SEAFLOOR CHARACTERIZATION

OF THE HISTORIC AREA REMEDIATION SITE USING

ANGULAR RANGE ANALYSIS

BY

LUIS A. SOARES ROSA

BSc in Geology, Faculdade de C iencias/Universidade de Lisboa, Portugal, 1993

THESIS

Submitted to the University o f New Hampshire

in Partial Fulfillm ent of

the Requirem ents fo r the Degree o f

M aster o f Science

in

Ocean Engineering (Ocean Mapping)

Septem ber 2007

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
 UMI Number: 1447908

INFORMATION TO USERS

The quality of this reproduction is dependent upon the quality of the copy
submitted. Broken or indistinct print, colored or poor quality illustrations and
photographs, print bleed-through, substandard margins, and improper
alignment can adversely affect reproduction.
In the unlikely event that the author did not send a complete manuscript
and there are missing pages, these will be noted. Also, if unauthorized
copyright material had to be removed, a note will indicate the deletion.

UMI
UMI Microform 1447908
Copyright 2007 by ProQuest Information and Learning Company.
All rights reserved. This microform edition is protected against
unauthorized copying under Title 17, United States Code.

ProQuest Information and Learning Company

300 North Zeeb Road
P.O. Box 1346
Ann Arbor, Ml 48106-1346

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
 This thesis has been exam ined and approved.

Thesis Director, QK^Gciano Fonseca,

Research Associate Professor of Ocean Engineering, UNH

M a
r
Dr. Lai(ryjA. Mayer/
Professor of Earth Sciences and Ocean Engineering, UNH

Dr. James V. Gardner,

Research Professor of Earth Sciences, UNH

L e O ___
Dr.. Larry G. Ward,
Research Associate Professor of Earth Sciences, UNH

Date

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
 ACKNOWLEDGEMENTS

I would like to thank my wife Ana and little Sofia for making this adventure

possible and Dr. Larry Mayer, while CCOM Director, for accommodating all the logistics

of a family moving between continents and living in two different states. I also would like

to express my thanks to my thesis advisor, Dr. Luciano Fonseca, and to the other

members of the committee, Dr. Larry Mayer, Dr. James Gardner and Dr. Larry Ward,

whose invaluable guidance and support helped me completing my studies.

Many thanks to SAIC, and especially to Tom Waddington for providing all the

multibeam sonar and SPI data used in this study, and also to Natasha Pinckard, Pamela

Luey, Michael Cole and the crew of the RA/ Beavertail, for all the help during data

collection in 2006. Thanks also to Dr. Larry Ward for the use of the Coastal Geology

Laboratory - UNH, to the Portuguese Hydrographic Institute, and especially to Dra.

Aurora Bizarro and to my colleagues at the Marine Geology Department, for the grain

size analysis with laser diffraction.

Many thanks to all my CCOM colleagues and friends, that in one way or the

other, made this a fruitful and unforgettable experience.

Finally a word of acknowledgement for my teachers because although I probably

learned less than they would expect, I learned much more than I ever imagined.

This work was only possible due to the support of the National Oceanographic

and Atmospheric Administration grant NAONOS4001153 and to the leave of absence

provided by the Portuguese Hydrographic institute.

iii

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
 TABLE OF CONTENTS

ACKNOWLEDGEMENTS.................................................................................................... iii

LIST OF TABLES................................................................................................................. vii

LIST OF FIGURES............................................................................................................ viii

ABSTRACT............................................................................................................................xi

CHAPTER PAGE

1. INTRODUCTION............................................................................................................1

1.1 Remote Seafloor Characterization.................................................................. 1

1.2 Interaction of Sound with the Seafloor.............................................................. 4

1.3 Geocoder and Angular Range Analysis (A R A )................................................ 5

1.4 Verification of Model Predictions and the Importance of Mean Grain Size .. 7

1.5 Historic Area Remediation Site (HARS) .........................................................10

1.6 O bjectives.......................................................................................................... 15

2. BACKSCATTER MOSAICS AND ANGULAR RANGE ANALYSIS (A R A ).............. 16

2.1 Backscatter Mosaics Assembled in G eocoder.............................................. 16

2.2 Angular Range Analysis (A R A ).......................................................................18

2.3 Multibeam Data Collection and Processing................................................... 22

3. GROUND TRUTH ......................................................................................................25

3.1 Position Uncertainty.........................................................................................26

3.2 Grab Sam ples.................................................................................................. 27

3.2.1 Surface sediment sam pling........................................................................27

iv

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
 3.2.2 Single sample versus composite sample ..................................................28

3.2.3 Grab sam pler...............................................................................................29

3.2.4 Sediment analysis (grain-size and organic content) .................................30

3.2.5 Uncertainties in sediment analysis.............................................................33

3.3 Sediment Profile Imaging (SPI) and Plan View Image .................................36

4. GROUND TRUTH AND ACOUSTIC REMOTE SENSING RESULTS,

COMPARISON AND DISCUSSION .........................................................................39

4.1 Mean Grain S iz e ..............................................................................................39

4.1.1 Calculation of mean grain size.................................................................. 39

4.1.2 Error introduced by drying the sam ple........................................................41

4.1.3 Error in the estimation of the mean particle s iz e ....................................... 42

4.1.4 Comparison between the SOP (sieve-laser diffraction) and thesieve-

pipette m e th o d ............................................................................................ 44

4.1.5 Estimated uncertainties in the determination of mean grain s iz e .............46

4.1.6 Mean grain size versus penetration of the optical p ris m ...........................47

4.2 Backscatter Strength .......................................................................................49

4.2.1 Backscatter strength versus mean grain s iz e ............................................49

4.2.2 Comparison of backscatter strength for the same type of sediment ....... 53

4.2.3 Differences in backscatter strength between 2005 and 2006 and

temporal evolution of the se a flo o r............................................................. 54

4.2.4 Backscatter strength versus penetration of the optical p rism ................... 58

4.3 Angular Range Analysis.................................................................................. 59

4.3.1 ARA in normal m o d e ................................................................................... 59

4.3.2 Comparison of ARA results with mean grain s iz e ..................................... 62

4.3.3 ARA in supervised segmentation mode (in areas of horizontal

heterogeneity)............................................................................................. 67

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
 4.3.4 Comparison of ARA results with the estimated mean grain size

for the top layer (in areas of vertical heterogeneity)................................. 70

4.3.5 Estimation of the consistency of ARA re s u lts...........................................73

5. CONCLUSIONS......................................................................................................... 74

Future W o rk .....................................................................................................79

APPENDICES...................................................................................................................... 80

APPENDIX A - PENETRATION OF THE SOUND INTO THE SEDIM ENT.................... 81

APPENDIX B - BACKSCATTER MOSAICS APPLYING DIFFERENT A VG s............... 86

APPENDIX C - PROCEDURES FOR GRAIN SIZE ANALYSIS ......................................93

C.1 Normal procedure for grain size a n alysis......................................................93

C.2 Measurement of the natural occurring particles iz e .......................................95

C.3 Measurement of the effect of drying the sam ple............................................ 96

C.4 Pipette analysis............................................................................................... 96

C.5 Loss On Ignition (L O I)...................................................................................... 99

APPENDIX D - EXAMPLES OF ANGULAR RESPONSE ............................................. 100

REFERENCES .................................................................................................................. 104

vi

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
 LIST OF TABLES

Table 3.1 - Analysis performed on each aliquot of sediment samples..............................32

Table 4.1 - ARA solutions within the area of the sediment samples................................ 73

Table C.1 - Pipettes withdrawal times and depths............................................................ 99

vii

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
 LIST OF FIGURES

Figure 1.1 - HARS location..................................................................................................11

Figure 1.2 - HARS 2006 bathymetry...................................................................................12

Figure 2.1 - Example of backscatter strength and angular response...............................18

Figure 2.2 - Histogram of backscatter strength................................................................. 20

Figure 2.3 - ARA in theme mode........................................................................................ 21

Figure 2.4 - 2005 backscatter mosaic and sampling location...........................................23

Figure 2.5 - 2006 backscatter mosaic and sampling location...........................................24

Figure 3.1 - Example of position uncertainty, station 16.................................................... 26

Figure 3.2 - Petit Ponar grab sampler................................................................................. 28

Figure 3.3 - Example of the criteria used to decide where to split the sample................. 32

Figure 3.4 - SPI system....................................................................................................... 37

Figure 4.1 - Histogram of grain size for all samples collected in 2 0 06 ............................40

Figure 4.2 - Mean grain size vs percentage of gravel plus sand for 2006 samples 41

Figure 4.3 - Influence of drying the sample on the mean grain size................................. 42

Figure 4.4 - Evolution of the grain size distribution with time on laser diffraction............ 43

Figure 4.5 - Influence of sample preparation on the mean grain s iz e ..............................43

Figure 4.6 - Relation between the organic content and particle aglomeration................. 44

Figure 4.7 - Difference in the caculation of Mz by laser diffraction and pipettes..............45

Figure 4.8 - Example of grain size distribuition using different techniques.......................47

Figure 4.9 - Penetration of the optical prism versus percentage of gravel plus sand 48

Figure 4.10 - SPI penetration vs mean grain size for 2006 data...................................... 49

Figure 4.11 - Percentage of gravel plus sand vs BS for 2005 and 2006..........................50

Figure 4.12 - Mz vs BS and mean particle size vs BS (2006)........................................... 50

viii

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
 Figure 4.13 - BS, SPI images and sample photos for stations 26, A4 and P2800...........52

Figure 4.14 - SPI images from samples 20028, 20064, L1200 and L2400...................... 54

Figure 4.15 - Grain size statistics for samples 20028, 20064, L1200 and L2400............54

Figure 4.16 - Difference in the backscatter strength between 2006 and 2005.................56

Figure 4.17 - Difference in bathymetry between 2006 and 2005......................................57

Figure 4.18 - Difference in BS draped over the difference in bathymetry........................58

Figure 4.19 - Mean BS vs penetration of the optical prism for 2005 and 2006................ 59

Figure 4.20 - ARA solutions for 2006 d a ta .........................................................................61

Figure 4.21 - ARA solutions and backscatter mosaic for station A12 (2006)................... 62

Figure 4.22 - ARA solutions vs Mz from grab samples...................................................... 63

Figure 4.23 - ARA solutions, SPI images and sample photos for station 17 (2006)........65

Figure 4.24 - Manual segmentation of the area around station A12 (2006).................... 69

Figure 4.25 - SPI images and grab sample photos for station A12 (2006)...................... 69

Figure 4.26 - Angular response and model fit for station A12 (2006).............................. 70

Figure 4.27 - Correlation between the mode and sediment layers....................................71

Figure 4.28 - ARA solutions vs Mz from grab samples after corrections..........................72

Figure A.1 - Mean grain size versus k.................................................................................81

Figure A.2 - Refraction of a ray at 45° grazing angle......................................................... 82

Figure A.3 - Mean grain size vs compressional sound speed........................................... 82

Figure A.4 - Mean grain size vs attenuation....................................................................... 82

Figure A.5 - SB versus grazing angle...................................................................................84

Figure A.6 - Estimation of acoustic wave penetration into the sediment.......................... 85

Figure B.1 - 2006 backscatter mosaic without any AVG....................................................87

Figure B.2 - 2006 backscatter mosaic with AVG Trend..................................................... 88

Figure B.3 - 2006 backscatter mosaic with AVG Flat.........................................................89

Figure B.4 - Difference between AVG Trend and AVG......................................................90

ix

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
 Figure B.5 - 2006 backscatter mosaic with a Lambertian correction................................ 91

Figure B.6 - Difference between AVG Trend and AVG Lambert.......................................92

Figure D.1 - Examples of angular response, stations A2 and N2000 (2006)................. 102

Figure D.2 - Examples of angular response, stations A9 and 97004 (2006)................. 103

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
 ABSTRACT

SEAFLOOR CHARACTERIZATION OF THE HISTORIC AREA REMEDIATION SITE

USING ANGULAR RANGE ANALYSIS
by
Luis A. Soares Rosa
University of New Hampshire, September, 2007

Angular Range Analysis (ARA) is a physics-based approach to acoustic remote

seafloor characterization. In order to better understand the capabilities and limitations of

this technique, ARA analyses were performed on multibeam sonar data collected at the

Historic Area Remediation Site, an area with high spatial variability. The remotely

derived results were compared to grain size information derived from grab samples and

Sediment Profile Imaging. Uncertainties in the determination of mean grain size from

ground truth were identified and when possible quantified. ARA proved to be an effective

remote sensing tool at a regional scale in its main operational mode that has a spatial

resolution limited to half-swath width of the sonar and to thirty pings. When the seafloor

is heterogeneous within half-swath width of the sonar, textural segmentation of the

backscatter mosaic allows the definition of “themes” out which ARA solutions can be

calculated, improving the correlation with ground truth.

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
 CHAPTER 1

INTRODUCTION

The use and exploration of the seas rely heavily on the propagation of

underwater sound. Light and radio waves, commonly used in remote sensing methods,

propagate well in air but are scattered and rapidly attenuated when penetrating the water

surface. Since direct measurements of seafloor properties are representative of only a

single point, are expensive and time consuming, and sometimes even impossible to

conduct, we depend on the acoustic remote characterization of the seafloor for a broad

range of disciplines including marine geology, offshore engineering and geotechnics,

benthic habitat mapping and mine warfare. In this thesis, a novel, physics-based

approach to remote seafloor characterization, Angular Range Analysis, is applied to a

well-studied area in order to better understand the capabilities and limitations of this

technique. Because any acoustically inferred sediment property will need to be validated

with direct measurements, an attempt is made to identify sources of uncertainty in

common methods of ground truth.

1.1 Remote Seafloor Characterization

Inferring seafloor properties through the use of acoustic means started with

qualitative studies of the echo character by marine geologists and geophysicists,

generally using single beam echo-sounder’s paper records to make inferences about the

nature of the seafloor. For that, the observed echo types were correlated with ground

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
 truth data (usually cores or grab samples) and this information was then extrapolated to

the entire survey area (Damuth, 1975; 1980). The same type of approach was extended

to analog sidescan sonar paper records, matching the observed acoustic texture with the

correspondent seafloor type. This type of approach is strongly dependant on

instrumental acquisition settings that often vary in and between surveys as well as on

subjective human interpretation.

Since the 1950s attempts have been made to quantify the relationship between

geoacoustic and physical properties of sediments (Urick, 1954; Hamilton et al., 1956).

By the 1970s, a substantial database of relationships between geoacoustical, physical

and geotechnical properties of seafloor sediment was developed (Hamilton, 1970; 1972;

1974; 1976; 1978). Thousands of measurements gave rise to a series of regression

equations, relating impedance, reflection coefficient and bottom loss with porosity and

density; attenuation with mean grain size and porosity; mean grain size with porosity and

density; and sound velocity with porosity, mean grain size and density.

Successful measurement of sediment geoacoustic properties from the acoustic

return started with the use of a calibrated chirp sub-bottom profiler. Sediment

classification models were developed based on the attenuation, impedance and volume

scattering (Mayer and LeBlanc, 1983; Schock et al., 1989; Panda et al., 1994; LeBlanc

et al., 1995).

A different type of approach examined the coherency of seafloor echoes as an

indicator of seafloor character, and the statistical analysis of echo fluctuations from ping-

to-ping was used as a remote sensing tool (Dunsiger et al., 1981; Stanton and Clay,

1986). Coherence in the signal was related with seafloor roughness and bottom type.

However, this method only allows a qualitative description in terms of bottom roughness.

By the 1980s, what had been long-used by fishermen in an empirical way was

implemented as a seabed classification system: discrimination of seafloor type based on

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
 waveform characterization of the first and second echoes (Orlowsky, 1984). Building on

that principle, commercial systems such as RoxAnn (Chivers et al., 1990; Murphy et al.,

1995) or QTC View (Collins et al., 1996) use the returned echo of a single beam echo-

sounder to remotely classify the seafloor. RoxAnn uses the final portion of the first return

as a measure of roughness and the second multiple of the acoustic signal as measure of

hardness. These two parameters are plotted against each other and clusters are

defined. QTC View extracts 166 parameters (full feature vectors) from the signal envelop

and then combines these into three primary parameters called Q-values. These values,

when plotted in a tri-dimensional space tend to group in clusters, each one

corresponding to a different type of acoustic response of the seafloor. Both approaches

only segment the seafloor into regions of similar response and need ground truth to

assign seafloor types to the segmented areas.

With the advent of digital oblique incidence systems (sidescan sonars, multibeam

echo-sounders) and the production of digital backscatter mosaics, classification systems

based on textural analysis of the image were implemented commercially (QTC Sideview

and Multiview, Triton Imaging SeaClass, GeoAcoustics GeoTexture, Arescon, GENIUS).

Most of these rely on the use of grey level co-occurrence matrices to segment the image

in regions with similar statistical properties (Pace and Dryer, 1979; Reed and Hussong,

1989). More elaborate classification techniques combine multibeam bathymetry with co­

registered backscatter data in a hierarchically supervised classification scheme, to

segment the seafloor into distinct facies (Dartnell and Gardner, 2004). Other approaches

use the signal envelope instead of the image, analyzing the power spectra (Pace and

Gao, 1988) or the probability density function of the echo amplitude (Stewart et al.,

1994).

All these methods require ground truth to identify the corresponding bottom type.

However, since there is no unique relationship between acoustic signature and seafloor

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
 type, an empirical relationship must be established for each survey site. In order to

produce a true classification system, the physics of the interaction of sound with the

seafloor must be properly understood.

1.2 Interaction of Sound with the Seafloor

Most of the systems used to remotely classify the seafloor fall in the monostatic

case, in which the transducer that receives the returned echo is located in the same

place where the sound wave was generated. The quantification of the returned echo was

first addressed by Urick (1954, p. 233) in terms of target strength, or reverberation

strength, per unit area of the bottom:

It remains now to convert these measurements into some sort of

scattering coefficient of the bottom. [...] Let a sound wave be incident on
a small area dA of the bottom at a certain grazing angle 0. At a distance
of 1 yard from dA back toward the source, let the intensity of
backscattering be Is. Then we will define a coefficient, which may be
called the “scattering strength” of the bottom at angle 0, to be the ratio of
Is to the incident intensity, per unit area of dA. It is convenient to express
scattering strength, referred to one square yard of bottom area, in decibel
units. Thus, scattering strength is the ratio of two intensities and is related
by the factor 2u to the backscattering cross section of a unit area.

In the same experiment, it was observed that backscattering strength is a

function of grazing angle and that this relationship changes with bottom type. In another

experiment, McKinney and Anderson (1964) recognized the dependence of

backscattering strength on the grazing angle, frequency and bottom type, as being of

“ primary interest”. However, only when interferometric sidescan sonars and multibeam

echo-sounders became available was it possible to collect acoustic backscatter versus

angle of arrival in a systematic manner. At that moment, the potential of this information

for remote classification of seafloor types was recognized (de Moustier and Matsumoto,

1993).

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
 One of the components necessary for the use of the backscatter angular

dependency as a remote characterization tool, is a high-frequency acoustic backscatter

model. Acoustic models predict the acoustic response of the seafloor from its physical

and geoacoustic properties. The list of properties can be extensive, due to the fact that

sediments are complex assemblages of a variety of particles, pore fluid, organic matter

and sometimes free gas. When an acoustic wave reaches the seafloor, part of the

energy is scattered back to the transducer due to irregularities at the water/sediment

interface and part is transmitted into the sediment and scattered by heterogeneities

inside and between the first layers. For a given frequency the amount of scattered

energy is mainly dependant on the seafloor roughness, the impedance contrast between

the water and the sediment, and the volume heterogeneities that may exist within the

sediment. All these processes must be taken into account in a comprehensive acoustic

backscatter model. In order to estimate the type of seafloor and remotely characterize its

properties from the angular response, an acoustic model has to be inverted. The Angular

Range Analysis (ARA) (Fonseca and Mayer, 2007), included in the Geocoder software

developed at the University of New Hampshire (Fonseca and Calder, 2005), implements

this concept and represents the newest contribution in physics-based approaches for

remote seafloor characterization.

