. 1143-1156. Wilding, T. A., Sayer, M.D.J., and Provost, P.G., 2003. Factors affecting the performance of the acoustic discrimination system RoxAnn. ICES Journal of Marine Science, vol. 60, p. 1373-1380. Yamamoto, T., 1996. Acoustic scattering in the ocean from velocity and density fluctuations in the sediment s. Journal of the Acoustic Society of America, vol. 99, p. 866869.