1.3 Geocoder and Angular Range Analysis (ARA)

The first step in the remote characterization of the seafloor using ARA is to obtain

accurate measurements of backscatter strength. Geocoder radiometrically corrects

backscatter intensities registered by the sonar and geometrically corrects and positions

each acoustic sample in a projected coordinate system, thus calculating the best

estimate of the actual backscatter strength returned from the seafloor.

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
 After the angular response of the seafloor (the variation of backscatter with

grazing angle) has been determined, it has to be linked to seafloor properties by an

acoustic model. Two modified models are implemented in Geocoder’s ARA (Fonseca

and Mayer, 2007): 1) a composite roughness model developed by Jackson et al. (1986)

and 2) an effective density fluid model (Williams, 2001) derived from the Biot theory

(Biot, 1956, 1962).

In the Jackson et al. (1986) model the sediment is idealized as an acoustically

refractive and lossy fluid, and the total backscatter strength is modeled as the sum of

two different processes: interface scattering and volume scattering. The acoustic

response of the sediment is modeled as function of frequency and grazing angle. This

model requires input parameters related to the impedance contrast (sound speed and

density in the water and in the sediment), attenuation (loss parameter), roughness

(spectral strength and the spectral exponent of bottom relief) and volume scattering

(volume parameter).

The Biot theory describes the propagation of acoustic waves in a porous elastic

matrix containing a viscous fluid. Models that consider the full Biot theory require

parameters related to sediment grains, pore fluid and the sediment frame. Williams’

(2001) acoustic propagation model approximates a porous medium as a fluid with a bulk

modulus and effective density derived from Biot theory. The implementation of this

model in Geocoder’s ARA requires, in addition to the parameters used in the

implementation of the Jackson’s model, the porosity, permeability and turtuosity of the

sediment.

The inversion of one of the acoustic models implemented in Geocoder’s ARA,

which is done by adjustment of the model to the observed variation of backscatter

strength with grazing angle, with model parameters constrained by relations between

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
 physical and geoacoustical properties of the sediment, results in a physics-based

prediction of seafloor properties.

1.4 Verification of Model Predictions and the Importance of Mean Grain Size

One possible benefit from the use of a remote sensing technique is to minimize

or eliminate the need for ground truth. However, before a remote sensing technique is

generally accepted, we have to make sure that the results accurately describe the real

world within an uncertainty or resolution that satisfies the objectives of the required

application.

The most accurate way of verifying model predictions is the direct measurement

of sediment properties, but this poses several problems. The measurement of

geoacoustic properties like sound speed and attenuation in the laboratory requires

undisturbed samples, but the coring and the sampling process often result in compaction

of the sediment and loss of water. Therefore, measured values may not reflect natural

conditions. Alternatively, measurements can be made in situ using probes pushed into

the sediment by divers, from submersibles or by remotely operated vehicles (Mayer et

al., 2002), but these properties are frequency dependant and the direct measurement

should ideally be done at the same frequency as the one considered in the model.

Another problem arises from the temporal variability of seafloor properties and the

impossibility of continuous measurements.

Some geoacoustic parameters are either very difficult or nearly impossible to

measure (Jackson and Richardson, 2007). For these, the preferred means of obtaining

them is through empirical regressions between geoacoustic and physical properties

(Hamilton, 1972, 1974; Richardson and Briggs, 2004; Jackson and Richardson, 2007).

But the measurement of physical properties is based mostly on core samples and the

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
 determination of properties such as bulk density and porosity, is subject to the same

sampling disturbance described above. These errors can be minimized through

techniques such as the collection of sediments by divers in slabs or in horizontal corers,

or with large box corers (Jackson and Richardson, 2007). Measurement of seafloor

roughness can be accomplished using digital stereo photogrammetry and volume

heterogeneity can be quantified with X-ray computed tomography (Pouliquen and Lyons,

2004). However, outside designed experiments, measures of seafloor properties seem

difficult to obtain.

Because of the difficulties described above, one of the most commonly measured

properties is grain size, often expressed by mean grain size or sediment type (e.g.

gravelly muddy sand). Jackson and Richardson (2007, p. 193 and p. 195) comment on

the relation between mean grain size and seafloor roughness:

When seafloor roughness, as represented by RMS roughness, is plotted as a

function of mean grain size, the result is a scattering of points across the
range of sediment types. [...] Sediment mean grain size alone may never
yield the types of predictive relationships for roughness required by high
frequency acoustic models unless the effects and rates of hydrodynamic and
biological process that create, modify and destroy roughness features are
incorporated into predictive relationships.

And between mean grain size and geoacoustic and physical properties of the sediment

(Jackson and Richardson, 2007, p. 150-151):

The coefficient of determination, r 2, between index of impedance and mean

grain size is much lower than the coefficient of determination between index
of impedance and sediment bulk density or porosity. This lower coefficient of
determination is reflected in the lack of a physical relationship between mean
grain size and either sediment bulk density or porosity. [...] Using values of
mean grain size as an index, especially in the silt-size range, may be very
misleading because of major differences in sorting (standard deviation of the
particle size distribution) or due to effects of compaction and packing. [...]
Given the aforementioned issues, it is perhaps amazing that empirical
regressions between grain size-related parameters and sediment density,
porosity, sound speed, or impedance have any predictive value.

Nevertheless, sediment type or mean grain size are typically used indirectly as

empirical predictors of acoustic behavior, probably because they are the most common

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
 descriptors found in sediment databases and are often the only sediment physical

descriptors available.

An acoustic model may be constrained by relations between physical and

geoacoustical properties of the sediment determined in carefully designed experiments,

but in a real-world application, when an acoustic measurement is made and the model

inverted to remotely characterize the seafloor, there will seldom be any seafloor property

data available to verify model results other than the mean grain size.

In this study, mean grain size as predicted by the ARA is compared with mean

grain size obtained through common methods of ground truth and sediment analysis. Of

particular concern is the use of mean grain size as the object of comparison. Mean grain

size as determined by a remote sensing tool such as ARA is always subject to suspicion

given the above mentioned apparent lack of physical relationship with other sediment

properties used to model the acoustic response of the sediment. However, mean grain

size as determined by common methods of ground truth and sediment analysis cannot

be considered as an absolute value either. It is affected by uncertainties that are present

from sampling to grain size analysis. But most importantly, we have to make sure that

we are comparing acoustic and physical samples in the same place, and this is

particularly true in areas of high horizontal and vertical heterogeneity.

Horizontal variability can be a limitation for ARA, which in its main operational

mode has a spatial resolution limited to half-swath width of the sonar in the across-track

direction, and to a certain number of pings (usually between 20 and 30) in the along-

track direction. Horizontal variability can also be a limitation for ground truth. The spatial

scale of sediment uniformity generally is not known and the size of the sediment sample

and the number of samples may not reflect the local variability. Additionally, the sampler

may not be precisely positioned.

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
 Vertical variability can led to errors in the comparison between ARA predictions

and mean grain size determined by grab sampling. The penetration of a high-frequency

acoustic wave is only in the order of a very few centimeters (Appendix A), and if the

sediment is layered at a centimeter scale, the grab sampler may transect other types of

sediment in the subsurface that are not being sampled acoustically.

This study was carried out on the Historic Area Remediation Site (HARS), off

New Jersey, because of the large amount of data available and because of the high

spatial variability (the result of anthropogenic actions).

1.5 Historic Area Remediation Site (HARS)

The area of study is located in the New York Bight, northwest of the head of the

Hudson Shelf Valley, six nautical miles east of Sandy Hook, New Jersey, with an

average depth of 25 m (Fig. 1.1). This area has been a place for disposal of assorted

material (garbage, city refuse, cellar dirt and sediments derived from dredging during the

maintenance, deepening and construction of new channels in New York Harbor) since

the mid-1800s and because of that, very little of the original shelf geology is preserved.

The United States Army Corps of Engineers (USACE) web site presents a brief history

of the HARS (http://www.nan.usace.army.mil/business/prjlinks/dmmp/benefic/hars.htm).

10

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
 . '' '

V i's -V 'r ■
■'T'.r'y?
MBMBj________
.... ■ <$.■**!
‘n j i ’ - K

■■’ : :« '.iii.|': ji
, .;v ^ >,1
na»sif!sa:s

Figure 1.1 - Perspective view from southeast showing the HARS location. With a size of
approximately nine by eight km the HARS is located in the Christiansen Basin, northwest of the
Hudson Shelf Valley, six nautical miles east of Sandy Hook NJ. Vertical exaggeration: 20 x.

The HARS was established in 1997 through an agreement among the U.S.

Environmental Protection Agency (EPA), the U.S. Department of the Army and the U.S.

Department of Transportation, in an effort to reduce the elevated contamination and

toxicity associated with some of the dredge materials in the area. The HARS comprises

the former Mud Dump Site (MDS) and some surrounding dredged material disposal

areas. It is divided into nine Priority Remediation Areas (PRAs) where remediation

material is to be placed (Fig. 1.2). There is also a Buffer Zone surrounding the PRAs and

a No Discharge Zone, which is an area outside the PRAs where no further disposal is

permitted.

11

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
 594000 595000 596000 597000 598000 599000 600000
1
------ 1
____ I____ 1
------ I
____ I
____ 1
------- ------

4476000
................................................... jg ..........................

4475000
m

4474000
4473000
4472000
4471000
m

&

4470000
4469000
4468000

594000 595000 596000 597000

J 1997 Category H Cap B oundary

200S Bathymetry
L- 1 1993 Dioxin Cap Boundary
i Depth (m)
j1Red Clay
{ Form
c High: -9
er MDS

□ HARS PRAs U TM Projection Zone 18N

] ] HARS Buffer Zone Boundary Low : -38 N AD 83 Datum

Figure 1.2 - Historic Area Remediation Site 2006 bathymetry. The HARS comprises the former
Mud Dump Site (MDS) and some surrounding dredged material disposal areas; it is divided into
nine Priority Remediation Areas (PRAs) where remediation material is to be placed.

12

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
 When surveyed in September 2006, the central region of the HARS was

dominated by an elevation produced by dumping, running approximately north-south

over depths of 23 to 25 m to the west and 27 to 34 m to the east (Fig. 1.2). The

topographic high in the north between PRAs 1 and 9, 15 m below sea level, is

designated Castle Hill and corresponds to material dumped prior to the 1930’s. In the

center of the region, another elevation at a depth of 15 m is the result of material

dumped between the 1930's and 1975. To the south, where the shallowest point in the

area occurs, at 10 m depth, mounds almost 10 m high were produced by recent disposal

of dredged material (Butman et al., 1998, 2002).

North of PRA 4, there are several individual features 50 to 100 m long, about 40

m wide and 1 to 2 m high, aligned northwest-southeast. Butman et al. (1998)

hypothesize that these features are individual dumps from barges. To the northeast,

there is a clay deposition area. Throughout the region to the northwestern corner of the

HARS, there are several features with the same alignment. These may correspond to

historic dumps because the orientation is the same as that taken by vessels when

entering the New York Harbor (Butman et al., 1998; Schwab et al., 2000).

Two, approximately circular, smooth regions, lie in the south of the site. They

correspond to the capping of dioxin mounds with more than 1 m of clean sand (SAIC,

2005b), the 1993 Dioxin Project (to the west) and the 1997 Category II Project (to the

east). Between 1997 and 1998 1.83x10® m3 of sand were used to cap 0.5x10® m3 of

Category II sediments previously deposited (Butman et al., 2002). Sediments classified

as Category II meet ocean dumping criteria but present a potential for bioaccumulation

(EPA, 1996).

PRAs 1, 2, 3 and 4 present well-preserved individual mounds that formed from

the placement of remediation material that has been ongoing since 1997. From 1997 to

2000 4.3x10® m3 of remediation material were placed in this area. The general

13

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
 morphology reflects the placement procedure, which follows sequentially the cells of a

grid in order to obtain a uniform coverage of 15 to 75 cm across the area of the grid.

The deepest area, in the southeastern part of the HARS, is aligned with the head

of the Hudson Shelf Valley and corresponds to a channel 300 to 400 m wide and 2 to 4

m deep. The channel is limited on the northeast by outcrops of southeastward-dipping

coastal plain strata, probably of Cretaceous age (Butman et al., 1998) and on the

southwest, by a smooth slope. The characteristic morphology of the channel gradually

disappears to northwest. All this area is relatively smooth and probably corresponds to

fine sediments winnowed from the northwest and transported eastward and downslope

toward the head of the Hudson Shelf Valley.

The rock outcrops (Fig. 1.2) probably extend in the subsurface to PRA 6. This is

an area with several topographic highs aligned northwest-southeast, which are

approximately 1 m high. To the north of this region, throughout PRA 9, there are several

circular depressions 30 to 50 m in diameter, up to 0.5 m deep, with a small high in the

center, with the same alignment. These may correspond to dumps of rocky material

(Butman et al., 1998) and/or to the disruption of the subsurface geology. The southwest

corner of the HARS is characterized by long-wavelength bedforms.

The HARS contains a wide variety of sediment types, ranging from clay to gravel,

in a relatively small area. This small-scale variability is particularly suited for a test of the

spatial resolution of the ARA approach. Datasets acquired by Science Applications

International Corporation (SAIC) in 2005 and 2006 were used for this study. Each

dataset consists of a multibeam sonar survey conducted with a Reason 8101 MBES,

Sediment Profiling Imaging (SPI), plan view images, and grab samples. All these data

were acquired over a relatively short period of time, which minimizes the effect of

temporal variability of the seafloor and, consequently, yields a better correlation between

remote sensing and ground truth data.

14

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
 1.6 Objectives

The primary objective of this study is to determine the applicability of ARA to an

area with high spatial variability such as the HARS in order to understand the capabilities

and limitations of ARA. Since any acoustically inferred sediment property will need to be

compared to direct measurements, an attempt was made to identify sources of

uncertainty in common methods of ground truth.

Chapter 2 outlines the acoustic methods used to predict seafloor properties in

terms of mean grain size. The ground truth methods used to verify ARA predictions are

detailed in Chapter 3. In Chapter 4, results from grain size analysis and from acoustic

remote sensing are presented, and a comparison is made between the two. A

discussion of the degree of confidence in these results is also included in Chapter 4.

Chapter 5 summarizes the conclusions and suggests new experiments and paths for

improvement.

15

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
 CHAPTER 2

BACKSCATTER MOSAICS AND ANGULAR RANGE ANALYSIS (ARA)

2.1 Backscatter Mosaics Assembled in Geocoder

As discussed in Chapter 1, backscatter mosaics have been used to remotely

classify the seafloor, usually through textural segmentation into several classes followed

by ground truth to define attributes for each class. Until the mid 1990s only backscatter

mosaics constructed from sidescan sonar data were available. However, the position

and attitude of the sidescan sonar towfish usually is not known with precision. Also,

backscatter intensities are recorded as a long time series for each side of the sidescan

sonar and there is no information about the angle from where the sound is being

received. This results in an ambiguity when two arrivals from different places reach the

sonar at the same time. Multibeam sonars provide a time series of backscatter values for

each beam in each ping, from which water depth measurements can be extracted. The

knowledge of the precise position and attitude of the multibeam sonar allows the

accurate determination of the position and the geometry of each beam over the seafloor.

However, what is usually recorded by the sonar is not the true backscatter strength but

the backscatter intensity as affected by system settings at the time of acquisition

(Beaudoin et al., 2002).

Geocoder (Fonseca and Calder, 2005) assembles a time series for port and

starboard sides from the backscatter intensity time series of each beam, accounts for

transmitting power, receiver gains and time varying gains applied by the system,

16

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
 corrects for transmitting and receiving beam patterns, as well as attenuation and

spherical spreading in the water column. It also accounts for the range to the transducer,

seafloor slope, transmit and receive beamwidths and pulse length in order to calculate

the area of ensonification. After these radiometric corrections are made, the backscatter

strength is calculated per unit solid angle per unit area. A filter is also applied to remove

speckle noise. Additionally, the geometric corrections account for refraction in the water

column and movement and attitude of the transducer, after which each backscatter

sample is mapped to a mosaic cell in a projection coordinate system. A mosaic obtained

with this process shows the spatial distribution of the best estimates of backscatter

strength, preserving its angular dependence, but as can be observed in Figure B.1

(Appendix B) the mosaic is difficult to interpret.

The final step in the assemblage of a backscatter mosaic is the removal of the

angular dependence of the backscatter. In order to apply the proper correction and

normalize backscatter intensities across the swath, detailed information about

geoacoustical and physical properties of the sediment is needed because the angular

dependency is a property of the seafloor type. If this information could be obtained for

each ping, along the entire swath and vertically into the sediment up to the depth of

acoustic penetration, it would then be possible to build a precise Angular Varying Gain

(AVG) table, different for each ping, and then obtain a truly normalized backscatter

mosaic. Inasmuch as it is virtually impossible to obtain detailed information about the

geophysical properties of the sediment concomitantly with the acoustic survey (and if

that type of information was available, the acoustic survey would not be necessary), it is

common to apply a generalized AVG filter to remove the effect of the backscatter

angular dependence and normalize the acoustic response across the swath. Examples

of the effect of using different AVG filters in the assemblage of backscatter mosaics are

17

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
 shown in Appendix B. For different AVG filters, the difference in backscatter values for

the same portion of the seafloor can be greater than 10 dB in certain areas.

Figure 2.1 - Right: backscatter strength for a portion of the line 06241 d27 with all corrections
applied and without AVG to normalize the acoustic response across the swath, the red and green
rectangles overlaying the backscatter mosaic correspond to the patch of the seafloor where ARA
is performed. Left: average angular response for each of the rectangles depicted on the right
panel, the green curve corresponds to the seafloor acoustic response on starboard, the red line to
port side. The blue curve is the modeled acoustic response fitted to the observations on the
starboard side, predicting a muddy sand seafloor. The angular resolution of the method does not
allow correct predictions when the seafloor varies significantly within half-swath as on the port
side, the angular response is a mixing between the characteristic curve of a fine sand/coarse silt
in the near range and coarse sand/gravel for the outer beams.

2.2 Angular Range Analysis (ARA)

The backscatter angular response is an intrinsic characteristic of the seafloor and

is used by the Angular Range Analysis implemented in Geocoder (Fonseca and Mayer,

2007) for its characterization. The analysis is performed for each patch of the seafloor

defined by the half-swath width in the across-track direction and by a stack of

consecutive pings in the along-track direction, normally between 20 and 30 (stacking

18

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
 several pings is important to reduce speckle noise), and this determines the spatial

resolution of the method (Fig. 2.1).

From the observed backscatter angular response, a set of ARA-parameters is

extracted, containing the slope and intercept for different parts of the angular response

curve, and the orthogonal distance (the distance of a point to the general trend in an

intercept-slope plane). The variation of these parameters is strongly influenced by

seafloor roughness, impedance and volume heterogeneities, respectively.

As described in section 1.3, two high-frequency acoustic propagation models are

embedded in the software, modeling the angular response for a given frequency as a

function of seafloor properties. Several combinations of seafloor properties may lead to

the same angular response so the acoustic models implemented in the software are

constrained by relations between seafloor properties determined by Hamilton (1972;

1974; 1976; 1978).

The prediction of seafloor properties from the observed angular response is

accomplished by model inversion, iteratively adjusting the model ARA-parameters to the

ARA-parameters calculated from the observations, but not in a free way. The model

ARA-parameters are constrained by the parameters of the forward model that are tied by

Hamilton’s empirical relations.

Given accurate measurements of backscatter strength, the main obstacle for

ARA is the nature of seafloor across the swath, which is assumed to be uniform. One

possibility to overcome this limitation would be to segment areas in the backscatter

mosaic with similar tones and textural patterns (Haralick, 1979) and then calculate

average angular responses for the segmented areas. Fonseca and Calder (2007) point

out that this sort of reasoning is difficult to justify since an assumption about the angular

response of the sediment was already made when building the first mosaic and applying

an AVG filter. In order to combine the spatial resolution of the backscatter mosaic with

19

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
 the angular resolution of the ARA, Fonseca and Calder (2007) clustered ARA

parameters as a feature vector in a multidimensional space using an unsupervised

clustering algorithm. Each cluster is expected to represent areas on the seafloor with

similar properties and similar angular responses. This technique implies an a priori

definition of the number of sediment types present in the area.

For highly complex areas like the HARS, it is difficult to define a priori how many

sediment types are present. One possibility would be to take sediment samples on a

dense and evenly spaced grid but that would destroy the purpose of a remote sensing

approach. Another option, although subjected to a priori assumptions about the angular

response, would be to use a backscatter mosaic to define the number of classes, but in

the case of the HARS, the size of the surveyed area compared with the degree of spatial

heterogeneity presents a scaling problem. The area is too big to differentiate classes

and all transitional classes seem to exist. So, the best that can be done is to classify

areas of high and low backscatter. Figure 2.2 depicts the backscatter histogram for the

entire 2006 backscatter mosaic presented in Figure 2.4.

,400,000

1,200,000

1,000,000

600,000

400,000

200,000

Figure 2.2 - Histogram of backscatter strength across the HARS using the average between the
two more reliable values that fall within each 1 m cell, as calculated by Geocoder.

If, in a limited area, a marked contrast between different seafloor types can be

observed, it may be assumed that the applied AVG did not mask that limit, but only

harmonized the textural appearance of the deposits. Within each deposit, sediment

heterogeneity may exist but that will not be resolved even with ground truth

20

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
 measurements. Therefore, average properties for the deposit must be assumed as much

on the ground truth as on the remote sensing side. For those areas where different

sediment types can be clearly identified, segmenting areas on the backscatter mosaic

with similar tones and textural patterns may be a valid method to determine the average

angular response for each deposit.

Figure 2.3 - ARA in theme mode. In this example two themes were defined based on the
observation of the backscatter mosaic, each one is expected to correspond to areas of the
seafloor with similar geophysical properties. For each theme the average angular response is
calculated (green lines). The ARA-model is then fitted to the angular response of each theme
(blue lines) and seafloor properties are predicted. In this example theme 1 corresponds to gravel
and theme 2 to medium sand.

The Angular Range Analysis for a specific area of the seafloor defined by the

user is implemented in Geocoder’s ARA with the designation “theme mode” (Fig. 2.3).

The first step in the ARA is the manual segmentation of the backscatter mosaic in areas

with similar tonal and textural patterns, called “themes” . Each theme is expected to

correspond to portions of the seafloor with similar geoacoustic and physical properties.

The second step is the calculation of the average angular response for each theme. This

calculation considers all the snippets that fall within each theme, independently of the

21

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
 line they belong to, and is not restricted to the half-swath width of the sonar and to a

stack of pings as in the normal mode (Fig. 2.1). The final step, extraction of ARA-

parameters, model inversion and prediction of seafloor properties, follows the same

process as for the “normal mode” .

2.3 Multibeam Data Collection and Processing

Multibeam sonar data was acquired by SAIC onboard MA/ Atlantic Surveyor with

a Reson 8101 multibeam system (240 kHz, 101 beams 1.5° x 1.5°) on August 26th to 31st

2005 (SAIC, 2005b) and on August 29th to September 14th 2006. Data was processed by

SAIC and archived in .gsf file format. Bathymetry post processing was done following a

CARIS HIPS processing pipeline.

The backscatter time series for each beam in each ping (snippet) was imported

into Geocoder directly from the processed .gsf files. Since digital numbers on the 8101

Reson system do not represent the true backscatter strength, Geocoder software

corrected for radiometric and geometric distortions and positioned each acoustic sample

in a projection coordinate system (Fonseca and Calder, 2005). The result is the correct

angular response of the seafloor. An AVG filter (option “Trend” in Geocoder) was

applied to normalize the backscattering strength across the swath and the final

backscatter mosaic was produced with 1 m cell resolution (Fig. 2.4 and Fig. 2.5).

Backscatter strength, as normalized by Geocoder when assembling mosaics, is the

average between 35° and 55° grazing angles (Fonseca, L. personal communication,

2007). ARA was applied in its normal mode, averaging a stack of 30 pings along-track to

calculate the angular response for an across-track range of half the swath width of the

sonar. In this study the effective density fluid model (Williams, 2001) implemented in

22

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
 Geocoder’s ARA was used; model results in terms of mean grain size are presented in

Chapter 4.

594000 595000 596000 597000 598000 599000 600000

20052

13W413

20054
20018 H2000

20026
20019
^210120027

I *' ■‘ 120028
97008U L i
20029

. . K5BL1600 L2400
97002197 1
M2800

200361 v 20016 20022

20023
20035 200241N3200
20013 NOREMED5
20041120042
200410
20043 20
J e> as© ms 20049
emcctttgvi ®G223 GsE•SiD
S » ‘s
970010

Q2400

9 7 0 0 5 *9 7 0 ^

NOREMED4

1--- 1--- 1--- 1--- 1----------------1---

594000 595000 596000 597000 598000 599000 600000

CD 1997 C ategory II Cap Boundary 20 0 5 - B a C k S C a tte r S tr e n g th

f } 1993 Dioxin Cap Boundary
dB
j Red Clay
High : -3.9
£ J Former MDS

□ HARS PRAe 2005- Grab Samples UTM Projection Zone 18N

; HARS Buffer Zone Boundary L o w : -53.27 2005 • SPI N AD83 Datum

Figure 2.4 - Position of the grab samples and profile images acquired in 2005 over the
backscatter mosaic assembled with Geocoder.

23

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
 594000 595000 596000 597000 598000 599000 600000

P
A13*H
rA14'C
m brte %13*D 'A15iBjA15«C
aio I bW io te

a io I a

20062IBJ2W62TC
fV T |U R rm&m
tCf*1^f f 2O051f8l^05irC
2006lIBr^Snfit?rWKl3TBTl'3Tc A7H?A7lF
1 20 0lB to1200lE 20061 FA
.G120OJA

H20001D SwSily
^IBT»1200Tc
I1200TA ^TbTisJc
20G64TC
2®liPIe mcWlo
20064TA

K0800IE
JMstTIoSTb
.1200IBrwarna
20064TB
M lE fliiT P
2002870
20028TB

Q225M3(s3aiU3
20IBJ20TCT"
O70(i7ier97oo7
97007
wnBSr
L24U U _»A
L1200T
M28Q0TC
■24IB124TC
25 t t
A4TBTA4H N3200TF
N2000TA wwonT *
20041IE

P320Q TD JP 3200
m m P3200

97004TCI9M

.. a « n ' .
■y<y
g p w r

j 1997 Category II Cap Boundary 2006 - BaC kS C atter S tre n g th

dB
High : -2.17
[_ j Former MDS
2006 - Grab Samples
CD HARS PRAs UTM Projection Zone 18N
i HARS Buffer Zone Boundary 2006-SPI N AD83 Datum
m

Figure 2.5 - Position of the grab samples and profile images acquired in 2006 over the
backscatter mosaic assembled with Geocoder.

24

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
 CHAPTER 3

GROUND TRUTH

To evaluate the effectiveness of ARA, model results must be compared to

ground truth data. Although the empirical relations among mean grain size and sediment

physical and geoacoustic properties (Hamilton, 1972; 1974; Richardson and Briggs,

2004; Jackson and Richardson, 2007), and backscattering strength, are subject to large

errors (Jackson and Richardson, 2007), mean grain size and sediment type are the most

common descriptors of seafloor properties.

Mean grain size is widely used as an environmental and geotechnical proxy, but

values are often reported without reference to the method used for sample collection or

analysis, as well as the associated uncertainty. This study tried to identify, and estimate

when possible, the limitations of the methods most commonly used for sediment

collection and analysis.

In addition to sediment sample data, Sediment Profile Imaging (SPI) and plan

view images were collected in the HARS by SAIC to evaluate benthic recolonization

status and the degree of benthic habitat disturbance at each station (SAIC, 2005a).

Profile images of the undisturbed water/sediment interface and top centimeters of

sediment allow comparison of the observed sediment type with grab sample results and

an estimation of seafloor roughness and volume heterogeneity. The penetration of the

SPI optical prism, which acts like a static load penetrometer, also provides a

25

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
 geotechnical parameter that can be used for comparison with mean grain size and

acoustic measurements.

3.1 Position Uncertainty

SPI and grab sampling positions are affected by an uncertainty. In the 2006

survey the position of each individual sample was referenced to the position of the GPS

antenna. Considering the uncertainty of the differential GPS (±3 m), the distance from

the antenna to the place where the SPI system was deployed (6.7 m) and to where the

grab was deployed (3.6 m) and the maximum cable angle (10° for the SPI system and

30° for the Petit Ponar grab sampler), total position uncertainty varies between ±15 m to

±25 m for the grab and from ±12 m to ±15 m for the SPI system, depending on the

depth. Figure 3.1 shows the position uncertainty for station 16.

Figure 3.1 - Station 16. In 2005, data is referenced to the nominal position of the station, depicted
as a blue dot. The blue circle represents the estimated uncertainty and has a radius of 40 m. In
2006 all the replicas for the nominal station are depicted as red dots, the yellow circle (12 m
radius) corresponds to the estimated uncertainty for successful SPI samples and the green circle
(15 m radius) corresponds to the estimated uncertainty for successful grab samples. The 2006
backscatter mosaic with 1 m cell size is used as background to show the small scale variability of
the area and why there is an uncertainty about the sedimentary deposit that is being sampled.

26

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
 Positions for the 2005 data followed a different approach, as all the replicas

within each station were referenced to the position of the planned station (SAIC, 2005a).

Therefore, the grain size and the penetration of the optical prism (SPI) can be

considered as average properties within the radius of uncertainty for each station.

Considering that the survey vessel was the same, maneuvered by the same people, and

in similar weather conditions, one can assume the observed dispersion of replicas

around each planned station in 2006 to be on the same order of magnitude of what

occurred in 2005. The uncertainty associated with this practice of referring

measurements to the planned station was estimated to be approximately 40 m (Fig. 3.1).

3.2 Grab Samples

3.2.1 Surface sediment sampling

Surface sediments were collected on September 1st and 2nd, 2005, with a 0.1 m2

Van Veen grab sampler, onboard the R/V Beavertail, to determine the toxicity of surface

sediments. The grab was deployed one or more times (usually no more than two)

around the nominal position of the stations shown in Figure 2.3. If the sediment was

acceptable (bucket more than half full and without evidence of washout or disturbance),

the entire content was placed in a large mixing bowl and the sampling continued until

enough sediment was collected (SAIC, 2005a).

The sampling strategy for 2006 was a compromise between revisiting primary

stations from 2005, in order to observe the temporal evolution of benthic recolonization

at a given station, and the need to obtain data in areas that were not previously sampled

(Fig. 2.4). On August 21st and 22nd, 2006, onboard RA/ Beavertail, sediment samples

were collected with a Petit Ponar grab sampler (Fig. 3.2) at the same time as SPI

27

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
 images were obtained. This grab sampler allowed for the collection of surface sediments

to a depth that varied between the thickness of the first layer of grains in gravely bottoms

up to approximately 10 cm in muddy sediment. The grab did not trip in some stations,

even after three attempts (stations 15, 18, 27, 28, 29, 20063, A1, K0800, M2800, N2000

and Q2400). When available, sediment that was stuck to the frame was collected and

labeled with the station name (samples 27_C, A1_C, K0800_C and N2000_C). At other

stations, the sampler closed without any sediment (stations A8 and A14) or collected

only a single pebble or cobble (samples A6_B, A10_B and P3200_B).

Figure 3.2 - Petit Ponar grab sampler used to collect sediment samples in 2006.

3.2.2 Single sample versus composite sample

The two different approaches followed in 2005 and 2006 bring up the issue of the

number of replicas needed to characterize the deposit at the sample locality. One

approach considers that a given deposit has an internal spatial heterogeneity that cannot

be resolved. In this case, the best sampling is a composite sample that reflects average

characteristics. Krumbein and Pettijohn (1938) suggest that at least four discrete

samples (replicas) should be combined into a single composite in order to reduce the

ratio, of the error of the mean of a set of observations n ( Em), over the error of a single

28

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
 observation ( E ) to 0.5, the point when the error starts to decrease more slowly as more

samples are taken. The error of the mean of a set of observations n varies inversely

with the number of observations, Em = e / 4 n .

In the other approach, position uncertainty is considered to be good enough to

attribute the sample to a homogenous patch of the seafloor. Since the position of a

sample is hardly repeatable, a second replica would sample a different patch of the

seafloor and describe sediment characteristics in a different place.

Collier and Brown (2005) report a minimum error in the calculation of the mean

grain size from four repeated grabs at the same station with position uncertainty ±15 m

to vary from ±1.2 O to ±0.3 O in gravelly silt sediments of the Loch Linnhe on the west

coast of Scotland. Although this example is valid only for a specific sediment type and

spatial variability of the seafloor, this adds to the concern about sampling

representativeness.

3.2.3 Grab sampler

The type and size of the grab sampler is another contributor to the uncertainty in

a grain-size measurement. For instance, the size of the grab may not be large enough to

collect a representative sample of the sediment population. Larger grains may not be

sampled at all, and if that is the case, frequency distributions of grain size will be open

on the coarser size, or coarser grains may be sampled in a way that does not reflect the

true distribution of the population. As an example, a 20 mm single pebble represents 9%

in weight of a 1 kg sample and sampling one more or one less pebble may shift

significantly the mean grain size.

29

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
 Krumbein and Pettijohn (1938) present a table of practical sample weights,

based on Wentworth’s (1926) work and suggest that 125 to 500 g should be collected in

order to obtain a representative sample for mechanical analysis if the coarsest size is

within the sand size (0.06 to 2 mm), 1 kg for granule size (2 to 4 mm) and from 2 to 16

kg for pebble size (4 to 64 mm). Unfortunately, the efficiency of a grab is in the opposite

direction, as the volume of sampled sediment decreases with larger grain sizes.

The type of grab used may allow washout of fine-grained material, as the Petit

Ponar sometimes does. The grab may also produce a shock wave that can disturb the

bottom and some less-dense material may be selectively undersampled, as has been

noted for the Van Veen (Wigley, 1967). Biased results can be expected in both cases.

Finally, if the objective is to evaluate the validity of an acoustic measurement,

one must consider that what is actually being calculated is the mean grain size for all the

sediment layers that were sampled by the grab sampler and not the mean grain size for

the top layer. In some situations, this is a very important issue, as the acoustic energy

will not necessarily penetrate to the same extent, but will be restricted to the top layers.

3.2.4 Sediment analysis (grain-size and organic content)

In 2005, sediment grain-size analysis was done by SAIC using sieves for the

sand fraction (US standard sieve seizes: 10, 20, 40, 60, 100 and 200) and a hydrometer

for silt and clay fractions, but only the percentage of gravel, sand, silt and clay was

reported. The moisture content of the sediment was also determined (SAIC, 2005a).

The 2006 sediment-sample analyses followed procedures based on the methods

used at the UNH Coastal Geology Laboratory and at the Marine Geology Department of

the Portuguese Hydrographic Institute.

30

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
 Laser diffraction was selected for particle-size analysis since it is relatively fast,

more precise (Sperazza et al., 2004) and permits determination of finer grain sizes, and

most importantly, because it allows the analysis of the entire sample with the same

method. The available laser diffraction equipment was the Malvern Mastersizer 2000

with a Hydro G dispersion unit from the Portuguese Hydrographic Institute. This

equipment allows particle-size measurements from 0.02 to 2000 pm with an accuracy

±1% on the Dv50, using the Malvern Quality Audit Standard, and an instrument-to-

instrument reproducibility better than 1% Relative Standard Deviation. Even though a

maximum grain size of 2000 pm is specified, the stirrer pump of the dispersion unit only

holds 1500 pm. After several tests it was found that the 500 pm sieve limits particles to a

size that does not clog the pump, because elongated particles with dimensions larger

than the square mesh will pass through the sieve. Because of this, every sample was

wet sieved at 500 pm. For the coarser samples, this meant that the samples needed to

be split at the upper limit of medium sand (500 pm) leaving the rest of the sample to be

measured with sieves.

HARS samples have a wide range of grain sizes so a pre-evaluation was made

to decide whether to split the sample at 500 pm or 62.5 pm. Photos were taken on all

collected samples using a SLR 6MP digital camera with a Micro Nikkor 105 mm f/2.8

lens at a fixed distance from the sediment surface (0.75:1 magnification). A 500 pm

mesh was put in place of the sample surface as a reference. From the photos, 75% of

the samples fall below the 500 pm size, and 80% of the samples have more than 90% of

their weight within that limit (based on a visual estimation of the area). Based on that, the

decision was to sieve at 500 pm in order to have most of the samples, or at least the

most significant part of them, analyzed with only one method. The same procedure was

applied to all the samples for consistency. Figure 3.3 illustrates the criteria used.

31

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
 Figure 3.3 - Portion of the photograph taken to sample 20028_A showing more than 90% of the
area below 500 pm.

In addition to laser diffraction, grain-size analysis was also performed with

classical sieve and pipette methods for comparison. The weight Lost On Ignition (LOI)

was also determined as a rough estimation of the percentage of organic material present

in the sediments.

Each sediment sample was thoroughly mixed and two to five aliquots were

removed depending on the size of the sample and visual estimation of the grain size.

The remaining material was stored. The following table summarizes the analysis

performed on each aliquot; the detailed procedure is described in Appendix C.

Aliquot Procedure
1s' Standard grain size analyses procedure, wet sieved at 500 pm followed by
sieving of the coarser fraction and laser diffraction for the finer fraction.
2na Measurement of the natural occurring particle size and the effect of
organic content on particle agglomeration, wet sieved at 500 pm, finer
fraction analyzed with laser diffraction following a different procedure,
without particle dispersion and ultrasonication.
3ra Measurement of the effect of drying the sample, wet sieved at 500 pm,
finer fraction analyzed with laser diffraction following the normal procedure
but without drying the sample.
4in Comparison with the pipettes method, wet sieved at 62.5 pm, followed with
sieving of the coarser fraction and pipette analysis of the finer fraction.
5tn Organic content estimation by weight loss on ignition.

Table 3.1 - Analysis performed on each aliquot of 2006 sediment samples.

32

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
 To compare grain-size data with acoustic data, all particles collected in the

sample that act as scatterers were included, namely material of biogenic and

anthropogenic origin (glass, brick pieces, etc).

3.2.5 Uncertainties in sediment analysis

In standard grain-size analysis, “size” is defined as the diameter of a sphere of

the same volume as the particle - true nominal diameter (Wadell, 1932). Sieves,

pipettes, laser diffraction and other grain-size analysis methods involve the division of

the sediment sample into a number of bins, where particle-size limits are related by a

ratio of 2 (Udden, 1914; Wentworth, 1922). This scale was later modified to a logarithmic

scale in phi values: 0 = -log2cf, where d is the particle diameter in mm (Krumbein, 1934,

1936). Therefore, grain-size distribution is constructed from the weight or volume

percentage of each fraction. In this study, an interval of 0.5 0 was used for sieving and 1

0 for pipettes. The resolution of the laser diffraction is higher, but was binned into 0.5 0

groups to match the sieve interval.

Different analytical methods measure different properties that are all referred to

as size. The pipette method measures the weight of the particles with a given settling

velocity and relates it to particle radius by Stokes’ law. Particle radius refers to the

“ radius of a sphere of the same specific gravity and of the same terminal uniform settling

velocity as a given particle in the same sedimentation fluid” (Wadell, 1934, p. 281).

Below 6 or 7 0 , the settling velocity is affected by particle shape, degree of

disaggregation, electrical charges, etc. (Folk, 1980). Temperature affects the viscosity of

the fluid but it is usually well controlled.

The sieve method measures the weight of the particles retained in a sieve and

relates the weight to the diameter of a sphere with the same size as the mesh. However,

33

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
 elongated particles with larger volumes may pass through the sieve, so that grains are

sorted according to the smallest cross sectional area that may or may not have the same

volume as the reference sphere. Sieving time also influences the degree of sorting but

that is a well-controlled factor. Sieve and pipette analysis can be quite time consuming

and relatively imprecise with an error in excess of 40% (Sperazza et al., 2004).

Low-angle laser light-scattering systems (laser diffraction) measure the angular

distribution and intensity of a laser diffracted by particles in suspension. A theoretical

model based on the diffraction of particles with particular properties and grain-size

distributions is then fitted to the actual measurements. Grain size is reported as volume

percentage for each size bin. Sperazza et al. (2004) analyzed the size distribution of fine

grained sediments (1 to 50 pm) and quantified method and instrument uncertainty

associated with laser diffraction analysis of grain size as less than 6% (2a). The

standard operating procedure of the Portuguese Hydrographic Institute is different from

the one described by Sperazza et al. (2004) and is optimized for continental shelf

sediments. Several factors contribute to the uncertainty in the measurement.

The amount of material required for analysis ranges from 0.1 g for clay-size

samples to 1 to 2 g for sand-size samples. The way the aliquot is withdrawn from the

sample material is determinant, as laser diffraction is a volume-based measurement

technique, and therefore, sensitive to small changes in the amount of large material in

the sample. The procedure the Portuguese Hydrographic Institute found to produce the

most reproducible results is to remove the aliquot from a paste, which is easier to

homogenize than a liquid or a solid. Sperazza et al. (2004) used three other methods: 1)

drying a sample and removing an aliquot; 2) directly removing material from the original

sediment sample with a spatula and; 3) extracting the aliquot with a pipette from a

sample previously dispersed with sodium hexametaphosphate. Uncertainty among the

different methods varied between 2.4% for D50 to 4% for D90. The direct method caused

34

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
 the highest uncertainty in the measurement, 5.9% for D50, and a little more than 10% for

Dgo. Sperazza et al. (2004) obtained these values in clays; measurements of grain size

in silts and sands will have higher uncertainty.

Wet dispersion in a solution of deionized water with sodium hexametaphosphate

is commonly used to ensure particle dispersion and prevent particle agglomeration and

flocculation. The dispersant has to be transparent to the laser beam and needs to have a

different refractive index from the particles being measured. Dispersion increases with

concentration, but beyond a certain point the solution becomes unstable causing

flocculation. There is an optimum value for each type of dispersant and sediment (2 g/l is

used at the UNH Coastal Geology Laboratory, 2.55 g/l is suggested by Folk (1980), 3 g/l

is the value used at the Portuguese Hydrographic Institute, 4 g/l is used at the University

of Rhode Island and Sperazza et al. (2004) suggest 5.5 g/l). The use of a dispersant

may increase the formation of air bubbles during sample analysis and shift results

(Duarte, J. personal communication, 2006).

Ultrasonication is used to disaggregate clay agglomerates. Sperazza et al.

(2004) found an increase in clay dispersion up to 60 s. After 60 s, some samples

showed a decrease in grain size interpreted as grain fracturing, whereas others showed

an increase in grain size, interpreted as the flocculation of clays. Portuguese

Hydrographic Institute standard operating procedure defines a disaggregation time of

120 s but the average grain size of the samples is different than those of Sperazza et al.

(2004).

Another variable is the amount of material to be analyzed. For the Malvern

Mastersizer 2000, an obscuration level between 15 and 20% produces consistent results

(Sperazza et al., 2004; Duarte, J. personal communication, 2006). The pump speed also

influences results and an optimum value must be reached in order to keep the sample in

suspension and minimize turbulence.

35

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
 The Malvern Mastersizer 2000 system uses the Mie solution to adjust a

theoretical diffraction pattern to the measurements (Singer et al., 1988; Agrawal et al.,

1991; Loizeau et al., 1994). This model requires knowledge of the refraction and

absorption indices for the particles being analyzed. However, natural occurring

sediments contain diverse mineralogical compositions, with each mineral having a

unique set of optical properties that can affect the outcome of the laser diffraction

results. In this study, a value of 1.544 for particle refraction index (quartz), 1.330 for

dispersant refraction index and 1.5 for particle absorption, were used. Sperazza et al.

(2004) report variations in the outcome that ranged from unimodal to bimodal

distributions depending on the absorption index used and refraction indices varying from

1.43 to 3.22 for the minerals in the sediments analyzed.

Finally, when results for the entire sample are calculated, data from sieves

(percent weight) and data from laser diffraction (percent volume) have to be joined. In

this process all the grains are assumed to have the same density, which may add

uncertainty to the results.

3.3 Sediment Profile Imaging (SPI) and Plan View Image

Sediment Profile Imaging (SPI) was developed as a benthic monitoring tool

during the 1970s and the middle 1980s (Rhoads and Cande, 1971; Rhoads and

Germano, 1982, 1986) and it has been used extensively for monitoring the impacts of

open-water dredged material disposal (the Disposal Area Monitoring System/USACE

web site presents an extensive list of references). A camera is mounted horizontally on

top of a wedge-shaped prism with a Plexiglas faceplate at the front and a mirror placed

at 45° to reflect the profile image at the faceplate to the camera. A strobe mounted inside

the back of the wedge provides illumination and the chamber is filled with distilled water

36

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
 to provide a clear optical path. The assembly is mounted on a frame lowered to the

seafloor by a winch at a rate of approximately 1 m/s (Fig. 3.4). Immediately prior to

landing on the seafloor, a plan-view image is acquired. When the frame comes to rest on

the seafloor, the winch wire goes slack and the camera prism descends into the

sediment at a slow rate (approximately 6 cm/s) controlled by the dampening action of a

hydraulic piston so as not to disturb the sediment-water interface. A trigger activates a

13 s time delay on the shutter release to allow maximum penetration before acquiring

the profile image.

The optical prism penetrates the sediment under a static driving force imparted

by its weight, acting like a static load penetrometer. The weight of the camera prism is

adjusted according with the expected sediment type in order to maximize penetration

and maintain a record of the water/sediment interface.

Figure 3.4 - SPI used by SAIC during 2005 and 2006 sample collection.

37

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
 At least two replicate SPI casts and one plan-view image were acquired for each

station. Sediment profile images were acquired on August 29th, 30th and September 2nd

and 3rd, 2005. In 2006 profile images were simultaneously acquired with grab samples.

Images view an area of 20 x 13 cm (HxW) and were acquired with an 8-bit 6MP digital

camera at a resolution of 2240X1448 pixels. Plan-view images are sometimes difficult to

interpret due to turbidity and were only sparsely used in this study to confirm the

consistency of profile images and grab samples.

38

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
 CHAPTER 4

GROUND TRUTH AND ACOUSTIC REMOTE SENSING RESULTS,

COMPARISON AND DISCUSSION

4.1 Mean Grain Size

The number of variables that need to be controlled in each grain-size-analysis

technique is particularly large, as discussed in section 3.2.5. Therefore, it is important to

compare the final results and assess the discrepancies.

4.1.1 Calculation of mean grain size

Statistics using the method of moments, arithmetic, geometric and logarithmic, as

well as using Folk and Ward (1957) graphical methods, were calculated for each sample

using GRADISTAT software (Blott and Pye, 2001). Although moment measures are

more accurate and more sensitive to environmental processes than are graphical

measures (Friedman and Sanders, 1978), moment measures employ the entire sample

population and are more affected by outliers in the tails of the distribution than the other

measures. Therefore, moment measures should not be used unless the size distribution

is fully known (McManus, 1988). Folk and Ward (1957) graphic measures appear to

provide the most robust basis for routine comparisons of compositionally variable

sediments (Jackson and Richardson, 2007) and are used in this study.

39

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
 The difference in this dataset, between the mean grain size determined

logarithmically by the method of moments compared to that by the Folk and Ward (1957)

graphical method is on average 0.15 O ± 0.16 O (1o). The sample that presents the

highest difference (A11_C; 0.88 O) has the grain size distribution open on the coarsest

fraction, which means that the grab sampler did not capture the full range of grain sizes.

A plot of grain size distribution by weight for all samples collected in 2006 is

shown in Figure 4.1. Samples were analyzed with the standard operating procedure

(SOP), using sieves up to 1 O and laser diffraction for finer sizes. Samples range from

clay to gravel with the majority of the material in the medium and fine sand size classes.

40

£ 20

Grain s iz e (phi)

© ©
3.5

s* 2.5
©©

Mean grain size (phi)

Figure 4.1 - HARS sediment samples collected in 2006: frequency distribution by weight of grain
size in units (top) and sorting versus mean grain size in O units (bottom).

40

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
 There was no grain-size distribution for samples collected in 2005 and the only

data available was in percentage of major fractions (gravel, sand and mud), as is the

case for many databases constructed for benthic-habitat mapping studies. Data from

2006 shows a good correlation between percentages of gravel plus sand and the mean

grain size (Fig. 4.2), at least up to the point where no mud is present in the sediment and

mean grain size continues to increase. The same relation is likely present in samples

collected in 2005.

100

R2 = 0.95

Mean grain size (phi)

Figure 4.2 - Mean grain size versus percentage of gravel plus sand for sediment samples
collected in 2006. There is a good correlation between the percentage of gravel plus sand and
the mean grain size up to the point where no mud is present in the sample and the grain size
continues to increase. R2 = 0.95 for samples with a percentage of gravel plus sand from 10% to
90%.

4.1.2 Error introduced by drying the sample

During the normal SOP, samples were dried and concerns arose regarding

possible agglomeration of clay particles. A comparison was made using only the portion

of the sample that was analyzed with laser diffraction. In the standard procedure, the

material left in the pan of the sieves stack was added to the fraction that was going to be

analyzed with laser diffraction. It may be assumed that the material left in the pan would

be evenly distributed through a wide range of finer sizes. However, to minimize errors,

only those samples with material left in the pan that constituted less than 1% by weight

41

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
 of the fraction finer than 1 cp were considered. Figure 4.3 shows the effect of drying on

the mean grain size. As expected, drying the sample slightly increases the mean grain

size (lower O values) in finer sediments, with a difference of 0.2 O for sample 17_A ( M z

= 6.3 <t>) and 0.4 O for sample 97007_C ( M z = 7.5 O).

H W ithout drying ■ S tandard procedure

20061_A L 2 4 0 0 .A 20028_A L1200_A A12_B 1 4_C 17_A 97007_C

Sample

Figure 4.3 - Mean grain size for samples that were always kept wet (blue bars) and for those that
followed the SOP (red bars). For finer sizes, drying the sample increases the mean grain size
(lower <P value).

4.1.3 Error in the estimation of the mean particle size

In order to estimate the effect of particle desegregation caused by the SOP, an

attempt was made to characterize the naturally occurring particle size using minimal

processing of a subset of samples. Twenty three samples, ranging from sand to clay,

were not digested with hydrogen peroxide, thus preserving organic content, and not

washed and centrifuged to remove salts. Nevertheless, samples had to be wet sieved at

500 pm in order to prevent clogging the pump of the laser diffraction system, and since

deionized water was used, salts were partially washed, therefore promoting some grain

dispersion. Sodium hexametaphosphate was not used to disperse the samples and

ultrasound was not applied in the laser diffraction system. Twenty measurements were

made at 20 s interval. Figure 4.4 shows the evolution of the grain-size distribution by

42

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
 volume for samples 14_C, 17_A and 97007_C caused by mechanical stirring. The

agglomeration effect is more pronounced for smaller grain sizes with the extreme shown

in sample 97007_C as a clear bimodal distribution.

14 C 17_A

Qrain (phi) Grain aiza (phi)

Figure 4.4 - Frequency distribution by volume of grain size in O units for samples 14_C, 17_A
and 97007_C (from left to right). Each line corresponds to the distribution measured with 20 s
interval. As time progressed the distribution moved to smaller sizes. Arrows point the evolution of
the grain size with time.

Figure 4.5 compares the mean grain size calculated using the standard

procedure with the mean grain size obtained using the first measurement of those

samples that went to the laser diffraction without any treatment. Again, in order to

minimize errors only those samples with material left in the pan that constituted less than

1% by weight of the fraction finer than 1 O were considered.

Directly to the laser diffraction ■ Standard procedure

Sample

Figure 4.5 - Mean grain size for samples that followed the SOP (red bars) and for those analyzed
by laser diffraction without any treatment (blue bars). The difference in mean grain size increases
as samples become finer (higher O value).

43

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
 Samples that represent the sediment closer to its natural occurrence have a

larger mean grain size (lower 0 value) and this difference is accentuated in finer

sediments. Particle agglomeration is also directly related to the organic content of the

sample (Fig. 4.6).

% Dry weight LOI

Figure 4.6 - Difference between the mean grain size in <J> units of the samples that followed the
SOP and those analyzed by laser diffraction without any treatment versus the organic content
(percentage in weight of material lost on ignition). The mean particle size in the natural sediment
is higher than the mean grain size determined by normal grain size analysis, and this difference is
directly related with the organic content of the sample, with a coefficient of determination R2 =
0.73.

4.1.4 Comparison between the SOP (sieve-laser diffraction) and the sieve-pipette
method

A subset of samples was also analyzed with the conventional sieve-pipette

method in order to compare results from a method that measures distributions by weight

with one that measures distributions by volume.

Figure 4.7 shows mean grain size and sorting obtained with sieve-laser

diffraction and sieve-pipette methods. Sieve-pipette consistently biases the mean grain

size towards the finer fractions and results in poorer sorting of the finer samples. This

may be due to the lack of resolution of the pipette method where the concentration of

44

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
 grains smaller than 10 0 is divided by bins 11 0 to 14 0 . Another explanation lies in the

way grain size is measured; sizes between 1 0 and 4 0 were measured with sieves in

the sieve-pipette method and sieves tend to retain coarser grains than those measured

for the same class in laser diffraction.

The difference in mean grain size between sieve-pipette and sieve-laser

diffraction is on average 0.35 O ± 0.14 0 (1a). In sample 17_A, this difference reaches

2 0 but the pipette method is more prone to human error so this measurement was

considered as a blunder.

mPipettes ■ Laser diffraction

20061_A L2400_A 20028_A L1200_A A12_B 11_A 20041_A 14_C 17_A 97007_C

Sample

mPipettes ■ Laser diffraction

20061_A L2400_A 2Q028_A L1200.A A12J 20Q41_A 14_C 17_A 97007_C

Sample

Figure 4.7 - Mean grain size (top) and sorting (bottom) in 0 units for samples analyzed with
sieve-laser diffraction (red bars) and sieve-pipette methods (blue bars). The sieve-pipette method
systematically results in a smaller mean grain size.

45

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
 4.1.5 Estimated uncertainties in the determination of mean grain size

Instrumental uncertainties were evaluated only for the automated method of laser

diffraction. For a population of 50 samples with three replicate measurements per

sample, the overall uncertainty in the measurement is 0.024 O ± 0.026 <t> (1a). The

maximum uncertainty found in the determination of mean grain size for a single sample,

with three replicate measurements, is 0.082 O ± 0.054 O (1a).

Uncertainties in mean grain size due to the effect of drying the sample in the

SOP increase as the grain size decreases (higher <t>). For medium or fine sands, the

effect of drying the sample is almost negligible whereas for silt and clay sizes the

difference can reach 0.2 O and 0.4 O, respectively.

Differences between sieve-laser diffraction (SOP) and sieve-pipette methods are

on average 0.35 <t> ± 0.14 O (1a) towards smaller grain sizes (higher O) for sieves-

pipettes.

The highest degree of uncertainty seems to arise from the basic premise of

grain-size analysis. From a sedimentological perspective, the objective is to obtain the

grain size of the clastic particles as they were deposited (Folk, 1980) in order to get

information about the parental source, transport and depositional processes.

Decomposing the sample to its elementary particles accomplishes this objective and

sets a standard for data comparison. But this process may not adequately characterize

the mean grain size as a descriptor for geoacoustic properties.

In spite of not being representative of the natural sediment, samples that

received minimal preparation for analysis compared to those analyzed with the SOP,

show a difference in the mean grain size of 0.2 O or less for medium and fine sands,

approximately 0.3 <t> for very fine sands, 0.5 O for coarse and medium silts, and

approximately 1.5 <J> for fine silts to clays. The natural sediments are expected to show

46

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
 higher differences because it was not possible to analyze the samples without disturbing

the sediment.

Figure 4.8 summarizes results for samples 20061_A (medium sand), A12_B

(fine sand), 14_C (very coarse silt) and 17_A (medium silt).

•20061_A LOO (average) •A12_B LDD (average)

20061_A w/out drying A12_B w/out drying

— 20061_A pipettes — A12..B pipettes

Oraln site (phi) Grain s ite (phi)

17_A LDD (average)

14_C w/out drying

— 14_C pipettes

Grain size (phi) Grain size (phi)

Figure 4.8 - Grain size versus percentage of weight for samples 20061_A, A12_B, 14_C and
17_A. The distribution of grain sizes obtained with the SOP (sieve-laser diffraction split at 1 O) is
shown for comparison with the sieve-pipette method (split at 4 <J>). Results considering only the
portion analyzed with laser diffraction are shown for comparison with the effect of not drying the
sample and of sending the sample directly to the laser diffraction without going through the
standard preparation (average of 20 measurements).

4.1.6 Mean grain size versus penetration of the optical prism

In order to estimate if mean grain size can be used as a proxy for geotechnical

properties of the seafloor, the penetration of the optical prism is compared with the mean

grain size. Figure 4.9 relates the penetration of the optical prism with the percentage of

gravel plus sand for data collected in 2005 and 2006. Although the penetration increases

for finer sediments, there is no apparent correlation with the force exerted over the prism

47

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
 (number of weights). Calculated penetration is the average of all the replicas in each

nominal station for 2005 data (SAIC, 2005a). For data collected in 2006, only those

stations where the SPI image and the grab sample were obtained in the same replica

were considered.

♦ 2 Weights ♦ 2 Weights
m3 Weights • 3 Weights
* 4 Weights a 4 Weights

4-
4 J

% Gravel*Sand % Gravel + Sand

Figure 4.9 - Penetration of the optical prism versus percentage of gravel plus sand for each year.
The weights in the camera were adjusted in order to obtain more penetration in harder bottoms.
Although an apparent correlation can be observed for data collected in 2005 there is no
differentiation by number of weights. Also, for data collected in 2005, the penetration of the optical
prism is the average for all profile images within the same station and the percentage of gravel
plus sand is calculated as the average of all the samples collected within the same station. For
data collected in 2006 the same correlation is not observed, in this plot the penetration of the
optical prism versus the percentage of gravel plus sand was calculated for each replica within the
station.

If mean grain size is plotted against the penetration of the optical prism, instead

of the percentage of gravel and sand (Fig. 4.10), a correlation is seen between the two,

thus explaining why mean grain size is sometimes used as a proxy for geotechnical

properties of the seafloor.

48

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
 2006

♦ 2 Weights
a 3 Weights
4 4 Weights

-4 -2 0 2 4 6 8 10 12 14
G rain s iz e (p hi)

Figure 4.10 - Penetration of the optical prism versus mean grain size for samples collected in
2006. Lines reflect the general trend for each number of weights.

4.2 Backscatter Strength

4.2.1 Backscatter strength versus mean grain size

The observed relationship between percentage of gravel plus sand and mean

backscatter strength, calculated within the uncertainty area of each sediment sample, is

depicted in Figure 4.11. A general trend is observed with backscatter strength increasing

with the percentage of gravel plus sand in 2005 but the same trend is not observed for

data collected in 2006. Sediment samples collected in 2005 are composite samples

whereas each sample collected in 2006 corresponds to a single deployment of the grab

sampler. The average backscatter strength is also calculated over different areas, the

size of the uncertainty area of the sediment samples collected in 2005 is on average 4

times the size of the uncertainty area of the sediment samples collected in 2006.

49

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
 Figure 4.11 - Percentage of gravel plus sand versus mean backscatter strength calculated within
the uncertainty area of each sample for 2005 (left) and 2006 (right).

The relation for data collected in 2006 seems to improve when mean grain size

rather than percentage of gravel plus sand is considered (Fig. 4.12, left), but though a

general trend can be observed, the correlation is weak (R2 = 0.14). Several factors may

contribute to the lack of a strong correlation: 1) uncertainty in sample position; 2)

measurement of a grain size that does not correspond to the size of the particles in the

natural sediment and; 3) inability of the grain size to represent the acoustic response of

the seafloor. As will be discussed below, an attempt was made to consider all these

factors with the results presented in Figure 4.12 right, showing an improvement in the

correlation with backscatter strength, with a coefficient of determination R2 = 0.74.

« Normal heterogeneity for the grain size

@More heterogeneous than normal
a Much more heterogeneous than normal

■i -3 -2 5 6 7 •2 -1 o 5 6 7 S 9
Mean grain «lz* (phi) Mtan partial* aiz* (phi)

Figure 4.12 - Relation between mean grain size and mean backscatter strength (left) and
between inferred mean particle size and mean backscatter strength for the deposit the sample
was attributed (right). The correlation with backscatter strength improves with supervised
positioning of the sample and estimation of the mean particle size. There is some differentiation
by degree of sediment heterogeneity.

50

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
 First, the uncertainty in sample position was considered. As an example, within

the area of uncertainty of sample 26_B (Fig. 4.13), two distinct backscatter areas can be

observed. In this case, based on the analysis of the entire deposit and on the

comparison with other similar sediment samples obtained in contiguous areas of high

backscatter, and with samples corresponding to finer sediments obtained in contiguous

areas of low backscatter, the sample was attributed to the area of high backscattering

strength. The average backscatter was recalculated considering only the area of high

backscatter strength. In station 4, replica 4_A, the analysis of the backscatter revealed

only one type of sediment within the uncertainty area of the sediment sample. The area

that corresponds to replica P2800_B presents some heterogeneity but it wasn’t possible

to clearly identify different sedimentary deposits, so the average backscatter within the

whole area of uncertainty of the grab sample was considered.

Next, two samples that presented significant differences from the SPI image

were removed (97007_C and K0800_C). The removed samples correspond only to red

clay whereas the SPI images show mostly sand. Although several cases exist where

there are differences between the sediment sample and what is observed in the SPI,

these cases represent the extreme situation where only one component of the surface

sediment was sampled.

The next step was taken to correct for the difference between measured mean

grain size and the actual mean particle size, using the relation shown in Figure 4.5 and

Figure 4.6 and assuming that the organic content is directly related with the degree of

agglomeration of finer particles. The mean grain size was corrected assuming the

relation between the mean particle size and the percentage of weight lost on ignition:

mean particle size = mean grain size + (% dry weight LOI x 0.2). This relation mainly

affects particles in the silt and clay range.

51

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
 BS 26_B A4_A P2800_B
Max. -14.7 dB -23.8 dB -12.1 dB
Min. -26.4 dB -29.4 dB -26.0 dB
Mean -18.7 dB ± 2.3 dB (1a) -26.9 dB ± 0.8 dB (1a) -18.8 dB ± 1.9 dB (1a)

Figure 4.13 - From top to bottom: backscatter strength in the uncertainty area of the grab sample,
SPI images and photographs of the grab sample for stations 26_B, A4_A and P2800_B. Each
pixel in the the backscatter mosaic corresponds to 1m. SPI images correspond to 20 x 13 cm
(HxW) and grab sample photos to 2.2 x 3.3 cm (HxW).

Finally, two of the factors that contribute to the scattering of the acoustic wave

were also considered; surface roughness and volume heterogeneity. In the HARS, the

52

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
 two main sources of heterogeneity are biological reworking and dumping of varied

material, both of which equally affect scattering on the surface and volume scattering in

the first few centimeters of sediment. Based on SPI images, the seafloor was classified

as: 1) having normal heterogeneity for the sediment type; 2) as being more

heterogeneous than expected or; 3) as having much more heterogeneity than expected.

All the SPI image replicas for the same station that presented the same type of sediment

as the grab sample were used in the decision process due to the “ point sample” nature

of the SPI compared with the spatial variability of the area. Figure 4.13 illustrates the

criteria; stations 26 and A4 have normal roughness and volume heterogeneity and

P2800 much more roughness and volume heterogeneity than expected for the type of

sediment.

Obviously, the criteria used are subjective, but illustrate the factors that may

contribute to the lack of correlation between mean grain size and backscatter strength,

which is sometimes observed in areas with high spatial heterogeneity.

4.2.2 Comparison of backscatter strength for the same type of sediment

One way of evaluating the consistency of results is to compare backscatter

strength for the same type of sediment. This comparison was made only for data

collected in 2006 because detailed granulometric distributions are not available for

sediment samples collected in 2005.

Samples 20028_A, 20064_A, L1200_A and L2400_A collected in 2006, are from

different locations but consist of the same sediment facies; a moderately well sorted fine

sand, symmetrical and mesokurtic (Fig. 4.14 and Fig. 4.15). Backscatter values

calculated as the average within the area of uncertainty of each sample are: -26.7 dB ±

1.7 dB (1a), -24.9 dB ± 1.0 dB (1a), -25.9 dB ± 1.4 dB (1a) and -25.8 dB ± 1.0 dB (1a),

53

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
 respectively. All the mean values fall within ±1 dB, which reflects the effectiveness of

Geocoder in applying the necessary corrections. A standard deviation between 1.0 and

1.7 dB for the backscatter strength in a circular area on average 40 m diameter,

suggests even better results if a point sample could be considered.

Figure 4.14 - SPI images from samples 20028_A, 20064_A, L1200_A and L2400_A (left to right)
obtained with a load of 4 weights. Photos correspond to 20 x 13 cm (HxW).

0 2OO28.A
■ 20064.A
□ L1200.A
□ L2400.A

SO RTING. FW_phi SKEWNESS.FW.phi KURTOSIS.FW.phi

Figure 4.15 - Mean, sorting, skewness and kurtosis determined by the Folk and Ward (1957)
graphical method, in <J> units, for samples 20028_A, 20064_A, L1200_A and L2400.A (left).
Percentage in weight of moisture and percentage of material LOI relative to the dry weight for the
same samples (right).

4.2.3 Differences in backscatter strength between 2005 and 2006 and temporal

evolution of the seafloor

Multibeam sonar data was acquired at the same frequency (240 kHz) and with

the same system (Reason 8101) in 2005 and 2006, so it is possible to calculate changes

54

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
 in backscatter strength from year to year and estimate the consistency of results.

Mosaics were assembled following the same procedure described in Chapter 3. The

average difference in backscatter strength between the two years is -1.0 dB ± 2.8 dB

(1o), this bias was corrected by leveling the average difference in backscatter strength to

0 dB. Figure 4.16 depicts the difference in backscatter strength between 2006 and 2005.

Variations in backscatter strength seem to be collocated with variations in

bathymetry (Fig. 4.17 and Fig. 4.18), corresponding to the deposition of new capping

material in PRAs 1 and 2. Considering the relation between mean grain size and

backscatter strength determined in section 4.2.2 it is possible to infer changes in

sediment type relatively to the previous year. Recently deposited coarser material

(positive difference in backscatter) in PRAs 1 and 2, seems to concentrate more than

recently deposited finer material (negative difference in backscatter) that has spread

over a larger area. Some scour depressions in PRA 1 and 3, which are probably caused

by anchoring (Fig. 4.16), seem to be filled with new material. There seems to be a

generalized transport of coarser material to the southeast across the HARS as

evidenced by the distribution of positive values on both sides of the mounds in the center

of the area. There is some noise concentrated around the sonar nadir, but that does not

affect the observation of major variations in backscatter strength.

55

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
 594000 595000 596000 597000 598000 599000 600000
» i--- I____ I ..

4476000
4475000
, . ' - ‘ . - , i

4474000
H i. .

4473000
• V i- /■ ’
i
—- - —
' ' /v ■ !
1

4472000
T ” ------- V - '

i
-
1
9liiiilis
/- ■ , w/i
**» *1" - s . . .
1 .
1
:L J
Mb

4471000
v.fl:, i 5 ' V .v:.''' :v <
••'sj1 s. 't;
a*

4470000
"*•-.* ..
5 D iffe re n c e
X '
■
.■ ■
'■: '.'i '' ; —' •

•1 4 9 -4 2 r \ V : -■
, \ . . a ............
N j
\ i , ----------

4469000
V - " ' ;
!
t __

3 ,1 - 6

4468000
6,1 - 8

1 2 ,1 -1 5

594000

j_____ 1 1997 C ategory II Cap Boundary

£ J 1993 Dioxin Cap Boundary

CIIJ RedC,ay
£ I Former MDS

CD HARS PRAs UTM Projection Zone 18N

| HARS Buffer Zone Boundary NAD 83 Datum

Figure 4.16 - Difference in the 240 kHz backscatter strength between 2006 and 2005, positive
values are depicted in pink and correspond to an increase in grain size. Negative values are
depicted in blue and correspond to deposition of finer material. Scours are indicated by arrows.

56

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
 o
s
£
3

o
S

©
8
2
3

O
8
r>

O
8
N
!"■
3

O
8
t'-
3

O
8
O
r-
3

O
8
3
3

©
s
8
5

594000 595000 595000 597000 598000 599000 600000

j j 1997 Category II Cap B oundary

2006-2005 Surface Difference
1993 Dioxin Cap Boundary
m
CLIJ RedC|ay
| I Former MDS jPjjjjjjjj H i^ i: 4
□ HARS PR As UTM Projection Zone 18N
[ { HARS B ufferZ one Boundary Low : -4 N A D 83 Datum

Figure 4.17 - Difference in bathymetry between 2006 and 2005, positive values are depicted in
blue and are mainly concentrated in PRAs 1 and 2 where new material was deposited. Some
erosion is also observed in the mounds that were already formed in 2005 (PRAs 1 and 2),
corresponding to negative values depicted in brown.

57

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
 Figure 4.18 - View from northwest over PRAs 1 and 2. The difference in backscattering strength
between 2006 and 2005 is draped over the difference in bathymetry. Positive values are depicted
in red and correspond to coarser sediments. Negative values are depicted in blue and correspond
to deposition of finer material. Vertical exaggeration: 15x.

4.2.4 Backscatter strength versus penetration of the optical prism

Considering the results shown in Figure 4.10 (correlation between mean grain

size and penetration of the optical prism), it would be expected that backscatter strength

should present the same degree of correlation when compared with the penetration of

the optical prism. The correlation observed in Figure 4.19 does not indicate backscatter

as a good predictor of the geotechnical properties of the sediment. The lack of

correlation might be explained by the uncertainty in sample position. In 2006 the optical

and the sediment sample were taken 10 m apart, plus the deviation derived from cable

angle. However, the mean backscatter was calculated as the mean within the

uncertainty area of the sediment sample, with an average area of 1257 m2. An apparent

correlation seems to exist in 2005 but the penetration of the optical prism was calculated

58

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
 as the average for all the replicas that fall within a radius of 40 m and the mean

backscatter was calculated as the mean within an uncertainty area of 5027 m2. In both

years, there is no differentiation by number of weights.

2006 2006
• 0 Weights « 2 Weights
* 1 Weight ■ 3 Weight*
♦ 2 Weights x 4 Weights
• 3 Weights
* 4 Weights

8 10 12 14 8 10 12 14 16 18
3P I penetration (cm) SPI penetration (em)

Figure 4.19 - Mean backscatter strength within the area of uncertainty of each sample versus
penetration of the optical prism for 2005 (left) and 2006 (right). The value presented for the
penetration of the optical prism in 2005 corresponds to the average of all replicas within a station.

4.3 Angular Range Analysis

4.3.1 ARA in normal mode

ARA was performed in normal mode resulting in one ARA solution for each

resolution cell, defined by the half-swath width of the sonar in the across-track direction

and a stack of 30 pings in the along track direction. Appendix D shows some examples

of the angular response for different types of sediments and model fits to the angular

responses. Thiessen polygons (a polygon defined by the perpendicular bisectors of the

lines connecting points in a triangular irregular network, whose boundaries delimit the

area that is closest to each point comparatively to all other neighboring points) were

constructed around each ARA solution and the mean grain size as predicted by the

59

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
 model is depicted in Figure 4.20. The observed segmentation of the seafloor is very

similar to the one that a geologist would have generated by visual interpretation of the

backscatter mosaic, but with the great advantage that the characterization of the

seafloor in terms of mean grain size is objective. It is derived from the measurement of a

geoacoustic property and is not dependant on the human interpretation of a backscatter

mosaic that can have a different outcome depending on which AVG was applied.

The 1993 dioxin, and the 1997 Category II capping projects are easily

recognized, and capping material is predicted to be medium to fine sands. A deposit of

fine sediment aligned with the axis of the Hudson Shelf Valley (Fig. 1.1) is also

recognized, with a predicted mean grain size of coarse to medium silt. Another

alignment of fine sediment exists to the west, and is more easily recognized than in the

backscatter mosaic, in agreement with sediment transport to the head of the Hudson

Canyon on the west flank of the topographic highs in the center of the HARS. The area

corresponding to rock outcrops on the east side of the HARS is predicted to have

M 2= -1 <t>, the lower limit of the ARA model. Mounds of remediation material placed on

PRAs 1, 2, 3 and 4 are predicted to have a mean grain size ranging from granules to fine

sand. Preexistent sediment that is exposed on the sides of these mounds is classified as

fine sand to coarse silt, which agrees with what is observed in the SPI images.

In spite of the good results at a regional scale, the limitations are evident when

comparing ARA solutions with the backscatter mosaic at a local level. Figure 4.21 shows

the predicted grain size at station A12 in PRA 9. A general trend can still be recognized,

but HARS variability at a decameter scale can not be resolved by the spatial resolution

of a method that employs the half-swath of the sonar (around 50 m for this region).

60

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
 594000 595000 596000 597000 598000 599000 600000
I____ I t ■■ - I * 1- - »

'• ..' 5; «£"> , ' -

4 r,i =■ cv '^b .. ; . •- r
*‘ *Y .
, '■ ■ fJ
*■ «

:■v?%
^ ersf*^.
C ".-

|5 -
'e r t -. ' Vt i ' jWJ l t- '-t C r
* ir.-i ;p - I I -a ■. .;0'<B;iV
*1, ' -

> 71.
■ ty * .5 . .m s

■, ’l >v--:sf ; - . s o f
r

•' vC1
0 '1'’1
1*-■.
: • ■1 j.o
M O • '
-• . r-
A
wMmSUBBmM
«■ 1
0

h . -■ .. .
mm
*'iji

"... £ * x "
. O .'

. c ' -s
o

•y # 4 f •■■-j v :

2 0 0 6 -A R A - '-M sg
phi
. -£*; Tii i - t' I ,*» , t .' \
:sei'_&: / ’ \ J t -Vinf-.c ?,
-fe J. X- *5-y ^rb.— ....I;— -----------------------
. - ' . V j t ' j y l ' J l v . .wciy.fS7o.j.F )

34 B3fr<e
e > :. ^ • " 1 M r - * * " ■'

t o
r -*- -

1--- 1---
594000 595000 596000 597000 598000 599000 600000

C~ J 199^ Category II Cap Boundary

C Z IJ 1993 Dioxin Cap Boundary

C H j Red clay
[" I Former MDS
2006 - Grab Samples
□ HARS PRAs UTM Projection Zone 18N
j HARS B ufferZ one Boundary 2006-SPI NAD 83 Datum

Figure 4.20 - Predicted mean grain size from the Angular Range Analysis for data collected in
2006. The area marked with a dark blue rectangle is zoomed in Figure 4.21.

61

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
 Figure 4.21 - Left: predicted mean grain size from the ARA for station A12 (2006). Polygons were
constructed using ARA solutions represented by black dots (the color scheme is the same as the
one used in Figure 4.20). The green circle represents the area of uncertainty of the grab sample
and yellow circles the SPI uncertainty. Right: backscatter mosaic for the same area, the red
rectangle corresponds to 150 x 200 m and marks the area where ARA was applied in “theme"
mode.

4.3.2 Comparison of ARA results with mean grain size

Comparison of the mean grain size as predicted by the ARA with ground truth

data is made considering the measured mean grain size and not the inferred mean

particle size as in Figure 4.17. It is assumed that sediment samples used in the original

measurements, that gave rise to the regression equations used in the ARA model,

followed the normal preparation used in the majority of the grain-size analysis methods.

When the average of the ARA solutions inside the area of uncertainty of the grab

sample is plotted against the measured mean grain size the result is a scatter of points

(figure 4.22).

62

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
 • Homogeneous patch of the seafloor not layered in the first centimeters
u Captured only part of a layered sediment
1 Penetration into a layered sediment
mDid not capture the full range of grain sizes
▲ Heterogeneous patch of the seafloor

!c 5

2 3
-S
<

-2 -1 0 1 2 3 4 5 6 7
Mean grain s iz e (phi)

Figure 4.22 - Predicted mean grain size from the Angular Range Analysis for 2006 data versus
measured mean grain size from sediment grab samples for sizes finer than -1 O. Stations where
the seafloor is homogeneous and without sediment layering in the first centimeters show a good
correlation between ARA solutions and mean grain size determined by grab sampling and
sediment analysis. Other points deviate from this trend, when the seafloor is heterogeneous,
when the grab sample did not capture the full range of grain sizes or when the seafloor is
homogeneous but the grab penetrated into a layered sediment.

The biggest difference is for station K0800 located in PRA 2 at the base of a

mound of remediation material. ARA solutions that fall inside the area of uncertainty of

the grab sample are relatively consistent (1.20, 2.26, 2.50 and 2.69 O) but far from the

measured mean grain size of 8.2 O. Backscatter strength inside the area of uncertainty

of the sediment sample varies between -16 dB and -27 dB but most of the region falls in

areas of higher backscatter and although some heterogeneity exists it is not enough to

explain this discrepancy. However, the sediment was recovered from the SPI frame in

the absence of a successful grab sample and represents only the upper millimeters. SPI

images are smeared by red clay that exists on the surface but it is possible to observe a

63

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
 few centimeters of remediation material (medium to fine sand on top of silts) that agrees

well with ARA predictions.

The same correspondence of ARA prediction and grain size occurs at station

97007 situated near the base of a mound in the red clay deposition area inside PRA 6.

Even though collected with a grab sampler, the recovered red clay only corresponds to a

surface layer less than 1 cm thick as observed in the SPI. It is difficult to evaluate the

grain size of the subsurface material on the SPI due to smearing by red clay but it

appears to be sand. Two ARA solutions that fall in this area are both sands (0.63 O and

1.15 <t>). Considering an attenuation of only 0.2 dB/cm for the first layer of red clay at 45°

grazing angle (Appendix A), it is likely that the contribution to scattering is mostly from

the subsurface sand layer.

Grab samples 27_C, A1_C and N2000_C are coarse and medium silts recovered

from the frame and are situated in relatively homogeneous areas. The recovered

sediment agrees relatively well with what is observed in the SPI images, except for a thin

surface layer of fine sand that is not represented in the sample from the SPI frame. This

probably explains why ARA predictions are coarser by 1 to 2 O. These are examples of

sampling procedures that failed to adequately characterize the sediment to the extent

that is penetrated by the acoustic wave.

64

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
 Figure 4.23 - ARA mean grain size solutions (the color scheme is the same as for Figure 4.20)
over the backscatter mosaic for the area of station 17 (2006). The area of uncertainty for each
sample is indicated by yellow circles for SPI images and with a green circle for the grab sample.
Sediment profile images are 20x13 cm (HxW) and the grab sample image corresponds to 2.2 x
3.3 cm (HxW).

The opposite effect seems to arise when the material is sampled to a depth

beyond what is penetrated by the acoustic wave. Station 17 (Fig. 4.23) is situated in an

area where backscatter variations are subtle (-21.7 dB ± 1.2 dB 1o) and different

deposits cannot be clearly defined. ARA solutions fall in the medium sand size class but

the determined mean grain size for the grab sample is in the fine silt size. SPI images

reveal from 2 to 5 cm of sand over fine grained material. The grab sampler penetrated

several centimeters into the bottom and the mean grain size is an average of more

sediment than what is contributing to scattering in the direction of the transducer

(approximately 2 cm in medium sands - Appendix A). Stations A9 and 11200 are similar

65

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
 cases. They are in areas of relatively homogeneous backscatter and ARA predictions

agree with what is observed in the first centimeters of the SPI.

Observations in stations 17, A9 and 11200 appear consistent with 2.3 dB/cm of

attenuation for medium sand ensonified at 45° grazing angle, and with an expected

acoustic penetration depth around 2 cm (Appendix A). They also seem consistent with

the vertical structure of the sediment. The second layer appears to be an homogeneous

very fine silt and the main contribution for scattering is at the wavelength scale,

0.1 A - 10 A (Greaves and Stephen, 2000), that has its lower limit for 240 kHz at medium

to coarse sand, close to the grain size of the top layer.

Other consequences of inadequate physical sampling may occur in coarse

material when the size of the grab is not enough to capture the full range of grain sizes.

Stations M1200, A2 and A3 are in relatively homogeneous areas and comparisons with

SPI images suggest that the mean grain size is probably higher than what was captured

by the grab sampler. ARA predictions are -1 0 but this is the lower limit of the model.

Still, results are within 1 0 of the measured mean grain size. Another example is from

station A16 where the grab sample spreads over the sand and silt size class with

M z = 3.9 O but the SPI did not penetrate into the bottom and on the surface reveals

2 cm pebbles. ARA results are 0.7 0 , a coarse sand. All areas where the measured

mean grain size is smaller than -1 0 are adequately classified by the ARA model.

When the backscatter strength within the uncertainty area of the grab sample is

relatively homogeneous and it is not possible to clearly identify different sedimentary

deposits, and when profile images show homogeneity in the first 2 to 5 cm, ARA

predictions are within ±0.4 0 of the measured value for mean grain size. This is

consistent for fifteen samples in the very coarse to fine sand range. Three other samples

66

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
 present worse correlations, two medium sands with 0.5 and 0.9 O difference, and one

coarse sand with 0.6 <P difference.

Station 20051 corresponds to a homogeneous area of the seafloor but there is

some striping in the backscatter caused by survey lines run on different days. ARA

results differ by 0.6 O from the measured value at this station.

In areas of relative horizontal and vertical homogeneity, ARA solutions correlate

well with the mean grain size determined by grab sampling except in the cases where

the grab sampler failed to take a representative sample of the sediment. When

horizontal and vertical heterogeneities exist, there is no apparent correlation between

the two determinations of mean grain size.

4.3.3 ARA in supervised segmentation mode (in areas of horizontal

heterogeneity)

Until now, only samples were considered that had been collected in places where

the backscattering strength is relatively homogeneous over a patch of the seafloor

approximately 50 m wide. In all these cases, the ARA model matched the observations,

or the physical sample did not characterize adequately the sediment that is being

ensonified.

The backscatter strength for 14 stations varies considerably within the

uncertainty area of the grab sample and this brings up two different issues. First, the

sample has to be attributed to the sedimentary deposit where it came from. This can be

accomplished by supervised positioning of the sample, using information from

contiguous areas. Second, the backscatter angular response uses half the swath width

of the sonar which does not provide enough spatial resolution for areas like the one

depicted in Figure 4.21. In order to overcome this limitation polygons of 200 by 150 m

67

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
 were defined around each station. Each polygon was manually segmented into smaller

areas with similar textural characteristics (themes) assuming that each theme

corresponds to areas on the seafloor with similar physical and geoacoustical properties.

In the normal ARA mode, the observation is the average angular response of a

certain number of stacked pings in the along track direction (30 in this study), whereas in

the ARA theme mode, the observation is the average of all the snippets that fall within

each theme, even if they come from different acquisition lines. ARA was executed for

each theme, which means that inside each rectangle there are as many observations as

the number of themes and as many solutions as themes.

Taking station A12 (2006) as an example, the backscatter mosaic allows

differentiating two major deposits (Fig. 4.24): one corresponding to areas of high

backscatter (depicted in light blue) and the other to areas of low backscatter (depicted in

dark pink). SPI replicas for the same station reveal high variation over a short range

(Fig. 4.25), showing a gravelly bottom and a layered sediment with 4 cm of sand over

silt. The grab sample with M z = 2.4 O was attributed to the deposit depicted in dark pink.

Two themes were defined for ARA, each one corresponding to a sedimentary deposit.

ARA solutions are -1 O (gravel) for the area in light blue and 1.61 <t> (medium sand) for

the area in dark pink. Angular responses for each theme are depicted in Figure 4.26.

68

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
 Figure 4.24 - Manual segmentation of the area marked by the red rectangle in Figure 4.21 right,
200 x 150 m. Although subtle variations can be observed in the backscatter mosaic only two
different types of deposits could be clearly identified, transitional terms are probably included in
each of the themes defined.

Figure 4.25 - 20 x 13 cm (HxW) SPI images and 2.2 x 3.3 cm (HxW) grab sample photo for
station A12 (2006). The distance between the two red dots corresponding to SPI positions is 5 m
and the uncertainty radius 15 m. The uncertainty radius for the grab sample is 24 m.

69

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
 A12 - Gravel
— A12 - Medium sand
-10 - — — Gravel - model fit
— — Sand - model fit

-15 -

2. -20 -

-25 -

-30 -

-35

G ra z in g an g le (degrees)

Figure 4.26 - Angular response and model fit in “theme” mode for station A12 (2006). The blue
line corresponds to the light blue area depicted in Figure 4.24, and the pink line to the area in
dark pink.

Although the segmentation method considerably improves ARA results in areas

where the normal mode does not have enough spatial resolution, when the medium is

layered in the first few centimeters, ARA predictions and grab samples do not match.

This is a similar case to the one already observed in section 4.3.2 for stations 17, A9 and

11200.

4.3.4 Comparison of ARA results with the estimated mean grain size for the top

layer (in areas of vertical heterogeneity)

If SPI observations are considered valid to a lateral extent of 10 to 20 m, it can

be assumed that the same sediment was collected by the grab sampler, if no significant

difference between the two is observed. In the majority of the cases, the uppermost layer

of sediment corresponds to remediation material, usually sand, deposited over silt. The

mean grain size varies to the extent the grab penetrated the sediment. If the sampler

collected mostly the top layer of the sediment, then the mean grain size will be larger. If

the sampler collected mostly the subsurface layer, then the mean grain size will be

70

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
 smaller. Therefore, the mean grain size is not a measure of sediment properties that will

influence its acoustic behavior but only a measure of grab efficiency. Although there

seems to be a general trend, if the penetration of the grab fluctuated as much as the SPI

penetration for the same type of sediment (Fig. 4.14 and Fig. 4.15), mean grain size can

hardly be used as a geotechnical indicator either.

ARA solutions seem to characterize well the few first centimeters of the seafloor

for the frequency of interest (240 kHz). Considering that when layering is present, each

layer is relatively homogeneous and corresponds to a different type of sediment, it is

possible to infer the grain size of the top layer by looking at the grain-size frequency

distribution. The grain size histogram usually has one mode for each type of sampled

sediment (Fig. 4.27) and the agreement with ARA solutions is remarkable as seen in

Figure 4.28.

20041

M*an grain *lz*(p h l)

Figure 4.27 - Grain size frequency distribution and profile images for samples 20041_A (top) and
I1200_A (bottom), from 2006. A first layer of remediation material, medium to fine sand, with a
few centimeters, overlays a layer of medium to fine silt. The blue arrow represents the mean grain
size for the sampled sediment and the red and green arrows the mode corresponding to each
layer.

71

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
 7

• Relatively homogeneous patch of the seafloor, not laye in the first centimeters
A Heterogeneous patch of the seafloor, not layered
6 A Heterogeneous patch of the seafloor, layered
■ Relatively homogeneous patch of the seafloor, layered
□ Mode top layer
A Mode top layer
5

R 2 = 0 .8 9 5 7

2
_n

■1

•2
-2 •1 0 1 2 3 4 5 6 7
Mean grain size (phi)

Figure 4.28 - Predicted mean grain size from ARA for 2006 data versus measured mean grain
size from sediment grab samples. Samples depicted by squares in Figure 4.22 that did not
capture the full range of grain sizes were removed. ARA solutions for samples depicted by
triangles in Figure 4.22, corresponding to high spatial variability, were obtained using
segmentation and definition of themes. The mean grain size of the top layer was approximated by
considering the mode corresponding to the top layer. Green arrows connect the mean grain size
of a layered sediment to the mode of the top layer of the same sediment sample. The red line
represents the linear regression after correcting for horizontal and vertical heterogeneities.

After correction for vertical heterogeneities as well, the correlation between ARA

solutions and mean grain size is high (coefficient of determination R2 = 0.90). However,

since points are mainly concentrated between 1 and 3 <t>, more data is needed to

validate this correlation.

72

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
 4.3.5 Estimation of the consistency of ARA results

When considering samples 20028_A, 20064_A, L1200_A and L2400_A collected

in 2006 (Fig. 4.14 and Fig. 4.15) that were used to estimate the consistency in

backscatter strength, it is observed that ARA predictions calculated as the average of

the solutions that fall within the area of uncertainty of the sediment sample are

(Table 4.1):

Sample ARA solutions (0) Mean grain size - grab sample (O)
20028 A 1.9 ±0.6 (1a) 2.2 ±0.6 (1a)
20064 A 2.5 ±0.0 (1a) 2.3 ±0.6 (1a)
L1200 A 2.1 ±0.2 (1a) 2.3 ±0.6 (1a)
L2400 A 1.9 ± 1.4 (1a) 2.2 ±0.6 (1a)

Table 4.1 - ARA solutions calculated as the average within the area of uncertainty of the grab
sample and correspondent mean grain size determined from sediment analysis.

For these four samples, mean ARA solutions are within a range of 0.6 O and only

0.2 to 0.3 O apart from the correspondent value determined by grab sampling. These

values suggest that the entire workflow, from data acquisition by SAIC with the Reson

8101 multibeam system to ARA in Geocoder, yields consistent results within ±0.3 O for

the same type of seafloor. Considering all the samples in Figure 4.28, after correcting for

horizontal and vertical heterogeneities, the average difference between ARA solutions

and mean grain size determined by grab sampling is ±0.4 <t> (1a).

73

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
 CHAPTER 5

CONCLUSIONS

Before confidence can be acquired for a remote sensing tool, to the point where

the need for ground truth is minimized or eventually eliminated, a comparison has to be

made between remote sensing results and ground truth data in different types of

environments. The present study tried to estimate the applicability of the Angular Range

Analysis implemented in Geocoder (Fonseca and Mayer, 2007) to an area of high spatial

variability (the Historical Area Remediation Site - HARS).

The parameter chosen as the object of comparison was mean grain size

because it is the most commonly measured seafloor property, widely used to define

dredging areas for beach nourishment, to predict sediment transport, seafloor

geotechnical behavior, and to define benthic habitats. Although it may not yield the

desirable predictive relationship with other physical and geoacoustic properties of the

sediment, it can be easily measured with low cost. However, any measurement is

affected by an uncertainty and, before assuming mean grain size as determined by

common methods of ground truth and grain size analysis as an absolute value, we have

to determine the sources of uncertainty in the measurement and when possible, quantify

them.

The major source of uncertainty in the determination of mean grain size by grab

sampling derives from the peculiarities of the HARS and sampling efficiency. In a natural

environment on the continental shelf it is expected that the major sediment types have a

horizontal continuity on the order of decameters to kilometers and a vertical continuity on

74

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
 the order of decimeters to meters. The anthropogenic actions at the HARS shaped the

seafloor and variations are at a meter/decameter level horizontally and vertically at a

centimeter level. The grab sampler was not precisely positioned and position uncertainty

is on the order of decameters. Penetration of the grab sampler usually follows a trend

with sediment type but has variations in the order of centimeters. This brings up two

issues. Where is the sample from? And what is being sampled?

Uncertainties in sampling position on a small boat using DGPS were found to

vary between ± 15 m to ± 25 m in the HARS, and in several cases the area of

uncertainty falls between different deposits. In spite of not reflecting the true normalized

average backscatter strength for each pixel due to the application of an AVG, the

backscatter mosaic is suitable for the identification of different deposit types and even to

recognize subtle variations. Using ground truth information from contiguous areas that

are homogeneous within the uncertainty area of sediment samples it was possible, to a

certain extent, to execute a supervised positioning of samples that fall between deposits.

Another type of uncertainty due to sediment sampling arises when the grab is not

able to capture the full horizontal and/or vertical range of grain sizes. This particular type

of uncertainty would be minimized with the use of larger and heavier grabs.

It was found that instrumental uncertainties associated with an automated

method of grain size analysis, such as laser diffraction, appear to be negligible. Although

there were not enough analyses to make statistically meaningful conclusions, it was

observed that the procedure of drying the sample, in order to calculate its weight and

split it between different analytical methods, increased the mean grain size by 0.01 to

0.05 0 for sand, 0.1 0 for coarse silt, 0.2 0 for fine silt and 0.4 0 for very fine silt/clay

sizes. Determination of mean grain size by sieve-laser diffraction and sieve-pipette

methods was on average higher 0.35 0 ± 0.14 0 (1a) for the sieve-pipette method.

Although it seems that uncertainties associated with the instruments, procedures and

75

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
 methods are relatively low, it must be considered that mean grain size determined by the

Folk and Ward (1957) graphical method is much less affected than the full distribution

curve.

The major uncertainty in the determination of the size of the particles in the

sediment seems to be caused by the common analytical procedure of decomposing the

sample into its elementary particles. The standard analytical procedure is reasonable

from a sedimentary perspective but causes an underestimation of the mean particle size

of the sediment of at least 1.5 <1> for fine silts to clays, 0.5 <t> for coarse to medium silts

and 0.3 <t>for very fine sands.

How representative a grab sample is of the surficial facies a region is directly

associated with the spatial variability of the seafloor under investigation. Although good

results can be expected in many natural environments, this is not the case for the HARS.

Comparison of results with remote seafloor characterization proved to be difficult, but

possible only because of profile images and the full grain-size distributions, which

suggests that unless the medium is relatively homogeneous, grab samples by

themselves, and grain size expressed in terms of major fractions do not have much

value in describing the surficial sediment.

Remote seafloor characterization in terms of mean grain size using Angular

Range Analysis implemented in Geocoder (Fonseca and Mayer, 2007) showed good

results at a regional scale and the limitations associated with the resolution of the

method in normal mode (stack of pings x half-swath) do not affect the general regional

trend. At a local level, however, and given the heterogeneity of the HARS, the lack of

spatial resolution is evident. When there is an homogeneous patch of the seafloor where

several ARA solutions fall, results are in agreement with the measured mean grain size

provided the sediment is roughly homogeneous over the first centimeters of depth.

76

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
 When there is horizontal and/or vertical heterogeneity, the ARA’s solutions deviate

considerably from the measured mean grain size.

When the sediment is vertically homogeneous over the first centimeters but

varies horizontally within the half-swath width of the sonar, it was found that segmenting

the backscatter mosaic to define regions of the seafloor with similar properties yields

good results. The definition of different “themes” and calculation of the average angular

response for each theme, gave results similar to the ones determined by grab sampling.

Vertical heterogeneity in the HARS (here considered as layering between

sediment types with marked differences) consists of two main types; an upper layer of

red clay (approximately 1 cm) overlying sand, and a few centimeters of sand (2 to 5 cm)

overlying medium to fine silt. In both cases the angular response of the sediment is

characteristic of sand. This suggests that, in the first case the sound penetrated in the

clay and was mainly scattered in the sandy layer and that, in the second case, scattering

occurred mostly in the sandy layer without significant penetration and scattering in the

underneath silt. When the mean grain size for the sandy layer is inferred from the mode

of the grain size distribution, results agree relatively well with the mean grain size

predicted by the analysis of the sediment angular response.

In all these cases, 1) vertical and horizontal homogeneity; 2) horizontal

heterogeneity with determination of the angular response in “theme mode”; 3) vertical

heterogeneity with approximation of the mean grain size by the mode corresponding to

the top sandy layer; ARA solutions are within a maximum range of ± 0.9 cp of the mean

grain size determined by ground truth, with an average difference of ±0.4 <P (1o) and a

coefficient of determination R2 = 0.90. However, these values also reflect the uncertainty

in the determination of mean grain size by grab sampling and sediment analysis.

In order to better evaluate the consistency of ARA predictions and also of the

backscatter mosaics assembled in Geocoder, acoustic remote sensing results were

77

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
 compared with four sediment samples taken in different places but corresponding to the

same sediment. The areas that were sampled are horizontally and vertically

homogeneous. ARA predictions for the mean grain size are within ±0.3 0 and the

backscatter averaged over the area of uncertainty of the grab sample is within ±1 dB.

Angular Range Analysis proved to be an effective remote sensing tool at regional scale

in normal mode and at a local scale, when the seafloor is heterogeneous, in “theme”

mode.

Mosaics assembled by Geocoder, which were corrected for geometric and

radiometric distortions, allowed the observation of the temporal evolution of the seafloor.

Changes in backscatter are collocated with variations in bathymetry, mainly

corresponding to the deposition of new remediation material. Between two different

years, the same workflow, from multibeam data acquisition to processing in Geocoder,

showed a bias of 1 dB, and some noise in the order of ±10 dB concentrated near nadir

that did not affect the interpretation.

Special care has to be taken when trying to validate remote sensing

measurements, especially when a grab sample might be the only possible ground truth

method; results should not be assumed a priori as being true. One possibility to

overcome this limitation is to execute the acoustic survey first, build the backscatter

mosaic in the field and choose homogeneous areas of the seafloor within the predicted

radius of uncertainty of the grab sample. To ensure that samples represent vertically

what is being sampled by the acoustic wave, one possibility is to estimate the degree of

penetration of the acoustic wave, collect a relatively undisturbed sediment sample with a

larger grab, like a Smith-Mclntyre, and remove a core from the center of the grab.

The idea that “The most reliable method to obtain information on the ocean

sediment grain size is the gathering of bottom samples followed by a laboratory grain

78

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
 size analysis.” (van Walree and Ainslie, 2006, p. 2555) does not seem to be particularly

valid in the case of the HARS and probably in other regions of high spatial variability.

Future Work

A definitive test for the Angular Range Analysis, or for any acoustic remote

sensing tool, will probably never be completed unless all types of seafloor and all

possible variations are covered. Further comparisons of results should focus on common

oceanic environments with low spatial variability and/or on areas of high spatial

variability but using box corers with acoustic positioning. It would also be interesting to

see the effect of layering at different scales and frequencies in the angular response of

the seafloor.

The application of the ARA in “theme” mode yielded good results for areas of

high spatial heterogeneity but it is not feasible to manually segment regions of similar

properties in large mosaics. An automated segmentation method usually requires a priori

definition of the number of classes but for large areas or for areas with gradual

variations, that is almost impossible. It might be possible to overcome this limitation by

dividing the area of study into smaller regions, and using the backscatter mosaic and

bathymetric data, define statistically different populations for each region and correlate

populations from different regions to segment the seafloor into different “themes", each

one corresponding to an area of the seafloor with similar properties.

79

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
 APPENDICES

80

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
 APPENDIX A

PENETRATION OF THE SOUND INTO THE SEDIMENT

An estimation of how much the sound is attenuated in the first layer can be made

by considering regression equations between mean grain size and k (dB m'1 kHz'1), a

constant used to calculate the attenuation a expressed in dB/m, defined by Hamilton

(1972) as:

a = k fn (A.1)

where / is the frequency in kHz and n the exponent of the frequency.

0.8

0.7

0.6
0.5

0.4

0.3

0.2

0.1

0
0 1 2 3 4 5 6 7 8 9 10

Figure A.1 - Mean grain size versus k (adapted from Hamilton, 1972).

The path traveled by the sound inside the sediment can be estimated considering

Snell’s law for a grazing angle 0,:

cos dx _ cos 02
(A.2)
C\ C2

81

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
 where cx is the sound speed in the water (1500 m/s) and c2 the sound speed in the

sediment (Fig. A.2),

Figure A.2 - Refraction of the ray that describes the incident sound wave with 45° grazing angle.

and the relation between sound speed in the sediment and mean grain size referenced

by Hamilton and Bachman (1982):

1 1750
£ 1700

Figure A.3 - Mean grain size versus compressional sound speed in the sediment (adapted from
Hamilton and Bachman, 1982).

The attenuation in dB per each cm of layer thickness, for 45° grazing angle and

the 240 kHz frequency used in this study, considering « * 1 , becomes:

3.5

240 kHz
45° grazing angle

Mean grain size (phi)

Figure A.4 - Mean grain size versus attenuation per each cm thickness of the first layer for a
grazing angle of 45° and a frequency of 240 kHz.

82

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
 Given the attenuation of sound in the sediment, it is possible to roughly estimate

the maximum penetration of the acoustic wave making some assumptions and

considering the sonar equation for an active sonar (Urick, 1983):

SL - 2TL + TS = NL - D l + D T (A.4)

The source level (SL) is specified as 210 dB re 1 pPa@1 m (manufacturer’s data).

Calculation of the transmission loss (2TL) considers spherical spreading and absorption,

for scattering on a rough bottom it is:

2TL = 40 log R + 2aR (A. 5)

where R is the range to the target, considered here as 35.4 m at a depth of 25 m with

45° grazing angle, and a the absorption coefficient with a value of 0.55 dB/m for 240kHz

(Kinsler etal., 1999), using these values, 2TL = 101dB.

The noise level (NL) for 240 kHz is mainly due to the thermal noise effect (Urick,

1983) and is calculated by:

NL = N 0 +lO log(Bw ) (A.6)

N0 = 33 dB re 1 pPa/Hz1/2 (Urick, 1983), and the receiver bandwidth (Bw) can be

approximated by Bw = 0.88/ r , where r is the pulse length, equal to 75 ps in our case,

so Bw = 11.7 kHz and NL = 73 dB.

Not having information about the detection threshold (DT) and the receive

directivity index (Dl), it is possible to consider Dl - DT as the “ increase in signal-to-

background ratio produced by the entire receiving system” (Urick, 1983, p. 21), here

assumed to be equal to 5 dB.

The seafloor is the target, and the acoustic backscaitering strength BS is (Lurton,

2002 ):

5 S = Ss +101og04) (A.7)

83

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
 where A is the area ensonified by a sound pulse with length t , with beam width (p, at a

distance R and at a grazing angle 0, and with a speed in the water c:

C T
A = <pR- (A. 8)
2 sin 0

and SB is the angular dependant acoustic backscattering strength that can be

approximated considering the graph in Figure A.5, as - 21 dB for a medium sand. So

10log(A) = -10 dB and BS = -31 dB.

240 kHz

Grazing angle (degrees)

Figure A.5 - SBversus grazing angle calculated using the implementation of Jackson et al. (1986)
model in Geocoders’s ARA with default values of input parameters. Each colored line
corresponds to a different mean grain size, expressed in O units.

Considering the sonar equation and the assumptions made, an estimate can be

made on how much loss is allowed for bottom penetration (BL):

DT - Dl + NL = SL - 2TL - BL + BS (A.9)

which results in 10 dB for bottom loss in medium sands. This value corresponds to a

penetration of approximately 2 cm at 45° grazing angle (considering an attenuation of

2.3 dB/cm for medium sand at 240 kHz in Figure A.4, and two-way travel of sound in the

sediment). Given all the assumptions made, it is expected that the calculations are

subject to large errors but the order of magnitude of the results is consistent with the

84

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
 observations. Figure A.6 depicts results of the calculations for different types of

sediments.

5
E
o
co <
3

*. i7
Q

0
0 1 2 3 4 5 6 7 8

M ean g ra in size (phi)

Figure A.6 - Rough estimation of acoustic wave penetration into the sediment for different mean
grain sizes at 45° grazing angle and 240 kHz. The penetration is estimated to vary between 1 and
10 wavelengths depending on the type of sediment.

85

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
 APPENDIX B

BACKSCATTER MOSAICS APPLYING DIFFERENT AVGs

Several mosaics were assembled in order to compare the influence of applying

an Angular Varying Gain to normalize the acoustic response across the swath. Three

AVGs from Geocoder were selected for this comparison: “Trend” , “ Flat” and

“ Lambert” .

In the case of the “Trend” and “ Flat" AVG corrections, the AVG for each ping is

constructed from the average angular response calculated over 300 pings centered on

the ping. In the “ Flat” AVG correction, the reference level is the mean of the two

average backscatter values calculated for each side of the sonar, between 35 and 55°

grazing angles. In the “Trend” AVG correction, the reference level is the trend line

between the average backscatter values (calculated between 35 and 55° grazing

angles) of each side of the sonar. The “ Lambert” AVG correction considers an angular

variation of the backscatter following Lambert’s law (BS(0) = BS0 + 20log (cos0), where

0 is the incident angle and BS0 the average backscatter between 35 and 55° grazing

angles).

The choice of which AVG curve to apply is subjective. Figure B.1 depicts the

backscatter mosaic without any AVG correction, the interpretation of this image is

relatively difficult due to high backscatter near nadir (the result of coherent reflection),

and a rapid drop in backscatter beyond the critical angle. Figures B.2, B.3 and B.5 depict

the backscatter mosaic with the AVG “Trend” , “ Flat” and “ Lambert” corrections,

respectively. The standard Lambertian correction is not effective in removing the effect

86

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
 of the seafloor angular response. The AVG “Trend” and “ Flat” corrections produce a

backscatter mosaic with a smooth response across the sonar swath.

594000 595000 596000 597000 598000 599000 600000

till

I
91

R
i'll
I

596000
19 599000 600000

2006 WJout AVG

dB
. _ ^ High : -0 65

U TM Projection Zone 1SN

N AD83 Datum

Figure B.1 - 2006 backscatter mosaic assembled with Geocoder without any AVG. The angular
response of the sediment makes interpretation of the backscatter mosaic relatively difficult.

87

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
 594000 595000 596000 597000 598000 599000 600000
1-------------------------- 1-------------------------- 1-------------------------- 1_______________ I_______________ I_______________ L _

4476000

4476000
4475000

4475000
4474000

4474000
4473000

4473000
4472000

4472000
4471000

4471000
4470000

4470000
4469000

4469000
4468000

4468000

— I-------------------------- 1---------------------------1-------------------------- 1---------------------------1-------------------------- 1-------------------------- 1—

594000 595000 596000 597000 598000 599000 600000

2006 AVG Trend

dB
High: -2.17

■ Low: - 53.7
UTM Projection Zone 18N

N AD83 Datum

Figure B.2 - 2006 backscatter mosaic assembled with Geocoder with AVG Trend.

88

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
 594000 595000 595000 597000 598000 599000 600000
1-----------------------------I___ I
------ I____ I
____ 1 1____

4476000

4476000
4476000

4475000
4474000

4474000
4473000

4473000
4472000

4472000
4471000

4471000
4470000

4470000
4469000

4469000
't ih - i 4 0
4468000

4468000

M i* £ s i* r
I I I
598000 599000 600000

2005 AVG Flat

dB
W
SmHigh:-2.17
i

M L o w : -53.92
U TM Projection Zone 18N
NAD 83 Datum

Figure B.3 - 2006 backscatter mosaic assembled with Geocoder with AVG Flat.

89

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
 594000 595000 596000 597000 598000 599000 600000
I __ I___ 1
------ I____ I____ ---
1 I
____

4476000
<0 -
r-
5

4475000
H
iV

4474000
4473000
u

4472000
111! Ill

4471000
4470000
200$ AVG T r*n d -F la t
dB := u

4469000
■ ■ - 5 3 --3
— -t9~f.8 : ' . K
( i -1 * - -0.5
ri-0.«-Q6
jT O f j a e -1 .5 ; U[

B ill -e 4468000
M B1-9
m ai -12 , ;

Mean = 0 .1 2 dB UTM Projection Zone 18N

Standard deviation = 0 .52 dB N AD83 Datum

Figure B.4 - Difference between AVG Trend and AVG Flat for the 2006 backscatter mosaic. The
difference can reach more than 10 dB.

90

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
 594000 595000 596000 598000 599000
J ---------------------------- 1 800000

4476000
ISlil
I
4475000
4474000
4473000

56
4472000
4471000

K
4470000
4469000

598000 599000

2006 AVG Lambert

dB
High :-0.43

I Low: -53.27 U TM Projection Zone 16N

N AD83 Datum

Figure B.5 - 2006 backscatter mosaic assembled with Geocoder with a Lambertian correction.

91

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
 594000 595000 596000 597000 598000 599000 600000
I

4476000
I

4475000
' ! ti i

4474000
“!s ■j|i I
i
h:r*

4473000
"-In: I: 'u vUi
— I

4472000
W NJ- jI

4471000
4470000
4469000
tilm
.J f - .i,i •*. ■?;! m ■>!!.., j i. • 1 I
i ~i------------------------r
596000 598000 599000 600000

^lean, = "4 -45 dB UTM Projection Zone 18N

Standard deviation = 2 .29 dB NAD 83 Datum

Figure B.6 - Difference between AVG Trend and a Lambertian correction for the 2006 backscatter
mosaic.

92

Reproduced with permission o f the copyright owner. Further reproduction prohibited without permission.
 APPENDIX C

PROCEDURES FOR GRAIN SIZE ANALYSIS

C.1 Normal procedure for grain size analysis

C. 1.1 Removal of organic material

Each sediment sample was placed in a large beaker, covered with 10 ml of 30%

hydrogen peroxide and stirred. Deionized water was added to remove any sample

material from the beaker walls and to prevent foaming over the top of the beaker. The

sample was left to digest for 2 to 5 days until the foaming stopped, then another 10 ml of

hydrogen peroxide were added. The procedure continued for about two weeks until all

the organic material was oxidized, or until a total 40 ml of hydrogen peroxide had been

added and no significant bubbling was observed.

C. 1.2 Washing salts from the sample

Deionized water was used to transfer the sample from the beaker to centrifuge

bottles; each bottle was filled with deionized water up to 80%, closed and thoroughly

shaken. Centrifuging took place for 30 min and the clear supernatant water was

decanted.

C.1.3 Wet sieving at 500 pm

A 500 pm sieve was placed over a funnel and a large pre-weighed beaker to

collect the finer fraction passing through the sieve. The sample was transferred from

93

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
 the centrifuge bottle to the sieve with the aid of a squeeze bottle with deionized water.

The sediment was rinsed with deionized water and stirred on the sieve with a “rubber

policeman” until no small grains were observed on the sieve and the water passing

through the sieve remained clear. Care was taken to limit the volume of deionized water

plus finer sediment to the size of the beaker. The sieve was turned upside down over a

glass bowl and rinsed with deionized water to collect any coarse sand. The excess water

was decanted and the sediment rinsed into a small pre-weighted beaker. Both beakers,

with sediment fractions larger and smaller than 500 pm were dried at 55 °C.

After all the water evaporated (usually after 3 days for the large beakers with the

finer fraction), the beakers were left to cool to room temperature in a desiccator and

reweighed. The total weight of the sample is the sum of the two fractions.

C.1.4 Dry sieving

The fraction coarser than 500 pm was disaggregated and dry sieved in 0.5 O

intervals for 15 min on a sieve shaker. The content of each sieve was transferred to a

pre-weighed aluminum dish and reweighed. The fraction left in the pan was also

weighed and added to the sediment to be analyzed with laser diffraction.

C.1.5 Laser diffraction (Malvern Mastersizer 2000 with the dispersion unit Hydro G)

Dried samples smaller than 500 pm were transferred to plastic jars and a solution

of 3 g/l sodium hexametaphosphate was added until the dried sample became a paste.

Samples were thoroughly mixed until homogeneous and shipped to the Portuguese

Hydrographic Institute for analysis. A small portion of this homogeneous paste was

removed and put in the tank of the dispersion unit of the laser diffraction system until the

obscuration level settled between 10 and 20 %. Ultrasound was applied to the sample

for 120 s and stopped immediately prior to analysis. Three measurements were made,

94

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
 each with 12 s integration time. Results are reported as the average of the three

measurements.

Configuration values used in the Malvern Mastersizer 2000 were 1.544 for

particle refraction index (quartz), 1.330 for dispersant refraction index and 1.5 for particle

absorption.

C.2 Measurement of the natural occurring particle size

In order to estimate the particle size as close as possible to the original sediment,

samples underwent minimal processing. The sample was not digested with hydrogen

peroxide and was not washed and centrifuged to remove salts, but was wet sieved at

500 pm to prevent clogging the pump of the laser-diffraction system.

The fraction coarser than 500 pm was discarded. The fraction finer than 500 pm

was collected in the beaker and left to rest for more than a week, then partially decanted,

left to settle again for a few more days and the remaining supernatant liquid removed

with a pipette.

Each sample was analyzed with the laser-diffraction system without adding any

dispersant and without ultrasound. The sample was only stirred in order to keep grains in

suspension.

Comparison with the samples that followed the normal SOP was made

considering only the portion that was analyzed in the laser diffraction. There is one

difference, for the samples that followed the normal SOP the sediment left in the pan

when sieving was added to the portion analyzed with laser diffraction. It is assumed that

the effect of adding dried sediment is worse than the errors introduced if the fraction left

in the pan doesn’t follow the same distribution as the entire fraction to be analyzed with

laser diffraction. In order to minimize errors the comparison was made only for those

95

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
 samples analyzed with the normal SOP with material left in the pan less than 1% by

weight of the fraction finer than 1 <t>.

The samples considered for this experiment have on average 97 % of their

weight distributed below the 1 O size class.

C.3 Measurement of the effect of drying the sample

Drying the sample is part of the standard procedure used at the Portuguese

Hydrographic Institute in order to determine the weight of the two portions to be

analyzed with sieving and laser diffraction. It is assumed that the sample analyzed with

laser diffraction will be separated in its constituent particles when dispersing the sample

with sodium hexametaphosphate, applying ultrasound and stirring. To evaluate the error

introduced by this method, some samples were analyzed with the standard procedure

but without drying for weight determination. The comparison was made only for the

portion analyzed with laser diffraction with the same assumptions as in C.2.

C.4 Pipette analysis

A subset of samples was analyzed with pipettes to compare results obtained with

two methods that measure different properties of the sediment. This technique used the

following steps:

C.4.1 Removal of organic material

Each sediment sample was placed in a large beaker, covered with 10 ml of 30%

hydrogen peroxide and stirred. Deionized water was added to remove any sample

material from the beaker walls and to prevent foaming over the top of the beaker. The

96

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
 sample was left to digest for 2 to 5 days until the foaming stopped, then another 10 ml of

hydrogen peroxide were added. The procedure continued for about two weeks until all

the organic material was oxidized, or until a total 40 ml of hydrogen peroxide had been

added and no significant bubbling was observed.

C.4.2 Washing salts from the sample

Deionized water was used to transfer the sample from the beaker to centrifuge

bottles; each bottle was filled with deionized water up to 80%, closed and thoroughly

shaken. Centrifuging took place for 30 min and the clear supernatant water was

decanted.

C.4.3 Dispersing the sample

The sample was placed in a 1000 ml beaker and filled up to 66% with dispersant,

a solution of 2.18 g/l sodium hexametaphosphate in deionized water, stirred until

dispersed and left to sit for one day. After verification that flocculation did not occur, the

sample was ready for wet sieving.

C.4.4 Wet sieving

After rinsing a 1000 ml graduated cylinder, the funnel, a 2mm sieve and a 62 pm

sieve (rinsing the 62 pm sieve is also important to reduce surface tension on the sieve’s

mesh), the dispersed sediment was poured over the 2 mmm sieve and rinsed with

dispersant in the squeeze bottle and with the aid of a “rubber policeman” . The sediment

that passed through the 2 mm sieve was removed and the gravel fraction was rinsed into

a glass bowl using deionized water and then to a pre-weighted beaker. Rinsing

continued in the 62 pm sieve until the dispersant passing through the sieve was clear.

The sieve was removed and the sand fraction was transferred to a glass bowl and then

97

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
 to a pre-weighted beaker with deionized water. Both beakers were placed to dry at

55 °C.

The volume of dispersant plus mud was limited to less than 1000 ml. The

graduated cylinder was topped up to 1000 ml with dispersant.

C.4.5 Dry sieving

The gravel and sand fractions were weighted after they were completely dried

and cooled to room temperature. The sand and gravel fraction were sieved together for

15 min in 0.5 O intervals, the content of each sieve was placed on a pre-weighed

aluminum dish and reweighed and the portion remaining in the pan was also weighed

and added to the mud fraction to be analyzed with the pipettes.

C.4.6 Pipettes

Each sample in the graduated cylinder was thoroughly mixed during 15 min

less than 0.5 hr from the start of the analysis and stirred with a rod for the last 2 min. The

first aliquot was removed with a 20 ml pipette 20 cm below the surface, 20 s after stirring

stopped, and expelled into a pre-weighted beaker. An additional 20 ml of deionized

water was used to rinse out the remaining sediment and expelled into the same beaker.

The beaker was placed to dry at 55° C in an oven for one day, cooled in a desiccator to

room temperature and reweighed. The first withdraw is a representative sample of the

entire sediment present in the graduated cylinder. Additional withdraws were taken at

the times and depths presented in table B.1, each sequential withdrawal sample all the

sediment at 1 O finer intervals.

98

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
 Time (h:mm:ss) Depth (cm) Fraction
0:00:20 20 <4 cp
0:01:56 10 <5 0
0:07:44 10 <6 0
0:31:00 10 <7 0
2:03:00 10 <8 0
8:01:00 10 <9 0
23:00:00 7 <100

Table C.1 - Withdrawal times and depths used for 20° C water temperature.

Samples were dried, weighed and corrected for dispersant weight and related

to the volume of the suspension from which the aliquot was drawn. The last aliquot

represents sediment finer than 10 O and was evenly divided between 10 O and 14 O.

C.5 Loss On Ignition (LOI)

Ten to 20 g of sample were placed in a pre-weighted aluminum dish and

weighed. The sample was dried in an oven at 55° C for 24 hr, removed and cooled in a

desiccator to ambient temperature for one hour, then weighed again. The difference

between the dry weight and the initial sample weight is a function of the percentage of

moisture in the sample. The sample was heated in a muffle furnace during 4 hr at

450 °C, removed and cooled in a desiccator to ambient temperature and reweighed. The

weight difference corresponds to the loss on ignition, assumed to be organic matter.

99

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
 APPENDIX D

EXAMPLES OF ANGULAR RESPONSE

The variation of backscatter strength with grazing angle is an inherent property of

the seafloor. Figure D.1 depicts the acoustic backscatter angular response for gravel

and silt. Figure D.2 shows the angular response for medium sand in two different areas

of the HARS. The sedimentary cover in one station is layered in the top few centimeters

whereas in the other it is homogeneous up to the penetration depth of the optical prism.

A perfectly flat seafloor would behave as a partially reflecting acoustic mirror,

reflecting sound at an angle equal to the angle of the incident sound, and no energy

would return to the transducer except at normal incidence. However, at high frequencies,

the seafloor has substantial irregularities at the scale of the acoustic wavelength and

acoustic waves are scattered randomly. These irregularities include the roughness of the

water-sediment interface, spatial variations in sediment physical properties and discrete

inclusions such as shell pieces or bubbles (Jackson and Richardson, 2007).

Silts and clays usually have a smooth surface compared with the acoustic

wavelength (A = 6 mm for 240 kHz) and backscatter near normal incidence is mainly

dominated by coherent reflection. These sediments also have high porosity, and the

acoustic impedance contrast between the sediment and the water is low. At oblique

incidence, backscattering due to surface roughness may not be important due to the lack

of impedance contrast and to the small surface roughness, the sound penetrates into the

sediment more than in sands and gravels and scattering due to heterogeneities that

exist within the sediment is expected to be stronger than scattering due to surface

100

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
 roughness. In a gravelly seafloor, the impedance contrast between the water and the

sediment is high and the water-sediment interface is rough; therefore, backscatter is

dominated by scattering due to surface roughness at all incident angles. Beyond the

critical angle there is a sudden drop in backscatter since almost no energy is transmitted

into the sediment. A part of the small amount of energy that is scattered back beyond the

critical angle comes from the surface whereas the other part may be due to some

penetration into the sediment and scattering by volume heterogeneities within the

sediment. Energy may penetrate into the sediment beyond the critical angle due to

downward scattering at the sediment surface and/or to the Biot slow wave, which does

not have a critical angle (Pouliquen et al., 2000).

101

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
 594000 595000 596000

Model
Angular response: gravel

Grazing angle (degree*)

Model
'Angularresponse: medium silt

Grazing angle (degrees)

594000 595000 596000

CZj 1997 Category II Cap Boundary 2006 BaCkSCatter Strength

£ j 1993 Dioxin Cap Boundary .p

C I . 1 Former MDS M ig h i-Z U

rn h a rs p ra . ^ * 2006 Grab Samples UTM Projection Zone 18N
i HARS Buffer Zone B oundary Low ' -53 7 ^ 2006 SPI NAD83 Datum

Figure D.1 - Examples of the angular response in the area of stations A2 (gravel) and N2000
(silt) from 2006. SPI images correspond to 20 x 13 cm (HxW) and sediment sample photos to 2.2
x 3.3 cm (HxW). The angular response, depicted in red, was obtained in normal mode (half-swath
with x 30 pings) for one patch of the seafloor near each station. The blue line is the adjustment of
the model to the observations.

102

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
 595000 596000 597000 598000

Model
-Angular response: medium sand

Model
Angular response: medium sand

595000 596000 597000 598000

C Z j 1997 Category ii cap Boundary 2 0 0 6 B a c k s c a tte r s tre n g th
(j 11993 Dioxin Cap Boundary
| | RedClay
C f ] Former MOS Hi9h > 2' 17
□ hars PRAs fe M * 2006 Grab Samples UTM Projection Zone 18N
: HARS Buffer Zone B oundary Low ' -53 7 ® 2006 SPI N AD83 Datum

Figure D.2 - Examples of the angular response in the area of stations A9 and 97004 (2006),
corresponding to a medium sand seafloor. The grab sample in station A9 captured also the
underneath silty layer. SPI images correspond to 20 x 13 cm (HxW) and sediment sample photos
to 2.2 x 3.3 cm (HxW). The angular response, depicted in red, was obtained in normal mode
(half-swath with x 30 pings) for one patch of the seafloor near each station. The blue line is the
adjustment of the model to the observations.

103

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
 REFERENCES

Agrawal, Y. C., McCave, I. N. and Riley, J. B., 1991, Laser diffraction size analysis, in:
Syvitski, J. P. M., (Ed.), Principles, methods and application of particle size analysis,
Cambridge University Press, p. 119-128.

Beaudoin, J. D., Hughes Clarke, J. E., Van Den Ameele, E. J. and Gardner, J. V., 2002,
Geometric and radiometric correction of multibeam backscatter derived from Reson
8101 systems, Canadian Hydrographic Conference Proceedings, p.1-22.

Biot, M. A., 1956, Theory of propagation of elastic waves in a fluid-saturated porous

solid. II. Higher frequency range, The Journal of the Acoustical Society of America, Vol.
28, No. 2, p. 179-191.

Biot, M. A., 1962, Generalized theory of acoustic propagation in porous dissipative

media, The Journal of the Acoustical Society of America, Vol. 44, No. 9, p. 1254-1264.

Blott, S. J. and Pye, K., 2001, GRADISTAT: A grain size distribution and statistics
package for the analysis of unconsolidated sediments, Earth Surface Processes and
Landforms, Vol. 26, p. 1237-1248.

Butman, B., Danforth, W. W., Knowles, S. C., May, B. and Serret, L., 2002, Sea floor
topography and backscatter intensity of the Historic Area Remediation Site (HARS),
offshore of New York, based on multibeam surveys conducted in 1996, 1998 and 2000,
U.S. Geological Survey Open File Report 00-503, 5 sheets.

Butman, B., Danforth, W. W., Schwab, W. C. and ten Brink, M. B., 1998, Multibeam
Bathymetric and Backscatter Maps of the Upper Hudson Shelf Valley and Adjacent
Shelf, Offshore of New York, U.S. Geological Survey Open File Report 98-616, 4 sheets.

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 backscatter with grain size
distribution of surficial seabed sediments, Marine Geology, Vol. 214, p. 431-439.

104

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
 Collins, W., Gregory, R. and Anderson, J., 1996, A digital approach to seabed
classification, Sea Technology, August 1996, p. 83-87.Damuth, J. E., 1975, Echo
character of the western equatorial Atlantic floor and its relationship to the dispersal and
distribution of terrigenous sediments, Marine Geology, Vol. 18, p. 17-45.

Damuth, J. E., 1980, Use of high-frequency (3.5-12 kHz) echograms in the study of
near-bottom sedimentation processes in the deep-sea: a review, Marine Geology, Vol.
38, p. 51-75.

Dartnell, P. and Gardner, J. V., 2004, Predicting seafloor facies from multibeam
bathymetry and backscatter data, Photogrammetric Engineering & Remote Sensing, Vol.
70, No.9, p. 1081-1091.

Dunsiger, A. D., Norman, A. C. and Vetter, W. J., 1981, Seabed characterization from
broad-band acoustic echosounding with scattering models, IEEE Transactions of
Oceanic Engineering, Vol. OE-6, No. 3, p. 94-106.

Environmental Protection Agency (EPA), 1996, New York-New Jersey Harbor Estuary
Program, Final Comprehensive Conservation and Management Plan (CCMP), 280 p.

Folk, R. L., 1980, Petrology of sedimentary rocks, Hemphill Publishing Company, TX,
184 p.

Folk, R. L. and Ward, W. C., 1957, Brazos River bar: a study in the significance of grain
size parameters. Journal of Sedimentary Petrology, Vol. 27, p. 3-26.

Fonseca, L., and Calder, B., 2005, Geocoder: an efficient backscatter map constructor,
Proceedings of the U.S. Hydrographic Conference 2005, San Diego, CA.

Fonseca, L. and Calder, B., 2007, Clustering acoustic backscatter in the angular
response space, Proceedings of the U.S. Hydrographic Conference 2007, Norfolk, VA.

Fonseca, L. and Mayer, L., 2007, Remote estimation of surficial seafloor properties
through the application of angular range analysis to multibeam sonar data, Marine
Geophysical Researches, Vol.28, No. 2, p. 119-126.

Friedman, G. M. and Sanders, J. E., 1978, Principles of sedimentology, John Wiley &
Sons, NY, 792 p.

105

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
 Greaves, R. J., Stephen, R. A., 2000, Low-grazing angle monostatic acoustic
reverberation from rough and heterogeneous seafloors, The Journal of the Acoustic
Society of America, Vol. 108, No. 3, p. 1013-1025.

Hamilton. E. L., 1970, Reflection coefficients and bottom losses at normal incidence
computed from pacific sediment properties, Geophysics, Vol. 35, No. 6 (December
1970), p. 995-1004.

Hamilton, E. L., 1972, Compressional-wave attenuation in marine sediments,

Geophysics, Vol. 37, No. 4 (August 1972), p. 620-646.

Hamilton, E. L. 1974, Prediction of deep sea properties: state-of-the-art, in: Inderbitzen,

A. L. (Ed.), Deep-sea sediments, physical and mechanical properties, Plenum Press,
p. 1-43.

Hamilton, E. L., 1976, Sound attenuation as a function of depth in the sea floor, The
Journal of the Acoustical Society of America, Vol. 59, No. 3, p. 528-535.

Hamilton, E. L., 1978, Sound velocity-density relations in sea-floor sediments and rocks,
The Journal of the Acoustical Society of America, Vol. 63, No. 2, p. 366-377.

Hamilton, E. L., Bachman, R. T., 1982, Sound velocity and related properties of marine
sediments, The Journal of the Acoustical Society of America, Vol. 72, No. 6, p. 1891-
1904.

Hamilton, E. L., Shumway, G., Menard, H. W. and Shipek, C. J., 1956, Acoustic and
other physical properties of shallow-water sediments off San Diego, The Journal of the
Acoustic Society of America, Vol. 28, No. 1, p. 1-15.

Haralick, R. M., 1979, Statistical and structural approaches to texture, Proceedings of

the IEEE, Vol. 67, No. 5, p. 786-804.

Jackson, D. R., Winebrenner, D. P and Ishimaru, A., 1986, Application of the composite
roughness model to high-frequency bottom backscattering, The Journal of the Acoustical
Society of America, Vol. 79, No. 5, p. 1410-1422.

Jackson, D. R. and Richardson, M. D., 2007, High-frequency seafloor acoustics,

Springer, 616 p.

106

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
 Lurton, X., 2002, An introduction to underwater acoustics: principles and applications,
Springer, 347 p.

Kinsler, L. E., Frey, A. R., Coppens, A. B. and Sanders, J. V., 1999, Fundamentals of
acoustics, 4th ed. , John Wiley & Sons, 548 p.

Krumbein, W. C., 1934, Size frequency distributions of sediments, Journal of

Sedimentary Petrology, Vol. 4, No. 2, p. 65-77.

Krumbein, W. C., 1936, Application of logarithmic moments to size frequency

distributions of sediments, Journal of Sedimentary Petrology, Vol. 6, No. 1, p. 35-47.

Krumbein, W. C. and Pettijohn, F. J., 1938, Manual of Sedimentary petrography,

Appleton-Century-Crofts, NY, 549 p.

LeBlanc, L. R., Shock, S. G., DeBruin, D. L., Jenkins, M. and Munro, L., 1995, High-
resolution sonar volume scattering measurements in marine sediments, The Journal of
the Acoustical Society of America, Vol. 91, No. 1, p. 107-115.

Loizeau, J. -L., Arbouille, D., Santiago, S. and Vernet, J. -P., 1994, Evaluation of a wide
range laser diffravtion grain size analyser for use with sediments, Sedimentology, Vol.
41, p. 353-361.

Mayer, L. A. and LeBlanc, L. R., 1983, The chirp sonar: a new quantitative high-
resolution profiling system, in: Pace, N. (Ed.), Acoustics and the Sea-Bed, Bath
University Press, 436 p.

Mayer, L. A., Kraft, B. J., Simpkin, P., Lavoie, P., Jabs, E., and Lynskey, E., 2002, In-situ
determination of the variability of seafloor acoustic properties: an example from the ONR
GEOCLUTTER area, in: Pace, N. G. and Jensen, F. B. (Eds.), Impact of Littoral
Environmental Variability on Acoustic Predictions and Sonar Performance, Kluwer
Academic Publishers, The Netherlands, p. 115-122.

McKinney, C. M. and Anderson, C. D., 1964, Measurements of backscattering of sound

from the ocean bottom, The Journal of the Acoustical Society of America, Vol. 36, No. 1,
p. 158-163.

McMannus, J., 1988, Grain size determination and interpretation, in: Techniques in
Sedimentology, Tucker, M. (Ed.), Blackwell, Oxford, p. 63-85.

107

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
 de Moustier, C. and Matsumoto, H, 1993, Seafloor acoustic remote sensing with
multibeam echo-sounders and bathymetric sidescan sonar systems, Marine Geophysical
Researches, Vol. 15, p. 27-42.

Murphy, L., Leary, T. and Williamson, A., 1995, Standardizing seabed classification
techniques, Sea Technology, July 1995, p. 15-19.

Orlowski, A., 1984, Application of multiple echoes energy measurements for evaluation
of sea bottom type, Oceanologia, No. 19, p. 61-78.

Pace, N. G. and Dryer, C. M., 1979, Machine classification of sedimentary sea bottoms,
IEEE Transactions on Geoscience Electronics, Vol. 17, No. 3, p. 52-56.

Pace, N. G. and Gao, H., 1988, Swathe seabed classification, Machine classification of
sedimentary sea bottoms, IEEE Journal of Ocean Engineering, Vol. 13, No. 2, p. 83-90.

Panda, S., LeBlanc, L. R. and Schock, S. G., 1994, Sediment classification based on
impedance and attenuation estimation, The Journal of the Acoustical Society of America,
Vol. 96, No. 5, p. 3022-3035.

Pouliquen, E. and Lyons, A. P., 2002, Backscattering from bioturbated sediments at very
high frequency, IEEE Journal of Ocean Engineering, Vol. 27, No. 3, July 2002, p. 388-
402.

Pouliquen, E., Lyons, A. P. and Pace, N. G., 2000, Penetration of acoustic waves into
rippled sandy seafloors, The Journal of the Acoustical Society of America, Vol. 108, No.
5, p. 2071-2081.

Reed IV, T. B. and Hussong, D., 1989, Digital image processing techniques for
enhancement and classification of SeaMARC II side scan sonar imagery, Journal of
Geophysical Research, Vol. 94., No. B6, p. 7469-7490.

Rhoads, D. C. and Cande, S., 1971, Sediment Profile Camera for in situ study of
organism-sediment relations, Limnology and Oceanography, Vol. 16, No. 1, p. 110-114.

Rhoads, D. C. and Germano, J. D., 1982, Characterization of organism-sediment

relations using sediment profile imaging: an efficient method of remote ecological
monitoring of the seafloor (Remots TM System), Marine Ecology - Progress Series, Vol.
8, p. 115-128.

108

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
 Rhoads, D. C. and Germano, J. D., 1986, Interpreting long-term changes in benthic
community structure: a new protocol, Hydrobiologia, Vol. 142, p. 291-308.

Richardson, M. D. and Briggs, K. B., 2004, Empirical predictions of seafloor properties

based on remotely measured sediment impedance, in: High Frequency Ocean Acoustic
Conference, Porter, M. B. and Siderius, M. (Eds.), AIP, Melville, NY, p. 12-21.

SAIC, 2005a, Results of the summer 2005 sediment toxicity and sediment-profile
imaging survey at the Historic Area Remediation Site, SAIC Report No. 696, 90 p.

SAIC, 2005b, Results of the summer 2005 sediment multibeam, bathymetric and
backscatter surveys at the Historic Area Remediation Site and the Shark River Reef,
SAIC Report No. 697, 51 p.

Schock, S. G., LeBlanc, L. R. and Mayer, L. A., 1989, Chirp sub-bottom profiler for
quantitative sediment analysis, Geophysics, Vol. 54, No. 4, p. 445-450.

Schwab, W. C, Denny, J. F., Butman, B., Danforth, W. W., Foster, D. S., Swift B. A.,
Lotto, L. L., Allison and M. A., Thieler, E. R., 2000, Sea floor characterization offshore of
the New York-New Jersey metropolitan area using sidescan-sonar, U. S. Geological
Survey Open-File Report 00-295, 3 sheets.

Singer, J. K., Anderson, J. B., Ledbetter, M. T., McCave, I. N, Jones, K. P. N. and

Wright, R., 1988, An assessment of analytical techniques for the size analysis of fine­
grained sediments, Journal of Sedimentary Petrology, Vol. 58, No. 3, p. 534-543.

Sperazza, M., Moore, J. N., Hendrix, M. S., 2004, High-resolution particle size analysis
of naturally occurring very fine-grained sediments through laser difractometry, Journal of
Sedimentary Research, Vol. 74, No. 5 (September 2004), p. 736-743.

Stanton, T.K. and C.S. Clay. 1986, Sonar echo statistics as a remote-sensing tool:
volume and sea floor, IEEE Journal of Oceanic Engineering, Vol. 11, p. 79-96.

Stewart, W. K., Chu, D., Malik, S., Lerner, S. and Singh, H, 1994, Quantitative seafloor
characterization using a bathymetric sidescan sonar, IEEE Journal of Ocean
Engineering, Vol. 19, No. 4, p. 599-610.

Udden, J. A., 1914, Mechanical composition of clastic sediments, Bulletin of the

Geological Society of America, Vol. 25, p. 655-744.

109

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
 Urick, R. J., 1954, The backscattering of sound from a harbor bottom, The Journal of the
Acoustical Society of America, Vol. 26, No. 2, p. 231-235.

Urick, R. J., 1983, Principles of underwater sound, 3rd ed., Peninsula Publishing, CA,
423 p.

Wadell, H., 1932, Volume, shape, and roundness of rock particles, The Journal of
Geology, Vol. 40, p. 443-451.

Wadell, H., 1934, Some new sedimentation formulas, Physics, Vol. 5, p. 281-291.
van Walree, P. A. and Ainslie, M, 2006, Mean grain size mapping with single-beam echo
sounders, The Journal of the Acoustical Society of America, Vol. 120, No. 5, p. 2555-
2566.

Wentworth, C. K., 1922, A scale grade and class terms for clastic sediments. The
Journal of Geology, Vol. 30, p. 377-392.

Wentworth, C. K., 1926, Methods of mechanical analysis of sediments: Univ. Iowa

Studies in Nat. Hist., Vol. II, No. II.

Wigley, R. L., 1966, Comparative efficiencies of Van Veen and Smith-Mclntyre grab
samplers as revealed by motion pictures. Ecology, Vol. 48, No. 1, p. 168-169.

Williams, K. L., 2001, An effective density fluid model for acoustic propagation in
sediments derived from Biot theory, The Journal of the Acoustical Society of America,
Vol. 110, No. 5, p. 2276-2281.

110

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.